Topic I · Career Research

Private-Sector FP&A: Process and Terminology

Estimated read: ~520 min across multiple sessions · v1.0 · 2026-09-03

1. Where FP&A sits and who does what

You already run a budget office. What changes in the private sector is not the work but the map: which function owns which piece, who you negotiate with to move a number, who actually decides, and what the plan is optimized for. This section walks the CFO org, the two ways FP&A is structured inside it, how an ask really moves, the contested boundaries with adjacent functions, and the title ladder, then maps all of it onto OFM and your own agency budget office.

The CFO org: same building, two opposite jobs

One distinction organizes everything else: controllership faces backward and is rule-bound; FP&A faces forward and has no external standard-setter. Controllership produces statements from past periods and "cannot deviate from the guidelines or rules" set by GAAP, tax law, and Sarbanes-Oxley; FP&A's output is "predictive (forward-looking)" and its practitioners "can explore different analytics methodologies and procedures" (FP&A Trends, 2018). That freedom is narrower than it sounds. There is no GAAP for a forecast, but you are bound by the chart of accounts, the planning system's dimensions, metric definitions already committed to a board or the street, and the prior plan's structure. The latitude is in analytical approach, not in how the plan is shaped or what the metrics mean. The same source's other split holds: controllership runs on internal ERP data, FP&A blends internal data with external market and macro inputs.

A blunter version: "The Finance Manager plans. The Financial Controller records and controls" (CFO Connect). Accounting owns the number; you own the story about it. The line is messier in three ways. Accounting does write variance commentary — flux analysis, prepared for auditors and the 10-Q, and under a few hundred people the Controller often writes what FP&A writes elsewhere — but yours aims at a decision rather than an auditor. FP&A also negotiates accruals, reclasses, and cutoff during close, which moves the reported number, so when R&D lands $31k over plan the first question is whether that is an overspend or a timing accrual. And accounting owns the chart of accounts and cost-center hierarchy; a new director's first structural fight is usually getting that hierarchy changed so the plan can be reported against. The bookings, billings, revenue, and cash distinction under ASC 606 sits on the same boundary and is the biggest source of confusion arriving from cash-adjacent public budgeting (section 9).

Around those two sit the specialists, mostly part-time CFO responsibilities until the company can staff them. Two are worth knowing precisely. Investor relations is "typically a department or person reporting to the chief financial officer (CFO) or treasurer," and in smaller companies "is often outsourced to independent investor relations firms" (Wikipedia) — at a private company there is no IR function and the investor-facing work is yours: the quarterly investor update, the board package, diligence materials for the next round. Internal audit is deliberately not the CFO's, since the chief audit executive usually "has a direct functional reporting line to the board and an administrative reporting line to a member of senior management" (IIA, Standard 1110), meaning the audit committee. A 2012 inventory names most of the rest — "treasury, financial reporting, tax, financial planning and analysis (FP&A), risk management, investor relations," plus a note that "increasingly, real estate and procurement are falling into the reporting line, too" (Hoffelder, CFO.com, 2012) — and misses the Chief Accounting Officer, corporate development, business systems, stock administration, and the deal desk.

Function	What it owns	Orientation	When it becomes a standalone seat
Controllership	Books, close, GAAP reporting, controls, chart of accounts	Backward	Early; often the first finance hire
FP&A	Plan, forecast, variance explanation, management reporting	Forward	Typically 51-250 employees
Chief Accounting Officer	Technical accounting, SEC reporting, the Controller's org	Backward	Large or public only
Treasury	Cash, liquidity, banking, debt, FX	Forward, short	Larger or multi-currency; CFO holds it before that
Tax	Returns, indirect tax, transfer pricing, nexus	Backward	Outsourced until several hundred employees
Corporate development	M&A sourcing, diligence, integration modeling	Forward	Once the company buys rather than builds
Investor relations	External narrative, guidance mechanics, analyst contact	Forward	Public only; FP&A does investor reporting before that
Business / finance systems	ERP, planning tool, reporting layer, data pipeline	Present	Once the planning tool outgrows spare time
Procurement	Vendor selection, negotiation, contracts	Present	Once third-party spend is material
Internal audit	Controls testing, SOX, risk-based audit plan	Backward	Public; reports to the audit committee, not the CFO
People / HR (outside the CFO org)	Comp bands, merit and bonus cycle, recruiting, the HRIS	Present and forward	Early; reports to the CEO, so this runs on influence

One structural rule matters more than it looks. AFP's guidance is that "FP&A reports directly to the CFO or the business, rather than other finance departments," because "FP&A needs to be a forward-looking organization, whereas others are generally rear-facing positions" (AFP, 2019). Where FP&A is nested under the Controller, expect the forecast treated as an accounting deliverable: late, tied to close, defended for tying to the ledger rather than for being right. That reporting line is a real signal when evaluating a company.

Corporate FP&A, embedded FP&A, and the business-partner model

FP&A splits on a second axis: who you sit with. Corporate FP&A "supports an entire organization's financial health and growth by planning, budgeting, and forecasting as well as management and performance reporting," and owns the consolidated plan the CFO, CEO, and board see. Business-unit FP&A does the same "with a focus on a particular unit or division of the business," where "a deep connection to the overall execution of plans is achieved" (Anaplan). AFP frames the structural choice as three levels companies grow through:

Model	Shape	Fits	Weakness
Level 1: centralized	One HQ team "responsible for all FP&A activities, including budgeting, forecasting and planning, creating scenarios for the CFO"	Small companies	AFP's: it "lacks the capacity to provide decision-making support and advanced analytics to operations"
Level 2: embedded	"HQ FP&A staff embed FP&A practitioners in the business units to support rapid decision-making"	Companies "increasing in size and sophistication"	Mine: divided loyalty, and the consolidated view gets harder to hold
Level 3: distributed	A shared service center, a Center of Excellence for "standardized and ad-hoc analytics," and embedded "business consultants"	Mine: large, complex enterprises	Mine: distance between the analytics and the decision they were built for

All quoted text is AFP's. AFP names a weakness only for Level 1 and attaches no size label to Level 3, so cells marked "mine" are my characterization, not the source's. The Center of Excellence builds models, templates, and recurring analysis once, so twelve business units do not each invent their own; vendors call the arrangement "hub and spoke," which I could not verify in a primary source, so treat it as working vocabulary. Finance business partnering is the behavior layer on top of whichever structure you land in: "the collaboration between finance and other business units to enhance decision-making and optimize resource allocation," where partners are "impartial advisors, providing objective insights, challenging decisions and ensuring good metrics" rather than report producers (AFP glossary). In postings it is a title of its own, embedded in Sales, Marketing, or Operations (CFO Connect).

One ask, end to end: how a $1.01M request moves through the org

Take a 250-person SaaS company on a calendar fiscal year. It exits the prior year at $40M ARR, plans $48M of GAAP revenue, and plans to exit the year at $52.0M ARR (section 4 builds that ramp). Keep the two measures apart: ARR is a point-in-time annualized run-rate of recurring subscription revenue, revenue is what the income statement recognizes over a period, and for a growing company plan-year revenue sits between prior-year ARR and plan-year exit ARR — here between $40M and $52.0M. Expense ratios are struck against revenue, capital-efficiency metrics against ARR; using them interchangeably marks you as new to SaaS.

Nine people sit in finance: a CFO, a Controller with three staff accountants, an AP and payroll specialist, a billing and collections analyst, a Director of FP&A, and one senior FP&A analyst. Tax is outsourced; RevOps sits under the CRO. That is richer than Aleph's benchmark average of 5.29 finance FTEs at 51-250 employees and lighter than its 10.86 at 251-500, as you would expect at the top of the first band (Aleph). Where you draw the boundary of "finance" moves the count by two or three seats, much of why benchmarks disagree.

Where the envelope comes from

The board set the constraint first: free-cash-flow breakeven by Q4 of next year, capping this year's net burn at $6M. At a 78% gross margin, $48M of revenue throws off $37.4M of gross profit, so operating expense cannot exceed about $43.4M, allocated at peer-benchmarked ratios: R&D 33% of revenue ($15.8M), S&M 40% ($19.2M), G&A 17% ($8.2M), leaving roughly $200k unallocated. The R&D baseline run-rate, existing team annualized plus merit, is $14.6M, leaving $1.2M of room in R&D for everything new. Note the direction: the ratio is derived from the burn target, not the reverse. What you are solving for is the burn commitment, not the 33%.

The ask

The VP of Engineering wants six backend engineers at $185,000 base. At a 1.35 fully loaded factor for benefits and payroll taxes (section 5), each seat is $249,750 a year, so six is $1.50M annualized. Start dates stage two in February, two in May, two in September: 46 person-months of the 72 a full year carries, so in-year cost is $957k, not $1.50M. Add $9,000 per head for laptops, tool seats, and dev environments and the ask is $1.01M in-year, $1.50M annualized. Three costs ride along and are routinely missed: agency recruiting fees (commonly 20-25% of base where agencies are used), signing bonuses on competitive engineering offers, and the cloud consumption six more engineers generate. Load those in and you have eaten essentially all of the $1.2M. Every function then touches the number:

1. The embedded FP&A partner builds it with the VP, position by position and month by month. Here that is the senior analyst wearing an R&D hat, not a separate team.
2. Talent and Total Rewards supply the evidence for start dates and band: Recruiting says two of the six carry a 90-day time-to-fill at current pipeline, so February is optimistic, and Total Rewards confirms $185k is the band midpoint. Start dates move because Recruiting says so, not because Finance argued.
3. Corporate FP&A tests it against the $1.2M, consolidates with eleven other departments, loads the planning system, and produces the board version. Closing the gap when asks exceed the envelope is section 3.
4. The Controller touches none of that, then books the accrual and the actual once hires land (section 8). Treasury, here the CFO with a spreadsheet, asks what $1.01M of added burn does to runway and whether the cash timing works, since payroll is monthly and collections are not.
5. Tax is in it because two hires will be remote in states where the company has no presence, raising registration and nexus questions you flag before offers go out even though you do not own them. Procurement negotiates the cloud and tooling expansion. RevOps would own the equivalent build for quota-carrying sales heads (section 4); at a public company a step-up this size also lands in the opex walk supporting guidance (section 10).

How the ask really moves

That sequence is the process, not the politics. Carry the WA process over unchanged and you will prepare for the wrong meeting: the fight happens earlier, less formally, and with far less documentation. None of what follows is citable, but any FP&A director will recognize it.

• The envelope is announced, not discovered. The CEO and CFO set the burn or margin target against the board's constraint before kickoff. The bottom-up build allocates inside the envelope; it does not size it. Showing up to kickoff planning to establish the total from the sum of departmental needs means you have misread the exercise.
• Submissions arrive padded. Every VP has learned that submissions get haircut, so asks come in above what the VP expects to get, and nobody measures or admits it. The counter-moves: demand position-level detail rather than dollar totals, ask for a ranked cut list at 80% of the ask before negotiating, and hold start dates rather than cut headcount, which is less visible to the VP and just as effective on the in-year number.
• The decision venue is a CEO staffing review, not the board meeting. FP&A staffs the numbers and owns the deck, the CEO adjudicates between VPs, and the outcome is usually pre-wired in one-on-ones so the meeting ratifies it. Being in those conversations is most of what separates a Director who influences the plan from one who assembles it.
• Your instruments are the calendar, the template, the model of record, and the pre-read. You hold no approval authority. You hold the power to set when submissions are due, in what shape, against which assumptions, and which version of the numbers everyone argues from. A VP who misses the template deadline loses standing in the room.
• Escalation is normal. A VP going over your head to the CEO is not a process failure. Plan for it: make sure the CEO has seen your version of the trade-off first.

Adjacent functions and where the boundaries are contested

Strategic finance is a time-horizon split from FP&A, not a seniority one: 12 to 60 months out on long-range models, scenario packs, board materials, and investment cases, where FP&A works 13 weeks to 12 months out on whether you are on plan and what to adjust now (Nimbl). That source's staging rule, one firm's model rather than an industry census, is a combined function from roughly $5M to $30M ARR and separate leaders only above $30M. Below that, "strategic finance" on a posting usually means the whole finance job (section 10).

People and Total Rewards is your highest-volume partnership in tech and the one the org chart hides, because People reports to the CEO, not the CFO. Personnel is the large majority of operating expense, the headcount plan is co-owned, and the req-approval workflow is jointly run. FP&A owns the dollars and the plan of record (the frozen budget; section 2 and section 7 on the naming); People owns the bands, offer approval, the merit and bonus cycle, and the HRIS (Workday, Rippling, HiBob) that is the system of record for who actually works here. The standing disagreement: Finance counts approved reqs, HR counts people in seats, and the two never agree without a defined reconciliation. Build it in month one (section 5).

RevOps and sales finance own the operating mechanics of the revenue motion: pipeline, sales productivity, comp structure, renewal and expansion, and the CRM and GTM systems generating the data. FP&A owns the financial model, budget, scenarios, and board reporting. Where RevOps reports is genuinely unsettled. Across 746 companies with RevOps employees, Operations is the most common line at most sizes (22-40%), Finance peaks at 20% in the 51-100 band and levels near 9% at 200+, and at 3,001+ employees Sales (33%) overtakes Operations (22%) (Pave) — "majority" in Pave's own phrasing means plurality. The case for the finance line, from the same analysis: "Finance has less stake in GTM narratives than Sales or Marketing does." Both Pave datasets here come from its compensation-software customer base, which skews to venture-backed tech, so large enterprises and PE-backed companies are underrepresented.

BizOps is the fuzziest label here. Sourced: it exists to "drive growth in the business by either launching and scaling new initiatives or by optimizing day-to-day operations," works across "sales & marketing, product, finance, analytics, human resources, and operations," and runs projects whose length "varies from anywhere between a couple of weeks to 9-month engagements" (RocketBlocks). Not sourced, because no neutral primary source draws it, is the line between BizOps and FP&A; commentary frames BizOps both as FP&A plus strategy and as something broader and non-financial, and both are true somewhere. The reliable read is structural: FP&A is a standing function with a recurring calendar, BizOps a project function with a queue. Read the responsibilities, not the title.

Two newer boundaries. With data and analytics, FP&A owns the financial metric definitions that go to the board and the data team owns the pipeline and product metrics, while shared operational metrics (ARR, retention, pipeline coverage) get fought over. Procurement executes vendor contracts while FP&A owns whether the spend fits the budget, overlapping on spend analytics (CFO Dive, sponsored by Deloitte, 2026; vendor-sponsored, so directional only).

What FP&A does not own, and who does

• Pricing is product or RevOps; you own the revenue impact, not the decision.
• Commission calculation and payment is sales comp or payroll; you own the accrual and the plan cost.
• Cap table and equity administration is legal and stock administration; you own dilution modeling.
• Stock-based compensation expense is the Controller's under ASC 718; the SBC forecast is yours.
• Chart of accounts and cost-center hierarchy is accounting's; you can only request changes.
• The board deck is often a chief of staff or strategy function, with FP&A supplying the finance pages.

The title ladder, and how big the team actually is

Level	Typical experience	What the level owns	Where the day goes
Analyst	1-3 yrs	"The workhorse of FP&A": data gathering, model building and maintenance	In the model and source systems
Senior Analyst	3-5 yrs	Runs projects, directs junior analysts, still hands-on	Half building, half reviewing
Manager	5-10 yrs	Leads analyses and planning cycles, quality-controls analyst output, runs department-head meetings	Assembling the budget; still modeling
Senior Manager	varies	Manager scope plus a larger portfolio or a management layer; not a universal rung	As Manager, wider span
Director of FP&A	10+ yrs	Runs corporate planning cycles, sets process, reviews rather than produces	Executive communication, review, recruiting
Director, Strategic Finance	8+ yrs	Long-range model, board materials, fundraising and capital allocation; frequently an IC	Building the model himself
Head of Finance	8-12 yrs	Everything including close, under roughly 150 people; the standard first finance-leader title at Series B	All of it, alone or with one analyst
VP Finance	10-15 yrs	FP&A plus controllership; usually the CFO's operating deputy	Running the finance org
VP FP&A	varies	The plan as a whole and the CFO's board relationship	Capital allocation, board and earnings prep
CFO	-	Financial strategy, long-range planning, investor relations, capital allocation	Outside finance more than inside it

Analyst, Senior Analyst, Manager, and a combined Director-or-VP band are per Wall Street Prep, which gives four rungs, puts Director/VP at 10+ years, and names neither Senior Manager, Senior Director, nor a VP-specific year count. Mergers & Inquisitions does place Senior Manager in the progression to Director. Year ranges for Head of Finance, VP Finance, Director of Strategic Finance, and VP FP&A are my read of the tech market and appear in neither source: orientation, not data. Tech compresses the ladder relative to a large enterprise, and unlike banking or consulting "there is typically no set time frame or up and out policy" (same source).

Team size, and why FP&A teams are smaller than you expect

Two benchmark studies converge. Pave, from org charts of "over 1,400 customers in Pave's dataset with at least 50 employees and at least one FP&A employee," finds the ratio settles near 130 employees per FP&A team member between 200 and 2,000 employees, widening at 3,001+ to a median of 161 (Pave). That filter matters: it describes staffing among companies that already have FP&A, not the odds a given company has any. Aleph, from 218 Y Combinator B2B companies and 3,597 finance FTE records, finds the whole finance function averages 0.35 people at 5-50 employees, 5.29 at 51-250, 10.86 at 251-500, and 49.2 at 500+, with the first full-time finance hire late in the 30-50 range, FP&A emerging in the 51-250 band, and an FP&A hire arriving before a Controller in over 20% of cases (Aleph). Do not merge the two: Pave counts FP&A only, Aleph counts all of finance and samples YC-backed B2B startups. Practically, a 400-person company with a $48M plan probably has three FP&A people, one of them the Director. Team size by company type is section 13.

Translation: OFM, the Legislature, and the same split you already work in

The harder mapping is authority, and it is easy to get pointed at the wrong room. Inside a company, corporate FP&A plays OFM's budget division: no appropriation authority, but it runs the calendar, issues templates, applies one set of rules to every submitting unit, tests each ask against top-down constraints, and holds real approval power over the plan it collects. Departments play agencies. The CEO plays the appropriator — the staffing review is where money is assigned. The board is the ratifier and constraint-setter, not the Legislature. At a venture-backed company it approves the annual plan as a scheduled agenda item, often in well under an hour, the substance settled in advance between CEO, CFO, and lead investor; boards engage with the burn or FCF target, runway, the top-line commitment, the headcount envelope, the option pool, and anything above the delegation-of-authority threshold, rarely with departmental opex detail. Scrutiny varies: line-level and monthly at PE-backed portfolio companies against a lender-facing model with covenants, moderate at public companies where the binding constraint is guidance already given to the street, light at venture-backed companies between raises (section 13). At plenty of private companies no body outside management formally ratifies the operating plan at all (section 11).

The analyst mapping splits in two. The corporate FP&A analyst who owns a department's envelope is the OFM analyst assigned to your agency: outside the unit, above it, applying executive constraints, your helper and your reviewer at once. The embedded finance business partner has no OFM equivalent; the nearest thing is one of your own DDA budget analysts assigned to a program area, inside the shop and close to the program but expected to challenge it anyway. That seat carries the dual-allegiance problem the OFM analyst does not have. This framing is my construction from sourced facts on both sides, not a claim any source makes, and it breaks in four places.

One term deliberately not translated: decision package, "an agency's specific request for additional funding, or any proposed changes to the agency's budget, through the legislative budget process" (DES), tied to a legislative mechanism no company has. The private equivalents (business cases, investment requests, initiative budgeting) are close in function, not form; the baseline-versus-new-ask split that carry-forward and maintenance level create is section 3.

2. The planning calendar

You already run an annual cycle inside a two-year cycle, with a formal correction window and a monthly monitoring rhythm. The private-sector calendar has the same skeleton and a different metabolism. Two things will feel foreign: frequency (the plan gets re-cut four to twelve times a year, not once a biennium) and authority. Nobody outside the company votes on the plan. Filing deadlines are fixed by regulation once a company is public, but the planning calendar itself is set by whoever the company answers to — a board, a lender, or the public markets.

This section covers the when. The how — target-setting, baselines, zero-based versus driver-based, gap-closing, contingency — is section 3.

2.1 The layered cycle

The Association for Financial Professionals splits the work into three activities companies routinely confuse: planning asks what is possible, budgeting asks what is expected, and forecasting asks what is happening. Planning produces the long-range plan, budgeting the AOP, forecasting everything after the AOP locks.

The 13-week cash flow

The layer most planning guides omit, and the first you will meet at a company burning cash, carrying debt, or owned by a sponsor. It is built direct — actual receipts less actual disbursements, by category and by week, rather than indirect off the P&L — and it rolls weekly: drop week one, add a new week thirteen, replace the closed week with actuals. Wall Street Prep's restructuring and DIP-financing version must be updated weekly. Accuracy is not uniform: weeks one and two should be near exact, weeks nine through thirteen directional.

A company with $48M in the bank and $2.0M a month of burn still runs one, because runway is an average and payroll is not. Two lines move it at a SaaS company: collections against invoiced ARR (annual invoices land in lumps, so one enterprise renewal slipping two weeks moves a month) and the semimonthly payroll dates. A P&L reforecast models neither collection timing nor payment dates, so several decisions the table attributes to it get made here.

The weekly flash

A short, deliberately uncontrolled report — often one email with six to ten numbers — issued days before anything is reconciled. Its job is to prevent surprises, not to be right to the dollar. Publishing numbers you know are provisional is a real adjustment from a world where a published figure carries the weight of a legislative document. Two intensifications to expect: in the last two or three weeks of a quarter the bookings flash goes daily and deal-level, a commit list with names, amounts and close dates; and a company under cash pressure runs a separate weekly cash flash, usually owned by treasury. One discipline makes a flash useful: freeze the numbers and their definitions for the year. Its value is week-over-week comparability, and redefining a line mid-quarter discards every prior week.

Reforecast naming

Reforecasts are often named by how much of the year is actual and how much forecast: a 3+9 is three months of actuals plus nine of forecast, a 6+6 the mid-year re-cut, a 9+3 the late-year landing estimate. The convention is thinly documented — the only published explanation located for this guide is a planning-vendor article — so treat it as shop vocabulary, commonest in enterprise and manufacturing finance. At a SaaS company you are likelier to meet POR (plan of record, the frozen budget), LE (latest estimate) or CV (current view) for the live forecast, F1 / F2 / F3 for numbered versions, and EA (estimated actual) for a month in progress before close. These are scenario names on system dropdowns and deck footers; "FY27 POR vs LE" is unreadable until you know them.

How much of this a real company actually does

Less than the diagram suggests. FP&A Trends' 2025 benchmark write-up reports that 29% of organisations need more than ten days to produce a forecast and only 15% can do it in under two, only 11% have fully aligned strategic, financial and operational planning, only 17% use fully driver-based models, and 46% of FP&A time still goes to collecting and validating data. On horizon, Wall Street Prep is blunt: most organizations forecast with reasonable certainty over one to three months, and accuracy wanes past three. A company whose reforecast takes two weeks sits in the slower third of the field, not on a broken team.

Where the cycle matures, it matures into a rolling forecast — always looking the same distance forward, so the horizon does not shrink to nothing in December. Horizons track how long decisions take to bite, from airlines at roughly six quarters updated monthly to pharmaceuticals at ten quarters updated quarterly. It sits alongside the annual budget rather than replacing it, because the budget is what the board approved and what bonuses pay against. The case for abolishing the annual budget outright is in section 3; section 7 covers what these documents are.

2.2 Fiscal-year conventions

Washington's fiscal year is fixed by statute at July 1 through June 30 (OFM glossary). A company picks its own, and two distinct rationales get cited, pointing at different months: end the year shortly after the highest-revenue period so it captures a complete cycle, or close during the slow season when inventory on hand is below average (both attributed to Pamela Drake).

• Calendar year (Jan 1 – Dec 31). The default across US tech and this guide's frame unless stated otherwise.
• The 4-4-5 family. Retail and much of manufacturing run 13-week quarters of two 4-week "months" and one 5-week "month"; which slot gets the long month names the variant — 5-4-4, 4-5-4 or 4-4-5. Retail's is the NRF 4-5-4 calendar, built so comparable months hold the same number of Saturdays and Sundays; the year ends in late January because December is peak and January carries returns (retail explainer). Walmart's registered year end is January 31.
• Mid-year enders. Microsoft ends June 30, Apple in late September (a 52/53-week year, so the date moves). The usual explanations, enterprise purchasing cycles and launch timing, could not be traced to a company statement; treat them as folklore.

2.3 A calendar-year AOP timeline, month by month

No dated, high-authority source gives a canonical AOP calendar for a generic SaaS company. Sources agree only on the shape, and put the start three to four months before fiscal year end. Made concrete at a scale you might plausibly target — $40M ARR, 250 employees, Series C, calendar fiscal year, planning FY27 during 2026:

When	What happens	Who drives it	Output
May – June	Refresh the long-range plan. Test whether the 3-year revenue and margin path still holds. Set the FY27 envelope: growth rate, ending burn, target ending headcount. With no living LRP — common at this stage — build a light three-year envelope rather than run the AOP with no multi-year constraint.	CFO, strategic finance	LRP refresh, one-page envelope
Late August	AOP kickoff. Guidelines and templates go out. Each department gets a preliminary opex and headcount envelope and the assumptions it must use (merit %, benefits load, cloud unit costs).	FP&A	Kickoff deck, planning calendar, templates
September	Revenue plan built first and in parallel: capacity, quota, ramp, pipeline coverage. Departments build v1 against the envelope. See section 4.	CRO / RevOps with FP&A	Revenue plan v1, department v1 submissions
Early October	Consolidation. The gap appears: v1 comes in over the envelope, almost always on headcount. FP&A quantifies it and builds the trade-off menu.	FP&A	Consolidated v1 P&L, gap analysis
Mid-to-late October	Exec review and gap-closing. Hire-date slippage, program cuts, a reserve carved out. Two or three iterations, most of it settled in 1:1s beforehand (see 2.4).	CEO, CFO, exec staff	v2 / v3 plan
November	Board preview at the Q3 meeting. Pushback on growth, burn or hiring pace. One more revision.	CFO	Draft board plan
Late November	Board pre-read due, typically 3–7 days before the meeting. December meetings often sit in the first half of the month, so the real drop-dead is here. The plan is effectively locked when the pre-read ships.	FP&A	Board deck, locked plan
December	Board approval. Comp plan design and quota modeling run in parallel, not after.	Board, CFO, CRO, People	Approved AOP
Late December – January	Load to the planning system: map the approved plan to the chart of accounts, cost centers and headcount plan. One to three weeks of work, not a step. Distribute departmental budgets. Issue quota letters and commission plans.	FP&A with RevOps	Budget of record, departmental budgets, signed quota letters
Early February	First BvA against the new plan, running alongside the Q4 and full-year close and the start of audit.	FP&A	First BvA

Three timing facts there do most of the damage in a first cycle. The board pre-read moves the real deadline a week earlier than the calendar says. System loading is where a December approval becomes a February budget, and it is the source of "why does my department budget not match what the CFO approved." And comp plans and quota letters should be signed at or before the start of the plan year, which cannot happen until the revenue plan is final by segment and territory; in practice they slip to late January while reps sell against last year's plan and finance absorbs the blame. An unsigned commission plan is also an employment-law exposure in several states. Cycles slip from collision, not modeling: the AOP build overlaps Q3 close, the Q4 forecast and audit prep.

That is roughly a four-month build — faster than average, slower than good. The Hackett Group benchmark cited in AFP's guide to shortening the budget cycle puts the average annual budget process at about six months and world-class finance organizations at 60 to 90 days. The guide is from 2015; weight it as a durable rule of thumb, not a current measurement. Its profile of slow versus fast planning:

Attribute	Traditional	Best practice
Cycle time	> 120 days, kicks off in month 4 of the year the planning happens in	< 70 days, kicks off in month 8 or 9
Approach	Bottom-up	Middle-up against top-down targets
Detail	Very detailed, no materiality filter	< 150 line items, key drivers only
Iterations	5+	< 3
Effort on data collection	80%+	< 50%
Tooling	Spreadsheets	Integrated planning application

Read the cycle-time row against intuition: starting earlier is the failure mode, not the virtue. A budget begun in April rests on assumptions eight months stale before the year it governs opens. The August kickoff above is the best-practice column, not a corner cut.

One case study in that guide is worth more than the benchmark, because it reports an interviewed practitioner rather than an idealized calendar — and it runs longer, not shorter. An anonymized large manufacturer on an October–September fiscal year took nine months: timeline set in February, five-year plan in April, first detailed submission in July, preliminary budget to the board in September, divisions revise, and the board approves in the second week of November, roughly six weeks into the year the budget governs.

2.4 What the timeline does not show

That table describes an analytical process that finds a gap and resolves it. The real process is a negotiation whose outcome is largely set before the bottom-up work starts. Two of its habits belong here rather than in section 3, because they move dates on your calendar rather than numbers in your model.

The move	What actually happens	What to do about it
The real deadline is the pre-wire, not the meeting	Neither the exec review nor the board meeting decides anything. The CFO settles it beforehand, one at a time — with each VP, and with the directors who will have opinions — and changes the plan before it is presented. By the time the packet ships, the argument is over.	Build the calendar backward from the pre-wire date, roughly two weeks before the board packet, and have the revenue plan, headcount roll-up and scenario set stable by then even though nothing is formally approved. Know every VP's number before the meeting. That, not the model, is year-one credibility.
Board approval is not a released requisition	A December-approved role can sit unhireable into February. Some of that is system loading. Some is deliberate: headcount gets approved-but-gated, in the plan but needing separate sign-off — the CEO's, or a monthly hiring committee's — before recruiting can open it. It is how a CEO says no while appearing to say yes.	Track gated versus open requisitions separately from the approved plan, or plan headcount and actual hiring diverge in Q2 with nobody having changed the plan.

The rest of the fall's politics belongs with the decisions themselves: where the top-down number really comes from, why v1 submissions arrive 10–25% over the envelope and what to do about it, why the across-the-board cut beats a better allocation model, and who reopens the fight at the first reforecast are all in section 3.2.

2.5 One plan, three numbers

The AOP does not produce one revenue number. It produces at least three, they are supposed to differ, and a new hire who reconciles them and reports the difference as an error gets corrected in public. For the $40M ARR company above, entering FY27:

Number	FY27 figure	Owner	What it is for
Board plan	$12.0M gross new ARR	CFO	The commitment. Variance analysis, bonus funding and departmental accountability all run against it.
Internal / stretch plan	$13.2M	CEO / CFO	What the exec team manages to, typically 5–15% above the board plan, so hitting the commitment is the floor rather than the goal. The difference between the two is the CFO's cushion and is rarely written down as such.
Quota capacity	$16.0M assigned	CRO / RevOps	About 21 AEs at a $750K list quota — the plan opens with 14 and hires toward 22 (section 4). Deliberately over-assigned, because not every rep hits.

The over-assignment is arithmetic, not padding: most companies over-assign quota by 20–30%, and typically five or six reps in ten hit 100% while seven in ten reach 80% or more, with more cushion in enterprise than SMB. The identity is exact: assigned quota ÷ plan = 1 ÷ planned attainment, so planning at 75% attainment is over-assigning by 33%, and $12.0M ÷ 0.75 = $16.0M of quota on the street. That is a little past the headline 20–30%, and lands where the same source's own split puts a company of this shape — roughly 120% for SMB against 135% for enterprise. Section 4 builds the whole capacity plan out of that one identity. The same source gives ramp times that constrain the hiring calendar — roughly 2–3 months in SMB, 4–6 in mid-market, 6–9 in enterprise — which is why reps hired after month eight contribute almost nothing to the current year. At a public company there is a fourth number: guidance, set at or below the board plan so the company can beat and raise. When three different revenue figures surface in your first week, ask which of the four each one is before assuming the model is broken.

2.6 What the calendar is anchored to

The timeline looks arbitrary until you see what it is bolted to. Companies build backward from a fixed external commitment, and which one changes the pressure on everything.

Public companies: the earnings and guidance clock

Quarterly reports have been required since 1970. The 10-K deadline runs by filer status: 60 days after fiscal year end for large accelerated filers, 75 for accelerated filers, 90 for all other registrants (Form 10-K, General Instruction A.(2)). A recently-public company you might join is very often an accelerated or non-accelerated filer, so the deadline you plan around may be 75 or 90 days, not 60.

The date that shapes the calendar is not the filing deadline anyway. The earnings release goes first, furnished on a Form 8-K under Item 2.02, Results of Operations and Financial Condition, usually weeks before the 10-K, timed by when the audit is far enough along for the audit committee to approve it. That is why a calendar-year company announces Q4 and full-year results in late January or February and files the 10-K after. Work backward from the earnings date: board and audit committee pre-read about a week earlier, guidance range agreed a week before that, and the AOP is the input. Inside the surrounding quiet period, investor relations turns the budget's key messages into guidance language for the Q4 call. The December-approved plan becomes public commentary six weeks later, which is the biggest behavioral difference from private planning: sandbagging is not just a game against your CFO, it is a game against a number the company must hit in public.

The quarterly requirement is under review. In May 2026 the SEC proposed letting companies report semiannually instead (File S7-2026-15, Releases 33-11414 and 34-105368, still at proposal stage). Nothing has changed. If it lands, the external clock loosens while the internal one does not — boards, lenders and comp plans run quarterly regardless.

The public commitment has already narrowed. Per the Harvard Law School Forum on Corporate Governance, the share of S&P 500 companies issuing quarterly EPS guidance peaked around 50% in 2004 and fell to 19% by 2024; as of Q3 2025, 112 gave quarterly EPS guidance while 264 gave full-year guidance — a structural shift toward the annual plan as the externally-committed unit.

Venture-backed companies: the board meeting and the milestone

No regulator. A board and cash. One vendor's framing has seed companies meeting monthly, Series A shifting to quarterly nine to twelve months after the round, and Series B and later settling into quarterly meetings (directional, vendor source). The budget lands in one of those slots, almost always the last meeting of the fiscal year.

The approval is contractual rather than customary. The NVCA Model Investors' Rights Agreement (October 2025 form), section 3.1 — the industry-standard financing document — requires an annual budget and business plan for the coming fiscal year, prepared on a monthly basis, submitted to the board for approval, in bracketed drafting at least 30 days before year end. The board-approved version, "as may be revised by the Board of Directors," is a defined term: the Approved Annual Budget. The same section, again bracketed, has the annual financials compare actuals against that budget with an explanation of material differences.

Two qualifiers. Section 3.1 runs to Major Investors, a defined holder threshold, with a bracketed carve-out excluding one that is a Competitor — not the whole cap table. And these covenants are negotiated deal by deal and honored loosely; a late budget is a common, tolerated breach in a way a missed OFM deadline is not. What survives is the structure: the expenditure-against-allotment monitoring you already live inside, in AFRS and Enterprise Reporting with OFM watching, has a direct contractual analog here, owed to investors instead of a central budget office.

Cash, not the calendar, sets the real deadline. CRV's guidance is to open a round with 12 to 18 months of runway left, budget three to six months for the raise, and target 18 to 24 months of post-close runway — enough to reach the milestone the next round gets priced on. Backward:

• A Series B company opens 2027 with $48M cash and $2.0M/month net burn — 24 months of runway, to January 2029.
• Fundraising opens with 12–18 months left: July 2027 through January 2028.
• The raise takes 3–6 months, so the round closes between October 2027 and July 2028.
• Diligence starts at the top of that window, so the milestone — say ARR from $22M to $40M — must be visibly on track by mid-2027.
• Sales capacity takes two to three quarters to hire and ramp, so those reps sit in the FY27 plan built in autumn 2026, with Q1 start dates.

That chain is the venture-backed planning calendar: anchored to a milestone and a cash-out date, not a date on a wall. If burn runs $200k/month hot the whole chain moves left, which the 13-week forecast catches months before the quarterly reforecast would. Section 13 covers PE-backed and large multi-BU anchoring.

2.7 Biennium plus supplementals versus annual plan plus reforecasts

Dimension	WA state	Private company
Plan period	2 years (biennium), with four-year and six-year outlooks around it	1 year (AOP), inside a 3–5 year LRP
Revised expectation, no change to authority	Quarterly allotment amendment, roughly 3.5x/year	Reforecast 4 to 12x/year
External assumption set that feeds it	Revenue forecast 4x/year, caseload forecast 3x+, both statutory	Market, pipeline and renewal assumptions, mostly built in-house from your own data
Change to the number everyone is measured against	Supplemental, most years, by bill	Re-baseline: rare, board-approved, needs a triggering event
Within-year administrative re-plan	The same amendment: money moves between months and objects inside the appropriation	Reallocation across cost centers, CFO approves
Who approves the revised expectation	OFM	CFO always; CEO when the landing number moves; board at the next meeting; sponsor and lender if levered; disclosure and audit committees at a public company when guidance moves
Can outsiders amend it	Yes, any legislator, on a supplemental	No
What the plan authorizes	Legal spending authority	Nothing legally; internal accountability only
Consequence of overspending	Statutory violation	A difficult conversation, possibly a job

The row on who approves the revised expectation matters because "no vote" does not mean "no approval." There is no such thing as an unapproved reforecast. Approval is serial and private rather than parliamentary: each department's new number goes past that budget owner, then the CFO, then the CEO if the landing number moves, and reaches the board deck already agreed. One that first meets a VP in a room full of peers is dead on arrival whether or not the math is right. Budget about as much time for pre-wiring as for building.

Why the private cycle re-plans so often

Because the budget stops being useful and everyone knows it. One practitioner in AFP's guide: "by Q3 and Q4 the budget variance is not relevant" (Tom Woods, quoted in AFP's guide). A December-approved plan built on September assumptions describes a company that no longer exists by June. In the state's world that staleness is absorbed by three separate machines: forecast councils update the assumptions on a statutory cadence, the allotment amendment re-projects the agency against them, and the supplemental repairs the authority. A company has no forecast council and no supplemental, so the reforecast does all three jobs at once — which is why it happens more often and carries more weight than an allotment amendment does.

The distinction that takes longest to internalize: the budget is frozen by default and unfrozen only on purpose. Freezing is what makes it usable as an accountability baseline — bonuses, quotas and variance all reference it, and a budget that moves to meet the forecast measures nothing. But re-baselining is a real, named event with real triggers: a reduction in force, an acquisition or divestiture, a materially missed first quarter, a down round, a financing that changes the burn envelope. The board approves the new plan, and the compensation committee usually resets bonus targets with it. There is a routine smaller version: many companies re-cut in late January once Q4 actuals land, because the AOP was built on a Q4 estimate and exit ARR feeds every driver. Two rules. A re-baseline is a decision someone above you makes, never a modeling convenience. And you keep the original plan visible beside the new one for the rest of the year, so performance against the original commitment stays auditable.

3. Building the AOP

The annual operating plan (AOP) is the one document a company agrees to be measured against for a year. Farseer's working definition is as good as any: it "translates the company's three-year or five-year ambition into an annual P&L projection and the supporting plans behind it" (Farseer). You will also hear it called the annual plan, the budget, the operating plan, or just "the plan." Where people are careful, "budget" means the locked financial detail and "AOP" means the budget plus the headcount plan, the revenue plan, the initiative list, and the assumptions behind them. Most people are not careful.

This section answers three of the guide's four core questions. Q1 (the end-to-end AOP process) is the whole section. Q3 (whether there is an analog to Washington's carry-forward level, and how companies separate the baseline from new asks) is answered in 3.4. Q4 (the budgeting approaches and when each is used) is answered in 3.3. Q2, on the revenue plan, belongs to section 4; the calendar this runs on belongs to section 2.

One framing note before the mechanics. In WA state the budget you build is a request — an ask that goes to OFM, then the Governor, then a Legislature that decides. In a company the plan you build is a commitment the company makes to itself and, at public companies, indirectly to the market. There is no appropriating body outside the firm; the negotiation is entirely internal and ends when the board says yes. That does not make the plan easy to get approved — a board can and does send a plan back, and at a venture-backed company an investor director's objection is close to a veto — but the people who decide are the same people who will be held to the result. That changes the politics far more than it changes the arithmetic, and most of what follows is a consequence of it.

3.1 Target-setting and the envelope

Planning starts with a number that did not come from the plan. Somebody decides what the company is trying to do next year before anyone models how to do it. That number is the target, and the range of acceptable outcomes around it is the envelope: the growth you have to hit, the loss or margin you are allowed to run, and the cash you are allowed to consume. The rest of the cycle is an argument about how to spend the envelope.

Three ways to set it

Under top-down budgeting, "executives or company leadership set the budget and then allocate portions of it to different departments" (Vena). Executives fix revenue and margin, finance splits the envelope by function, and department heads are told their number. Bottom-up budgeting inverts it: the process "starts at the department level, with each department creating a budget and moving it up to the top" (CFI), and leadership sets objectives from that input. Hybrid is what nearly everyone actually runs: leadership sets revenue and margin targets, departments build detailed expense plans, and those plans have to land inside the targets. No source found for this page measures how common hybrid is, so treat "nearly everyone" as the consistent shape of the practitioner accounts rather than a counted statistic — Vena, Farseer, Pigment and Protiviti all describe a process with a leadership-set frame and departmental build inside it, and none describes a pure form of either.

The reason hybrid won is time. Vena's comparison table puts top-down at 3 to 6 weeks and bottom-up at 2 to 6 months — roughly three to four times longer at the midpoints, and up to nine times at the extremes, for the same deliverable. A growth-stage company that cannot spend five months planning, but also cannot credibly tell a VP of Engineering what her infrastructure bill will be, splits the difference structurally. The practical consequence: "most large organizations need two to three months for the full cycle" (Farseer), which is enough for one bottom-up pass and two rounds of negotiation, and not enough for three.

Cycle times for top-down and bottom-up from Vena's comparison table; the two-to-three-month full-cycle figure from Farseer, which describes a hybrid process. The fit and failure columns are common practice, not sourced claims.

Where the top-down number comes from

The envelope is not arbitrary, and knowing its sources is most of what makes you sound fluent in a kickoff meeting. Four inputs dominate, in different mixes by company type (section 13 has the variation):

• Growth rate. The board or CEO fixes an ending-ARR or revenue growth number. At venture-backed companies this is usually anchored to what the next round will require rather than to what the business would naturally produce.
• Profitability or margin. An operating-income floor, an EBITDA target, or a maximum loss. At public companies this becomes guidance (section 10).
• A combined efficiency rule, most commonly the Rule of 40: growth rate plus profitability margin should be at least 40%. What makes it a planning constraint rather than a scorecard is that it is one number the board can hold fixed while letting management choose the mix. Telling a CEO he may run at 30 is telling him he can have 40% growth at a 10% loss, or 20% growth at 10% profit, and the AOP is the argument over which. Current benchmark bands by ARR scale are in section 9.
• Runway. For anyone not yet self-funding this can override the other three. Cash runway is "the number of months a company can operate using only the liquid cash it has in reserve," cash divided by net burn, "the amount of money the company loses monthly, after income" (Cube). Cube's 12-to-18-month figure is a statement about the historical gap between funding rounds, which has run "up to 22 months" — not a published board policy. The planning rule that follows is arithmetic, not convention: planned cash consumption cannot exceed cash on hand, minus whatever balance the board wants still standing at the point the company would start raising.

Worked example: setting the envelope

Take the guide's default company and give it a name, because this section carries it through several tables: Meridian, a Series C SaaS business exiting FY26 at $40M ARR with 250 employees (about $160K of ARR per head), $34M of cash, and a calendar fiscal year. The board wants FY27 ending ARR of $52M, up 30%, and will fund a Rule of 40 of 15 measured as year-over-year ARR growth plus EBITDA-excluding-SBC margin on GAAP revenue. Growth of 30 therefore buys a margin of −15. Everything else derives from those two numbers.

Line	FY26 actual	FY27 envelope	How it was derived
Ending ARR	$40.0M	$52.0M	Board target, +30% (FY26 grew 25% from $32.0M)
GAAP revenue	$36.0M	$45.0M	Ramp of the ARR plan through the year (section 4)
Gross margin	78%	78%	Held flat; any change is a separate argument
COGS	$7.9M	$9.9M	Revenue × (1 − GM)
Gross profit	$28.1M	$35.1M	Revenue − COGS
Allowed EBITDA ex-SBC	($6.9M)	($6.8M)	−15% of revenue in FY27, from Rule of 40 = 15
Cash opex envelope	$35.0M	$41.9M	Gross profit $35.1M + allowed loss $6.8M
Stock-based compensation	$2.5M	$3.2M	Held at 7% of revenue; non-cash, below the envelope
Depreciation and amortization	$0.7M	$0.9M	Held at 2% of revenue
GAAP operating loss	($10.1M)	($10.9M)	Gross profit − cash opex − SBC − D&A
Rule of 40	6	15	ARR growth + EBITDA ex-SBC margin

The bolded row is what the company will spend the fall arguing about, and the three rows under it are why finance has to be careful about which number it quotes to whom. The board approved a −15% margin. The auditors will report a −24% GAAP operating margin. The department heads manage to $41.9M. All three are correct, and confusing them in a meeting is the fastest way to lose credibility in a first quarter. (Meridian's gross margin here is stated on a cash basis, with SBC and D&A held below the gross-profit line; real companies push some of both into COGS, which moves gross margin a point or two. Ask which convention a company uses before you compare its margin to anyone else's.)

Notice what the envelope implies before anyone has submitted anything: cash opex may grow 20% while revenue grows 25%. That gap is the plan. Every hiring decision, vendor renewal, and marketing program in FY27 has to fit inside a 20% expansion of the cost base. An FP&A lead who walks into kickoff able to say that sentence has done the most useful thing available in week one.

Making the runway test bite

The margin rule and the cash rule are separate constraints and the tighter one wins. Test Meridian's envelope against cash. It holds $34M. The board wants $15M still standing when the company opens a Series D process, planned for month 21. That allows $19M of cumulative burn over 21 months, about $10.9M a year.

Planned FY27 cash burn is the EBITDA loss of $6.8M plus roughly $1.5M of capex and $2.0M of working capital build as receivables grow with revenue — about $10.3M, which clears with $0.6M of headroom. So for Meridian the Rule of 40 binds and cash does not, but only just. Change one input and the answer flips: at $28M of cash instead of $34M, the same test allows only $13M of cumulative burn, $7.4M a year, which after capex and working capital permits an EBITDA loss of $3.9M and a cash opex envelope of $39.0M — $2.9M below what the margin rule allowed, which is about fifteen fewer full-year roles. When the two rules disagree, cash wins, every time, and the Rule of 40 becomes the aspiration you explain to the board rather than the constraint you plan to. Run this test before you publish an envelope, not after, and note that the v1 submissions later in this section (3.5) at $47.3M of opex would have produced $15.7M of burn and failed the cash test outright even at $34M of cash.

Splitting the envelope by function

An opex envelope is not a plan until it is divided. Finance splits $41.9M across sales, marketing, customer success, R&D and G&A before departments are told anything, because a department head who is asked "what do you need?" with no ceiling attached will answer a question nobody wanted asked. The split starts from three inputs: last year's actual mix, external benchmarks, and the strategy the plan is supposed to fund.

The most usable public benchmark set is SaaS Capital's annual spending survey, whose 2026 edition drew more than 1,000 private B2B SaaS respondents. It publishes departmental medians as a percentage of ARR: 15% for sales, 8% for marketing, 9% for customer support and success, 22% for R&D, and 15% for G&A. It also reports that "the total median spend across all departments is 96% of Annual Recurring Revenue (ARR) for bootstrapped companies while equity-backed are spending 101% of ARR," with equity-backed companies "spending 70% more on sales, 64% more on general and administrative costs, 100% more on marketing, 56% more on R&D, and 100% more on customer success" — buying growth of 25% a year against the bootstrapped median of 20% (SaaS Capital).

Three cautions before you use any of that in an argument, and the first is the one that catches people. The survey publishes no combined COGS median. It reports COGS only as four separate components — hosting 5%, DevOps 4%, professional services COGS 5%, other COGS 3% — and adding them to reach 17% produces a figure the survey never states. Medians are not additive across a survey population: the company sitting at the median on hosting is not the company sitting at the median on professional services, so the sum of component medians is not the median of the sum, and the real combined COGS median could fall either side of 17%. The same arithmetic is why the five departmental medians do not add to the survey's own total-spend median. Second, the denominator is ARR, while your own P&L percentages will be computed on GAAP revenue, which for a fast grower sits nearer mid-year ARR than ending ARR and therefore makes every ratio look larger. Third, bootstrapped and equity-backed companies in the same survey spend differently enough that a blended figure is the wrong comparison for either — Meridian is venture-backed, so its peer number is the 101%-of-ARR equity-backed total, not the 96% bootstrapped one.

Function	Meridian FY27 plan	% of FY27 revenue	SaaS Capital 2026 median, % of ARR
COGS	$9.9M	22.0%	none published
Sales	$9.2M	20.4%	15%
Marketing	$5.4M	12.0%	8%
Customer success and support	$4.1M	9.1%	9%
R&D	$13.4M	29.8%	22%
G&A	$9.8M	21.8%	15%
Total opex	$41.9M	93.1%	—

COGS sits above the gross-profit line and outside the $41.9M opex envelope; the five opex functions sum to the envelope exactly. Total spend including COGS is $51.8M. The benchmark column is empty for COGS because SaaS Capital publishes no combined COGS median — only the four component medians quoted above, which cannot legitimately be summed into one.

Restate Meridian on the survey's own denominator — ending ARR of $52.0M — and the picture changes: sales 17.7%, marketing 10.4%, customer success 7.9%, R&D 25.8%, G&A 18.8%, COGS 19.0%, total spend 99.6% of ARR against the 101% equity-backed median. Meridian is planning to spend about what a typical venture-backed peer spends, which is what a Rule of 40 of 15 looks like. Notice the shape of that result: every departmental line sits above its own published median while the total lands just under the published total. That is the non-additivity above seen from the other side, and the practical lesson is to argue the total and the individual lines as separate questions rather than assembling a target by adding medians together. Marketing running two and a half points hot and G&A running nearly four points hot are worth a question, neither worth a conclusion, since a company running a land-and-expand motion in a competitive category legitimately spends above the median on demand generation. Customer success runs a point light, which is the line to look at first if net retention comes in below plan. COGS has no published median to be hot or cold against, so the only defensible sentence about it is that 19.0% sits in the neighborhood of the survey's four components taken together — a weak claim, and one to present as weak rather than dress up as a benchmark miss.

Benchmarks set the opening position. What settles it is an efficiency metric that ties spend to outcome, and the standard one for the go-to-market half of the envelope is CAC payback, "the number of months needed by a company to recoup the initial costs incurred in the process of acquiring a new customer," computed as sales and marketing expense divided by new MRR times gross margin (Wall Street Prep, whose worked example runs both terms on a single month).

The denominator is where people slip, and the ARR-versus-MRR distinction is the whole of it. Meridian's plan spends $14.6M on sales and marketing to add $12.0M of net new ARR at a 78% gross margin. Convert the ARR to MRR before dividing: $12.0M of net new ARR is $1.00M of net new MRR ($12.0M ÷ 12), because Meridian's MRR rises from $3.33M to $4.33M across the year — and the average month contributes about $83K of that $1.00M. Both terms then have to cover the same period. Annual over annual: $14.6M ÷ ($1.00M × 0.78) = 18.7 months. Month over month: ($14.6M ÷ 12) ÷ ($83K × 0.78) = $1.22M ÷ $65K = 18.7 months, the same answer, because both terms scale by twelve. Get the pairing wrong and the error is an order of magnitude, not a rounding difference: full-year S&M over one month's new MRR gives 225 months, and full-year S&M over $12.0M treated as though ARR were MRR gives 1.6 months. Say ARR or MRR every time, and say over what period.

Against Wall Street Prep's rule of thumb that "most viable SaaS startups have a CAC payback period of fewer than 12 months," Meridian's plan is more than six months outside it. There are exactly two ways to close that inside the envelope — cut sales and marketing to $9.4M, a 36% reduction that would take the pipeline with it, or raise net new ARR to $18.7M, which means exiting FY27 at $58.7M rather than $52.0M, seventeen more points of growth on the same spend. Neither is available, which is the useful finding rather than a failure of the plan: check the number against a current benchmark before treating it as a miss, because a 12-month payback is an SMB or product-led result today rather than a general bar (section 9). That is the real conversation the functional split exists to force, and it is why the cost side of the plan cannot be settled without the revenue side (section 4). The related metrics that do the same job for other slices — the magic number, sales efficiency, revenue per employee — are in section 9.

The case against setting targets this way

Top-down target-setting has a documented failure mode worth knowing, because it will come up. Michael Coveney's position is that a stretch target set in isolation from the manager who has to deliver it "will be seen as 'not my numbers'" (FP&A Trends). His alternative separates forecasting from target-setting: establish what is realistically achievable from operational staff, analyze separately what the market would permit if the company stretched, identify the process changes that would bridge the two, fund both the run-the-business base and the change initiatives, then track whether those initiatives are actually delivering the stretch — so that "improved performance can be managed in a collaborative way, rather than relying on hope." That structure is the private-sector version of a split you already run, and it reappears in 3.4.

3.2 How the number actually gets decided

Everything above is the arithmetic, which is the easy half. What follows is practice rather than doctrine: it does not appear in vendor process diagrams, and it is what an experienced FP&A lead is actually managing between September and December. Where a source exists it is given; where one does not, the claim is flagged as observed practice and should be checked against the specific company.

The plan is pre-wired before it is presented

A board meeting is not where the plan gets decided. It is where a decision already reached in private gets ratified. Between the last review and the board packet, the CFO walks the plan individually through the directors who will have opinions — the lead investor, the compensation committee chair, the audit chair at a public company — surfaces objections, and changes the plan before it is presented. The word for this is pre-wiring (also socializing, or pre-reading the deck). A plan that arrives at a board meeting genuinely unseen is a plan whose CFO has made a mistake. Unsourced; standard practice.

The consequence for you is procedural and specific. The board packet's deadline is not the real deadline. The real deadline is roughly two weeks earlier, when the CFO needs a defensible version to walk around, and every dependency behind it — the revenue plan, the headcount roll-up, the scenario set — has to be stable by then even though nothing is formally approved. Build the calendar in section 2 against the pre-wire date, not the meeting date.

There is more than one number, on purpose

Companies carry three versions of the same year at once — the board plan, the higher internal plan the executive team actually manages to, and the higher still quota assigned to the sales organization — and 2.5 works through what each is for and why the over-assignment is arithmetic rather than padding. The operational point that belongs here is narrower: only one of them can be the plan of record. Load the board plan into the planning system for budget-versus-actual reporting (section 8) and carry the internal plan as a separate scenario, because a company that loads all three ends up unable to explain its own variance. Unsourced; standard practice.

The plan is a compensation document

The single fact that most changes how the AOP behaves is that people's pay depends on it. Protiviti's budgeting best-practice checklist puts the mechanism plainly: performance metrics chosen during the budget process "can be used as a basis for incentive planning to drive behaviors that may allow the organization to achieve strategic and financial goals" (Protiviti). Read that in reverse and you have the politics of the fall: whatever metric you propose as the plan's headline becomes the metric someone's bonus keys off, and everyone in the room knows it before you do.

The consequences are direct. Executive bonus targets are set against plan, so a plan set 10% higher lowers the expected payout for every executive on the plan — which means the argument about the target is also an argument about their pay, whether or not anyone says so. At a public company the compensation committee approves the metrics and thresholds, often at the same board meeting that approves the plan, which is why the metric definitions in 3.1 get contested so hard. And the stakes run past the bonus pool: at a public company a materially missed commitment carries job and legal exposure for the people who made it, which 3.5 takes up under the budget-as-contract framing (AFP).

Some decisions were made before the cycle opened

A meaningful share of what looks like an open planning question was settled months earlier and is travelling through the process for form's sake. A multi-year enterprise contract signed in June commits FY27 spend. A reorganization announced in September has an implied headcount shape. An acquisition in diligence has a plan attached that cannot be discussed. A commitment the CEO made to a customer or to the board in a prior quarter is not going to be reopened by a departmental submission.

Two behaviors follow. Find out early which lines are actually open — the fastest route is to ask your CFO directly which parts of the envelope are already committed, a question that gets a straight answer far more often than new arrivals expect. And when a submission argues against something already decided, say so at once rather than routing it through a review cycle that cannot change it; running a department through three weeks of modeling to reach a foregone conclusion is how an FP&A team loses its business partners. Unsourced; standard practice.

3.3 The starting point, and the ways to build

This subsection answers Q4. There are four named methodologies plus a fifth cadence that replaces the annual cycle altogether, and a starting point everybody uses and nobody names. A real company mixes them by line item rather than choosing one.

The run-rate baseline (the thing nobody names)

Before any method is applied, somebody establishes what next year costs if nothing changes. That is the run-rate baseline: take the most recent month's or quarter's actual spend, annualize it, and treat the result as the do-nothing number. The refinement is the exit rate — not the year's average but December's level annualized, which is the honest baseline when the company grew during the year. A company that spent $33M in FY26 but exited December at a $38M annualized run rate starts FY27 $5M above its own prior-year total before hiring anybody. Missing that distinction is a common first-year error.

Be aware that "run-rate baseline" is working vocabulary, not a documented methodology. No source found for this guide defines it the way ZBB and activity-based budgeting are defined; it shows up implicitly as the thing incremental budgeting assumes and zero-based budgeting refuses. The nearest sourced statement of the practice is Pigment's: finance produces "an initial projection grounded in strategy and prior performance, creating the starting point against which bottom-up inputs can be reconciled" (Pigment). Note the word reconciled. FP&A proposes the baseline; the department argues with it.

Incremental budgeting

Incremental budgeting starts from last year's budget and "usually implements incremental percentage increases or decreases," typically 1% to 10% (CFI). Only proposed new spend gets examined; the base rides through untouched. It is fast, politically cheap, and the default where cost structures are stable. Its defect is that it compounds: anything wasteful in the base is re-funded forever, and a department that won an argument in 2022 never has to win it again.

Zero-based budgeting

Zero-based budgeting (ZBB) "allocates funding based on efficiency and necessity rather than budget history." Every expense must be justified to qualify for inclusion, and there are "no expenses that are automatically added to the budget" (CFI, same source). The operational unit is the decision unit — a department, program, or activity rebuilt from $0 each cycle rather than adjusted (Cube).

ZBB is the technique most likely to be misdescribed, so be precise about two things. It is not annual: almost nobody rebuilds every cost center from zero every year, because ZBB needs "qualified personnel and specialized training, which can be time-consuming and costly" and is "substantially more complex and tedious" than the alternative (CFI). Most companies apply it to a rotating subset of the cost base, or once, hard, during a cost transformation. And it is not the same as cost-cutting.

One caution that matters specifically at the companies you are targeting: Cube's account of ZBB's weaknesses is that its short-term focus can undervalue long-horizon investment such as R&D, whose revenue attribution is indirect, and can squeeze functions like HR whose benefits are real but hard to price (Cube; paraphrased, not quoted). A justify-every-dollar regime starves anything whose payoff is diffuse, which at a software company is most of engineering's platform work.

Driver-based budgeting

Driver-based budgeting builds the plan out of operational quantities rather than dollar line items. It "focuses on the key factors that drive financial performance," and "instead of just looking at past data, it builds forecasts using real-time business metrics" (CFI). A driver-based cloud-hosting line is not "$1.2M, up 15%"; it is monthly active accounts times average compute per account times unit price, and it recalculates when any of the three moves. This is the dominant method at tech companies for anything volume-sensitive, and it is what makes the reforecasts in section 7 cheap: change three assumptions, the model rebuilds.

It is also the format FP&A pushes onto departments. Farseer's third step of ten is "design and share driver-based templates," and its recommended submission rule is that "a cost increase beyond inflation or volume growth needs a business justification" (Farseer) — a rule only enforceable if the template makes the volume driver explicit. The corresponding question FP&A asks on every submission is one sentence: "Where does this number come from?"

Activity-based budgeting

Activity-based budgeting (ABB) analyzes the activities that create cost and prices them per unit of activity. It "does not take historical costs into account," and runs in three steps: identify the cost drivers, project the units of each driver required, calculate the per-unit cost (CFI). Be honest about its scope. ABB is a cost-accounting technique with real currency in manufacturing and much less in software. CFI's own fit note is that ABB is "better suited to new businesses that lack historical costing data," while established businesses that already hold years of cost history find it less necessary — a statement about company maturity rather than about industry, though the practical effect in software is the same. Know the term; you are unlikely to build one at a SaaS company. Its closest live cousin there is cost-per-ticket or cost-per-transaction modeling inside COGS (section 6).

Rolling forecasting, as an alternative to the cycle

The fifth option is not a way of building the annual plan but a way of not depending on it: a rolling forecast re-projects a fixed horizon — commonly four, six, or eight quarters — every month or quarter, so the planning horizon never shortens the way an annual budget's does in November. It is the operating mechanism behind Beyond Budgeting (3.9) and, in a far more common compromise form, it runs alongside an annual budget that is kept for board approval and external reporting. The mechanics, horizons and cadence choices are section 7; it appears in the table below because in a planning conversation it will be offered as an alternative to the other four and you should be able to say what it does and does not replace.

3.4 Baseline versus new investment

This subsection answers Q3. Yes, there is an analog to the carry-forward-level / maintenance-level / policy-level stack. Companies build the same three layers, but by convention rather than instruction, calling each layer something different at every company, with the boundaries settled by internal argument rather than by a definition somebody else issued. How rigorously the layers are kept apart varies enormously: at a disciplined company the baseline is modeled and defended line by line, while at a company running a light incremental process the three collapse into one departmental number and the separation exists only in the CFO's head. Do not assume the structure is in place; ask what the company's baseline is and see whether anyone can produce it.

1. Run-rate baseline. Existing headcount at existing salaries, existing contracts, existing infrastructure at existing volumes. Also called the base, the base case, the do-nothing case, or run-the-business — and frequently KTLO, keep the lights on, particularly in engineering organizations where "KTLO versus new product work" is a standing capacity argument.
2. Mandatory true-ups. Cost growth on the existing base that nobody chose: the merit and promotion cycle, benefits and payroll-tax inflation, contractual vendor escalators, cloud spend rising because usage rose, and the annualization of people hired part-way through last year. This layer has no standard name — you will hear "annualization," "the merit ask," "run-rate creep," "inflation."
3. New investment. Everything discretionary — new headcount, new programs, new products, geographic expansion, a platform migration. This is where individual asks compete against each other and against the envelope.

The artifact that makes the layers visible: the bridge

The deliverable that carries this structure is the bridge, also called the walk or the waterfall: a single view that starts at last year's actual and ends at next year's plan, with every intervening line explaining one reason the number moved. It is usually presented as a waterfall chart in the board deck and as a table in the working model, and it is the most-requested artifact of the whole cycle, because it is the only one that answers the question a CEO actually asks — "why is next year $6.9M more expensive?"

Build it for Meridian, whose cash opex goes from $35.0M to $41.9M.

Bridge line	Amount	Layer	Where the number came from
FY26 actual cash opex	$35.0M	Actual	Closed year
Annualization of FY26 hires	+$2.0M	True-up	Headcount went 215 to 250 during FY26, so the exit level is ~8% above the average that was expensed; applied to the $25.5M of payroll sitting in opex
Merit and promotion cycle	+$1.0M	True-up	3.5% on the annualized payroll base, effective on the merit date (section 5)
Benefits, payroll tax and insurance inflation	+$0.4M	True-up	Renewal quotes and statutory rate changes
Vendor, software and hosting escalators	+$0.5M	True-up	Contractual uplifts on ~$9.5M of non-people opex
Attrition credit, 18 planned departures	−$1.4M	True-up	Partial-year effect of expected voluntary attrition
Run-rate baseline (KTLO)	$37.5M	Baseline	What FY27 costs if the company does nothing new
Hiring: 36 gross hires, phased	+$3.7M	New	18 net adds plus 18 backfills at $190K fully loaded, phased (3.6)
New programs and initiatives	+$0.7M	New	Approved business cases net of the ones cut
FY27 plan	$41.9M	Plan	Ties to the envelope in 3.1

Read the bridge and the plan explains itself. Of the $6.9M increase, $2.5M — 36% — is consumed by the baseline before a single new person is approved, and only $4.4M is available for everything the company actually wants to do next year. That is the sentence that ends most arguments about whether a department is being treated fairly, and it is the private-sector equivalent of the moment you show a program manager how much of the biennial increase CFL and ML already ate.

Two mechanical notes, because both are common first-year mistakes. Annualization is not a percentage you look up — it is derived, and a serviceable rule is that it runs at roughly half the year's percentage headcount growth, applied to the payroll base, when hiring was spread evenly. A company that grew headcount 16% carries about 8%; a company that grew 40% carries about 20%; a company that hired everyone in Q4 carries nearly the full 40%. And an attrition credit is only honest if the plan also contains the backfills, which is why Meridian's 18 departures and 18 backfills both appear. Netting them to zero hides the timing, and the timing is where the money is.

Keeping the layers apart

Companies keep the third layer structurally separate from the first two because merging them lets departments hide new asks inside the base. Vena's recommended counter to bottom-up padding is to "set aside an overall strategic budget for new initiatives and make sure departmental padding doesn't encroach" (Vena). Pigment's version is that strategic initiatives — M&A, product launches, geographic expansion — need to be modeled discretely, with their own assumptions and phasing, then integrated into the overall projection rather than smuggled into a departmental line (Pigment; paraphrased). The parallel to OFM's structure is real, but note that the private-sector motive is stated by practitioners as anti-padding discipline, while OFM's own instructions do not state a motive for the CFL/ML/PL split at all — so read the resemblance as convergent design, not as evidence that both were built for the same stated reason.

The business case, and an honest caveat

The vehicle for a new-investment ask goes by several names: business case, investment request, funding request, initiative brief, or at companies with a formal process a capital request or CAR (capital appropriation request), mostly at industrial and large enterprise companies. A typical one states the objective, the quantified problem or opportunity, resources required by year, expected financial return, the assumptions the return depends on, and a named owner accountable for delivering it.

This is the weakest-sourced topic in the section, and it is worth saying so rather than manufacturing precision. No authoritative source found while researching this page documents a standard business-case format, standard ROI hurdle rates, or standard approval thresholds for private-sector operating investments, because there is no standard — they are set per company, usually by the CFO, and vary enormously. The mathematics that goes inside one (NPV, IRR, payback) is section 10; the delegation-of-authority matrices governing who can say yes are section 11. What matters here is only the structural point: new asks are documented, competed, and approved separately from the base, and a department that wants more money "explain[s] why" in a formal adjustment request (Farseer).

3.5 Rounds, versions, and closing the gap

The mechanical sequence is settled practice. Farseer's version runs: develop guidelines, build the calendar backward from board approval, design and share driver-based templates, train stakeholders, collect operational inputs, run weekly check-ins, consolidate department plans into one company view, review assumptions and variances, complete FP&A, business, CFO, and CEO reviews, then get board approval and lock the final version (Farseer). Vena's process article and Protiviti's ten-step list differ in labeling, not substance, though it is worth knowing what each is: Farseer and Vena describe how a cycle runs, while Protiviti's is a best-practice checklist of what a good process should include — data integrity, cross-functional participation, technology and AI enablement, performance metrics — rather than an account of what companies typically do (Vena; Protiviti). The instruction inside Farseer's list worth memorizing is the second: build the calendar backward from board approval. Every other date is derived from one somebody else owns — and, per 3.2, the date you actually work back from is the pre-wire, not the meeting.

V1 and the gap

The first consolidated roll-up is v1, and it never fits. Pigment gives the canonical shape of the mismatch: "Sales might project $50M in revenue when the CEO's target is $65M. Research and development might request 30 new engineers when the CFO has budgeted for 20" (Pigment). Farseer is blunter: "Bottom-up almost never matches top-down on the first pass. Departments plan conservatively. Leadership plans ambitiously. Closing that gap through honest conversation, rather than arbitrary cuts, is where FP&A earns its seat."

Two behaviors produce the gap. On the revenue side, sales leaders submit low; on the cost side, department heads submit high. Both are budgetary slack, which CFI defines as "the practice of overestimating the expenses and/or underestimating the projected revenues when preparing a budget statement for the next financial period" (CFI). The colloquial term for the revenue half is sandbagging — a word you will hear constantly and see written down rarely; Pigment uses it in passing when describing what finance has to navigate in bottom-up inputs, alongside "the optimism, the competing priorities." CFI names three causes: genuine uncertainty, information asymmetry (the department knows more about its own costs than you do and can exploit that), and reward structures — when "employee awards and payoffs are dependent on budget attainment," managers create slack deliberately to guarantee they beat the number. That third cause is a design problem, not a character problem. If a VP's bonus depends on hitting the plan she submits, she will submit a plan she can hit. CFI's own remedy is to decouple performance evaluation from budget attainment; few companies do, so FP&A absorbs the problem instead by knowing the business well enough to price the padding.

Pricing the padding

"Know the business well enough" is not a technique, so here is what the technique actually looks like. None of the following is documented in a source found for this page; all of it is ordinary practice, and all of it is checkable against data you already have. Compare each department's submitted line to its own trailing twelve months of actuals, not to its prior budget, since a department that underspent its budget by 12% and is now asking for last year's budget plus 8% is asking for a 20% real increase. Look at the prior year's budget-to-actual by department: a team that has come in 10% under three years running has a known slack rate and can be discounted by it without a debate. Check whether the volume driver behind a line moved as much as the dollars did. Ask for the phasing, because a padded line is usually padded in Q4, where it is least visible and most deniable. And treat a submission with no line-item detail as a submission that has not been made.

The professional way to raise it matters more than the finding. "Your number looks padded" ends the partnership; "your submission is 20% above your trailing run rate and I can only see 8% of drivers behind it — walk me through the rest" gets you the answer and keeps the relationship (section 14).

Gap-closing, worked

Return to Meridian. The opex envelope is $41.9M. The CFO holds $1.5M back (see 3.6), so departments are allocated $40.4M. V1 submissions total $47.3M. The gap is $6.9M, 17% of what was allocated — an entirely ordinary v1 gap.

Lever	In-year savings	Effect on FY27 exit run rate
Move 12 of the 44 requisitions from Q1 to Q3 starts	$1.1M	None
Cut 8 requisitions outright (weakest business cases)	$1.1M	−$1.5M
Reduce marketing program spend $4.2M → $3.0M	$1.2M	−$1.2M
Software and vendor consolidation at renewal	$0.6M	−$0.6M
Contractor conversion and T&E policy change	$0.4M	−$0.2M
Phase a platform-infrastructure program over two years	$0.9M	None
Three asks deferred to a "fund if we beat H1" queue	$1.6M	−$1.6M
Total	$6.9M	−$5.1M

Four things there are worth naming. The 44 requisitions — 26 net-add requests plus 18 backfills — less the 8 cut are the 36 gross hires in the bridge in 3.4; the two views have to reconcile, and if they do not, one of them is wrong. The two largest levers — start-date phasing and multi-year program phasing — save in-year money without cutting anything, which is why timing is the first place an experienced FP&A lead looks. The exit-run-rate column exists because the CFO cares about both numbers: in-year cost makes FY27's P&L work, and the exit run rate is what FY28 starts from, so a gap closed entirely with timing tricks leaves FY28 with a problem before it opens. And the marketing cut has a consequence that travels backward — $1.2M less program spend means less sourced pipeline, which changes the revenue plan, which changes the envelope, which changes the CAC payback calculation in 3.1. Cost-side gap-closing is not independent of the revenue plan (section 4), and pretending otherwise is the most common way an AOP becomes internally inconsistent.

Controlling the rounds

Version proliferation is the operational risk. Farseer names it exactly: "the final approved plan must be clearly identified and locked. In Excel, this is where 'AOP_v21_final_FINAL_approved(2).xlsx' is born." Its six submission rules are the standard countermeasure: approved templates only; justify any increase beyond inflation or volume growth; allow post-submission changes only at defined checkpoints with the right approval, and not "through a Tuesday night email with 'final_v7' in the subject line"; require a formal adjustment request for extra money; phase expenses into a realistic monthly or quarterly split, because "a full-year number split evenly across twelve months is lazy and usually wrong"; and "define unused budget treatment," deciding up front whether unspent quarterly funds carry forward or return to a central pool. That last rule is the one most often skipped and the one you should recognize fastest: it is the private-sector answer to the use-it-or-lose-it problem, and a company that has not decided it has decided it by default (3.9).

In a planning system rather than a spreadsheet, each round is held as a named scenario — top-down target, bottom-up submission, each review iteration — instead of a new file (section 12). The review chain runs departmental consolidation, FP&A review, business review, CFO, CEO, board; each layer is a real gate that produces changes, which is why the calendar has to hold two to three weeks between the last submission and the board packet.

Locking the plan and the "budget of record"

Once the board approves, the plan is locked, and the version everything is measured against for the rest of the year acquires a name — most commonly the budget of record, sometimes the approved plan, the plan of record, or just "the budget." That exact phrase is working vocabulary; no source found for this page defines it formally, so treat it as a term to recognize rather than one to quote authoritatively.

The mechanism behind it is well documented even where the phrase is not. Prince Oppong, a Senior Director of Strategic Finance at PayPal, puts it directly: "A budget isn't just an aspirational exercise or an expression of strategy; it's literally a contract between multiple parties." The consequences are personal and, at a public company, legal: "CEOs and executives can lose their jobs, lose a significant portion of their remuneration or face legal exposure if facts were misrepresented" when performance deviates materially from what was committed. And the plan does not quietly move afterward — reallocating against a locked plan "requires CEO and CFO support. Full stop" (AFP). The locked budget then stays fixed all year even as the forecast moves away from it; that gap is section 7, and explaining it monthly is section 8.

What happens when the plan is late

Plan the calendar assuming approval slips, because it often does. A December board meeting that ends without a decision, a January re-cut after a bad Q4, a re-plan triggered by an acquisition or a down-round conversation — any of these leaves the company operating in January against no approved budget. What companies do in that window is uniform enough to be worth knowing in advance: departments run on the prior year's spend level or on a provisional allocation, new requisitions freeze except for named exceptions, non-committed program spend pauses, and finance reports against the last submitted version with a clear label saying it is not approved. The two mistakes are letting an unapproved version circulate without that label, and loading a provisional plan into the system of record, which produces a year of variance reporting against a number nobody ever agreed to. Unsourced; standard practice.

3.6 Contingency, phasing, and the release gate

No plan is built to be spent exactly. Companies hold back capacity in four forms, and only the last is well documented.

An unallocated pool. A slice of the envelope the CFO does not distribute — called the unallocated pool, the contingency, the CFO reserve, the CEO reserve, or "corporate." It funds mid-year opportunities, absorbs surprises, and gives the CFO something to release when a business case that lost in November turns out to have been right in April. This practice is real and common but not standardized: no source found for this page defines it as a named budget line, states a typical size, or documents who releases it. Pools at a company Meridian's size generally run low single-digit percentages of opex, and Meridian's $1.5M is 3.6% of the envelope. Treat that as illustrative, not a benchmark, and ask what a specific company does rather than assuming.

Timing levers. The larger and more reliable contingency is that the plan's own phasing contains slack, and nearly all of it lives in headcount, because headcount drives most of opex (section 5). Take Meridian's 36 gross hires — 18 net adds plus 18 backfills — at an average fully loaded cost of $190,000. If all 36 started January 1 they would cost $6.8M. Hired evenly at three per month starting on the first of each month, they generate 234 person-months of the 432 in a full year — 54% of the annualized cost, or $3.7M in-year against a $6.8M exit run rate. That single ratio is the most useful arithmetic in expense planning: it holds whatever the headcount number is, because it depends only on the shape of the hiring curve, and it means the phasing decision is worth more than almost any line-item negotiation.

It cuts both ways. Assume every requisition fills on its planned start date and the plan overstates spend, because real hiring slips: a 45-day average slip across 36 hires is 54 person-months, about $0.9M of underspend that looks like brilliant cost control and is actually a recruiting problem. Assume slip and then hire on time and you are $0.9M over. Experienced teams apply an explicit vacancy or slip factor, state it as an assumption, and track it, rather than leaving it implicit.

Accrual and effective-date levers. A set of assumptions inside the plan move cost without changing anything anyone would call a decision, and they are worth knowing because they are where a plan gets quietly tightened in the last week. The bonus accrual rate is the largest: a plan that accrues the annual bonus pool at 100% of target and one that accrues at 85% differ by real money on a Meridian-sized payroll, and both are defensible. The merit effective date is the second: moving the annual increase from January 1 to April 1 keeps the full annualized cost but removes a quarter of it from the plan year. Capex versus opex classification is the third — see below. Each is legitimate, each has to be stated as an assumption rather than buried, and each is the first thing an experienced CFO asks about when a plan lands $500K under the envelope with no visible cuts. Unsourced; standard practice.

Pre-committed response playbooks. The best-sourced contingency practice costs nothing up front: deciding in advance what you will do if the plan misses. Pigment's argument is that "performative downside scenarios just show lower revenue without defining how you'd actually respond," and that a usable one answers specific questions — "At what revenue threshold do you freeze hiring? Which initiatives get paused in what sequence? What costs can you cut in 30 days versus 90 days?" The payoff is speed: "The teams that respond fastest have contingency plans ready before conditions shift. They're acting on predefined playbooks rather than building strategy in real time" (Pigment).

Triggers only work if the assumptions they hang on are falsifiable. Pigment's examples of the right grain are assumptions stated as specific, checkable numbers — a win rate moving from 22% to 28%, an average deal size holding at $85K while the company moves upmarket — specific enough that a miss is unambiguous and the pre-agreed response fires without a debate about whether the plan is off track (Pigment, same source; paraphrased). A plan whose stated assumption is "revenue will grow 30%" has no triggers, because by the time you know it is wrong it is too late to act. The related pairing on the revenue side — a commit number treated as near-certain and a stretch number above it, often with different compensation consequences — belongs to sales capacity planning in section 4. Its relevance here is that expense planning is normally built to the commit number, with the spend a stretch outcome would require held as a conditional release rather than funded in the base.

An approved headcount plan is not permission to hire

This is the single most common misunderstanding a new FP&A hire carries into the first quarter, and it has an exact public-sector analog you will find reassuring. A position in the approved plan is a budgetary provision, not an authorization. What authorizes hiring is a released requisition. AIHR's definition: "A job requisition is a formal request to create a new position or to fill a vacant role in a company." It travels an approval chain the hiring manager starts and HR closes: "As soon as it is approved, HR finalizes the job req and assigns a job requisition number to the document or online form. Then, the job intake (an initial meeting between HR and the requesting manager) takes place, after which the visible recruitment process starts." The budget question sits inside that chain explicitly: "Often, the question is also whether or not the required funding for the role is available. If it's available, the job requisition can be approved quickly. However, if the budget is unavailable, there will be a more stringent analysis of whether or not the role is needed" (AIHR).

The planning consequence is that a company has two levers, not one: what is in the plan, and when the requisition is released against it. Companies use the second constantly. A common structure releases requisitions quarterly against performance — Q1's reqs open at the start of the year, Q2's release only if Q1 revenue lands within some percentage of plan — which converts a fixed headcount plan into a conditional one without renegotiating anything. A hiring freeze is the same lever pulled all the way, and it does not require a re-plan because the plan was never the authorization. When you build the headcount plan (section 5), find out who releases requisitions and on what trigger, because that person controls the largest line in your budget regardless of what the plan says. The mirror image at DSHS is the gap between an FTE authorized in the appropriation and a position you can actually fill given hiring freezes, vacancy savings targets, and position-control review — same two-lever structure, different vocabulary.

3.7 What you actually ship

The AOP is a set of artifacts, not a document. Farseer's component list is representative: a revenue plan by customer, product, channel, or geography; production and inventory plans where relevant; a COGS plan; a marketing and sales plan; a "headcount and salary plan: existing workforce, new hires, and total employee cost"; an administrative plan; an R&D plan; and a capex plan covering "equipment, facilities, and systems." The financial output is "a full P&L for the coming year, phased by month or quarter," and the plan "mirrors the structure of the income statement, plus key balance sheet items" (Farseer). At a software company the production and inventory plans drop out and a cash and runway plan becomes more prominent.

Deliverable	Grain	Primary audience	Lives in
P&L by month by department	Month × cost center × account	CFO, department heads	Planning system; the system of record for the year
Headcount plan	Position × start month × department × location	CFO, CHRO, hiring managers	Planning system, tied out to the HRIS and the requisition list
Revenue plan	Month × segment × motion; ARR waterfall	CRO, CFO, board	Planning system, reconciled to CRM
Cash and runway plan	Month; balance sheet and cash flow	CFO, board, lenders	Model or planning system
Opex-to-plan bridge	Prior-year actual to plan, one line per driver	CEO, CFO, board	Board deck as a waterfall; working model as a table
Budget book	Full detail by department and account	Department heads and their finance partners	Distributed after approval; the reference each owner manages against
Board deck	Summary P&L, KPIs, assumptions, scenarios, hiring plan	Board	Slides; the version that gets approved

Two caveats. "Budget book" and "board deck" are working terms — every company has both artifacts, but no source found for this page defines either or itemizes its contents, so the descriptions above are practice rather than citation. And what gets loaded to the planning system is narrower than what gets built: the approved P&L at the grain the company reports at, the approved headcount plan, and the driver assumptions the reforecast will flex. Working files, department rationale, and scenario variants stay outside it. If the planning system does not hold the locked plan at reporting grain, budget-versus-actual reporting (section 8) cannot be automated, and someone will spend the following year rebuilding it in Excel every month.

Two structural choices that decide what a department's number even means

Direct versus allocated cost. Some of what a department consumes is bought centrally and pushed out — facilities, IT, security, recruiting, shared cloud infrastructure — so a departmental target means nothing until the allocation policy is settled: which costs are pushed, on what basis, and whether the receiving VP is held to them or merely shown them. Section 6.3 covers the mechanics, the basis choices, and the showback-versus-chargeback distinction. What belongs here is the timing. Settle it before the cycle opens, because a company that changes its allocation basis mid-plan has destroyed its own year-over-year comparability.

Capitalized versus expensed software. At a software company part of engineering payroll never reaches the operating expense line at all, so the R&D line inside the envelope is stated net of a capitalization-rate assumption. That makes the rate a lever: change it and operating income moves with no change in cash, in headcount, or in what anybody builds. State it in the plan's assumption list rather than letting it drift, and settle whether the board's margin metric is computed on EBITDA, which is insensitive to the whole question, or on GAAP operating income, which is not (3.1). The stage rules under ASC 350-40, the amortization that brings this year's capitalization back as next year's expense, and the cloud computing arrangement extension are section 6.5.

3.8 Common failure modes

The failure catalog is consistent across sources, which is a good sign it is real. AFP's practitioner pain points are worth reading in the practitioners' own words: "Top-down expectations conflict with bottom-up builds"; "The metrics used to manage the business are different than the budget drivers"; "So many iterations! One more change … last-minute decisions … effects that cascade through the model"; "Operations sometimes wants to avoid accountability and considers the budget to be a 'finance activity'"; "Reality quickly makes the budget out of date"; and, most memorably, "If I do budgeting the same way again next year, fire me" (AFP).

Failure mode	What it looks like	What experienced teams do
Skipping the setup phase	Templates go out before guidelines, targets, or a calendar exist; every submission comes back a different shape	Farseer's observation is that "the teams that struggle usually skipped steps 1 through 4 and went straight to collecting numbers." Publish guidelines, calendar, templates, and training before requesting a number.
Version chaos	"AOP_v21_final_FINAL_approved(2).xlsx"	Named scenarios in a planning system, or one owned file with a change log and defined edit checkpoints
Blanket percentage increases	Every department submits last year plus 8% with no driver behind it	Driver-based templates plus the rule that any increase beyond inflation or volume growth needs written justification
Divide-by-twelve phasing	Annual costs spread evenly, so every month's variance is wrong and Q1 looks like an overspend	Require a real monthly split; phase hires to start dates and programs to their actual quarters
No bridge	The plan is presented as a total, so nobody can say why next year costs more and every department believes it was treated worst	Build the prior-year-actual-to-plan bridge first and present it before the departmental detail (3.4)
Consolidation eats the cycle	Weeks spent merging spreadsheets, no time left to analyze what they say	Fix the intake format; the analysis, not the assembly, is the job
Accuracy obsession without variance discipline	Elaborate precision in the plan, no rigor about explaining misses afterward	CBH's diagnosis: "Budget processes often fail when companies focus too much on accuracy and not enough on variance tracking." Design the plan so variances are explainable at the driver level.
Imposed stretch targets	Departments deliver the number they were given and disown the result	Separate the realistic forecast from the stretch, and fund the specific initiatives meant to bridge them
Generic downside scenarios	A "low case" that is the base case times 0.85 with no response attached	Trigger thresholds and pre-agreed response sequences, per 3.6
Plan headcount read as hiring authority	Managers open requisitions on January 2 for the whole year's plan and the burn overshoots by Q2	Explicit requisition-release gates tied to performance, agreed before the plan is published (3.6)
Managing to the budget	Departments spend the remainder in December so the base is not cut next year	The use-it-or-lose-it problem, addressed structurally in 3.9

Sources for the quoted diagnoses: Farseer, CBH, Coveney, Pigment, and AFP. The unquoted rows are common practice.

AFP frames the fix as seven decisions finance should settle deliberately before the cycle starts rather than discover during it: how much risk the company will take; how the baseline gets set; how plans translate into operational targets; who makes which decisions; whether the process is optimized for control or agility; what changes are permitted during the year; and what level of granularity is right (same source). That last is the most commonly botched — a plan built at a finer grain than the company can report against produces variance analysis nobody can perform.

3.9 Beyond Budgeting and the case against the annual budget

There is a serious, decades-old argument that the annual budget should not exist. Know it, because it is intellectually respectable and because the vocabulary shows up in job descriptions at companies that consider themselves modern.

The best-known statement of the case is Jeremy Hope and Robin Fraser's "Who Needs Budgets?" in Harvard Business Review, February 2003, whose central claim is that "budgeting, as most corporations practice it, should be abolished." The argument is not that planning is bad but that the annual budget perpetuates "the command and control culture" and centralized hierarchies at companies that need devolved networks, and actively prevents the other management tools they have already adopted from working (HBR). Call it the movement's best-known articulation rather than its founding document: the authors' own bylines on that 2003 article identify them as directors of the Beyond Budgeting Round Table, which by then already existed. The organization that continues the work is the Beyond Budgeting Institute, which describes Beyond Budgeting as "a progressive set of leadership principles and management processes proven to free organisations of 'command and control' cultures and improve performance," and names Handelsbanken, Coloplast, David Lloyd Clubs, Roche, Ferrer, and Human Rights Watch among adopters (BBI).

The specific dysfunction it targets should be immediately recognizable to you. Bjarte Bogsnes, a long-time practitioner and writer on Beyond Budgeting, states it as: "the budget bank is open once a year only, and unspent budget funds are lost." That is the same use-it-or-lose-it dynamic that produces end-of-period spending sprees in state agencies, arising from the same mechanism — a fixed allocation for a fixed period, forfeited if unused. (The WA version keys to the biennium rather than to the calendar year: "All appropriations shall lapse at the end of the fiscal period for which the appropriations are made to the extent that they have not been expended or lawfully obligated" (RCW 43.88.140), and for the operating budget that period is the biennium — which is why the pressure concentrates in the second June of a biennium rather than every June.) What replaces it, in Bogsnes' account, is three things: rolling forecasts instead of the fixed annual budget; relative targets, where performance is judged against peers, the market, or the prior period rather than an absolute number fixed a year in advance; and dynamic resource allocation instead of pre-allocated line items, including what he calls burn-rate guiding — "operate at an activity level that expressed in money is in the range of x until something else is decided" — explicitly to stop finance micromanaging travel and consultant spend line by line (Bogsnes, FP&A Trends).

The practical takeaway for someone entering private-sector FP&A in 2026 is that you will build an annual plan, and a meaningful part of the job is reducing how much that plan constrains decisions after February. The mechanisms for that — rolling forecasts, quarterly reallocation, relative rather than absolute targets — are section 7. What this section covers is the plan itself, which remains, at essentially every company you will interview with, the thing the year is measured against.

4. The revenue plan (built separately, and why)

This section answers core question Q2 directly: yes, revenue is planned separately from expenses. The trigger is not company size but structure. Once there is a quota-carrying sales team and a CRM of record, the top line is built by a different group, from a different data set, and it finishes first, because the expense and headcount plans are sized against it.

In WA state the top-line numbers you build against are handed to you, but not all by the same body and not all at your level. The Caseload Forecast Council's entitlement forecast drives your maintenance level, and the Legislature's appropriation, allotted by month, sets the envelope you spend inside. The ERFC revenue forecast binds one level up: the governor's budget document proposes expenditures "based upon the estimated revenues and caseloads as approved by the economic and revenue forecast council and caseload forecast council" (RCW 43.88.030(1)). In a company you build the spending plan against a revenue plan your own sales organization commits to. The shape of the dependency is familiar. Who owns the number, whether it is a forecast or a target, and whether the organization can go change it, is not.

Who owns the number, and why the split exists

Three groups touch the revenue plan, and the boundaries determine who you argue with.

• The CRO (Chief Revenue Officer) or VP of Sales commits to the number and tells the board the team will deliver it — a CRO owning all revenue sources including renewals, partner and self-serve, a VP of Sales owning the quota-carrying team and new bookings.
• RevOps (Revenue Operations), or Sales Ops at smaller companies, builds the math underneath that commitment: pipeline, stage conversion, win rates, territory and quota assignment, ramp curves, comp mechanics, and the CRM data it derives from. RevOps owns "the leading indicators and the operational mechanics that turn the plan into reality" (RevSearch, a search firm placing RevOps leaders — practitioner content, not survey research). Below roughly Series B, RevOps often does not exist and a sales-ops analyst or the CRO's chief of staff does this part time, pushing more of it onto you.
• FP&A owns the financial translation: whether the number is credible against history, what it costs to deliver, how it converts to billings and cash, the sensitivity cases, and how it lands in the board deck. Same source: FP&A "owns the financial outcomes and the long view of the financial plan."

On capacity planning that source splits it cleanly: RevOps supplies "productivity, ramp, and quota inputs; FP&A owns the budget envelope." The joint decisions where the two must agree or the plan breaks — comp design, quota setting, capacity planning, renewal and expansion forecasting, the marketing budget, forecast reconciliation before the board narrative — describe your own job wherever there is no real RevOps function.

Why the two plans are built separately

No source states this as a formal design principle; it is a convention rather than a rule. Four reasons hold up under questioning.

1. Different source data. The expense plan comes from a headcount roster, vendor contracts and the general ledger; the revenue plan from the CRM — open opportunities, stage conversion, cycle length, renewal dates, usage.
2. Different math. Opex is additive and controllable: you decide to hire someone, the cost appears. Revenue is probabilistic: you hire a rep and eight months later you may or may not have revenue. One is a set of decisions, the other a forecast of behavior you influence but do not control.
3. Sequencing. The revenue plan is an input to the expense plan, not a peer. Sales headcount derives from the new-ARR target, customer success headcount from ending ARR and account count, hosting cost from usage, marketing spend from the pipeline requirement. See section 3 and section 5.
4. Different accountability. The CRO is measured against the revenue plan personally and the team is paid on it through commission, while the expense plan is spread across dozens of budget owners with no variable comp attached. Merging them would put two incompatible incentive structures in one document.

Three numbers, not one

Every company past roughly thirty reps runs at least three revenue numbers at once, and nobody introduces them to you. Hearing "our plan is $12M" and "we have $16M of quota out there" in one meeting and concluding somebody is lying is a first-month mistake.

• The board plan (or external plan) — what the CEO commits to the board, and at a public company the basis for guidance. This is the number that must not miss.
• The internal plan — commonly a few points above the board plan. This is what FP&A sizes the expense and headcount plan against, so a small miss consumes cushion rather than forcing an in-year cut.
• Quota on the street — the sum of every individual quota actually assigned, deliberately larger than either. Bill Binch, an operating partner at Battery Ventures, defines it as "the exact amount of quota, including ramps and attrition, that is deployed at any given moment" and puts it on the CEO dashboard weekly beside business plan quota (Battery Ventures).

The gap between plan and assigned quota is over-assignment. "Most companies over-assign quotas by 20-30%," with more cushion where the risk is — roughly "120% for SMB and 135% for Enterprise," because enterprise reps ramp longer and large deals slip (Mostly Metrics, CJ Gustafson's annual-planning series, from the SaaS CFO seat).

Both failure modes are yours to flag. Over-assign too little and a normal rep-level miss flows straight to the board number with nothing absorbing it. Over-assign too much and quotas stop being credible: attainment collapses, commission stops paying, and your best reps leave in Q2, costing you the capacity in the build below.

Bookings, billings, revenue, cash: four numbers, one deal

A single subscription contract produces four numbers on four timelines, and finance people use them interchangeably in conversation while meaning entirely different things. When someone says "we did $4M last quarter," ask which one.

• Bookings — the value of a signed contract, recorded once, at signature. Committed future revenue, and a leading indicator of sales execution. TCV (total contract value) is the whole multi-year commitment; ACV (annual contract value) is the annualized figure, computed by "taking a company's TCV bookings and dividing the metric by the term of the contract" (Wall Street Prep). Most companies exclude one-time and professional-services fees from ACV, so it is rarely exactly TCV ÷ term; read your company's definition before quoting it.
• Billings — the value of invoices actually sent, and a real working-capital lever: annual upfront billing on a multi-year deal pulls cash forward without touching revenue.
• Revenue — the GAAP number, recognized only as the service is delivered (Wall Street Prep).
• Cash — collections, lagging billings by your DSO. See section 9 for the working-capital mechanics.

Worked example: one $6M deal, five answers

A company signs a $6M contract on 1 July for four years of service, billed annually in advance at $1.5M. The term runs to 30 June of the fifth calendar year, so the columns below do not stop at year four.

Monthly recognized revenue is $6,000,000 ÷ 48 months = $125,000, or $1.5M annualized; a July start puts only six months, $750K, in the signing year (Wall Street Prep). The CRO can truthfully tell the board the team booked $6M while you truthfully report revenue moved $750K. That gap is why investors treat bookings as "a more accurate indicator of the growth profile of a SaaS company" than GAAP revenue (Wall Street Prep): revenue lags sales execution by design, so an accelerating company looks slower than it is, and one that has stopped selling looks healthy for about a year.

The ARR waterfall

The ARR waterfall (also ARR bridge, rollforward or walk) is the most-used artifact in SaaS planning: a rollforward structurally identical to the fund-balance rollforwards you already build, answering how we got from last period's recurring base to this one's.

Ending ARR = Beginning ARR + New ARR + Expansion ARR − Contraction ARR − Churn ARR

• New ARR — from customers who were not customers before. Also called new logo ARR.
• Expansion ARR — more seats, more usage, tier upgrades, cross-sell into existing customers.
• Contraction ARR — customers who stayed but shrank: seat reductions, downgrades, renegotiated pricing.
• Churn ARR — customers who left entirely.

ARR is the annualized value of recurring contracts in force at a point in time. It is not a GAAP measure and appears in no audited statement. MRR is the same thing monthly, used where the motion is month-to-month or SMB.

Worked example: a $40M ARR company planning to $52M

ARR walk component	Plan year ($M)	% of beginning ARR
Beginning ARR (1 Jan)	40.0	—
New ARR	12.0	30.0%
Expansion ARR	5.2	13.0%
Contraction ARR	(1.8)	(4.5%)
Churn ARR	(3.4)	(8.5%)
Ending ARR (31 Dec)	52.0	130.0%

New ARR, gross new ARR, net new ARR

One word, three quantities, and the walk above contains all three. People use the phrases loosely in conversation and precisely in comp plans, which is where the ambiguity starts costing money.

Term	In the walk above ($M)	What it counts	Where it is the right number
New ARR (new-logo ARR)	12.0	Customers who were not customers before. Expansion, contraction and churn all excluded.	Quota, sales capacity, pipeline coverage, AE commission
Gross new ARR	17.2	Everything added: new ARR plus expansion ARR, before any loss is netted off.	Combined sales and customer-success production; gross-add trends
Net new ARR	12.0	The change in ARR: new + expansion − contraction − churn, which is ending minus beginning.	Efficiency ratios — burn multiple, net-new-ARR CAC ratio (section 9)

The two 12.0s are an accident of this example, and exactly why you say which one you mean. Net new ARR equals new ARR only when expansion precisely offsets contraction and churn, which is what NRR of 100% means; the walk above was built that way. Hold the $52.0M exit and move NRR to 110% and they separate at once: net new ARR is still $12.0M, since ending minus beginning has not moved, while new ARR falls to $8.0M. That is lever 3 of the capacity build below, seen from the other side.

The rule to carry out of this: quota-bearing work is sized on new ARR — AE capacity, ramp, pipeline coverage, commission — because new logos are what a quota-carrying rep is assigned and paid on. Sizing the sales team against net new ARR charges reps with expansion and churn they do not carry: at 110% NRR you would hire half again more reps than the plan needs, and below 100% NRR you would under-hire. Keep net new ARR for the places where it is the metric's definition, and for the ARR growth rate itself.

Vendor glossaries disagree about this vocabulary, so read your company's own metric dictionary before putting any of the three in writing. One defines net new ARR as "the sum of New Logo ARR and Expansion ARR, minus Contraction ARR and Logo Churn ARR" while treating unqualified "New ARR" as "inflows only (New Logo + Expansion)," which is the gross-new sense (PortfolioIQ); a billing vendor's glossary restricts new ARR to revenue "generated exclusively from newly acquired customers during a specific period," expansion excluded (Ordway). Both are vendor content, not a standard. This guide uses the second reading throughout: new ARR means new logos.

Churn and contraction are not percentages, they are a renewal schedule

Showing churn as 8.5% of beginning ARR is a presentation, not a build: the only dollars at risk in a given year are the ones whose contracts come up. Build it from the renewal base — the subset of beginning ARR whose renewal dates fall inside the plan year, segmented and phased by quarter, with a gross renewal rate applied to each segment.

That reproduces the $5.2M in the walk and makes two things visible the percentage hides. The blended rate on the whole base (87.0% retained) is far more comforting than the rate on the dollars actually at risk (83.8%), because a fifth of the base was never up for renewal. And the renewal calendar is lumpy — say $6M, $7M, $8M and $11M by quarter — so a heavy renewal quarter produces a worse-looking walk at identical customer behavior, quarter-to-quarter churn comparisons are close to meaningless unless you normalize for the base, and a large multi-year cohort renewing inside the plan year is a concentration risk for the downside case.

Two retention ratios, and the difference between planning them and reporting them

• Gross revenue retention (GRR), sometimes gross dollar retention: what you kept from the existing base before expansion. (40.0 − 1.8 − 3.4) ÷ 40.0 = 87.0%. It excludes expansion by construction, so it can never exceed 100%, and measures whether the product holds customers.
• Net revenue retention (NRR), also net dollar retention or NDR: includes expansion. (40.0 − 1.8 − 3.4 + 5.2) ÷ 40.0 = 100.0%. It can and often does exceed 100%, and measures whether the installed base grows on its own.

Computing them straight off the walk is fine for setting a target and wrong for reporting an actual. The walk's churn line includes logos acquired inside the year, which belong in neither the numerator nor the denominator of a retention ratio; the reporting construction fixes a cohort live at the start of the period and measures it at the end. If $0.6M of the $3.4M churn came from in-year logos, cohort GRR is (40.0 − 1.8 − 2.8) ÷ 40.0 = 88.5%, not 87.0%. Your walk and your reported NRR never tie exactly, so know which construction the board deck uses.

Read the level the way an experienced director would. NRR of exactly 100% means expansion precisely offset losses, so the entire 30% growth from $40M to $52M is bought with new-logo sales, the expensive kind; at 110%, roughly $4M arrives without a single new customer and the capacity plan below gets much smaller. No other single assumption in a SaaS plan moves as much money, which is why FP&A owns "the NRR target in the plan" while RevOps owns the leading indicators driving it (RevSearch).

Benchmarks, and why a single NRR number is meaningless

The KeyBanc Capital Markets and Sapphire Ventures private SaaS survey is the closest thing this field has to an annual reference; a new edition lands each November, so check for a newer one before quoting. The 16th edition, published 13 November 2025, has "ARR growth ... projected to accelerate from 15% in 2024 to 20% in 2025" with companies "maintaining gross retention near 90% and net retention above 100%" (KeyBanc / Sapphire, 2025; press release). Treat even that as provisional: the 15th edition projected "ARR growth in 2024 is expected to slightly decelerate to ~19%" (KeyBanc / Sapphire, 2024) and the year came in at 15%. Neither landing page discloses sample size or methodology. The quota and attainment figures in the capacity build below come from that 2024 edition — quotas "risen from ~$650K in 2022 to ~$750K in 2023 and 2024," attainment "projected to increase to ~75% in 2024" — vintage inputs to a worked example, not this year's benchmark. Now compare two real filings:

Sources: Snowflake 10-K, HubSpot 10-K. A 21-point spread between two well-run companies is a segment and business-model gap, not a quality gap: enterprise customers on consumption pricing grow inside the account almost automatically, SMB seat-based customers do not and churn far more. Any NRR figure cited without its segment, ACV band and survey population is not a benchmark, it is a number.

Two companions to those ratios. A cohort analysis groups customers by acquisition period and tracks that group forward — the waterfall says churn was $3.4M, the cohort view says it concentrated in customers acquired eighteen months ago on a discount promotion. And logo retention is a customer count, not a dollar-weighted ratio: a company can retain 70% of logos and 105% of dollars at once.

From the ARR walk to the revenue line

The ARR walk is not the revenue plan: the board approves a P&L, and the P&L carries GAAP revenue. Converting one into the other is the deliverable FP&A owns outright, and where an AOP most often fails to tie to its own assumptions. Three steps.

1. Phase the new ARR. SaaS bookings are back-loaded: "typically speaking, 60% of a software company's business is in the second half of the year… Of which, 40% can be in Q4" (Mostly Metrics). A 15 / 25 / 20 / 40 quarterly split of the $12.0M is a reasonable default until you have three years of your own actuals.
2. Apply a provisioning lag. Revenue starts when the service is made available, not at signature. Two to four weeks is common for standard SaaS; longer where implementation is real.
3. Count only live months. Each cohort earns revenue for the fraction of the year it is live.

Component	ARR ($M)	Avg. live fraction of year	In-year revenue ($M)
Beginning base, net of losses timed to the renewal calendar	40.0	~97%	38.7
Expansion ARR	5.2	~40%	2.1
New ARR (15/25/20/40, mid-quarter starts, 3-week lag)	12.0	35%	4.2
In-year recognized revenue	—	—	45.0

So the company exits at $52.0M of ARR and reports $45.0M of revenue. The rule of thumb worth carrying into a meeting: with a Q4-weighted book, in-year revenue from in-year new bookings runs about a third of the new ARR added, rising toward half if bookings are genuinely linear. That is how a company plans 30% ARR growth and 25% revenue growth in one document without either being wrong.

Capacity planning: hiring the sales team backward from the number

This connects the revenue plan to the headcount plan, and it has only a narrow government analog. In WA state caseload drives cost and you staff to serve it; the nearest thing to the math below is a revenue-generating staffing package — Office of Financial Recovery collections staff, DOR compliance and audit FTEs — justified by recoveries per FTE, which wins one decision package but does not size the agency. Here staff drives revenue, so you fix the number and solve backward for the people required to produce it, deterministically, once four assumptions are set: quota, attainment, ramp and attrition. Continue with the $40M company, which needs $12.0M of new ARR — the new-logo line of the walk, not net new ARR, which happens to carry the same value in this plan.

Step 1: convert list quota into productive capacity

A quota is the annual new-ARR target assigned to an account executive (AE). Attainment is actual bookings over quota. Nobody plans at 100%, because most reps do not hit quota: "as a rule of thumb, typically five or six out of ten hit 100%, and seven out of ten achieve 80% or more," and that share falls as the company grows and territories get crowded (Mostly Metrics). Fullcast puts the healthy target at "60% to 70% of reps at or above quota" (Fullcast) — explicitly a share of reps, whereas the KeyBanc survey's ~70-75% attainment is a different, undefined measurement. Take the $750K list quota and 75% planning attainment:

Productive capacity per fully ramped AE = $750K × 75% = $562.5K of new ARR per year.

Step 2: count the capacity you already have

Start the year with 14 fully ramped AEs and assume 25% annual attrition spread evenly — an assumption, not a benchmark, and one of the inputs the build is most sensitive to, so replace it with your own trailing rate. Four departures, each costing on average half a year, is roughly two AE-years lost. Carried capacity is about 12 productive AE-years, or 12 × $562.5K = $6.75M.

Now separate gross hires from net adds, because the requisition plan is what recruiting and comp consume. Those four departures need four backfills before a single incremental rep is added, and a backfill carries the same ramp penalty as a growth hire — so attrition costs twice, once as lost AE-years here and again as ramp on the replacement. A plan expressed only in net adds understates recruiting load, onboarding capacity and in-year comp cost.

Step 3: find the gap, then try to hire into it

The gap is $12.0M − $6.75M = $5.25M, needing 9.3 more productive AE-years. Here ramp — time from start date to full productivity — destroys the naive answer. Typical ramps run "SMB: 2 to 3 months, Commercial / Midmarket: 4 to 6 months, Enterprise: 6 to 9 months" (Mostly Metrics), and Fullcast makes the operational point: "If average ramp is six months but your plan assumes impact in month three, you will miss" (Fullcast). Take six months of ramp producing nothing, hire 12 (four backfills plus eight growth hires, ending at 22 AEs) and try again with twenty:

Hiring scenario	Productive AE-years earned in-year	New ARR delivered
12 hires, 4 each in Jan / Feb / Mar	5.0	$2.81M
20 hires, 7 / 7 / 6 in Jan / Feb / Mar	8.4	$4.73M

That binary treatment — nothing, then everything — is a simplification. Real models use a productivity curve and prorate the rep's ramped quota to the same curve so capacity and commission cost stay consistent: at 0% in months 1-3, 40% in months 4-6, 75% in months 7-9 and 100% thereafter, the 12-hire scenario yields 5.45 AE-years rather than 5.0, worth about $0.25M more in-year ARR. Back-test the curve against your last two cohorts rather than assuming it.

Even twenty reps in Q1 — an unrealistic pace for a team of fourteen, requiring requisitions opened the previous autumn — falls short of the $5.25M gap. That is the central lesson of sales capacity planning: this year's new-ARR number is mostly determined by hires made last year. As a hiring rule, "anytime you hire a rep after month eight of the year, you are basically just adding capacity for the following fiscal year" (Mostly Metrics). When the AOP lands on a number the existing team cannot reach, there are four honest levers, and naming them in the room is your job:

1. Lower the new-ARR target and take the growth rate down.
2. Raise assumed attainment or quota. This is where plans go to die. Planning at 90% attainment when history says 72% does not find capacity, it hides a miss.
3. Shift the burden to expansion. Raising planned NRR from 100% to 110% adds $4M from the existing base, cutting the new-ARR requirement to $8.0M and the gap to $1.25M. Customer success capacity is cheaper per dollar than AE capacity, which is why NRR gets fought over. Apply the skepticism of lever 2: a ten-point raise is a lever only if something concrete changes — a scheduled price increase, a cross-sell product shipping in H1, a renewals team being stood up, a favorable shift in the renewal-date mix. Otherwise it is lever 2 wearing different clothes, and the more dangerous of the two, because expansion carries no requisition anyone can audit and the miss surfaces eleven months later. Ask what trailing-twelve-month NRR is and what accounts for the difference.
4. Accept a two-year build. Hire in Q1 knowing the in-year yield is a down payment on next year's carried capacity, and plan the P&L around that honestly.

Putting a price on lever 4

Start with a consistency check most people skip: quota-to-OTE ratio, fully ramped quota divided by on-target earnings. "As a rule of thumb, this starts around ~5x, and grows over time to ~7x or more," with enterprise reps at 6x to 9x (Mostly Metrics, citing ICONIQ). A $750K quota therefore implies OTE near $150K, not the $250K an enterprise rep elsewhere might carry — and pairing a survey quota with an unrelated OTE is the fastest way to make a sales model look uneconomic for no reason.

Assume $150K OTE split 50/50 between base and variable, and a fully loaded factor of about 1.25 for employer taxes and benefits (see section 5; confirm your company's own factor). Twelve hires starting in January, February and March are 11.0 FTE-years in the plan year, not 12:

• Base: 11.0 × $75K × 1.25 = $1.03M.
• Commission, earned on actual bookings rather than at target: the $75K variable half against a $750K quota is 10% of bookings, so about $0.31M on $3.07M of in-year new ARR.
• In-year cost $1.34M; annualized run-rate cost of those twelve reps, landing in full next year, is 12 × $150K × 1.25 = $2.25M.

Do not stop there and compare $1.34M of cost against $3.07M of new ARR: the ARR persists at run rate into next year's beginning base while the cost does not, and the AE line is only part of what it took. The tool that answers the question is CAC payback: fully loaded sales and marketing cost to acquire the new ARR, divided by that ARR times gross margin, in months. If the AE line is roughly 40% of the S&M this motion requires, call it $3.3M against $3.07M of new ARR at 75% gross margin: $3.3M ÷ ($3.07M × 0.75) = 1.43 years, about 17 months. The 2025 KeyBanc survey has account-executive payback "expected to shorten to 18 months by 2026" (KeyBanc / Sapphire), so 17 months is defensible. Presented that way it is a multi-year investment whose in-year P&L cost is real and whose return sits in next year's beginning ARR — an investment decision, not a headcount line.

Step 4: seasonality, and why it compounds with ramp

The same 15 / 25 / 20 / 40 shape used in the revenue bridge governs capacity, and the two assumptions interact badly: reps hired in Q1 with a six-month ramp become productive right as the heavy quarters arrive, while reps hired in Q3 contribute almost nothing. That is also why quotas are phased rather than divided by four. Seasonality is market-specific — selling to state government or school districts inverts it, which you know from the June 30 scramble.

Step 5: pipeline coverage

Pipeline coverage is qualified open pipeline divided by the target it must cover, measured per selling period, not against a year. Clari states the formula plainly: required coverage = 1 ÷ win rate, with the traditional 3x rule being "a starting point, not a standard" that assumes a 33% win rate (Clari). Fullcast agrees that the 3x rule "falls apart for enterprise teams with lower win rates and longer cycles" (Fullcast); both recommend weighted coverage built from historical stage-by-stage conversion.

Derive it from your own win rate, and be specific about which: dollar-weighted, not opportunity-count, measured from a named CRM stage using RevOps' stage definitions, not yours — the two differ by a factor of two at plenty of companies. At a 22% dollar-weighted win rate from Stage 2, required coverage is 1 ÷ 0.22 = 4.5x, applied quarter by quarter:

Quarter	New-ARR target ($M)	Qualified pipeline needed entering the quarter ($M)
Q1 (15%)	1.8	8.2
Q2 (25%)	3.0	13.6
Q3 (20%)	2.4	10.9
Q4 (40%)	4.8	21.8

Two things follow. The annual figure — $12.0M ÷ 0.22 = $54.5M — is a pipeline creation requirement spread across the year, net of what exists and what carries over, not a balance anyone holds on 1 January; the useful version of that observation is that the demand-generation plan is sized to create $36M when the win rate demands $54.5M. And coverage is a live in-period control that stale deals inflate: Clari recommends discounting or removing deals "aged beyond two times your average sales cycle length," and evaluating enterprise motions with 120-day-plus cycles across a rolling two-quarter window rather than a single quarter.

The rest of the go-to-market headcount

The AE build is the spine, and the supporting roles are where a capacity plan quietly under-resources itself. Ratios to calibrate locally rather than import: business development reps "in the midmarket usually pull double or triple duty supporting multiple reps (2:1 or 3:1)" while "Enterprise and Federal AEs usually get their own BDR," solutions engineers run at "a similar ratio," and about a third of the sales organization ends up as individual quota carriers — drift well below that and there is too much supporting cast (Mostly Metrics). Customer success is sized off ending ARR and account count rather than bookings, which is what makes the expansion lever cost money too, just less per dollar than AE capacity. Quota is set by segment; a blended average is a benchmark input, never a planning input.

Marketing-sourced pipeline and the demand-generation budget

The pipeline requirement splits between what marketing generates and what sales generates by outbound, and the marketing share sizes the demand-generation budget. Practitioners commonly discuss a 30-50% marketing-sourced share for mixed mid-market motions, higher for product-led companies, lower for enterprise outbound. I found no source for these ranges I would put in front of a CFO — the available material is vendor content, not survey research — so use your own trailing twelve months. The chain of arithmetic below is how the marketing budget gets defended in the AOP.

The vocabulary varies by company, and the definitions belong to RevOps rather than marketing: a lead becomes an MQL (marketing qualified lead) when it clears a scoring threshold, an SAL (sales accepted lead) when a rep agrees to work it, an SQL or Stage-2 opportunity when it is qualified into the pipeline, and eventually closed-won. Every conversion rate between those stages must come from your own trailing twelve months, with the sample size stated, or the model is decoration.

Worked, off the coverage numbers above and an average new-logo ACV of $60K:

• $54.5M of Stage-2 pipeline to be created across the year ÷ $60K = about 908 qualified opportunities.
• At a 40% marketing-sourced share, marketing owns roughly 363 of them.
• At a 25% MQL-to-opportunity conversion rate, that is about 1,453 MQLs.
• At a $450 blended cost per MQL, roughly $654K of demand-generation program spend — media, events, content syndication — before any marketing headcount, brand or tooling.

Then the timing. With a five-month average sales cycle, pipeline that closes in Q3 must be created in Q1, so the pipeline-creation plan runs one to two quarters ahead of the bookings plan, and cutting marketing program spend in Q2 to protect a quarter does not damage Q2, it damages Q4. Insist too on the distinction between sourced pipeline (marketing created the opportunity) and influenced pipeline (marketing touched it). They differ by a factor of two or more and get conflated constantly, in the direction that makes the budget look better.

When it is not SaaS

The default frame here is subscription software, but your target roles include marketplaces, usage-priced infrastructure and services businesses. The driver tree changes; the discipline of building the top line separately from operating drivers does not.

Units times price

The base case, closest to what you already do: volume times price, decomposed by product, channel and geography, with price and volume forecast separately so variance can later be split into price, volume and mix effects (see section 8). The work is all in the volume driver — retail traffic and conversion, manufacturing order book, e-commerce sessions.

Marketplace take rate

A marketplace does not own the transaction, it taxes it. GMV (gross merchandise volume) is total value transacted and is emphatically not revenue, since most of it flows to the seller; the take rate is "the fees collected by a third-party service platform" (Wall Street Prep):

Referral fee (revenue) = Take rate × GMV, or for payments, Transaction fees = Take rate × TPV (total payment volume).

Product marketplaces run take rates "between 5% to 25% (but most pay ~15% on average), whereas service-oriented marketplaces are usually priced marginally higher" (Wall Street Prep); payment providers run far thinner. So a marketplace revenue plan is two plans moving independently — a GMV plan from supply and demand drivers, a take-rate plan from pricing and mix. On $600M of GMV at 15%, revenue is $90M; a quarter with GMV up 20% and take rate down to 14% grows revenue about 12%, so presenting only the GMV number gets you corrected in public.

Name the accounting question underneath, because it decides whether the reported top line is $600M or $90M. Reporting gross or net is the ASC 606 principal versus agent determination: an entity is a principal "if it controls the specified good or service before that good or service is transferred to a customer" — indicated by responsibility for fulfillment, inventory risk and pricing discretion — and a principal "recognizes revenues at the gross amount received for the goods and services, while an agent recognizes revenue at the net amount" (RevenueHub, published by the BYU School of Accountancy). The same source notes the guidance "is unlikely to change an entity's net income," but "can significantly impact the top-line revenue and gross profit percentages." That is how a company appears to lose most of its revenue with no change in cash, and why a business-model change that shifts control of the good is a conversation with technical accounting before it is a line in your model.

Usage and consumption pricing

Here the customer commits to spend but revenue recognizes as they consume, which breaks the tidy bookings-to-revenue relationship ratable SaaS enjoys. That is why consumption companies report RPO (remaining performance obligations) so prominently: "the amount of contracted future revenue that has not yet been recognized, including (i) deferred revenue and (ii) non-cancelable contracted amounts that will be invoiced and recognized as revenue in future periods" (Snowflake 10-K).

The forecasting difficulty shows in the same filing. Snowflake held $9.8B of RPO at 31 January 2026 but expected only about 46% to convert within twelve months, "based on historical customer consumption patterns," against a weighted-average remaining contract life of 2.7 years, because "the amount and timing of revenue recognition are generally dependent upon customers' future consumption, which is inherently variable at our customers' discretion." A CRM-and-rep-commit forecast tells you what sold; it cannot tell you what will be consumed, so the revenue forecast runs on product telemetry and drawdown curves rather than sales stages.

Professional services utilization

Revenue = billable headcount × available hours × billable utilization × realized bill rate

Billable utilization is billable hours over total available hours. Practitioners commonly quote 70-80% for delivery staff and 50-60% for partners who also carry business development; the sourcing for those bands is vendor content rather than survey research, so use your own trailing utilization by role. Worked: 40 consultants × 2,000 available hours × 72% utilization × $185 realized rate = $10.66M. Slip utilization to 66% and revenue falls to $9.77M, a $0.9M miss at identical headcount and rate card. Utilization absorbs everything the firm fails to sell and fails to staff, which makes it the first place any problem elsewhere becomes visible. A services line inside a software company is usually run near breakeven and defended as a driver of software attach and retention, not as a profit center.

Revenue recognition FP&A must respect

You will not make revenue recognition decisions — technical accounting owns them, and at a public company the auditor reviews them — but you will build a plan whose revenue line must obey them, and getting this wrong is the fastest way to lose your controller's confidence.

ASC 606 is the US GAAP revenue standard (IFRS 15 is its international twin). Its five-step model, in the words of an audited filing rather than a textbook:

1. Identify the contract with a customer. Both parties have approved it, rights and payment terms are identifiable, the customer has the ability and intent to pay, and the contract has commercial substance.
2. Identify the performance obligations. Promises are separate obligations when "capable of being distinct" — the customer can benefit from the service alone or with readily available resources — and distinct in the context of the contract.
3. Determine the transaction price from the consideration expected in exchange for the services, including estimated variable consideration where significant reversal is not probable.
4. Allocate the transaction price across the performance obligations on a relative standalone selling price (SSP) basis.
5. Recognize revenue when or as each performance obligation is satisfied, that is, when control transfers to the customer.

Source: Snowflake's 10-K revenue policy note. The corresponding HubSpot note applies the same five steps to a ratable model, recognizing software revenue "ratably over the subscription period beginning on the date the online software product is made available to customers." The Codification itself is the authority and it is free: a Basic View "is available to the general public" at asc.fasb.org (Baruch College research guide). A company's own audited policy note is usually more useful, since it shows the standard applied to a specific business model.

Over time or at a point in time

Step 5 hides the fork that matters most to a plan: "when or as" means some revenue spreads and some lands in one quarter. Ratable SaaS spreads — HubSpot recognizes subscriptions over the term from the date the product is made available. Consumption recognizes on usage — Snowflake says plainly that "unlike a subscription-based business model, in which revenue is recognized ratably over the term of the subscription, we generally recognize revenue on consumption." Term licenses, common at infrastructure and data-platform companies, do neither: a large license component is often recognized at delivery with support and maintenance ratable, producing lumpy, quarter-shaped revenue. Establish the recognition pattern by product line before modeling anything, and never assume a new product recognizes the way the flagship does.

Deferred revenue and the reconciliation you will be asked about

Deferred revenue (also contract liability, or unearned revenue) is billings received ahead of delivery. It sits on the balance sheet as a liability and releases to the income statement as service is delivered. HubSpot carried about $1.0B of it at 31 December 2025 and recognized $802.7M of 2025 revenue out of the balance existing at the end of 2024 (HubSpot 10-K). The identity you tie out monthly:

Opening deferred revenue + billings − revenue recognized ± change in contract assets = closing deferred revenue

Worked, for the $40M company: opening $18.0M, billings $51.0M, revenue recognized $45.0M — the in-year figure from the revenue bridge, not the $52.0M exit ARR — closing $24.0M. Two cautions. The two-term version holds only where every invoice passes through deferred revenue; where revenue is recognized ahead of billing, as in ramped multi-year deals and consumption contracts, the offset lands in contract assets instead, which is why Snowflake states its "accounts receivable include billed and unbilled receivables." And most companies have no clean billings figure in the ledger — billings is typically derived as revenue plus the change in deferred revenue, making the identity circular if used as a check rather than a definition. Confirm with your controller how billings is computed before quoting a billings growth rate, and note HubSpot's warning that because billing terms vary, "we do not believe that change in deferred revenue is an accurate indicator of future revenue."

Three consequences for the plan you build

• Multi-year deals do not flatter the plan year. Upfront payment for later delivery cannot be recognized upfront, so a great Q4 for bookings can be an unremarkable Q4 for revenue — and an AOP that assumes otherwise is wrong before it is approved.
• Start dates matter more than close dates. A deal closing 28 December and provisioning 15 January contributes nothing to the closing year, so your model needs a provisioning-lag assumption; sales forecasts almost never carry one.
• Sales commissions are usually capitalized, not expensed as paid. Under ASC 340-40, incremental costs of obtaining a contract are deferred and amortized over the expected benefit period — unless that period would be a year or less, in which case the standard's practical expedient permits expensing as incurred, which plenty of smaller companies on annual contracts elect. The benefit period, not the contract term, drives it: HubSpot's subscriptions are "one year or less" yet it amortizes commissions "on a straight-line basis over a period of approximately two to four years," reasoning from the customer relationship period (HubSpot 10-K). Ask which treatment your company applies before modeling either. Where commissions are capitalized, a sales hiring surge raises cash outflow immediately but hits the P&L over years, so the S&M expense line and the sales cash outflow diverge, sometimes materially. Model both.

Translation: the revenue plan against CFC, ERFC and caseload budgeting

5. Headcount and compensation planning

At a software company, people are the budget. Once you have the revenue plan (section 4), the rest of the annual operating plan is mostly an argument about how many people you hire, when they start, and what they cost. Everything else — software, travel, marketing programs, hosting — is real money but it is not where the argument is. Take the running example used throughout this section: a growth-stage SaaS company entering the year with 250 employees and $40M of ARR, average base salary $140,000 (so $35.0M of base payroll) and a 1.35x cash load factor, giving an average fully loaded cash cost of $189,000 and $47.3M of personnel cash cost. That is commonly two-thirds to four-fifths of all operating expense and cost of revenue at this scale, depending mostly on paid marketing and cloud hosting. (That share is typical practice, not a surveyed benchmark. The $140,000 average assumes a mostly-US roster; a company with substantial offshore engineering runs well below it.) Hence headcount planning gets its own process, system, approval chain, and monthly reconciliation.

Keep those two numbers next to each other, because a 2026 planning conversation starts there: $40M of ARR across 250 people is $160,000 of ARR per employee, and that is thin. For a public comparison you can check yourself, Zscaler's fiscal 2025 10-K reports $2,673.1M of revenue and 7,923 employees, or about $337,000 per head, at 23% growth. Since the 2022 correction, the binding constraint on a headcount plan at most growth-stage and public software companies is not what the departments need. It is an efficiency target — ARR or revenue per employee, a burn multiple, an opex-as-percent-of-revenue glide path, a Rule of 40 number — that the board or the CEO sets before the bottom-up round opens, and that converts into a total headcount envelope and an exit-headcount commitment. The plan below adds 50 net heads; if ARR only grows to $52M, ARR per employee goes to $173,000, and a board will ask why that is enough of an improvement. Expect the same challenge in the form of an AI productivity assumption: boards now routinely ask why a function scales linearly with revenue, and many plans carry an explicit uplift, most often in support and in parts of engineering. Nearly all of those assumptions are asserted rather than measured. Your job is not to referee whether the uplift is real; it is to make it an explicit, named, testable line — which function, what percentage, tested against what — instead of a silent haircut applied to somebody else's ask.

The vocabulary shift you need to make is small but total. In Olympia, the thing that is appropriated is dollars. FTEs ride along: an appropriation is "a legal authorization to make expenditures and incur obligations for specific purposes from a specific account over a specific time period," and the 2025-27 Allotment Instructions define allotments as "a detailed plan of expenditures authorized in the budget, the assumed revenue estimates, and the related FTE estimates required by law (RCW 43.88.110)". The hard edit is on money — "internal agency expenditure allotments cannot exceed the appropriated EA control numbers" — while an FTE mismatch against the EA control number is a warning you correct or explain, and in the interagency-agreement case the same instructions expressly permit an agency to "exceed their allotted FTE control number if they provide an explanation with the allotment packet." In a company you plan approved headcount: a list of specific seats, each with a title, a level, a department, a location, a start month, and a fully loaded dollar cost. Not a ceiling — a list. The unit of authorization is not an hour. It is a requisition.

The position-level build

The headcount plan is a table with one row per seat, not a headcount number by department. This is the single biggest mechanical difference from a state budget build, where FTE is usually a decimal in a program cell. Each row carries roughly: employee name or requisition ID, title, level, department or cost center, manager, location, employment type (FTE or contractor), status (filled, open req, planned), start month, base salary, target bonus or commission, equity grant value, a backfill/net-new flag, and two fields people forget until the auditors ask — expense classification (cost of revenue, R&D, S&M, G&A) and a capitalization percentage for engineering roles. Abacum's finance-led framing of the process is the standard one: start from current employees rather than a blank sheet, tie new roles to business drivers, collect structured requests from department heads, cost each role using salary bands and location, route through finance and leadership for approval, then compare the plan to real hires every month.

Baseline, backfill, net new

Every row is one of three things. Existing filled seats are the baseline — you carry them forward at their current comp, adjusted for the merit and promotion cycle (below). A backfill replaces someone who left; in HR operations the word is sometimes reserved for a temporary cover where the original employee may return, but in FP&A usage it just means a replacement hire that holds total headcount flat. Net new (or incremental) headcount adds to the total. The distinction is load-bearing because backfills are usually pre-approved inside the plan and net-new roles are not: a CEO who says "we are freezing headcount" almost always means net new is frozen and backfills still flow, and a CFO who says "we are down to critical backfills only" means something much harder.

That asymmetry creates the most common quiet leak in a headcount plan: backfill-as-upgrade. A senior analyst leaves at $140,000 and the role is reposted at $190,000 with a level bump, because the manager has wanted the upgrade for a year and a vacancy is the only moment they can get it. Nothing about that is dishonest, and it will not appear anywhere in the plan, because the row is flagged "backfill" and backfills are pre-approved. Set a rule up front: a backfill is a backfill at the same level and the same band; anything above it re-enters as a partial net-new ask for the delta. On a plan with 38 backfills, a $30,000 average creep is $1.1M of run-rate nobody approved.

How the number actually gets set

The process description above is accurate and incomplete. It reads as though department heads submit honest estimates, finance costs them, and an approval chain runs. What actually happens at most growth-stage companies, as common practice rather than published doctrine, is closer to this.

• The CEO and CFO agree a total headcount envelope and a rough departmental split before the bottom-up round closes, derived from the efficiency constraint above. The bottom-up submission is real input — it reveals priorities, gathers justification, and surfaces roles leadership had not thought of — but it is not the decision, and treating it as the decision is the fastest way for a new FP&A director to lose the room.
• Managers submit two to three times what they expect to receive, because they expect to be cut. Arguing role by role against an inflated list is a losing game. The counter is to hold the envelope and make the department choose: here is your number, rank your asks, tell me what does not get done at 60% of the list.
• Requested start dates are bids, not estimates. Hiring managers front-load to January for exactly the reason the phasing table below makes obvious — in-year cost is what gets negotiated, and a January start captures twelve months of it. The slippage haircut exists because of that behavior, not because hiring managers are bad at calendars. Hiring managers know the haircut exists and pull their dates earlier in response, which is why it has to be applied centrally by FP&A and disclosed as a plan assumption, never negotiated row by row.
• Expect a block of unallocated reserve headcount held by the CEO or CFO. It is not a plug; it is how leadership keeps optionality for a role they cannot yet name. Model it explicitly, with an assumed start month, so it does not arrive in Q3 as a surprise.
• Recruiter capacity is often the binding constraint, not budget. A plan for 88 requisitions is roughly 22 hires a quarter; at a working assumption of four to six hires per recruiter per quarter that is four to five recruiters, who are themselves seats in the same plan. A hiring plan that does not scale recruiting with it is not a plan, it is a wish, and it will show up in Q2 as slippage you did not assume.

Keep three states distinct in your own vocabulary, because people use "approved" for all three and then argue past each other: budgeted (the seat is in the AOP), approved (the requisition has been released), and open (the role is posted and being worked).

Driver ratios: how you know an ask is too big

Tying roles to business drivers is the principle everyone states. The instrument is a driver ratio: a number the plan holds constant or deliberately moves, against which a department's ask can be judged without arguing about any individual role. The single most useful question in a headcount review is "what ratio does this hold constant?" A bottom-up ask with no ratio behind it is the one you push back on first.

The cleanest example is sales capacity, and it is not an ask at all — the AE count is an output of the revenue plan. Section 4 builds it end to end and the headcount plan does not get to restate it; what belongs here is the shape of the ratio and what it does to your requisition count. The ratio is productive capacity per fully ramped rep: a $750,000 list quota at 75% planning attainment is $562,500 of new ARR a year, so the $12.0M of new ARR the plan needs is 21.3 productive AE-years, against 14 ramped reps at the start of the year less what 25% sales attrition takes out of them. Section 4 lands that at roughly 12 carried AE-years, a $5.25M gap, and 12 hires — four backfills owed before a single growth rep, ending the year at 22 AEs. The ramp assumption is doing most of the work there, and it is worth measuring rather than assuming: Drivetrain defines ramp-up time as the time it takes new account executives to reach quota and recommends deriving it from your own reps' history.

Two consequences are this section's rather than section 4's. Those four backfills are requisitions, recruiter load, and cash, so a sales plan expressed only in net adds understates all three. And an AE hired in Q4 delivers approximately nothing in the plan year while carrying a full year of cost into the next one, which makes late-year sales hiring a next-year investment that should be argued as one. The binding rule across the two plans: the headcount plan and the revenue plan must use the same start dates, or the company has booked quota it did not hire for.

The supporting ratios follow the same logic — SDRs per AE, sales engineers per AE, ARR or account count per customer success manager, tickets per support agent, span of control for managers (commonly six to eight direct reports), G&A as a percentage of revenue. Treat every one of those as a shape, not a benchmark. They vary enormously by company, motion, and segment, and none of them is a published standard. Your job on day one is to pull the employer's own trailing ratios and plan against those.

Attrition, and why the plan needs it

A headcount plan that ignores attrition overstates cost and understates hiring need. The convention is to assume a departure rate against the existing base and decide, by department, what share gets backfilled. The formula is unremarkable — departures divided by average headcount, where average headcount is beginning plus ending over two. What is contested is the rate. Economy-wide, BLS JOLTS put the July 2026 monthly quits rate at 1.9% and layoffs and discharges at 1.0%, which annualizes to roughly 23% voluntary and 12% involuntary across all industries. (Multiplying a monthly rate by twelve slightly overstates the annual figure, because the denominator is an average employment base rather than a headcount snapshot. Treat it as an order of magnitude.) Software companies typically plan below the national quits rate, often 10-15% voluntary, but that is a company-by-company number set from its own history, not a benchmark you can look up. Use the employer's own trailing twelve months; if they do not have one, say so in the plan assumptions rather than importing a number.

Three refinements separate an attrition assumption that survives review from one that does not.

• Apply it by department, not blended. A 25% sales rate and an 8% G&A rate average to a company number that is wrong for both, and it misplaces the departures and the backfill dollars into the wrong cost centers.
• Split regretted from unregretted. The backfill decision and the urgency differ, and first-year attrition runs far above the tenured rate, so a plan that hires 88 people should expect a visible share of them to leave inside the year.
• Model the vacancy gap, not just the rate. This is where the savings actually live. A plan assuming 15% attrition with immediate 100% backfill saves nothing. A March departure backfilled at 43 days of time to fill plus a notice period leaves roughly two to three months of open salary. FP&A usually shows this as a line called open position savings or a vacancy credit; some companies budget it up front as a negative contra-line rather than letting it accumulate all year as favorable variance that departments then try to spend.

Monthly phasing, time to fill, and start-date slippage

A hire approved in the plan does not cost a full year; it costs the months after it starts. Every row is phased by month, and the phasing assumption is worth more than the salary assumption. Two conventions do the work. The first is time to fill, defined as the calendar days from requisition approval to offer acceptance — distinct from time to hire, which starts when the candidate applies. Workable's data, as summarized by AIHR, puts the US average at 43 days, with information technology at 50 and engineering at 58. Add two to four weeks of notice period at the candidate's current employer and a requisition opened on January 1 produces a start date in late February to mid-March for a typical role, and mid-to-late March for engineering.

Note what time to fill measures: it is an average over requisitions that closed. Requisitions that get cancelled, re-scoped, or simply sit open for six months never enter it. The interval you are actually forecasting is plan-to-desk, which is longer, and that gap is why the slippage haircut sits on top of the time-to-fill assumption rather than instead of it.

The second convention is start-date slippage (also called a hiring lag, hiring haircut, phasing discount, or — the usual phrasing in corporate FP&A, and the one section 13 uses — a vacancy factor): the deliberate assumption that approved requisitions fill later than the hiring manager asked for. Practice splits between pushing every planned start one month later than requested and applying a global attainment factor — budgeting, say, 85% of hires landing on their requested date. Both get to the same place. The time-to-fill data above is the sourced part; the haircut itself is common practice, not a published standard, and I found no practitioner source that names and quantifies it. Treat any specific slippage percentage you hear in an interview as that company's own calibration.

Worked example: 250 people, $40M ARR

The identity that governs the whole plan is: opening headcount, less departures, plus hires, equals exit headcount. Our company enters at 250 and plans to exit at 300. Attrition is assumed at 15% of the opening base, or 38 departures, and the plan backfills all 38. Gross hiring is therefore 50 net new plus 38 backfills = 88 requisitions, and 250 - 38 + 88 = 300. New hires are budgeted at an average fully loaded cash cost of $180,000 — slightly below the $189,000 existing average, because the plan deliberately places part of the engineering hiring outside the US. That gap is itself a planning choice, and you should state it rather than let it look like an error; a plan weighted to senior US engineering would budget new hires above the existing average, not below. Derive that requisition count once, in one place, from the attrition rate and the net-add target: the baseline bridge in section 3.4 and any in-year reforecast (section 7) should quote the same arithmetic rather than rebuild it, or the plan ends up carrying two hiring numbers and no way to say which one recruiting is working against.

Phasing assumption	Avg. months of cost per hire	In-year cost of the hiring plan	Run-rate entering January
All 88 start January 1 (wrong, but this is what a naive plan does)	12.0	$15.84M	$15.84M
Evenly spread, starting the first of each month	6.5	$8.58M	$15.84M
Same, with one month of start-date slippage applied	5.5	$7.26M	$14.52M

Three things to take from this. First, the same 88 approved seats can cost $7.3M or $15.8M in the plan year depending only on phasing — an $8.6M swing on a $40M-ARR company. Second, one month of slippage across the plan is worth $1.32M in-year, which is why the slippage assumption gets argued about in exec review and why FP&A owns it rather than the hiring managers. Third, and most important, the exit run-rate barely moves. Rows one and two both leave the company carrying $15.84M of annualized new payroll into January; row three carries $14.52M, lower only because the last cohort has slipped past December 31 and is not on the payroll at year end. Slippage removes the tail cohort from the exit base. It does not undo the commitment. In-year cost is what you negotiate; run-rate is what you inherit.

Tie it out at the exit: 212 surviving original employees at $189,000 is $40.1M, plus 88 new hires at $180,000 is $15.8M, for $55.9M of personnel cash run-rate on 300 people, or about $186,000 a head — before the merit cycle, which is modeled separately below.

Fully loaded cost

Fully loaded cost is what a seat costs the P&L, not what the offer letter says. You should be able to build it from components rather than quote a multiplier, because the components are what get argued over. AIHR's model, which loads equipment and allocated overhead on top of pay and benefits, reports about 1.56 times base for recurring cost and 1.64 times including one-time first-year expenses. Treat those two figures with care: the published example uses a $50,000 base and a $93,390 total, which is 1.87x, so the page's own arithmetic does not reconcile against its stated multipliers. The useful part of that model is not the number, it is the reminder that equipment and overhead exist and that a cash load factor deliberately excludes them.

The authoritative component data is the BLS Employer Costs for Employee Compensation release. For private industry workers in March 2026, total compensation averaged $46.60 per hour worked, of which wages and salaries were $32.60 (69.9%) and benefits $14.01 (30.1% of total compensation). Expressed the way FP&A needs it — as a percentage of wages rather than of total compensation — the components look like this.

Component (private industry, March 2026)	$ / hour	% of total comp	% of wages
Wages and salaries	32.60	69.9%	100.0%
Paid leave	3.54	7.6%	10.9%
Supplemental pay (bonus, overtime, shift)	1.90	4.1%	5.8%
Insurance	3.62	7.8%	11.1%
Retirement and savings	1.57	3.4%	4.8%
Legally required benefits (FICA, UI, workers' comp)	3.38	7.2%	10.4%
Total compensation	46.60	100.0%	143.0%

Component figures from BLS ECEC Table 4; the percent-of-wages column is arithmetic on those figures. Two adjustments turn this into a headcount-plan load factor. Paid leave comes out — for a salaried employee you pay the salary whether they are at their desk or on vacation, so it is already in base and counting it again double-counts. Supplemental pay comes out of the load factor too, because bonus and commission are modeled explicitly as their own rows (below). What remains — insurance, retirement, and legally required benefits — is about 26% of wages, which is why the common FP&A shorthand for a cash-only load factor is 1.25x to 1.4x base.

One seat, built up

Line	Amount	Basis
Base salary	$150,000	Band midpoint for the level
Target bonus	$15,000	10% of base, accrued monthly at expected attainment
Employer payroll taxes	$13,200	7.65% FICA on $165,000 of cash, plus FUTA and state UI
Health and other insurance	$18,000	Per-head premium, not a percentage
Retirement match	$6,000	4% of base
Fully loaded cash cost	$202,200	1.35x base
Stock-based compensation	$50,000	Annualized grant-date fair value, non-cash
Total P&L cost	$252,200	1.68x base

Two notes on the payroll-tax line, because it is the row a CFO or a comp analyst will poke first. It sits below the BLS 10.4%-of-wages figure not because of the Social Security cap — the 2026 wage base is $184,500, so all $165,000 of this seat's cash is taxed at the full 6.2% plus 1.45% employer rate — but because BLS's legally-required-benefits average is taken across all private workers at about $67,800 of annual wages ($32.60 x 2,080), where unemployment insurance and workers' compensation, both capped or rate-based, are a much larger share of pay. The cap does start to bite above $184,500 of cash compensation, where the employer's marginal rate drops from 7.65% to 1.45%; on an executive plan that is a real modeling difference, and on a plan full of $150,000 seats it is not.

For scale, Pave's dataset puts the median US employee at $150,000 of total target cash and $57,000 of unvested equity value, about $207,000 combined (published October 2024, updated August 2026), and the median non-US employee at about $133,000. That geographic gap is why location is a required field on every plan row and why "we will hire this role in Bengaluru or Warsaw instead" is a real budget lever in exec review rather than a rhetorical one. It is also not a fringe strategy: Zscaler disclosed that approximately 63% of its full-time employees were located outside the United States as of July 31, 2025.

Cost to hire

The build above prices a seated employee. It does not price the act of hiring, and a plan with 88 requisitions has to. The lines are: external agency fees, which as common practice run roughly 20-25% of first-year base and get used on some share of senior and hard-to-fill roles; sign-on bonuses, near-universal at senior levels in tech and paid as cash in the month of start; referral bonuses; relocation where applicable; and a per-head equipment and software allowance. On our plan, assume agencies on 15% of the 88 requisitions at 22.5% of a $140,000 base ($416,000), a sign-on bonus averaging $15,000 on 30% of hires ($396,000), and $4,000 of laptop and software seats per head ($352,000). That is about $1.16M — a seven-figure line a naive plan omits entirely. Those percentages are common practice, not surveyed benchmarks; get the employer's own agency usage and average sign-on from recruiting. Note also where the money sits: recruiting cost usually lands in G&A or in a People cost center, not in the hiring department's budget, so the hiring manager never sees it and finance has to be the one to raise it.

Where a headcount dollar lands

"Department or cost center" implies but does not equal expense classification, and the difference decides numbers the CFO reports externally. Support engineers, professional services, and the infrastructure staff who run the production platform typically sit in cost of revenue, while the rest of engineering sits in R&D — and the same manager can own both. That means a support hiring plan is a gross margin decision and belongs in the margin bridge, not only in the opex bridge. FP&A owns the classification convention, because auditors and the board expect it held constant year over year; changing it mid-year restates a metric the market watches.

The second field is capitalized software development labor. Costs incurred during the application development stage of internal-use software can be capitalized rather than expensed. Zscaler's policy is a clean, public statement of the mechanics: "We capitalize certain costs incurred during the application development stage... Costs related to preliminary project activities and post-implementation activities are expensed as incurred... Capitalized internal-use software is amortized on a straight-line basis over its estimated useful life, which is generally three to five years, and is recorded as cost of revenue" — and their fiscal 2025 capitalization, inclusive of stock-based compensation, was $124.5M. For planning purposes it is set as a per-person or per-project percentage during the headcount build. The consequences are worth internalizing: capitalizing reduces current-year R&D opex, creates amortization that comes back in later years, and often lands in COGS rather than opex, so an engineering department can appear under budget with nothing operationally different. It is also a favorite target in diligence. Details of the expense categories are in section 6.

Variable comp: bonus and commission

Bonus and commission come out of the load factor because they behave differently from salary, not because they are small. Get them wrong and you produce the most predictable variance surprise of a first year.

Corporate bonus is accrued monthly, but accruing at 100% of target is the naive version. The convention is to accrue at an expected corporate attainment factor, reset each quarter as the year's revenue and EBITDA picture firms up. Worked example: our company carries a $2.0M bonus pool and accrues at 100% of target through Q3, so $1.5M sits in the accrual. In Q4 the compensation committee lands corporate attainment at 60%, making the correct full-year expense $1.2M. Instead of the $500,000 of Q4 accrual the plan assumed, you book a $300,000 credit — an $800,000 swing in one month, spread across every department line, showing up in the close package as a wave of favorable variances. None of them are operational. It is a prior-period accrual correction, and if you do not say so in the variance commentary (section 8), somebody will spend it.

Sales commission is not really a headcount line at all — it is a rate applied to the revenue plan's new ACV or bookings, which means it is owned jointly with the revenue plan (section 4) and it self-funds part of a revenue miss. Three complications. Accelerators above quota make the expense convex, so a plan that lands at 115% of bookings costs more than 115% of the commission budget. Ramping reps usually get a guaranteed draw, which is real cash paid against no productivity during the ramp months, and it belongs in the plan next to the ramp assumption. And the accounting diverges from the cash: incremental costs of obtaining a contract are capitalized and amortized rather than expensed when paid. Zscaler states the practice plainly — "We capitalize our sales commissions and associated payroll taxes that are incremental to the acquisition of customer contracts and recognize them as expenses over the estimated period of benefit," using an estimated period of benefit of five years for initial contracts, with renewal commissions amortized over the renewal term. The practical consequence: the commission number in the S&M line is not the commission checks that went out, the difference sits on the balance sheet as a deferred contract acquisition cost, and if you forecast the expense line off the commission plan you will be wrong every month. Forecast the cash and the amortization separately, and get the period-of-benefit assumption from the controller.

Forecasting stock-based compensation

Stock-based compensation is a real GAAP operating expense, allocated to the department where the employee sits, so it inflates every departmental opex line you manage — and it is not cash, so it does not touch the burn or runway model. It is also, at software companies, the dominant non-GAAP add-back, and the size of it surprises people. In fiscal 2025 Zscaler bridged a GAAP loss from operations of $128.5M to non-GAAP income from operations of $580.1M, of which $685.5M was stock-based compensation and related payroll taxes against $16.8M of intangibles amortization, $4.9M of restructuring, and $1.3M of acquisition costs. SBC was roughly 97% of the add-back.

The forecast is a layered build, not a per-head percentage:

1. Opening expense from unvested grants already in flight — prior-year new-hire grants and refreshes that keep amortizing whether or not you hire anyone this year.
2. New-hire grants, amortized from start date over the vesting period. A four-year grant with a one-year cliff still expenses straight-line from the start date, so a $200,000 grant is $50,000 a year at steady state but only about $25,000 in a year the person starts at mid-year. That is the same phasing arithmetic as cash, applied to a non-cash line.
3. The annual refresh pool granted at the comp cycle. At most established tech companies this is a larger planned number than the cash merit pool and is the main year-over-year driver of SBC growth. You will be asked to size it.
4. A forfeiture assumption, trued up when actual forfeitures differ.

For the reader targeting private growth-stage companies, the exception matters more than the rule. Where RSUs are double-trigger — vesting on both a service condition and a liquidity event, which is standard at late-stage private companies — no expense is recognized until the liquidity trigger, and the company then takes a large catch-up charge at IPO. GAAP opex at a private company therefore understates the true cost of a seat, and the constraints that actually bind are the option pool percentage and the annual dilution burn rate, not the P&L line. FP&A usually owns the SBC forecast jointly with stock administration, who hold the grant ledger. Sourcing caveat: the double-trigger and refresh-pool mechanics above are standard practice rather than something I could source live; Carta publishes the underlying data and blocked automated access.

Practical consequences for how you report: never mix SBC into a load factor you also use for cash forecasting; expect to publish departmental opex twice, GAAP and "cash opex excluding SBC"; and expect the CFO to care much more about the second one. The mechanics of the add-back and the SEC's rules on non-GAAP measures are in section 9.

Bands, merit, and promotion

A salary band (or pay band) is the range a company will pay for a level in a job family and a location, defined by three points: a minimum, a midpoint set at the market target, and a maximum, with broadbanded structures running as wide as 75% to 150% of the median. Most companies target the 50th percentile of market for total compensation, with deliberate variation by element — a company can pay cash at the median and equity at the 75th, and many do, because equity distributions are much wider and more right-skewed than cash distributions. The gap widens with seniority: Pave finds the spread from the 50th to the 90th percentile of new-hire equity is 131% at entry level and 283% at senior level, while the base salary spread narrows from 45% to 24%. The term you will hear for where an individual sits inside their band is compa-ratio: salary divided by band midpoint, so 0.90 means paid 10% below midpoint. FP&A does not own bands — compensation does — but you will be asked to cost a band change, and the cost of moving a band is the number of people below the new minimum times the gap, annualized.

The merit cycle (also comp cycle or comp review) is the annual event where performance ratings turn into pay changes. Most companies run it once a year on either the calendar or fiscal year, and a lighter mid-year cycle for promotions is not uncommon. The sequence is: set objectives, benchmark roles against market, calibrate performance ratings, model merit and promotion scenarios, align the total rewards package, route through managers, HR, finance, and executives, then communicate and audit — finance approves the overall budget and monitors spend.

How FP&A budgets it

The merit budget is expressed as a percentage of payroll and split into separate pools. Payscale's 2026-2027 survey has US employers planning 3.5% salary increases for 2027, against 3.4% actual in 2026, with government at 3.0% and business services at 4.5%. Pave's survey of about 100 total rewards leaders at large US tech companies, asked about their plans for the 2025 cycle, splits the pools: a 3.5% median merit budget plus a 1.0% promotion budget against a 5.0% median overall salary budget, with a planned promotion rate of 15%. In 2024 actuals from the same source, 14.9% of employees were promoted at a median increase of 10.5%, 69% of employees received a raise at all, and the median non-promotion raise was 4%. (The article was published in November 2024 and updated in August 2026; read the year labels before you quote any of it in an interview.)

Two or three pools is the usual structure, and the ones people forget are the ones that break the plan. Besides merit and promotion, expect a market adjustment pool for people who fall below a re-benchmarked band, and a parallel equity refresh pool running on the same calendar. Pave notes that comp teams "often use merit cycles to make market adjustments to situate employees within their compensation bands" and that equity refreshers are sometimes included in merit cycles. Pool percentages also commonly differ by geography, which means one blended number applied to global payroll is wrong in both directions.

The modeling point is the same annualization problem as hiring. On our 250-person company with $35M of base payroll, a 3.5% merit pool plus a 1.0% promotion pool is $1.575M annualized. If increases take effect April 1, the in-year cost is nine months of that, $1.18M, and the remaining $394k lands in next year's baseline before anyone has been hired. Getting this wrong — budgeting the annualized number in-year, or the in-year number in the run-rate — is one of the most common errors in a first AOP.

It is also worth knowing, before you are surprised by it, that merit is a discretionary pool sized in the AOP and re-decided in Q1. When revenue misses, it is one of the first levers pulled: deferred a quarter, halved, or converted to a promotion-only cycle. Deferring the effective date from April 1 to July 1 quietly recovers about a quarter of the pool. Model the effective date as an assumption with a sensitivity, not as a fact.

Contractors and the flex layer

Contractors, agency staff, and outsourced teams sit outside the headcount plan's FTE count but inside the same department's opex. Finance cares for three reasons: they are faster to add and remove than employees, they carry no benefits load or severance exposure, and they are the standard workaround during a headcount freeze — so a freeze that counts only FTEs leaks spend through the contractor line unless someone watches it. Expect a separate contractor line by department and a policy on whether it counts against the headcount budget.

Two operational details that trip people up. First, contractors usually sit in a separate GL account outside payroll and are excluded from headcount metrics, which means any headcount-derived efficiency number — ARR per employee above all — needs a stated convention on whether contractors are in or out, applied consistently every period or the trend means nothing. Second, the "hire this role in Warsaw instead" lever usually runs through an employer of record when the company has no entity in that country: a third party employs the person locally and bills the company, adding a per-head fee on top of local salary and statutory burden. Cost the lever at the EOR-inclusive figure, not the raw local salary, and know that at some headcount per country it becomes cheaper to establish an entity — which is a legal and tax project with its own timeline, not a budget line.

The classification test is the IRS's, a facts-and-circumstances weighing across behavioral control (does the company control what the worker does and how), financial control (payment, expenses, tools), and type of relationship (contracts, benefits, permanence, whether the work is a key aspect of the business). The IRS is explicit that there is no "magic" or set number of factors. FP&A does not make the call, but you should recognize the risk pattern — a long-tenured "contractor" sitting on a team, working set hours, using company equipment — and route it, because a reclassification is a retroactive payroll-tax liability, not a budget variance.

Modeling a reduction in force

Cost reduction runs an escalating ladder: hiring freeze, then internal redeployment, then reduction in force. A hiring freeze is a temporary halt on recruitment where open positions go unfilled and no new roles are created, though exceptions may be granted for key positions, and typically runs three to six months. Freezes are the cheapest lever because they work through attrition: on our 250-person company at a 15% attrition rate, departures run about 3.1 people a month, so a full freeze retires roughly $590,000 of annualized run-rate for every month it stays in force — about $1.8M after a full quarter — with no severance cost and no announcement.

That is the gross arithmetic, and the realized number is always smaller. Some of those departures are in roles the company must refill within weeks regardless, the exception list starts filling the day the freeze is announced, and demand for the frozen work reappears in the contractor line. Model the freeze at well under half the arithmetic unless the company has actually held one before and you can look up what it delivered.

When a RIF is on the table, FP&A models three separate things and must not conflate any two of them: the P&L charge, the cash outflow, and the run-rate savings. They have three different sizes and three different timings, and the CFO, the controller, and the board each ask for a different one. The example below cuts 25 seats at the $150,000-base level priced earlier, so each removes $202,200 of fully loaded cash.

The PTO row is the one people add to the charge by reflex. If the vacation liability is already accrued on the balance sheet, paying it out is cash leaving and a liability being relieved, not incremental expense — so the P&L charge here is $784,000 while the cash is $1.0M. With a March 1 separation date, in-year cash savings are ten months of $5.055M, or $4.21M, against $1.0M of cash out, for about $3.21M of net cash in the plan year — and $5.055M of run-rate carried into the following year. Executives usually ask for the second number and are shown the first. The SBC reversal reduces GAAP opex without producing a dollar of cash, so it belongs in the GAAP bridge and nowhere near the runway model.

One timing caveat, stated more carefully than most guides state it. The charge is usually recognized when the plan is communicated to employees, but not always: where employees are required to keep working through a notice period, the benefit is generally recognized over that service period instead of all at once, and contractual severance or an established severance practice follows different criteria again. I could not reach a citable text of the recognition standards, so treat this as a flag rather than a rule: get the recognition date from the controller before you put a restructuring charge in a forecast. It routinely straddles a quarter end, which is exactly when it matters.

Notice law: the federal floor and the state law that usually binds first

The federal WARN Act sets the floor, not the answer. Under 29 U.S.C. § 2101 an employer is covered at 100 or more employees excluding part-timers (or 100 or more who together work at least 4,000 hours a week); a plant closing is a shutdown of a single site of employment causing employment loss for 50 or more employees, excluding part-timers, in any 30-day period; and a mass layoff is a reduction in force at a single site of employment affecting at least 33% of employees and at least 50 employees, or at least 500 employees regardless of percentage. Covered employers must give 60 days' notice. Two qualifiers change who is covered and are usually the first thing counsel checks: every count is measured at a single site of employment, and part-time employees — averaging under 20 hours a week, or employed fewer than six of the preceding twelve months — are excluded. A distributed or heavily remote company can cut a large absolute number of people and fall outside the federal test entirely.

State mini-WARN statutes frequently bind first, and Washington's is a live example the reader should model before the federal one. Chapter 49.45 RCW, enacted as SB 5525, chapter 277, Laws of 2025, effective July 27, 2025, defines an employer as "a person who employs 50 or more employees in this state, excluding part-time employees" and a mass layoff as "a reduction in employment force that is not the result of a business closing and results in an employment loss during any 30-day period of 50 or more employees, excluding part-time employees." There is no percentage test. Notice is 60 days, in writing, to the affected employees or their union and to the Employment Security Department. Run that against our 250-person company: the 25-head RIF above triggers neither statute. A 50-head RIF is 20% of the workforce, so it triggers nothing federally, and it triggers a full 60 days of Washington notice. A reader who modeled that cut off the federal thresholds alone would put the savings start date two months early. (Washington's statute expressly excludes "the state or any political subdivision thereof," so it is a private-employer rule; California, New York, New Jersey, and Illinois have their own.) For FP&A the practical effect is timing: a notice-triggering action means 60 days of full payroll after the announcement, so the run-rate savings start two months after the charge, and the charge itself may spread across that window.

Governance: requisitions, freezes, and the HRIS tie-out

A requisition ("req") is the formal request to open a specific role — a formal request to create a new position or fill a vacant one. The approval chain AIHR documents is the one you will see, with finance in the middle of it: hiring manager submits, HR reviews for completeness, the department head evaluates need and budget impact, finance or a budget committee evaluates the financial ask, executives sign off on senior roles, HR assigns a requisition number, the job description is built, and the posting goes live. The budget gate is explicit: if budget is available the req moves quickly, and if it is not, there is a harder conversation about whether the role is needed at all.

Being in the approved plan is necessary but not sufficient: the req is a second approval, taken at hiring time against current conditions. Resist mapping that onto appropriation versus allotment — an allotment is "an agency's plan of estimated expenditures, revenues, cash disbursements, and cash receipts for each month of the biennium," filed once and amended quarterly, not a per-hire gate. The closer analog is the freeze you are living in: a position can be fully funded in the enacted budget and still not fillable without an agency-head exception logged to OFM. Same structure, private version. The plan says the seat exists; the req says you may open it today. The broader authority framework is in section 11.

The monthly headcount reconciliation

Headcount reconciliation is the recurring tie-out between the plan and reality, described by Drivetrain as tracking and aligning a company's headcount with actual hiring and staffing numbers, run at month-end or quarterly, across the HRIS or HCM as the employee system of record, payroll, finance systems, and the spreadsheets in between. The causes of variance they list are the ones you will actually chase: delayed start dates versus plan, unexpected departures, transfers between departments, promotions changing cost allocation, lag between system updates, manual entry errors, and inconsistent categorization across departments.

In practice it is a three-way tie: approved headcount in the planning system, active employees in the HRIS, payroll dollars in the general ledger. They never agree on the first pass, and the job is to classify every difference rather than force the number.

Difference type	What it means	Action
In plan, no req opened	The hiring manager has not started	Push the start date; release the in-year dollars to the forecast
Req open past plan start date	Slippage is real, not assumed	Re-phase; check whether the slippage assumption is too optimistic
In HRIS, not in plan	Usually a level or band upgrade on a "backfill"; sometimes a transfer in or a genuinely unbudgeted hire	Compare title and band to the row it replaced; re-enter the delta as a net-new ask, or escalate
In plan, terminated in HRIS	Departure not yet reflected	Decide backfill or not; that is a real budget decision, not a data fix
Department mismatch	Transfer or reorg not mirrored in the plan	Re-map the cost center; restate prior months if material
Payroll dollars without a plan row	Contractor coded to payroll, or a timing lag	Reclassify; check the freeze is not leaking into the contractor line

Do this monthly and headcount variance explanations in the close package (section 8) write themselves. Do it quarterly and you will spend the quarter arguing about whose number is right.

Public-sector translation: FTE authority and workforce planning

6. Operating expenses, COGS, and capex

Fully loaded headcount — salary, bonus, commissions, benefits, payroll taxes, equity — is 70–80% of a software company's cost base, and section 5 covered it. The definition moves that number fifteen points: base salary alone is closer to 57% of operating expenses, per Lighter Capital across 83 private B2B SaaS companies between $250K and $22M of ARR (mean 57.7%, median 57.4%). Ask which definition a quoted percentage uses.

This section is everything else: the coding grid, the non-headcount build and its politics, how shared costs reach consuming teams, what sits in cost of revenue, and the capitalize-or-expense decision that moves reported margin without moving cash. The mechanics you know. The grid, the vocabulary, and gross margin you do not.

6.1 The cost-center grid and the chart of accounts

Every transaction is coded on several independent dimensions at once. Two matter constantly:

• GL account (the "natural" account) — what was bought: salaries, employer taxes, software, travel, legal fees, hosting. The chart of accounts proper.
• Cost center (department) — who bought it and whose budget it consumes.

Others in common use: legal entity, region, project, product line, campaign. One AWS invoice line carries account 5100 (Hosting), cost center 3200 (Platform Engineering), entity 001, product PLT. Reporting is then a question of which dimensions you sum across.

Department taxonomy

Public SaaS income statements are near-identical in shape. Datadog's FY2025 statement of operations runs Revenue, Cost of revenue, Gross profit, then Research and development, Sales and marketing, and General and administrative to operating income. That is the whole external taxonomy, and your internal cost-center hierarchy has to roll up into it.

• COGS (cost of revenue) — the cost of delivering to customers who already pay: hosting, the engineers keeping production up, support, third-party fees, implementation services.
• R&D — product, design, engineering, building what does not exist yet. The largest line at engineering-heavy companies; at the private median it runs level with sales and marketing combined. Which is biggest tells you how the company competes.
• S&M — sales, sales engineering, marketing, and often customer success when CS is treated as growth rather than delivery.
• G&A — finance, legal, HR, IT, executive, plus audit, D&O insurance, and the corporate lease.

Two benchmarks, because interviewers use them as shorthand. SaaS Capital's 2026 survey of more than 1,000 private B2B SaaS companies, fielded in March 2026, gives medians as a percent of ARR. Datadog is public hypergrowth at roughly $3.4B of revenue. The difference between them is the lesson.

Three corrections you will hear. The columns are not measured alike: Datadog is GAAP and carries $751M of stock-based compensation in FY2025 — $469.5M in R&D, $156.5M in S&M, $94.9M in G&A, $29.7M in cost of revenue, 21.9% of revenue, while the private survey is cash spend with effectively none. Ex-SBC is the comparable column, and it collapses the R&D gap from 23 points to 9. SaaS Capital's denominator is ARR, not revenue, so even the percentage columns share no base. And the private figures are medians of separate distributions rather than a company: they sum to 86% of ARR, so never subtract the 17% COGS median from 100 and call it a gross margin. Use the total-revenue figure in 6.4.

Scale explains the rest. G&A halves from the private median to a $3.4B public company, and falls two thirds ex-SBC, because audit fees, a CFO, and an HR system cost about the same at $40M of revenue as at $400M. R&D still runs nine points higher on a comparable basis, because a company competing on product velocity spends there — into a GAAP operating loss of $44.4M. Datadog manages to a non-GAAP margin, which is why the SBC distinction is not a technicality.

Who owns a cost center

The department head, who answers for the variance in the monthly business review. FP&A owns the process: builds the model, loads the plan, produces the variance report, chases commentary. Your standing comes from telling a VP what their own numbers mean, not from controlling their money, and the FinOps Foundation frames budgeting the same way, as a process that "ensures accountability from each budgeted cost center." Hold it at the right altitude, though: the envelope is set top-down from a burn or margin target before any VP is asked what they need, Finance holds the pen, and Finance sits in the approval chain on material purchases. Department heads own allocation within an envelope they did not set. Say the strong version to a VP whose number was just cut 12% and you will be corrected.

6.2 Building and controlling non-headcount opex

Non-headcount opex is a fifth to two fifths of the cost base depending on the definition above, and gets a fraction of the planning attention, which is why the surprises live here. Split each department's non-comp spend three ways, because each is built differently.

1. Committed and contracted — signed agreements with known amounts and renewal dates: software, the office lease, insurance, agency retainers, audit fees. Built from a contract register, not last year's actuals.
2. Driver-based variable — spend that scales with something already planned: hosting with usage, payment processing with transactions, commissions with bookings, recruiting fees with the hiring plan. A rate times a driver.
3. Discretionary program — the envelope the department head allocates: events, content, brand, T&E, training. Top-down as a number, detailed bottom-up inside it.

The baseline-versus-new-ask split from section 3 gets computed at line level here. Bucket 1 held flat is your carry-forward analog. Bucket 1 with escalators applied, plus bucket 2 re-run at next year's volumes, is maintenance level: same services, more of them, at higher prices. That is OFM's own split, where ML "reflects the cost of mandatory caseload, enrollment, inflation, and other legally unavoidable costs not contemplated in the current budget" and explicitly names current lease and contract payment increases. Bucket 3 above last year is the policy-level equivalent. Nobody uses those words, and nobody hands you the flat-continuation number the way OFM does, where "Agencies do not recalculate CFL for their budget requests" because the system arrives pre-populated. You build the baseline yourself, and keeping the flat number visible underneath the volume-and-price adjustments is a discipline you impose or lose.

Worked example: the Marketing non-headcount build

A $40M ARR company on a calendar fiscal year, planning against a $12M new-ARR target, marketing accountable for sourcing 40% of the qualified opportunities behind it.

Two details do real work. The renewal date: the 8% uplift applies for ten months of twelve, which is why a contract register carries dates and not just annual values. And the vocabulary. Marketing will call the paid-media assumption pipeline coverage — open pipeline dollars over the target they have to cover — and someone in the room will quote 3–4x as the standard. It is not one. Section 4 derives required coverage as 1 ÷ win rate, which at this company's 22% dollar-weighted rate is 4.5x, and the $654K above falls out of that chain rather than out of a rule of thumb. Coverage is a dollar ratio. The build underneath it uses a win rate, a conversion rate. Related, not interchangeable, and a plan needs both stated separately, because a coverage shortfall and a win-rate shortfall have different fixes.

Paid media matters in review because it is the only line with a defensible answer to "what happens if we cut it?" Cutting $200K at $1,800 per opportunity removes 111 opportunities: at a $60K deal size, $6.7M of pipeline, and at the 22% win rate about 24 wins, or $1.5M of ARR — 12% of the new-business plan. Expect the pushback, because it is correct. The last dollar of paid media performs worse than the average dollar, so cutting 30% of budget removes fewer than 30% of opportunities. Nobody has the marginal cost-per-opportunity curve, so you agree a step function — the first $100K of cuts at $2,100 per opportunity, the next $100K at $2,600 — and write it down. Brand and content lag two to four quarters and will not show in the cut quarter at all, which is why they go first. Committed lines cannot be cut inside the plan year without breaking a contract.

How the numbers you receive are actually built

Every bottom-up non-comp submission carries a pad, commonly 10–20%, mostly hidden in bucket 3. That is rational rather than dishonest: a haircut is coming, and a VP who submits their true need gets cut below it. Finance knows and sets the envelope accordingly, which guarantees the pad next year. The counter is the three-bucket split: bucket 1 is verifiable against signed contracts, bucket 2 against a driver planned elsewhere, and neither hides a pad. That pushes the negotiation into bucket 3 and turns "your marketing budget is too high" into "your events envelope is up 40% on flat headcount, what changed?" Two behaviors to plan for rather than discover: fourth-quarter spend-down is real, so publish in advance whether unspent budget carries over, and contract signature timing gets negotiated near quarter end to land expense in the period that needs it, so track commitments alongside expense.

Software spend and control after the plan locks

Software is the fastest-growing non-headcount category and the one nobody owns end to end. What works: a contract register listing every vendor, annual value, renewal date, auto-renew status and notice period; a named owner per contract; and a rule routing renewals above a threshold through Finance before the notice window closes. Renewal uplift, commonly 5–10%, belongs in the plan explicitly rather than as a surprise in month seven. T&E and marketing programs are best built from drivers (T&E per traveler per month by department, times the headcount plan) rather than trended forward, but be warned that the planning process here is thinly sourced in public: the best published example is a decade old, Mondelez modeling T&E by trips, travelers, and trip characteristics under a zero-based approach, reported by AFP in 2016. Do not assume the company you join has a documented process. Many do not.

After the plan locks, control runs through purchase orders and a delegation of authority matrix (section 11) plus softer guardrails. The FinOps Foundation's governance framing is the clearest published version and generalizes past cloud: policies are "Guidelines," which "are advisory rather than mandatory," and "Guardrails," meaning "formal processes, architectural controls, and structural constraints that define mandatory pathways" for compliant action. Its worked control is worth stealing: teams are notified monthly of resources at zero utilization, and those resources "will be decommissioned automatically the following Tuesday unless a documented retention justification is submitted." Default-off with an opt-out beats a quarterly license audit nobody runs.

The month-end loop is where non-headcount opex is actually controlled: purchase order, receipt, accrual, invoice, payment. FP&A's job at close is the accrual — for every open PO where the service was delivered but no invoice arrived, book the cost in the month it was consumed. Skip it and monthly opex sawtooths on invoice timing rather than activity, and cost-center owners stop trusting the variance report. The mirror case is prepaids: annual software contracts are usually paid up front and amortized monthly, so cash and expense on one contract diverge by up to twelve months. Keep the register with two columns, annual cash and monthly expense, or your opex and cash forecasts will disagree and you will not know which is wrong.

On tolerance, most companies set a dual threshold for mandatory commentary. Numeric's practitioner guidance is representative — "materiality thresholds of 5 percent to 10 percent per line item, paired with a fixed-dollar floor", its example flagging anything over 7% or $25,000. Set the floor for fifteen to twenty-five explanations a month, not two hundred. Cloud is the exception, looser because it is genuinely volatile: FinOps publishes 20% maximum budget-to-actual variance on cloud spend at the least mature stage, 15% at the middle, 12% at the most mature — the noise practitioners accept in one category and nowhere else. Say in an AOP kickoff that a 15% opex variance is within tolerance and you will not be asked again. The same page names holdback management, a reserve "not allocated to a specific purpose but available to a budget owner to account for changes, overages" and forecasting errors: the unallocated pool from section 3.

6.3 Allocations, shared services, and chargebacks

Some costs are incurred centrally and consumed by everyone: the AWS bill, the corporate lease, the IT and security stack, the finance team. Three ways to handle them, and the choice is a management decision rather than an accounting one.

• Leave them central. The cost sits in G&A or a shared cost center and nobody else sees it. Correct for genuinely company-wide costs like audit and D&O insurance.
• Showback. Consuming teams are shown their share; the cost does not move. FinOps treats showback as always required and notes the difference from chargeback is "the formality described in this capability of sending expenses to official accounting budgets."
• Chargeback. The cost is journaled onto the consuming cost center and hits that team's variance. Real teeth, real arguments, real overhead: you need a methodology you can defend monthly.

Choose by controllability: can the receiving team change its consumption? Cloud spend a team provisions is controllable, so chargeback changes behavior. The audit fee and the corporate lease are not controllable below the CFO, and charging them out generates argument without behavior change. Where you do allocate, report each cost center in two blocks — direct cost, controllable and scored, and allocated cost, shown but not scored. Say that in the first business review or re-litigate it for a year. The basis is political the same way: headcount penalizes people-heavy teams, revenue penalizes the teams that are working, and someone will argue principle for whichever you did not pick. Three methods are in general use, per the framework's allocation capability: fixed allocations (predetermined percentages), proportional distribution (by relative spend or usage), and proxy metrics (headcount, seats, revenue, square feet, when direct usage is unmeasurable). Its worked example distributes a $200,000 monthly shared pool across a $1.2M spend base, sending $75,000 to Marketing, $50,000 to Commercial, and $37,500 each to Finance and Logistics.

Worked example: allocating a cloud bill, and why it moves gross margin

A $200K monthly cloud bill. Tagging attributes $150K directly; $50K is shared plumbing (networking, the control plane, shared observability, security tooling) that cannot be tagged to a workload. Allocate the untagged $50K proportionally to the tagged base.

The last column is the one to internalize. One invoice splits across two income-statement lines, and the split is a tagging decision made by engineers, not an entry made by Finance. If dev and test get mis-tagged into production, $20K a month moves from R&D into COGS: on $42M of revenue, $240K a year, 0.6 points of gross margin. Small enough to miss, large enough that a diligence team will find it. Which is why allocation is described as assigning cost "using accounts, tags, labels, and other metadata, creating accountability among teams" — the tagging discipline is the accounting.

The other half of hosting cost: the rate

Hosting is usage times a rate, and Finance owns the rate. Commitment-based instruments cut it substantially: AWS Savings Plans provide savings beyond on-demand rates in exchange for "a commitment of using a specified amount of compute power (measured per hour) for a one or three year period," advertised at up to 72% off. The FinOps framework groups these under rate optimization and names the family: reserved instances and committed use discounts (resource-based), savings plans (spend-based, lower discount, more flexibility), spot instances, and above all of them a negotiated enterprise agreement on total spend. Two planning consequences. The hosting line needs a blended rate and a coverage assumption — what share of usage sits under commitment — not a single unit cost. And a commitment is a contractual minimum, so if growth undershoots you pay for capacity nobody used and the shortfall lands as a COGS variance no cost-center owner can explain. Model the commit as bucket 1 and on-demand overage as bucket 2, and on arrival ask what is committed, at what coverage, and when it expires.

Management view versus legal-entity view

Two parallel sets of books over the same transactions. The management view is how the business is run and measured: by product, region, and cost center, at plan or constant currency so FX noise does not obscure operating performance, with intercompany activity treated as ordinary internal cost movement. The legal-entity view is what gets audited and filed: transactions by incorporated entity, intercompany balances eliminated on consolidation, actual GAAP translation rates. For a US-only single-entity company they are identical. Add a UK subsidiary whose engineers serve US customers and they diverge: the intercompany service charge is real revenue and real cost in each entity's books and nets to zero on consolidation. FP&A plans in the management view and lets the controller bridge to the statutory one.

The two are less separate than they used to be. ASC 280 segment reporting uses a "management approach", requiring segment information through what KPMG calls the "eyes of management" — segments follow the internal reporting structure and the chief operating decision maker's view. ASU 2023-07 tightened this by "requiring disclosure of significant segment expenses and increasing the frequency of segment reporting to interim periods." At a public company your internal management reporting is therefore no longer purely internal: how you cut cost centers into segments, and which expense lines you show the CEO, can end up in the 10-K.

6.4 COGS, gross margin, and cost to serve

Gross margin is revenue minus cost of revenue, over revenue. Investors look at it first because they treat it as a property of the product rather than a spending choice, which is what makes it comparable across companies. Treat that as roughly true and specifically false: the COGS boundary, the services mix, the discount profile, and the capitalization policy in 6.5 each move it by points without touching the product.

What sits in SaaS COGS, with SaaS Capital's medians as a percent of ARR: hosting (5%), DevOps and the production-engineering labor keeping the platform up (4%), professional services delivery, the implementation team (5%), and other (3%) — third-party API fees, data licensing, payment processing. Customer support and success is reported separately at 9%.

Subscription margin versus blended margin

The most useful decomposition in SaaS. Benchmarkit's 2025 metrics report puts total revenue gross margin at a 77% median, subscription revenue gross margin at 81%, professional services gross margin at 30%, and professional services at roughly 15% of total revenue. The first gap says the median company carries enough services to cost four points of blended margin. The report also uses those figures as a diagnostic: if services revenue exceeds 15–20% of total revenue or services gross margin falls below 30%, total gross margin likely lands under 77%. Do not compose the medians and expect them to close, though, because they describe separate distributions rather than one company. Blend 81% against 30% and you need about 92% subscription and 8% services to reach 77%; at the reported 15% services share the same blend lands near 73%. What survives the arithmetic is the point that matters: a company with a heavy services attach rate reports a materially lower blended gross margin than a pure-subscription peer with an identical product and identical hosting cost. Always ask for the subscription-only line, and remember from 6.1 that the median gross margin is this 77%, never 100 minus SaaS Capital's 17% of ARR.

Worked example: the gross margin bridge

The running company, whose full P&L is laid out in section 9, carries $40M of ARR as a year-end run rate, and what gets recognized across the year is $38M of subscription revenue plus $4M of professional services, $42M in total. Run rate and recognized revenue never match on a growing book, so check which one a margin is quoted against.

• Subscription COGS at 19% of subscription revenue = $7.22M → 81% subscription margin
• Services COGS at 70% of services revenue = $2.8M → 30% services margin
• Total COGS $10.02M on $42M of revenue → blended gross margin 76.1%, the figure section 9 carries

Now move $2M of revenue from services to subscription by productizing implementation. Subscription $40M at 19% COGS and services $2M at 70% COGS gives total COGS $9.0M and a blended margin of 78.6%. Two and a half points at the same price. It is not free, and this is what the first draft of such an analysis always leaves out: $1.4M of services delivery cost comes out, roughly nine delivery FTE at a loaded rate, and about $0.4M of engineering and hosting goes in to build and run the productized onboarding. It is a headcount reallocation out of Services, so expect the CFO to ask for the transition-year bridge — services revenue falls the moment you stop selling implementations, the delivery reduction lags a quarter or two, and year one looks worse than the steady state. Quantifying the J-curve and the landing point is a strategic-finance task; see section 10.

Cost to serve, and why segment margins diverge

Blended gross margin is an average that hides the decision it should inform. Split the same $7.22M of subscription COGS by what drives it, then push it onto two segments: 500 SMB customers at $15K ACV ($7.5M) and 122 enterprise customers at $250K ACV ($30.5M).

An enterprise customer costs ten times more to serve and pays nearly seventeen times more, which is the argument for going upmarket in one line. Eleven points of margin separate two segments of the same product. Now apply the unsettled boundary above: move the $3.4M customer success organization into COGS, 80% of it on enterprise accounts with named CSMs, and enterprise margin falls from 83.2% to 74.3% while SMB falls to 62.9%. Same company, same customers, same cash. The convention decides the answer, which is why it has to be settled before anyone uses these numbers to decide whether to keep selling downmarket. Per-customer allocation also needs usage tagging or a defensible proxy metric — 6.3 arriving with consequences. Revenue-side unit economics are in section 9.

AI cost is the live gross-margin question

The SaaS Capital COGS medians predate material AI cost in most respondents' P&Ls, and this is the first thing a CFO or an interviewer raises about gross margin in 2026. Inference and model API fees sit in cost of revenue as usage-driven variable cost, alongside routing and gateway compute, vector databases and embedding pipelines, evaluation runs, and GPU infrastructure serving the product. CloudZero's breakdown puts AI-augmented SaaS at about 80% gross margin, AI-enabled SaaS at 60–79%, and AI-native products where the model is the product at 50–59%, against a 70–85% SaaS baseline, citing ICONIQ data showing AI gross margins moving from 45% in 2025 to a projected 53% in 2026.

The structural problem is that this converts a fixed line into a variable one and breaks the operating leverage story. Plan it as a bucket 2 driver: active users times actions per user per month times blended cost per action. At the $40M company, 8,000 active users running 120 AI actions a month at $0.05 each is $576K a year, 1.4 points of blended gross margin. Double usage to 240 actions and cost goes to $1.15M, another 1.4 points, with zero additional revenue because the contract is priced per seat. That is the seat-versus-usage trap, and The SaaS CFO's version is the same shape: $100 of revenue with $20 of traditional COGS is an 80% gross margin; add AI features that push COGS to $35 and the margin is 65%. Averages hide it, so watch top-decile user cost against median, not the mean. The levers Finance models are routing to cheaper models (the largest, given the price spread between tiers), caching stable context, batching asynchronous work, cutting context length, and committing model spend the way you commit on cloud; the pricing lever is a usage or credit component on top of the platform fee. Expect to be asked for an AI-adjusted gross margin or a broken-out AI COGS line.

Gross margin is also what makes operating leverage work — "the proportion of a company's cost structure that consists of fixed costs rather than variable costs." A high-gross-margin business converts each incremental revenue dollar to operating income at close to the gross margin rate, because the marginal cost of one more customer is small. That is the argument for the SaaS model, why an investor treats a 60% gross margin software company as a different asset from an 80% one, and precisely what a usage-priced inference bill erodes.

6.5 Capex, leases, and capitalized software

Capital expenditure is cash spent on an asset delivering benefit over more than one period. It does not hit the P&L when spent; the asset goes on the balance sheet and its cost is charged over its useful life as depreciation (tangible) or amortization (intangible, capitalized software included). Accumulated depreciation is "the cumulative reduction in the carrying value of a fixed asset (PP&E) since the date of initial purchase," a contra-asset netted against gross PP&E. Straight-line is (cost − salvage) ÷ useful life: $100M of PP&E over 10 years with no salvage is $10M a year, $50M accumulated after five.

Software companies are not capital intensive, so capex is usually small: laptops, office build-out, leasehold improvements, occasionally data-center hardware. It matters for three reasons. It is a cash outflow that never appears on the P&L, so a cash forecast built off net income and ignoring capex is wrong. Depreciation is a P&L expense with no cash behind it, which is why EBITDA adds it back. And capitalize-or-expense on software moves reported margin materially.

Capex usually runs on its own approval track with a lower threshold than opex: a capital request naming the asset, amount, in-service date, and useful life, approved separately from the operating budget and often by a different signer. The modeling artifact is a fixed-asset roll-forward — opening net book value, additions, disposals, depreciation, closing — feeding three statements at once: depreciation in the P&L, PP&E on the balance sheet, capex in investing cash flow. Build it once as its own schedule rather than re-deriving depreciation inside the P&L tab: when an in-service date moves, three numbers have to move together and only a schedule does that reliably.

Leases

Under ASC 842 a lessee puts a right-of-use asset and a lease liability on the balance sheet for essentially every lease. Datadog's policy note is the plain version: "Operating lease assets and liabilities are recognized at commencement date based on the present value of lease payments over the lease term," with leases of twelve months or less kept off balance sheet and expensed straight-line. Classification matters for what it does below the operating line. An operating lease produces a single straight-line expense inside opex, above EBITDA, never added back. A finance lease splits into amortization and interest, and EBITDA excludes both by construction — "the entire lease expense is considered an operating expense, while the interest portion of a finance lease appears below EBIT" — so classifying the same commitment as a finance lease raises reported EBITDA without changing a dollar of cash. Data-center and hardware commitments frequently land on that boundary. Ask for the lease schedule in your first month: it is the cleanest source for the committed-cost bucket in 6.2.

Capitalized software: the one that matters

Two standards, and you need to know which the company is under. ASC 985-20 governs software to be sold, leased, or marketed and keys capitalization on technological feasibility. ASC 350-40 governs internal-use software, which at a SaaS company means the hosted product itself. Almost everything here is 350-40, and per BDO, "No changes were made to the external use software guidance in ASC 985-20."

What most finance teams still use. ASC 350-40 has run on three project stages: preliminary project stage (expensed), application development stage (qualifying costs capitalized once management authorizes and commits funding and completion is probable), and post-implementation stage (maintenance and support expensed). Amortization begins when the software is ready for its intended use, over short lives — Datadog amortizes capitalized internal-use software straight-line over three years, raised from two in January 2025.

What replaces it. ASU 2025-06 deletes the stages. It "removes all references to project stages in ASC 350-40 and provides a principles-based recognition threshold": capitalization begins when "Management authorizes and commits funding for the software project" and "It is probable the project will be completed and the software will be used to perform the function intended," with nothing capitalized while significant development uncertainty remains — novel functionality unresolved after testing, or major performance requirements still undefined or changing. Effective for annual periods beginning after December 15, 2027; "Early adoption is permitted" at the start of an annual period. Agile teams that could never point to a discrete application development stage now capitalize on a funding-and-probability test instead. Stage vocabulary is what you will hear through 2027; authorization and probability is what you will argue about after. KPMG's February 2026 handbook covers ASU 2025-06 and the accounting for AI software development, including data costs.

The mechanics, at the $40M ARR company with $9.6M of R&D. Finance and Engineering agree that 12 of the 45 engineers spent an average of 55% of their time on qualifying development across two projects: 6.6 FTE-equivalents at a $190K fully loaded cost, so $1.25M is capitalized rather than expensed. Amortized straight-line over three years, $417K a year from the in-service date, roughly $208K in year one at a mid-year placement.

• R&D expense falls from $9.6M to $8.35M, or 3.0 points of revenue.
• Amortization of $208K lands in year one. At most SaaS companies internal-use software supporting the hosted product amortizes into cost of revenue — Datadog splits it "between paid customer accounts and free customer accounts not generating revenue," putting the paid share in cost of revenue and the free share in sales and marketing. On $42M of revenue that is about half a point of gross margin, moving opposite to R&D. Some companies amortize to R&D instead; ask, because identical spend produces visibly different-looking companies.
• Operating income improves by about $1.04M.
• Cash is completely unchanged. Not one engineer was paid differently.

That is the whole trick, and why investors compute "EBITDA less capitalized software" and diligence always asks for the capitalization policy. Capitalized software flatters EBITDA twice: the spend never hits opex, and the amortization is added back. A company that capitalizes aggressively shows better EBITDA and worse free cash flow conversion. Expect to produce both views.

Cloud versus capex

The shift from owned servers to rented capacity moved spending from capex to opex: a purchased server was capitalized and depreciated over three to five years, while the equivalent cloud workload is a monthly operating cost. ASU 2018-15 pulled the implementation costs of a service-contract cloud arrangement back into the ASC 350-40 framework rather than defining its own, and Deloitte records why — the EITF decided that "ASC 350-40 already contains sufficient explanatory guidance." So the capitalize-or-expense list is 350-40's list: configuration and interfaces, coding, testing including parallel processing, and installation are capitalized; training is always expensed; data conversion is expensed except for the narrow slice covering "data conversion cost that allows access of old data by the new system." Get that carve-out backwards and a controller will correct you. The subscription fee itself is pure opex with no depreciation schedule behind it, so a cloud-native company carries far less PP&E and far more run-rate opex than the same business would a generation ago. The accounting mechanics here are well sourced; the board-level "cloud trades capital efficiency for flexibility" narrative is not, so treat it as conventional wisdom rather than something verified for you.

7. Forecasting after the budget is locked

The plan is approved, the board deck is filed, and nobody will change the numbers in it for twelve months. Within six weeks the company will be running on a different set of numbers, produced by your team, updated on a cadence, and used to make every real decision of the year. That second set is the forecast. The AOP build is the loud part of the job; this is the part that fills the calendar.

This section owns the rolling-forecast half of the guide's fourth core question, "when is each budgeting approach used?" The budgeting methods themselves are in section 3; whether a rolling forecast replaces the annual budget is answered in 7.3.

7.1 Four documents, four jobs

You already run parallel numbers: appropriation, allotment plan, carry-forward level, maintenance level, policy level, the ERFC revenue forecast, the Caseload Forecast Council caseload. The difference is that exactly one of yours carries legal force and everything else is built to stay reconciled to it. Private-sector planning runs several numbers at once, none of which has legal force, and the fastest way to look new is to confuse them.

The budget (also the plan, the AOP, the plan of record, abbreviated POR in a lot of buildings) is a commitment: what the company told the board it intends to achieve. It doubles as the reference point for spend and headcount approvals and as the baseline for bonus and commission plans. Note the words "reference point." A budget line is not spending authority in the appropriation sense, a difference large enough that it gets its own bullet in 7.7 and a whole section in section 11. The budget is deliberately static: Corporate Finance Institute lists its flexibility as "Static — changes require executive approval", and Prophix puts the purpose split as budgets specifying what management hopes to achieve during a specific financial period, forecasts what a company is likely to achieve.

The forecast (or reforecast, or latest view) is the team's current best estimate of how the year actually lands. It moves on a cadence, and it is the number the CFO steers with. It carries no authority of its own: a forecast showing $2M of unspent marketing budget does not release that money, and a forecast showing an overrun does not authorize it.

The outlook is the same idea pointed outward. At a public company it is effectively synonymous with guidance: the revenue and EPS ranges management gives on the quarterly earnings call. Guidance is not the internal forecast published. Three revenue numbers coexist at a public company — plan, internal forecast, guidance — and guidance sits below the internal forecast on purpose, so the company can beat it and raise it. How much cushion is a CFO policy call, which is why "we guided to $X" and "we expect $X" are never the same statement. The internal number circulates to a deliberately short list: selective disclosure of material nonpublic information to analysts and institutional investors triggers Regulation FD's public disclosure requirement, so leaking it is a securities problem, not an etiquette problem. Section 13 has the rest. Inside a private company the word is loose: some call every quarterly reforecast "the Q2 outlook," others reserve it for a horizon beyond the fiscal year. No authoritative source fixes an internal distinction between "forecast" and "outlook." Ask on day one which word means what in your building.

The latest estimate (LE) is a within-period update between formal cycles: the month is half over, three deals slipped, and the number the business is working to is no longer the one submitted three weeks ago. It is common vocabulary in companies with industrial heritage and rare in software. Do not confuse it with LBE (latest best estimate), the house word for the whole reforecast at Amazon- and Microsoft-lineage companies and therefore at much of Seattle tech. I found no authoritative definition for either; treat both as common usage rather than standards, and confirm locally. The related flash is not a forecast at all: a flash report is a summary of the key operational and financial outcomes of a business, typically provided by accounting to management on a frequent basis, perhaps daily or weekly, off preliminary data that formal statements may later adjust. It reports on a period that has ended but not closed (section 2).

Finally, a projection is not a forecast. A forecast updates expectations from real data; projections explore hypothetical scenarios conditional on something that has not happened. When someone asks for "a forecast that shows what happens if we buy them," that is a projection, and mislabeling it is how a hypothetical ends up in a board deck as a commitment.

Why the budget is not allowed to move

The instinct from government is that when circumstances change you change the authorized number, because the authorized number is what controls behavior. In a company the budget stays frozen precisely so it can keep doing three jobs a moving number cannot do.

• It is the yardstick. Budget-versus-actual measures whether the commitment was any good; forecast-versus-actual measures whether your team can predict. Re-cut the budget whenever the business moves and you erase the first measurement and never build the second.
• It is the compensation baseline. Quotas, bonus targets and often executive equity are struck against plan. Moving the plan mid-year moves people's pay, which turns forecasting into a negotiation over compensation.
• It is the board's copy of the deal. Restating approved numbers is a governance event, not a planning routine.

So by November the budget is often visibly wrong and still gets printed in every report. That is not a defect: one number that is always approximately right tells you neither thing.

When the budget does move

"Never" overstates it, and saying it in a room where it happened last year looks naive. The rule is that a bad forecast is never grounds to move the plan. But four events legitimately are, and all four were common in growth-stage tech between 2022 and 2024: a reduction in force; an acquisition, divestiture or major product line change; a financing outcome materially off assumption (a down round, a failed raise, an unexpectedly large one); and a plan that Q1 proves unreachable rather than merely difficult. The resulting document goes by revised plan, re-plan, budget reset, reset case, or Plan B. Three mechanics matter to you:

• The original plan usually keeps appearing in variance reports alongside the revised one, at least until the next fiscal year, which is why immutable version naming matters, below.
• Sales quota relief is decided separately from the P&L re-cut, on its own timeline, and is a partial re-baseline by another name. Expect the two conversations to be confused.
• A re-plan is a board event with a deck and a decision, not a planning-calendar routine. If someone tells you "we re-cut the budget in April," ask which of the four things happened. If the answer is none of them, they mean the forecast.

7.2 Cadence, house names, and the n+m convention

The naming convention that trips people up first is a pair of numbers with a plus sign: months of closed actuals plus months of remaining forecast, always summing to twelve. A 3+9 is three months of actual results and nine months of forward projection. The plus sign separates actuals periods from forecast periods while holding a constant twelve-month view. The notation is inherited from consumer goods, industrial and pharma FP&A, and in software it shows up mainly where the finance leadership came from those industries.

What you are more likely to hear in Seattle tech, and what nobody will define for you: POR for the locked plan, used more often than "AOP"; LBE for the reforecast; current view or CV for the same thing; F1 / F2 / F3 for the first, second and third forecast cycles of the year; and the landing or landing zone for the expected full-year result. These are house conventions, not standards. Ask in week one which ones your company uses.

The mechanics are the same in all three: as each month closes it moves from the forecast side to the actuals side, and each remaining open month is updated based on the latest trends and assumptions. The 6+6 is the one that comes midyear, in July. One distinction to hold: within a fiscal year the window is the fiscal year and it resets each January. Farseer, the source above, is describing the rolling variant, where the twelve-month window keeps extending past the year end. That is 7.3, and it is a different animal.

A 6+6 end to end

Take a 250-person SaaS company on a calendar fiscal year that entered the year at $40.0M ARR with $22.0M in the bank. The board-approved FY plan: exit at $52.0M ARR, recognize $46.0M of revenue, spend $56.0M across COGS and opex, post a $(10.0)M operating loss, grow to 292 people, and end the year with $13.5M of cash. Subscription revenue is recognized ratably, so a half-year's revenue runs close to the average of opening and closing ARR for that half, times a half: the plan's $21.5M H1 and $24.5M H2 fall straight out of the $40.0M / $45.8M / $52.0M ARR ramp.

June closes; accounting finishes on business day six. What FP&A has in hand:

• New and expansion ARR booked in H1: $7.4M against an $8.6M plan.
• Gross churn and contraction: $(3.4)M against $(2.8)M planned.
• ARR at 6/30: $44.0M against $45.8M planned.
• H1 revenue: $21.0M against $21.5M.
• Net hires: 14 against 21 planned, average start-date slip about five weeks. Headcount at 6/30 is 264.
• H1 COGS and opex: $25.6M against $26.5M, almost entirely the hiring lag. H1 operating loss $(4.6)M against a planned $(5.0)M.
• Cash at 6/30: $18.2M, so $3.8M of net burn in six months, less than the operating loss, because deferred revenue builds on prepaid annual contracts and stock compensation is not cash.

The build takes about three weeks in a fixed order, because each step feeds the next. Sales and RevOps rebuild second-half bookings first, off pipeline coverage and current win rates rather than the plan's assumptions (section 4): $7.6M of H2 new and expansion against $9.4M planned, split $3.5M in Q3 and $4.1M in Q4, since coverage entering Q3 is 2.8x against the 3.5x assumed, plus churn of $(3.6)M against $(3.2)M on two known Q4 renewal risks. Recruiting and hiring managers then rebuild headcount off actual start dates and open requisitions, landing at 280 rather than 292, and FP&A rebuilds compensation from the new position-level file (section 5). Non-headcount spend gets a lighter touch: owners confirm or revise committed spend, $0.6M of program spend moves into next year, and anything nobody changes holds at budget.

The result:

Measure	FY plan	6+6 forecast	Delta
Exit ARR	$52.0M	$48.0M	$(4.0)M / (7.7)%
FY revenue	$46.0M	$44.0M	$(2.0)M / (4.3)%
FY COGS + opex	$56.0M	$52.8M	$(3.2)M
Operating loss	$(10.0)M	$(8.8)M	$1.2M better
Year-end cash	$13.5M	$14.6M	$1.1M better
Year-end headcount	292	280	(12)

Read that table the way an FP&A director reads it. Three of the four financial lines look better than plan. The business is materially worse. A $4.0M exit-ARR shortfall costs only $2.0M of revenue this year, because subscription revenue is recognized ratably and the missing bookings would have landed late, but it removes the full $4.0M from next year's opening base before a single new deal. The expense favorability is not thrift, it is twelve unfilled roles: capacity you did not get, which will cost more to buy next year. Run one efficiency metric across it and the disguise falls off. Burn multiple is net burn divided by net new ARR: the plan was $8.5M of burn against $12.0M of net new ARR, or 0.71x; the forecast is $7.4M against $8.0M, or 0.93x. Less cash burned, worse efficiency. So the commentary is one paragraph and it says we are beating the loss target by underperforming on the two things that matter, our efficiency got a third worse, and next year starts $4M in the hole. That paragraph is the deliverable; the number is the ticket to write it.

What the cycle actually produces

The table above is a variance table. The cycle produces two more artifacts, both core vocabulary you will otherwise be missing in your first forecast review.

The first is the bridge (also the walk): a decomposition of the delta into named, owned, driver-level buckets that sum exactly to the total, rather than one net number per line. The revenue bridge for the 6+6 above walks plan to forecast like this:

Bucket	Owner	Impact
FY plan revenue		$46.0M
New and expansion bookings below plan	CRO	$(1.4)M
Churn and contraction above plan	Chief Customer Officer	$(0.5)M
Go-live and start-date timing on signed deals	VP Services	$(0.2)M
Price and mix ahead of plan	CRO	$0.1M
6+6 forecast revenue		$44.0M

A bridge whose largest bucket is "other" has not been built. Each bucket names a driver and a person, which is what makes the next conversation possible: a $2.0M miss is an argument, and four named buckets with owners is an agenda.

The second is the risks and opportunities schedule, universally shortened to R&O. It sits below the forecast and lists what is not in the number: unbudgeted downside with a probability and an owner, and identified upside not yet committed. AFP's guidance is to describe each item from a probability and an impact perspective, to stop at ten or fifteen rather than chase a hundred ("after the first 10 or 15, the marginal benefit of adding more is reduced"), and to pair each with a contingency plan for how the company responds if it lands. So the reported position in the review is not "$44.0M." It is "$44.0M forecast, $0.7M of risk, $0.4M of opportunity": the risk being one large Q4 renewal, a services delivery slip and an EU data-residency requirement that could stall two deals; the opportunity being a partner-channel deal and an October renewal price increase.

The verb attached to this is calling it in: leadership deciding how much of the opportunity to move out of the schedule and into the committed number. "What is in R&O this cycle, and how much are we calling in?" is the sentence a forecast review runs on. A forecast submitted with an empty R&O schedule means either a genuinely quiet quarter or a team that is not looking.

The reforecast is a negotiation

Everything above describes the cycle as estimation: gather actuals, re-run drivers, publish. That is how the calendar looks, not how the inputs arrive. Every submission you receive has been shaped by someone whose pay or headcount depends on it, and a director who takes the roll-up at face value in his first cycle gets asked by the CFO why he believed it.

• Sales. The number arrives sorted into forecast categories, typically commit, best case, and pipeline. The commit is deliberately beatable, because the rep and the VP are both measured on hitting it. Your counter is not to argue with rep judgment; it is a haircut derived from historical stage-conversion rates and coverage ratios, applied consistently. The standing question in the room is which number the company is actually managing to: the sales commit, the FP&A-adjusted number, or the CFO's. Know which one your model publishes.
• Opex. A department head who honestly forecasts a $400k underspend usually watches finance sweep it. The rational response is to forecast full spend and find something to buy in December, exactly the December spend-down you know from agencies protecting next biennium's carry-forward base. The counter is structural, not moral: state up front whether savings are swept or retained, and hold a central contingency so finance does not have to raid departments to cover surprises.
• The CFO override. The bottoms-up roll-up is an input, not the answer. Forecasts routinely get topped down before they leave the building, and that is a legitimate exercise of judgment about which submissions are credible. Your job is to make the override visible, carried as a named adjustment line rather than blended into department detail, so next quarter you can tell whether the CFO or the roll-up was right.
• The diagnostic to run in week two. Pull the last four cycles and find the departments whose forecast equals budget every single time. That is not stability, it is non-participation, and it is the cheapest thing you will ever fix.

What is frozen inside a reforecast

Reforecast scope is company-specific and worth asking about, because it sets how much work a cycle is. Common conventions: closed months are locked; approved compensation for existing employees is not re-opened (only timing and open roles move); allocations and shared-service charges hold flat unless the basis changed; commission-plan targets never move. What is in play is bookings and pipeline assumptions, hiring timing, program and vendor spend, and anything with a decision behind it.

Version naming matters more than it sounds. A company keeping FY26 BUD, F1, F2 and F3 as immutable saved versions can answer "when did we first see this coming?" One that overwrites "Forecast_latest_v4_FINAL" cannot. The planning tools in section 12 enforce this natively.

Monthly versus quarterly

Quarterly is the mainstream cadence. Monthly reforecasting is common where cash binds or revenue is volatile, and the trade is real: it keeps the number current, eats most of a small team's capacity, and trains the business to renegotiate its number continuously rather than deliver it. One vendor survey, Aleph's own August 2026 poll with just 27 monthly reforecasters in it, put 77.8% of monthly reforecasters against 37.0% of quarterly ones on still having an accurate budget at mid-year; treat that as a vendor arguing for its product. The defensible claim is narrower: re-examine assumptions more often and you catch a broken one sooner, at the cost of FP&A time. A good default is a quarterly reforecast plus a monthly light-touch update moving only revenue and headcount timing.

7.3 Rolling forecasts

A rolling forecast drops the fiscal-year boundary, instead continuously replanning over a fixed horizon, typically the next twelve months at each quarter-end. Because the horizon never shortens, it removes the pathology of the static year: by October a calendar-year company can see three months ahead, so every decision gets deferred into next year's plan.

Horizons are industry-specific, not universal. The table everyone quotes originates with Larysa Melnychuk of FP&A Trends, in an AFP annual conference presentation reproduced by Wall Street Prep:

Industry	Horizon	Update cadence
Airlines	Rolling 6 quarters	Monthly
Technology	Rolling 8 quarters	Quarterly
Pharmaceuticals	Rolling 10 quarters	Quarterly

That is where the "8 quarters" figure you will hear in tech interviews comes from: one practitioner's conference slide, reproduced widely enough to feel like a standard.

Do rolling forecasts replace the annual budget? Almost always no. Two adoption figures circulate and both are secondhand: Wall Street Prep cites an EPM Channel survey finding only 42% of companies use a rolling forecast, with 20% having tried and failed, while Ramp cites the 2024 FP&A Trends Survey at 49%. Neither original survey was verified here, and the EPM Channel figure is undated and reads as pre-2020, so quote the 2024 number if you quote either. Both agree on the useful part: most adopters run the rolling forecast alongside the annual budget, because the traditional annual budget is still considered by many organizations to provide a useful guidepost connected to a long term strategic plan. The budget keeps doing its three jobs and the rolling forecast becomes the operating view. The minority who genuinely abandon the budget are doing Beyond Budgeting (section 3).

7.4 Scenarios and sensitivities

These two words get used interchangeably and should not be. A sensitivity analysis flexes one input and holds everything else constant. A scenario analysis flexes an internally consistent set of assumptions together to tell one story. Vena's article works both, on separate examples: for sensitivity, a baseline of 10,000 units at $50 producing $200,000 of net profit, with volume alone moving (+10% to $220,000, -10% to $180,000); for scenario analysis, three coherent stories, best case at 10% annual revenue growth with expenses down 5%, base at 3% growth with expenses stable, worst at a 5% revenue decline with a 10% expense increase (source).

They are sequential, not alternative: run sensitivities first to find which drivers move the answer, then build scenarios around only those. In the 6+6 above, flexing H2 new and expansion ARR by plus or minus 10%, $0.76M in each direction on the $7.6M forecast, moves exit ARR by that same $0.76M either way, a $1.5M spread. It moves FY revenue by less than $0.2M either way, because ARR booked across the second half earns an average of only about three months of in-year revenue. That asymmetry is the finding: in-year revenue is nearly insensitive to second-half bookings, and the reason to care about them is next year, where the same $0.76M is worth four times as much. Three pre-baked scenarios would never surface it.

Four rules separate scenarios that get used from scenarios that get admired in a deck:

1. Build from shared drivers, not separate totals. Every case runs off one model with different driver values. A downside typed in as a number cannot be interrogated.
2. Move two or three drivers, not twenty. A scenario that changes everything explains nothing.
3. Keep it internally consistent. A downside with lower revenue and unchanged hiring is not a plan; costs flex with revenue, commissions and hosting included.
4. Attach a decision. Each case carries a one-line story and a planned response. The deliverable is what you will do, not the spreadsheet.

These track Aleph's process description closely, including the line worth quoting to yourself before every scenario exercise: the value is having the decision made before the quarter forces it. FP&A Trends adds the governance layer most teams skip: unless the organization defines trigger points, decision rights and intervention options in advance, the analysis often stays in the planning deck and the decision system never changes.

Applied to the 6+6: the downside case assumes H2 new and expansion ARR comes in 20% below the reforecast ($6.1M instead of $7.6M) and gross churn runs a point worse over the half on the $44.0M base ($4.0M instead of $3.6M). Exit ARR lands at $46.1M, FY revenue at $43.5M, and year-end cash near $14.0M once the lost annual prepayments net against unpaid commissions. The trigger is written in advance and is specific: if Q3 new and expansion lands below $3.2M against the $3.5M forecast, freeze the eleven open requisitions and cut $1.2M of second-half program spend. Note what each lever is worth. The program cut is $1.2M of this year's cash. The hiring freeze is only $0.35M this year, eleven roles at roughly $190k fully loaded averaging two remaining months each, but it is $2.1M of annualized cost not committed, which is the number that matters for next year's base. Together they put year-end cash back near $15.6M and take the December run rate to about $0.60M a month, putting runway back over two years. When Q3 closes at $3.0M, nobody needs a meeting to decide what happens. That is the entire point.

Some teams weight the cases by probability (Carl Seidman's FP&A Trends piece uses 50% best / 30% base / 20% worst as its illustration). Weights are useful for your own expected-value math and dangerous in a board deck, where the average of three cases becomes a fourth number nobody planned for. Present the cases.

7.5 Cash and runway

At a public company the P&L forecast dominates. At a venture-backed company the cash forecast is the forecast and everything else is supporting detail, because the P&L cannot kill the company and the bank balance can.

How a cash forecast is actually built

The twelve-to-eighteen-month cash forecast is built by the indirect method: start from the P&L forecast, add back non-cash charges (stock compensation, depreciation), and roll working capital. Three levers do almost all the work in a software business. Receivables move with DSO (days sales outstanding) or, better, a collections curve fitted to how your invoices actually pay. Payables move with DPO. And the big one is the gap between billings and revenue: on a prepaid annual contract you collect twelve months of cash on day one and recognize a twelfth of it a month, so deferred revenue builds and the company burns less cash than its operating loss, exactly what happens above, where an $(8.8)M forecast loss produces $7.4M of net burn. Get this backwards and every cash number you publish is wrong in the same direction. What drives the cash line is billings and collections, not recognized revenue (section 9).

The 13-week cash flow is the other model, built the opposite way: the direct method, a weekly schedule of named receipts and disbursements, this customer's wire, that payroll run, the quarterly sales-tax remittance. It lives in a spreadsheet rather than the planning system because it is receipt-level and changes daily, usually maintained by one person, often the CFO. Companies add one as runway tightens; I found no published threshold and would not quote one. Know the connotation: the 13-week originates in restructuring and lender reporting, so a board member asking whether you run one is asking about distress, not modeling technique.

Runway

Runway is cash available divided by monthly net burn: $600,000 against $100,000 of monthly burn is six months. Two refinements matter. Use net burn (cash out minus cash in), because that is what walks the balance toward zero and what investors use. And use a trailing three-month average, recomputed every month, because collections are lumpy: a quarter with three large annual renewals collected in March makes March look like profitability and April like catastrophe.

Continuing the example: H2 net burn of $3.6M averages $0.60M a month, but it is back-loaded. The Q3 renewal cohort collects in July and August and the hires that slipped out of H1 all start in Q4, so Q4 runs at about $0.75M a month, making the trailing three-month average at 12/31 $0.75M. Year-end cash of $14.6M against that is roughly 19.5 months of runway, and that number is not neutral. CRV's 2026 framing is that companies from pre-seed through Series A and beyond now target 18 to 24 months, up from the older 12-to-18-month rule, because the interval between rounds has stretched; the process should open with 12 to 18 months of runway left and consumes three to six months to close. So 19.5 months at December is not a comfortable finding: it says open the process in the first half of next year, and it changes what the company decides in Q4 about which roles to backfill, whether to sign the three-year infrastructure contract, and how hard to push on December collections.

This is the forecast doing work the budget cannot. The plan said $13.5M at year end, and nothing about that number tells anyone to start fundraising. A forecast, a burn trend, an efficiency metric and a market norm together produce a decision with a date on it.

7.6 Accuracy, and how FP&A actually gets graded

Start with the fact that reframes the subject: per the 2026 AFP FP&A Benchmarking Survey, as cited in an AFP-published article, only 14% of finance teams formally track forecast accuracy. Coming from an agency where allotment variance is monitored quarterly by an external control agency and explained in writing, that will read as implausible. It creates an opening: a manager who institutes basic accuracy tracking in his first two quarters is doing something six of seven finance teams have not done.

Two decisions before the first metric

Both matter more than which metric you pick, and a first-time implementer usually gets both wrong.

1. Fix the vintage and the lag. Accuracy is only meaningful against a stated forecast version measured at a stated distance from the period. Score the Q1 forecast against actuals for the quarters it covered, one-quarter-ahead and two-quarter-ahead, and publish the lag with the number. Without a fixed vintage the metric improves by itself as the period closes, since the last forecast before close is always the best one, and a team can post a beautiful trend while forecasting no better than last year. "One-quarter-ahead revenue MAPE" is a metric. "Forecast accuracy" is not.
2. Fix the level. Measure at the level you manage, department, revenue line or headcount by function, and report the consolidated figure only as a summary. A consolidated MAPE of 2% routinely sits on top of every department being 15% wrong in offsetting directions, which is the most common way an accuracy program produces a comforting number and changes nothing.

Expect the first six months to mostly reveal which line items are structurally unforecastable: legal settlements, a single lumpy hardware purchase, a nascent product line. Removing those from the scorecard is legitimate, as long as you do it in advance and in writing rather than after a bad quarter.

The metrics

They are borrowed from demand planning and are unremarkable. MAPE (mean absolute percentage error) averages the absolute percentage miss and is the usual headline; it can produce misleadingly high error percentages for slow-moving products and is impossible to calculate when actual sales are zero, so it is poor for small departments and new products. Bias averages the signed error and says whether you run consistently high or low; it is the companion metric to MAPE in the same supply-chain literature, and the one that catches sandbagging. RMSE (root mean squared error) is the standard statistical companion, not part of the RELEX treatment: it punishes large misses disproportionately, which matters when one big error breaks a cash plan. The formulas transfer to FP&A cleanly; the numeric thresholds published in supply chain do not.

The most useful framework I found is Jason Brisbane's four-component score, published by AFP (disclosure: he runs an FP&A software company and the article is partly a pitch; the framework holds up anyway). It scores four dimensions rather than one: accuracy (how far forecasts deviate from actuals on average), bias (does the team consistently over- or under-estimate), volatility (how stable the error pattern is) and persistence (do the same errors repeat, which he treats as the most actionable finding, since a repeating error means the team is not recalibrating). They aggregate to a 0-100 index on which lower is better: below 25 is "institutional-grade forecast discipline," above 60 is "structural fragility." Calling it a "score" invites the wrong reading, so say "fragility index" when you present it. It is calculated per line item off six or more periods of paired data, monthly ideally, quarterly acceptable.

One caveat the source does not raise. As published, the framework runs on budget-versus-actual pairs, which measures whether the plan was any good. To grade the forecasting function instead, run the identical four components on forecast-versus-actual pairs at a fixed lag. Both are worth having; they answer different questions, and the distinction is the same one that keeps the budget frozen in 7.1.

Four components rather than one is the part worth memorizing, because it is the same dynamic you know from budget-padding in government. Measured on accuracy alone, teams game it: sandbagging forecasts, narrowing scenarios and avoiding ambitious targets in order to hit the number. Bias, volatility and persistence exist to make that gaming visible. A team that is never wrong is not forecasting well; it is forecasting conservatively, which is a different and worse thing.

How grading actually works

Formal mechanisms tying accuracy scores to performance reviews or compensation are rare and rarely documented; I found no source describing a standard one. What operates instead is informal, harsher than a metric, and has three parts.

• The surprise is the failure, not the miss. Missing the quarter is a business outcome; missing it when the forecast said otherwise two weeks earlier is an FP&A outcome. The norm everywhere serious is no surprises: a number that moves is fine if it moved in your commentary first (section 14). This is also what the R&O schedule buys you, since a risk you disclosed and quantified in July is not a surprise in October.
• The direction of the miss carries a signal. Forecasting consistently low reads as sandbagging, and costs you credibility with the business whose budgets you then cut on your own bad numbers. Consistently high reads as being captured by sales. Either pattern, once the CFO notices, outlives any single miss.
• At a public company the grading is external. Landing inside the range you guided is the de facto standard, and the market prices the miss within minutes. The internal quarterly reforecast is what produces the number the CFO is willing to guide to (section 13).

One more framing worth carrying, from FP&A Trends: an organization that forecasts the quarter to within 1% but fails to see a structural shift coming has been accurately wrong, precise about the number and wrong about the trajectory. Point accuracy is table stakes. Being the person who saw the trajectory is what gets you promoted.

7.7 Translation: allotments, supplementals, and re-baselining

You already run this process, and Washington's version is more legally constrained and more precisely scheduled than anything a company does. The translation runs mostly in your favor.

The core mapping: a quarterly allotment amendment is a reforecast and a supplemental budget is a re-baseline. Reforecasting changes your best estimate of what will happen; re-baselining changes the target you are measured against. An allotment amendment updates the monthly spending plan under an unchanged appropriation; a supplemental changes the appropriation itself, and only then do you re-allot against the new ceiling, which is why OFM carries separate 1st and 2nd Supplemental packet purpose types alongside the quarter-adjustment ones.

The cadence analogy is worth saying out loud in an interview, as long as you say it precisely. OFM's 2025-27 instructions run seven quarterly amendment cycles across the eight-quarter biennium (the first quarter is covered by the initial allotment, not an amendment), with fixed due dates from Quarter 2 on October 25, 2025 through Quarter 8 on April 25, 2027, each requiring a brief description of how the spending plan assumptions have changed. That is a quarterly reforecast with mandatory written variance commentary against a two-year authorized plan: a longer horizon and a harder discipline than the annual cycle most companies run.

The materiality test translates cleanly. OFM requires an explanation when assumptions about key budget drivers — caseloads, population and client patterns, or revenue change significantly, and expressly does not treat a cost shift from delaying the purchase of a small number of personal computers as significant. Same judgment as deciding whether a department's $40k vendor slip belongs in the reforecast: reforecast what a decision changed, not what a month's timing moved.

WA practice	Private-sector equivalent	Where it breaks
Appropriation	Approved budget / plan of record, as the spending reference point	A legal ceiling enforced outside the agency, versus an internal governance limit enforced by the approval process
Allotment (monthly plan of estimated expenditures, revenues, cash receipts and cash disbursements for the biennium)	Phased budget by month loaded to the planning system, plus the cash forecast of 7.5	Allotments are filed with and monitored by a control agency; the private-sector equivalents are internal only, and are two separate models rather than one filing
Quarterly allotment amendment	Quarterly reforecast (3+9, 6+6, 9+3)	Amendment deadlines are statutory; reforecast cadence is a management choice the CFO can change at will
Revenue allotment amendment after an ERFC forecast revision	Revenue reforecast after a change in bookings or pipeline	ERFC forecasts are externally produced and mandatory to adopt; a company's revenue forecast is built in-house, arrives pre-shaped by sales, and is argued over
Supplemental budget	Re-baseline / re-plan; board-approved plan revision	A supplemental is legislation on an annual cycle; a re-plan is an internal board decision triggered by an event (RIF, M&A, financing, an unreachable plan), not a calendar
Reversion / lapse	Underspend against plan	Reverted appropriation authority genuinely expires; an underspent budget mostly just changes next year's baseline conversation
State budget outlook (RCW 82.33.060): maintenance-level revenue and expenditure projection for the current and ensuing biennium, prepared by the budget outlook work group and approved by ERFC	Outlook / guidance	WA's outlook is a statutory, externally approved four-year balance projection tied to the balanced-budget requirement; companies use "outlook" loosely for an internal reforecast, and at a public company it means guidance to investors. Same word, three different objects

8. Monthly close, variance analysis, and management reporting

Sections 3 through 7 produce a plan. This section is the machine that runs twelve times a year and tells you whether it is holding. The analytical work is expenditure monitoring, which you already do monthly. What changes is tempo and deliverable: the re-plan is monthly rather than a quarterly amendment packet, the deadline is a business-day count after period end rather than the 25th of the fiscal month, and the output is a narrative with a chart rather than a variance table with an explanation field.

One reframe to carry: in WA state the variance explanation with a hard deadline is the one justifying a quarterly amendment, and OFM expects management action but does not put it on a calendar. In a company the explanation is itself the monthly deliverable — an input to a management meeting, not a filing.

8.1 The close calendar

The monthly close is accounting's process of finalizing the general ledger for a period: posting the last transactions, booking accruals, reconciling subledgers to the GL, and locking the trial balance. Until the books close, FP&A has no actuals, and everything downstream sits behind that lock.

Days are counted in business days after period end, written BD1, BD2 and so on. Large enterprises and companies with European finance operations more often write WD1 through WD5 (working day); it means the same thing. You will also hear T+3 or T+5, borrowed loosely from securities settlement. No authoritative source standardizes T+N, so treat it as shop shorthand.

How long it actually takes

The benchmark a CFO will recognize is APQC's. Across roughly 2,300 organizations, top performers close in 4.8 days or less, the median is 6.4 calendar days, and the bottom quartile needs 10 or more, measured as cycle time from running the trial balance to completing consolidated statements. Note the unit. APQC counts calendar days; internal close calendars count business days, so a six-day close quoted from a benchmark and a six-day close on the controller's schedule are not the same close. Ask which is meant.

Vendor sources cluster nearby with more spread. Farseer puts most mid-sized organizations at three to five business days, highly automated companies at one to three, and complex multi-entity businesses at five to ten. Vena gives five to seven for high performers and eight to ten for mid-sized companies and, citing a third party (Ledge), reports that 50% of companies take six days or more and 94% still run the close out of Excel — a vendor survey quoted by a vendor selling the alternative. AFP describes regional variation instead, with some regions closing within four days and others running from about the twentieth of the month to the sixth of the next. The working range is five to ten business days, three to five where it has been invested in. Quarter-end runs slower, and at a public company the constraint is the earnings call, with the CFO "prepping for an earnings call within 15 days after the quarter-end close."

Because the close takes a week and the CEO will not wait, a common practice is to publish a flash report first: an unaudited one-pager with revenue, cash, headcount and two or three CEO metrics. OnlyCFO recommends it be "ready by business day ~2 of the period" (prescriptive advice, not a measure of how many companies do it). The flash may be wrong at the margin. Its job is to kill the week of speculation between month end and real numbers.

An illustrative calendar for a 250-person, $40M ARR SaaS company. Day counts are mine, built to sit inside the ranges above; a real company's calendar is published and enforced by the controller.

Business day	Accounting owns	FP&A owns
BD1–2	Cut off AP, post revenue, sync payroll and expense systems	Submit accrual estimates; publish the flash report
BD3–4	Bank and card recs, AP/AR subledgers to GL, payroll, benefits, depreciation	Review preliminary P&L for anomalies; push miscoded spend back to accounting
BD5	Final journal entries; lock the trial balance	Build the package shell; pre-write commentary for known events; reconcile HRIS headcount to the payroll register; tie CRM bookings to the revenue schedule; confirm commission and bonus accruals with their owners
BD6–7	Documentation, close file, audit support	Load actuals, run variance, finish commentary, pre-brief budget owners
BD8	—	Distribute the monthly reporting package
BD9–10	—	Run the monthly business review; open the reforecast (section 7)

This is not a relay where FP&A waits for accounting to finish. The teams overlap, and the reason a good FP&A group publishes on BD8 is that most of the commentary was drafted before the books locked, from preliminary data and knowledge of the business; only the numbers changed. There is no idle day in a close.

One variant to recognize: not every month gets the same treatment. A soft close is "closing the books using an abbreviated closing procedure" — AccountingTools lists accruals, intercompany eliminations, allocations, inventory counts, account reconciliations and reserve updates among the steps commonly skipped, and warns the result can be materially inaccurate. Many mid-size companies run soft monthly closes with a full hard close at quarter end, which is how they report a three-day monthly number; a continuous close pushes the work through the month instead. Under a soft close some accruals are estimated rather than supported, so the monthly number is directionally right and the quarterly one ties.

Accruals, and why FP&A cares

An accrual records an expense in the period the activity happened rather than the period the invoice arrives. Without accruals the P&L is a function of vendor billing habits. AFP names the recurring ones: bonuses, business taxes, performance fees, and campaigns straddling periods. Vena adds unbilled expenses and services rendered but not invoiced, validated against timesheets and contracts.

Accruals are where FP&A does real work during close, because accounting does not know what the business committed to and the budget owner does. Marketing signs a $180,000 campaign running November 15 to January 31; the agency invoices in February. Without an accrual, November and December look $135,000 better than they are and February looks like a disaster. So you estimate delivery — 30% of impressions in November, 45% in December, 25% in January — and accrue $54,000, $81,000 and $45,000, the method an AFP contributor describes for campaigns crossing periods: "I estimated the percentage of sales received in each month and applied those costs." When the invoice lands you true up in the current month. The accrual was never meant to be exact, only to stop the P&L lying about which month the money was spent in.

AFP's pre-close check is worth stealing: test the top ten to twenty vendor expenses for payment gaps, specifically "a payment that is historically monthly but has not been paid for 3–4 months." That is almost always a missing accrual, and far cheaper to catch before the lock than to explain after it.

The FP&A / accounting split

The clean version: accounting owns whether the number is right, FP&A owns what it means. Ramp puts the handoff at the lock — "The handoff between the teams typically happens at month-end close, when accounting's books are final and FP&A can start its analysis on confirmed numbers" — and states the dependency that will define your first year: a late close means a late forecast. But AFP is explicit that practice varies: "Some FP&A teams take no action until the books are closed; others provide feedback to accounting before, during and after the close, and some...make journal entries into the general ledger." Ask in week one.

8.2 Variance analysis

Three comparisons share the word "variance" and answer different questions. Corporate Finance Institute separates them cleanly: budget-to-actual, forecast-to-actual, and forecast-over-forecast.

• Budget vs. actual (BvA) compares results to the locked annual plan, interpreting the reasons for each difference (Wall Street Prep). This is the accountability document. It does not move all year, which is the point: it is what the department head agreed to.
• Forecast vs. actual (FvA) compares results to the latest reforecast. This is the credibility document. Missing the budget you set in October is forgivable; missing the forecast you set three weeks ago is not.
• Forecast over forecast compares two vintages of forecast for the same future period, showing whether management's view of Q4 has drifted. Nobody asks for it until they distrust the forecast, at which point it is the only report that matters.

By month four or five, BvA on a fast-moving line is largely archaeology and FvA carries the operating conversation. Scope that carefully: it holds for the internal conversation and fails for the governance one. At a PE-backed or public company the plan stays the standing scorecard all year, because it is what was committed to the sponsor, the board or the Street, and in leveraged structures covenant tests are frequently defined against the budget rather than the forecast. So the monthly business review leads with forecast versus actual, the board deck leads with plan versus actual, and you maintain both. Wall Street Prep names the other standard cuts: prior period, same period prior year, YTD actual versus budget, and full-year forecast versus budget.

Flux analysis is the same arithmetic from the accounting side: a period-over-period movement review across GL accounts, balance sheet included, run as a close control and as audit evidence that someone looked. Variance analysis is decision-oriented, flux is control-oriented, and in practice they are often the same spreadsheet with two audiences.

Materiality thresholds

Be blunt about the evidence: there is no industry-standard materiality threshold, and any source claiming one is quoting its house convention. What is consistent is the shape. First, a dual criterion pairing a dollar floor with a percentage trigger, so a 400% swing on a $2,000 line does not consume review time and a 1% miss on a $3M line does not slip through. Second, tiering by account risk rather than one flat number. Ramp publishes a representative table:

Account type	Dollar range	Percent range	Ramp's rationale
Revenue	$50,000–$100,000	5–10%	High visibility, direct profit impact
Payroll	$25,000–$50,000	5–10%	Large, predictable base with clear drivers
Operating expenses	$10,000–$25,000	10–15%	Higher natural variability
Balance sheet	$25,000–$50,000	15–20%	Major movements, not routine activity
Sensitive accounts	$5,000–$10,000	5–10%	Higher risk or regulatory scrutiny

Even the operator is contested. Ramp uses OR (investigate a variance greater than $25,000 or 10%); Planir recommends AND, flagging a variance exceeding "$25K and 10% of budget." Not a trivial difference: on a large expense base, OR buries the team in $30,000 explanations of $4M lines. Planir's procedural advice, credited to Numeric, holds regardless of the number — set thresholds with your CFO and board chair before the first report lands.

Some teams calibrate the floor against external-audit convention instead of inventing a number. The recognized single-rule benchmarks are roughly 5% of pre-tax income, 1% of total revenue, 0.5% of total assets and 1% of shareholders' equity, with published ranges of 0.5–1% of revenue, 1–2% of total assets and 5–10% of net income. Borrow them only as a sanity check: audit planning materiality asks whether a misstatement could change an investor's decision, not whether a movement deserves a management paragraph, and those benchmarks are annual while yours are monthly.

For the $40M ARR company — revenue about $3.3M a month, payroll about $2.4M — $50,000 or 5% on revenue, $25,000 or 5% on payroll, $15,000 or 10% on everything else yields fifteen to twenty five explained lines a month, which a human can read. Three rules keep that from failing predictably.

1. Test at three levels: the month, year to date, and full-year forecast versus plan. A $30,000 monthly overrun never trips a $50,000 monthly threshold and is a $360,000 full-year problem.
2. Apply thresholds at cost-center level, not only consolidated. A $40,000 miss is invisible company-wide and is 20% of a small team's budget, and the person who has to act sits at the department level.
3. Annualize in the commentary template. "$30K over on cloud spend" becomes "$360K if unaddressed," which is the sentence that gets it fixed.

Expect thresholds to ratchet: after a missed quarter the CFO tightens them, and nobody loosens them later. Two coverage targets are worth adopting personally. OnlyCFO sets accuracy at "revenue and EBITDA variance percentage within +/- 2% on a quarterly basis" for a decent-size public company (nearer 5% for smaller ones), and explanation at "90%+ coverage in variance explanations of material accounts. So if a variance is $100K, the team should list drivers for net $90K." That 90% rule converts "we looked into it" into a number you hit or miss.

Timing or permanent

This classification ends every real variance conversation, because it decides whether the full-year number moves. Pigment's definitions are standard: a timing variance "reflects a shift in when something happened rather than whether it happened, like revenue that closed in the following quarter or a hiring plan that slipped by six weeks," while a structural variance "reflects something more fundamental, like a product category losing share to a competitor or a cost input that has permanently repriced." Timing changes the phasing of the forecast; structural, or permanent, changes its total and usually triggers escalation.

The test is one question: does the full-year figure change? Two readings of the same $180,000 favorable Q2 marketing variance. Timing — the campaign slipped to Q3, full-year spend is unchanged, Q3 goes up $180,000. Say that explicitly, or finance books a phantom $180,000 of savings and somebody spends it. Permanent — a vendor contract was renegotiated and two requisitions cancelled at the start of Q2, dropping the run rate about $60,000 a month, so the $180,000 repeats in Q3 and again in Q4, full-year spend improves by $540,000, and the headcount plan changes.

The political half matters more than the arithmetic. "Timing" is the most abused word in variance commentary, because it lets an owner explain a miss without conceding anything. The discipline: require the reversal month be named. Timing with no stated reversal month is not an explanation, and next month's report should show whether the reversal happened. Make the classification and the reversal month required fields in the commentary template.

Commentary standards

Ramp's line is the whole discipline: the value is in the commentary, not the calculation, and a number without a driver is unfinished work. Quantify in both dollars and percent — "travel expense increased $45,000, or 150%, due to attendance at two industry conferences."

Planir gives a four-question template: what happened (dollars and percent), why (quantified drivers), what it means going forward (effect on the reforecast), and what we are doing about it. Their two illustrations are separate examples about different lines. For question two, a revenue miss: "Q1 revenue came in $500K (8%) below budget," attributed to rep attrition, unfavorable product mix and customer credits, with the unexplained remainder stated. For question three, an expense line: "The $50K overspend reflects the pull-forward of H2 campaign costs. No incremental budget is required. Full-year marketing is projected to land within 2% of plan, with Q2 spend trending $30K below budget based on committed contracts." That one classifies the variance as timing, names the offset, and closes the budget question in three sentences.

Three rules from AFP's variance analysis glossary. Run management by exception: attention goes only to variances past threshold. Apply a three-step review to each — root cause, controllability, and a named person responsible for the response. And analyze favorable variances, the step almost everyone skips: a favorable variance may reveal a practice worth copying, or conceal a quality problem, a hiring failure, or an unfavorable variance parked elsewhere. Underspend on engineering payroll is a hiring plan that missed.

The timing norm governing all of it, treated in full in section 14: no budget owner should learn about their own variance by reading the management report. AFP's phrasing is "no one likes surprises." You pre-brief on BD6–7, before distribution. It is the single highest-return habit in the job.

8.3 Bridges and walks

A bridge (also a walk, drawn as a waterfall chart) decomposes the change in one metric between two points into the drivers that caused it. The words are interchangeable, and it is the format the board wants. Glacier Lake Partners: "The P&L tells the board what happened. The bridge tells the board what to do about it."

The EBITDA bridge

The standard structure decomposes plan-to-actual EBITDA into five drivers: volume, price, mix, cost, and non-recurring items. Glacier Lake's worked example:

Step	Driver	Impact	Running EBITDA
Start	Planned EBITDA	—	$800,000
1	Lower service volume	($60,000)	$740,000
2	Favorable pricing	$15,000	$755,000
3	Unfavorable customer mix	($20,000)	$735,000
4	Higher vendor and labor cost	($30,000)	$705,000
5	One-time legal settlement	($25,000)	$680,000
End	Actual EBITDA (total variance ($120,000))	—	$680,000

Read that politically. A $120,000 miss looks like one failure. The bridge says five things happened: $25,000 of settlement will not repeat, $15,000 of price is good news, $20,000 of mix and $30,000 of cost are separate management problems, and the real operating issue is $60,000 of volume. Five drivers, four owners, four conversations, only one of which is about missing the number.

The ARR walk

In a subscription business the monthly revenue bridge everyone builds is not price/volume/mix. It is the ARR walk, the most-shown chart in a SaaS management review and board deck. Annual recurring revenue (ARR) is the run-rate value of active subscriptions, and CFI states the roll-forward directly: "ARR = (New ARR) + (Expansion ARR) – (Contraction ARR) – (Churned ARR)" applied to a beginning balance. Expansion is upgrades, upsells and increased usage; contraction is downgrades and seat reductions; churn is cancellations.

Component	Amount	Running ARR
Beginning ARR	—	$39,400,000
New logo	$620,000	$40,020,000
Expansion	$310,000	$40,330,000
Contraction	($140,000)	$40,190,000
Churn	($290,000)	$39,900,000
Ending ARR	$500,000 net	$39,900,000

Two rates fall out of that table. Gross revenue retention (GRR) excludes expansion and caps at 100%: (39.4 − 0.14 − 0.29) / 39.4 = 98.9%. Net revenue retention (NRR) adds expansion and is uncapped: (39.4 + 0.31 − 0.14 − 0.29) / 39.4 = 99.7%. Those are monthly rates and boards quote trailing-twelve-month figures, so state the period — compounded, 98.9% monthly GRR is roughly 88% annual. Period is not the only gap. Both rates here are taken off the walk against the whole opening base, which is right for reading the month and wrong as a reported figure: the reported construction fixes a cohort at the start of the period and excludes logos acquired inside it, so the walk rate and the board rate never tie. The two constructions and the arithmetic gap between them are in section 4 and section 9. For the benchmark, Benchmarkit's median GRR fell from 88% for calendar 2024 to 84% for calendar 2025; section 9 carries the retention benchmarks by segment with the vintage on each. Price/volume/mix is not obsolete here; it moves inside the walk, explaining the expansion and contraction bars and covering usage-, seat- and unit-priced products, COGS and gross margin.

Then the reconciliation that trips up newcomers. "ARR is not a GAAP metric. It is defined by the software industry", Ordway notes, adding that no standard definition exists and neither FASB nor the IASB gives guidance on calculating it. ARR is forward-looking run rate at a point in time; GAAP revenue is historical and recognized ratably under ASC 606, so they never tie directly. The tie runs through the deferred revenue rollforward: beginning deferred revenue plus billings minus revenue recognized equals ending deferred revenue. FP&A owns that reconciliation at close, and it is the check on whether the ARR story and the income statement are compatible. Learn the four numbers, because executives use them interchangeably and they differ: bookings (contract value signed, from the CRM), billings (invoiced), ARR (run rate from the subscription schedule), and revenue (recognized, GL).

Price, volume, and mix

Glacier Lake states the two solid formulas and the sequencing rule that keeps them honest: price effect = plan units × change in price; volume effect = incremental units × plan price; and "Use prior-period price for the volume effect and prior-period volume for the price effect." Hold one driver at base while you move the other. Two products, plan versus actual for one month:

Total variance: +$7,500 favorable, 0.2%. Under any sane threshold this line is never looked at. The bridge says otherwise:

Effect	Calculation	Amount
Price (at plan units)	Core 800 × (−$150); Enterprise 100 × (+$200)	($100,000)
Volume (total, at plan average price)	(960 − 900) × $3,666.67	$220,000
Mix	Per-product volume at plan price ($210,000 − $90,000) less total volume effect $220,000	($100,000)
Joint (price × volume cross-term)	Core 70 × (−$150); Enterprise (−10) × (+$200)	($12,500)
Total	Ties to the revenue variance	$7,500

A quiet 0.2% variance hides a $220,000 volume beat, a $100,000 discounting problem, a $100,000 shift away from the high-price Enterprise product, and a $12,500 joint effect — the arithmetic consequence of discounting Core while selling more of it. Nobody owns a joint effect, which is exactly why some templates bury it in price and others in volume. Show it, and say in the commentary that it is a cross-term rather than a decision. The first three need an owner.

This is where the threshold rule and the bridge rule appear to collide, so resolve it once: materiality governs which lines get written commentary, not which get analyzed. Revenue, gross margin, payroll and cash are decomposed every month regardless of net variance, because they are always material to the business even when they are not material to the month, and because a small net number is where offsetting drivers hide. Everything below those four is exception-driven.

Rate and volume

Rate/volume variance is the same mechanic in different vocabulary, standard in banking, treasury and labor-cost analysis: revenue analysts say price, cost and labor analysts say rate. f9finance computes rate = (actual rate − base rate) × actual volume and volume = (actual volume − base volume) × base rate, where "base" is prior year, prior month, budget or forecast. The case you re-run constantly is a wage overrun driven by hours worked rather than by an unfavorable rate. Mix should be "shown as its own line, never folded into volume". One more AFP term you will meet in manufacturing and multi-product settings: a flexible budget variance recalculates spend at actual sales volume before isolating price and usage effects.

8.4 The reporting package

Close output is a layered set of deliverables with different audiences and depth. Cadence names follow one vendor framework (Rhythms.ai); meeting lengths are illustrative, but the shape — weekly is a pulse, monthly is trajectory, quarterly is strategy — matches every other source.

Vena notes monthly reporting is standard internally while "many organizations provide the Board with detailed quarterly or annual reports, with a simplified overview of the monthly reports." Venture-backed companies tend to run monthly board updates and quarterly full meetings; the tighter the runway, the more often the board wants cash.

What the CEO and board actually read

Every source lands in the same place: boards and CEOs read the narrative and the bridge, not the trial balance. The raw variance table is appendix material. It must exist, it must tie, and almost nobody opens it. Vena is concrete about why decks fail. Do not assume shared vocabulary — "Not all board members instantly know the jargon or acronyms specific to your business." Pair dollars with percentages, and show problem-solving rather than a tour of good news. Their named failure modes: irrelevant information, jargon, overcrowded slides, and charts needing verbal explanation.

Board process is where a new director gets caught, and it is mostly unwritten. Materials go out as a pre-read, conventionally 48 to 72 hours before the meeting; late materials guarantee a meeting spent walking through numbers instead of deciding. Bad news is pre-wired: the CEO or CFO calls directors ahead so a material miss is never first heard in the room. Audit and compensation committees get their own packages. And board materials are a durable record in a way an internal MBR page is not, so keep the support for every forward-looking number you show, and expect the CFO to rewrite your draft.

The headcount report

People cost is roughly two thirds to three quarters of operating expense at a company this size, which makes headcount the largest single driver of expense variance and a standing agenda item. The standing view: budgeted versus actual ending headcount by department, hires versus plan for the month and YTD, average headcount, open requisitions with expected start dates, attrition, and backfills. Farseer's structure is the usual one — assumptions hub, employee-level roster, a waterfall of "beginning count, plus new hires, minus attritions, equals ending count", payroll summary, department drill-down.

Two disciplines make it useful. First, average headcount drives payroll expense; ending headcount does not. A December hire barely moves the year, and a department can hit plan on ending headcount while running $300,000 under on payroll because everyone started late. Farseer frames it as two questions: "Headcount variance...and payroll variance...are separate questions that deserve separate answers." Second, FP&A owns the tie between the HRIS roster, the payroll register and the GL, which will disagree over contractors, interns and people on leave.

Decompose payroll variance as a rate-and-volume problem, using the same convention as the rate/volume rule above: everything stated as actual minus plan, so a positive number is spend above plan and unfavorable, and parentheses mean below plan. Volume is (actual FTE-months − planned FTE-months) × planned average cost. Rate is actual FTE-months × (actual − planned average cost). Taking rate at actual volume is what makes the two terms sum to the total with no residual and no joint line. Worked, for an engineering group planned at 60 FTE for the month at a fully loaded $16,000 each:

Line	Calculation	Amount
Plan payroll	60 FTE-months × $16,000	$960,000
Actual payroll	56 FTE-months × $16,600	$929,600
Volume (headcount)	(56 − 60) × $16,000	($64,000)
Rate (average cost)	56 × ($16,600 − $16,000)	$33,600
Total variance	Ties to actual less plan, no plug	($30,400)

The department ended the month at 60 heads, exactly on plan, so the ending-headcount report shows nothing. Four hires the plan had starting on the 1st started on the 29th, and that alone is the $64,000. Average cost ran $600 a head over plan on the roles that did get filled, giving back $33,600 of it. Net $30,400 favorable, or a $365,000 full-year number if the slippage persists — and two conversations, one with the recruiter and one about offer levels. The usual finding, as here: most of the gap is start-date timing, not rate.

Cash and the balance sheet

At a venture-backed company the cash pages sit at the front of the board deck. Four things belong in the monthly package. Net burn versus plan, with the reconciliation from EBITDA variance to cash variance — working capital, capitalized costs and prepaid timing routinely push the two in opposite directions, and the CEO will ask why a good EBITDA month burned more cash. Runway, recomputed at close on trailing-three-month average burn. DSO and AR aging, the check on whether a revenue beat is real or merely uncollected. And the deferred revenue rollforward, tying billings, recognized revenue and the ARR walk.

KPI dashboards are the self-service layer underneath all of this; which metrics belong is section 9's territory and which tools render them is section 12's. The close-cycle point is narrower: a dashboard that disagrees with the package destroys trust in both.

Closing the loop

The report is not the deliverable; the change in behaviour is. Every material variance should leave the review with an entry in an action register: the variance, a named individual, a committed action, a date, a status. Its current state is the first agenda item next month. An action assigned to a team rather than a person, or with no date, is not an action. When the same variance appears three months running with the same explanation, change the budget or change the owner.

A real miss escalates into a get-well plan: a hiring pause or freeze, discretionary spend held at a stated approval level (often every non-payroll purchase above a threshold routed to the CFO), tightened discounting through the deal desk, sometimes an out-of-cycle reforecast. FP&A builds it and then polices it, tracking whether the freeze is actually holding — the least popular and most useful part of the job.

8.5 Public-sector translation: allotments, TALS, and the Fiscal Status Reports

WA state	Private-sector equivalent	Where it breaks
Allotment (monthly-phased biennial expenditure, revenue and FTE plan)	Monthly-phased annual operating budget, the budget of record (section 3)	The allotment sits under an appropriation that is a legal ceiling; a company budget carries no legal force. Also a 24-month biennium, revised through formal quarterly TALS packets rather than an informal monthly forecast update
OFM monthly Fiscal Status Reports	Monthly reporting package / flux report	Public by agency rather than board-restricted, and with no narrative layer — closer to a raw BvA table than to a bridge with commentary
TALS "significant variance" edit-check warnings requiring explanation	Materiality threshold triggering variance commentary	State trigger is qualitative reviewer judgment with no published number but is a hard gate; private thresholds are explicit numbers producing a soft flag
Quarterly allotment amendment with narrative explanation, due the 25th of the fiscal month	Quarterly reforecast (section 7)	A gated filing: miss the AFRS cutoff and TALS returns the packet if it contains retroactive transactions or transactions for a closed fiscal month. A private reforecast has an internal deadline and no gate
Revenue amendments required whenever the ERFC forecast is revised	Reforecast triggered by a bookings or pipeline reset (section 4)	ERFC is an independent statutory forecaster you must adopt; a company's revenue forecast is produced in-house by the CRO and sales ops, and is negotiable
Amendment triggered by a revised Caseload Forecast Council forecast or other changed key driver	Reforecast triggered by a driver reset: customer count, usage volume, headcount ramp	The caseload number is produced outside your agency and you adopt it; a company's volume drivers are argued internally and you are one of the arguers
Enterprise Reporting agency reports (rp.ofm.wa.gov)	Self-service BI and KPI dashboards (section 12)	Same purpose, very different refresh expectations — company dashboards are expected near-real-time, not monthly

9. The statements and metrics FP&A lives in

You already read financial documents for a living: a DDA allotment report has a structure, a basis of accounting, ratios people argue about, and a political meaning on every variance. This section is not teaching you to read a statement. It is teaching you a different one.

Three things change. The P&L replaces the appropriation as the object everyone watches, because there is no authority to stay inside of. A second layer of measurement sits on top of the statements — ARR, retention, CAC payback, Rule of 40 — with no state analog, because it answers a question the state never asks: is the growth engine worth funding. And almost none of that layer is defined by a standard-setter. GAAP tells you what revenue is; nothing tells you what ARR is, so your first job in a new role is finding out how this company defines its own numbers.

One running example throughout: a private B2B SaaS company at $40M ARR, 250 employees, growing 30% on ARR and 25% on recognized revenue. Every figure below ties to it.

9.1 P&L anatomy

The income statement runs in one direction, and each subtotal answers a different question:

• Revenue — earned this period under GAAP, not what you sold and not what you collected (§4 covers bookings and billings).
• Cost of goods sold (COGS), or "cost of revenue" — the cost of delivering what you sold.
• Gross profit and gross margin — the test of whether the model itself is sound, since gross margin is the share of each revenue dollar available to fund sales, engineering, and profit. A company at 45% has half the fuel per dollar of one at 80%. SaaS Capital calls it "the fundamental health metric for the business model itself."
• Operating expenses, in three departmental buckets: S&M, R&D, and G&A (finance, legal, HR, facilities, the exec office).
• Operating income, also EBIT — the number most FP&A teams are managed against. Then interest, other income and expense, taxes, and net income, which FP&A inherits from accounting and treasury.

The COGS/opex boundary is where the judgment is

The three-bucket taxonomy is near-universal; the interesting decisions sit above the gross profit line. SaaS Capital's recommended income statement deliberately puts more line items in COGS than in operating expenses, because in a subscription business "those are the costs that really matter." A well-built SaaS COGS carries hosting, support, the delivery part of customer success, implementation labor, embedded third-party software, and payment processing.

The recurring fight is customer success, which both keeps customers alive (delivery, so COGS) and drives expansion (selling, so S&M). Move it from COGS to S&M and gross margin improves while sales efficiency worsens, with no change to operating income, so a margin that jumped 300 basis points in a quarter is more often a reclassification than a win. The second boundary question is whether engineering payroll is capitalized as internal-use software or expensed as R&D (§6): capitalizing moves cost into an asset that amortizes later, and it is among the first things a sophisticated investor normalizes back out.

Gross margin and the opex stack, by the numbers

Blended gross margin hides more than it reveals when a company sells both software and services, which is why SaaS P&Ls report margin by revenue line. These medians come from Benchmarkit's 2025 SaaS Performance Metrics, covering calendar 2024 for private B2B SaaS. Vintage matters, and I flag it every time.

Revenue type	Median gross margin (CY-2024)	Why it lands there
Total (blended)	77%	Mix-dependent, so not comparable across companies with different services attach rates
Subscription only	81%	Hosting, support, delivery against recurring revenue — the number investors price
Professional services	30%	Labor against a fixed fee, run near breakeven to enable subscription revenue

The same survey puts median S&M at 37% of revenue, R&D at 34%, and G&A at 24%. Stack all four and the composite runs 77 − 37 − 34 − 24 = a negative 18% operating margin. Four medians from four distributions never describe one real company, so never present that stack as a benchmark P&L, but the direction is the point: the middle of this market loses money on purpose, at scale, funded by investors. The successor edition, Benchmarkit's 2026 B2B SaaS & AI-Native Metrics for calendar 2025, moves one line materially: median R&D fell 8 points to 27%, top quartile 22%, which alone takes that composite to roughly −11%. Median software gross margin is 80% and "stable through 4 years."

One warning on margin. Inference and model-serving costs at AI-native companies sit in cost of revenue and are variable with usage, not largely fixed the way hosting capacity is, so the 80% anchor does not automatically carry to a company whose COGS scales with tokens served. Benchmarkit's 2026 edition reports that "industry-wide AI infrastructure costs have not yet compressed software margin" at the industry level, so ask what is in cost of revenue and how it scales.

The running example, laid out

Line	$000	% of revenue
Subscription revenue	38,000	90.5%
Professional services revenue	4,000	9.5%
Total revenue	42,000	100.0%
Subscription COGS	(7,220)	17.2%
Services COGS	(2,800)	6.7%
Gross profit	31,980	76.1%
Sales & marketing	(15,540)	37.0%
Research & development	(14,280)	34.0%
General & administrative	(6,300)	15.0%
Operating income	(4,140)	(9.9%)

Subscription margin is 81% and services 30%; S&M and R&D sit at the CY-2024 medians. G&A deliberately does not: 15% against a 24% median, because a lean 250-person finance, legal, and people function is realistic and a median-G&A example would bury the EBITDA arithmetic below under an 18-point loss. It is the one line here below its benchmark, so do not carry 15% around as the market rate.

The company loses $4.1M at the operating line while growing 25%, and nobody treats that as a problem. That is the largest cultural gap between the two worlds, and not only taste: RCW 43.88.055 requires the Legislature to enact an operating budget "that leaves, in total, a positive ending fund balance in the general fund and related funds" and to keep the "projected maintenance level" inside available resources for the ensuing biennium (RCW 43.88.055). The structure this company chose on purpose is one your employer is forbidden by statute to adopt.

9.2 EBITDA, adjusted EBITDA, and the SEC's rules on non-GAAP

EBITDA strips out financing structure and non-cash charges so companies with different debt loads and asset bases can be compared on operations alone — Wall Street Prep calls it "normalized operating cash flow generated by core business activities." It is the default profitability language in PE-backed companies and lending covenants, and common enough at growth-stage software companies to expect in any board deck. Its three standard weaknesses, from the same source: it ignores capital expenditure, ignores working-capital movement, and being non-GAAP, "the lack of standardization and inconsistency by which specific items are included (or excluded) offers the management team more discretion." Adjusted EBITDA adds company-specific add-backs — restructuring, acquisition costs, litigation, and at software companies almost always stock-based compensation (SBC). Give the example $1.2M of D&A and $5.5M of SBC and watch the sign change.

Measure	$000	Margin
Operating income (GAAP)	(4,140)	(9.9%)
Add back: depreciation & amortization	1,200
EBITDA	(2,940)	(7.0%)
Add back: stock-based compensation	5,500
Adjusted EBITDA	2,560	6.1%

One company, one set of actuals, a swing from a 9.9% operating loss to a 6.1% adjusted profit, driven entirely by definition. The first question to ask about any profitability number handed to you is "adjusted for what."

Both inputs need a caveat. $5.5M of SBC is 13% of revenue, a public-company ratio; a private company records SBC at grant-date fair value off its 409A, and mid-to-high single digits is more typical. It is high here because a refresh grant was priced after a 409A step-up, and SBC is lumpy and grant-timing-driven, so a jump in it is usually a compensation event rather than a business event. The $1.2M of D&A is low against the $2.9M of capitalized software and capex in the cash bridge below, because the capitalization policy was adopted last year and amortization has not reached steady state: left alone it climbs toward $2.4M and EBITDA worsens by that amount with no operating decision behind it. Know which of your moving metrics are arithmetic.

The rules that constrain non-GAAP presentation

Once a company is public, non-GAAP stops being a matter of taste. Three instruments govern, and conflating the first two is the tell that someone has not been through an S-1. Regulation G covers any public disclosure — press release, earnings call, investor deck — requiring the most directly comparable GAAP measure alongside plus a reconciliation, and prohibiting materially misleading presentation. Item 10(e) of Regulation S-K covers SEC filings and adds the equal-or-greater-prominence rule, a statement of why management believes the measure is useful to investors, and a bar on non-GAAP measures on the face of the financial statements. The Division of Corporation Finance's Compliance and Disclosure Interpretations, last updated in December 2022, hold the real constraints:

• Equal or greater prominence (Q102.10). Item 10(e)(1)(i)(A) "requires that when a registrant presents a non-GAAP measure it must present the most directly comparable GAAP measure with equal or greater prominence." This catches releases that lead with adjusted EBITDA in bold, or chart a non-GAAP measure with no GAAP equivalent beside it.
• No excluding normal recurring cash operating expenses (Q100.01). Such a measure "could be misleading," and the staff treats expenses recurring "repeatedly or occasionally, including at irregular intervals" as recurring — closing the loophole of calling a cost one-time because that exact cost will not repeat.
• No individually tailored accounting principles (Q100.04). Adjustments "that have the effect of changing the recognition and measurement principles required to be applied in accordance with GAAP" may be misleading. Named examples: accelerating revenue recognition, changing principal-versus-agent presentation, converting an accrual measure to a cash basis.
• Tax effects shown separately (Q102.11). The tax effect of non-GAAP adjustments is its own explained line, not netted inside the add-backs.

Memorize the third: you may not convert accrual to cash and call the result a performance measure. Given how much of a budget analyst's instinct runs toward "what actually got spent," that is a habit to unlearn. None of this binds a private company legally (§13), but companies planning an IPO behave as though it does a year or two ahead of filing, and FP&A builds the reconciliation history.

9.3 The balance sheet and cash flow you will actually forecast

Below roughly $100M of revenue, most teams do not maintain a full three-statement model in production. They maintain a P&L, a headcount plan, and a cash forecast that bridges from the P&L through a handful of balance-sheet drivers. That is practice, not a rule: debt covenants, an audit-driven board, or an imminent IPO all pull the full model earlier.

Working capital and the collection cycle

Days sales outstanding (DSO) measures collection lag; days payable outstanding (DPO) = (AP ÷ COGS) × 365 measures how long you take to pay. Days inventory outstanding (DIO) completes the set for physical businesses, and the cash conversion cycle is DIO + DSO − DPO (Wall Street Prep). Software drops DIO, so working capital reduces to collections versus payments, plus deferred revenue.

The textbook DSO denominator — revenue — is wrong in a subscription business. Receivables arise from billings while revenue is recognized ratably, so a customer billed $24,000 up front creates $24,000 of AR against $2,000 of monthly revenue, and identical collections work produces wildly different DSO depending on billing frequency. The convention is to compute on invoiced revenue: "use credit sales (invoiced revenue), not cash revenue," because customers paying by card on file generate no receivable and including them understates DSO for the slow-paying part of the book. The same source recommends a rolling calculation on trailing 90 days of billing, which "smooths out month-end spikes from large invoices issued late in the quarter."

For the example: $7.0M of AR against $42.0M of revenue is 61 days; against $44.4M of billings, 58 days. The gap is small here because billing is mostly annual, but at a company shifting to quarterly terms the two separate by weeks, and that movement is an artifact of billing policy, not of collections work. Cutting DSO from 61 to 50 days permanently releases about $1.3M of cash. Two conventions before you compare anything: there is no universal target cash conversion cycle, since it "differs substantially by industry," and in SaaS the textbook DPO misleads because COGS is a small fraction of total spend, so many teams compute it against total opex.

Deferred revenue and the billings identity

Deferred revenue is cash billed or collected that cannot yet be recognized because the service has not been delivered. Under ASC 606 the five steps are: identify the contract, identify the performance obligations, determine the transaction price, allocate it, and recognize revenue as each obligation is satisfied. ASC 606 was issued in 2014 as ASU 2014-09 and took effect for public companies in 2018 and private companies in 2019 (ChartMogul walks the mechanics and dates) — settled standard, not a recent change.

Worked: a customer signs a $24,000 twelve-month contract and prepays. Bookings are $24,000 now. Cash is $24,000 now. Revenue this month is $2,000, with $22,000 in deferred revenue releasing at $2,000 a month. Sales celebrates, cash improves, the P&L barely moves — three true statements about one event, which is why the private sector needs three words for it.

The identity tying them together is Billings = Revenue + Change in Deferred Revenue. In the example's fourth quarter, revenue of $11.4M with deferred revenue rising from $19.0M to $22.5M implies billings of $14.9M. Across the full year deferred revenue rose only $2.4M, because the balance drains through three quarters and rebuilds in the heavy Q4 renewal window. The identity is exact only when every deferred dollar arrived as a billing; it breaks on unbilled receivables and contract assets, balances acquired in an acquisition, FX translation, and write-offs. Derive billings this way to sanity-check, then reconcile to the billing system before anyone quotes it.

RPO and cRPO: what replaced billings as the forward read

Deferred revenue captures only what has been invoiced. A company signing three-year contracts billed annually can have its best bookings quarter ever and show flat deferred revenue, so treating the change in deferred revenue as the leading indicator is a view a cycle out of date. ASC 606 introduced the fix. Remaining performance obligations (RPO) is revenue from signed contracts not yet recognized: it "consists of both deferred revenue (invoiced but unearned) and unbilled, non-cancelable contract amounts." Current RPO (cRPO) is the portion "expected to be invoiced and recognized in the next 12 months," and it is what analysts model and many public software companies guide to (Ordway, walking the disclosures of Okta, Sprinklr, and Palo Alto Networks).

cRPO growth against revenue growth is the cleanest forward-demand signal at any company with mixed billing terms, because it is indifferent to invoice timing in a way billings and deferred revenue are not. RPO growing while cRPO is flat means longer contracts, not more sales next year. At a private company you have neither an RPO disclosure nor a clean billings series, so build the backlog off the contract file and reconcile it to deferred revenue monthly; that reconciliation is where bad contract data surfaces.

Deferred commissions: the balance-sheet item nobody warns you about

Sales commissions are not expensed when paid. Under ASC 340-40 an incremental cost of obtaining a contract — one that "only exists because the company obtained the contract" — is capitalized and amortized over the period of benefit, and "the amortization period is not always the contract term": where new-business commission exceeds renewal commission, the cost amortizes over the expected customer relationship including anticipated renewals, commonly three to five years. A practical expedient permits immediate expensing when that period is a year or less, and it is frequently misapplied to multi-year relationships that merely renew annually. The result is a deferred contract cost asset, usually called deferred commissions.

Two consequences you meet in month one. The commission line in the P&L is an amortization figure covering several prior years of bookings, not commissions earned on this period's bookings, so a sales leader comparing it to his team's payout schedule finds they do not tie and both numbers are right. And the amortization-period judgment is the same lever as capitalizing engineering: stretch the assumed customer life and current S&M expense falls with nothing real changing. For a CAC ratio you want commissions as earned on the period's bookings; ask which is in the model, because when bookings growth changes direction the two diverge widely.

The cash flow statement and free cash flow

The cash flow statement is built by the indirect method: start at net income, add back non-cash charges, adjust for the change in working capital. D&A "is a non-cash add-back because the real cash outflow via Capex already occurred," and an increase in net working capital "reflects that there is more cash tied up in operations; thereby the cash flow decreases" (Wall Street Prep). Free cash flow is operating cash flow minus capital expenditure, and FCF margin is FCF ÷ revenue. Two conventions to nail down: most software companies treat capitalized software as capex here and a few do not, and unlevered FCF (before interest) is the version used in valuation work. FCF displaced adjusted EBITDA as the public-software yardstick after 2022, which is why Bessemer's Rule of X below is built on FCF margin.

Free cash flow bridge (FY)	$000
Adjusted EBITDA	2,560
Cash interest and taxes	(300)
Increase in accounts receivable	(2,000)
Increase in deferred revenue	2,400
Increase in prepaid and other assets	(600)
Increase in deferred commissions	(900)
Decrease in accrued compensation	(700)
Cash from operations	460
Capitalized software	(2,300)
Property and equipment	(600)
Free cash flow	(2,440)

FCF margin is (5.8%). Note where the bridge starts: adjusted EBITDA, which already excludes $5.5M of stock compensation. SBC never leaves the bank, so it is invisible in every cash measure on this page. FCF is honest about working capital and capital spending and exactly as silent about dilution as adjusted EBITDA. The two correct different lies.

Burn and runway

Gross burn is total monthly cash out. Net burn is cash out minus cash in. Runway is cash ÷ net burn, in months. Neither is a GAAP concept; they are the operating vocabulary of any company not yet self-funding. The example reports +$2.6M of adjusted EBITDA and still consumes $2.4M of cash, an average net burn near $200K a month against $18M in the bank. The shape is what matters: this company collects most of its billings in Q4 and pays prior-year bonuses in Q1, so its worst month burns several times its average, and runway computed as cash ÷ average burn would miss that trough entirely. Forecast it off the billing calendar and the cash bridge, not off one-twelfth of the plan. Scenario and downside runway belong to §7.

9.4 The metrics layer

The statements say what happened. The metrics layer says whether the machine that produced it deserves more capital. ARR and MRR are the contracted run-rate of subscription revenue, and they are not GAAP revenue: ARR is a point-in-time snapshot of what today's book would produce over the next twelve months, while revenue is what was earned in a past period. The ARR waterfall is built in §4.

What consumption pricing does to ARR

That definition assumes a contracted run-rate exists. Under usage-based pricing it often does not, and companies substitute: annualizing the last month (Datadog, Sumo Logic, DigitalOcean), annualizing a trailing window (MongoDB uses the prior 90 days for Atlas usage and 30 for self-serve; Dynatrace annualizes daily revenue), or counting only contracted entitlements (Cloudera). Several of the best-known usage-priced companies — Amazon, Snowflake, Twilio, Fastly — report no ARR-style metric at all. There is no standard: consumption variability "requires companies to establish clear and transparent ARR definitions, which often differ significantly across organizations" (Ordway).

Three consequences. A mixed book is reported as committed ARR plus variable overage on separate lines, so a customer with a $150K commitment consuming at $220K annualized is $150K contracted and $70K variable; collapse them and you destroy the distinction the board needs. Retention moves with consumption month to month rather than at renewal events, so it runs higher and far more volatile — Benchmarkit's 2026 edition puts median NRR at 108% for usage-priced against 98% for seat-priced companies. And a company changing pricing models mid-year restates ARR; the restatement, not the growth rate, is the interesting conversation.

Who actually owns these numbers

None of this is automatically yours. Past roughly $30M ARR, RevOps or Sales Ops usually owns bookings, ARR, pipeline, and the retention cohorts, and they live in the CRM; FP&A owns the P&L, the cash forecast, and the reconciliation between the two worlds. Three candidate sources of truth disagree by construction: the CRM says what was sold, the billing system (Zuora, Stripe Billing, Maxio, Chargebee) says what was invoiced, and the general ledger says what was recognized. An outside hire who assumes the metrics layer is his loses a turf fight in month one; one who owns the reconciliation ends up stronger, because that is where discrepancies surface and whoever holds it decides which number is the error. Watch for two patterns, both common and neither fraud: a definition quietly changed mid-year to make a board target reachable, and prior periods restated on it with no footnote.

Retention: NRR and GRR

Both are cohort measures: take only the customers who existed twelve months ago and compare what they pay now to what they paid then. Net revenue retention (NRR) includes upsell, cross-sell, and price increases, so it can exceed 100%. Gross revenue retention (GRR) excludes all three, capturing pure churn and contraction, and cannot exceed 100%. Customers acquired during the year are excluded from both, which is the usual way an amateur calculation goes wrong. Worked: a cohort worth $30.0M of ARR a year ago added $3.6M of expansion and lost $0.9M to downgrades and $0.9M to churn. NRR = 31.8 ÷ 30.0 = 106%; GRR = 28.2 ÷ 30.0 = 94%. The 12-point spread is the expansion engine: at those rates the company grows even if it never sells another new logo.

Benchmarkit's CY-2024 medians are NRR 101% and GRR 88%; the CY-2025 edition moves median GRR down to 84%, with the 75th percentile falling from 95% to 91%. Segment matters more than the headline. SaaS Capital's 2025 retention survey reports by contract size rather than one number, with the $25K–$50K ACV band at a median NRR of 102% across a 97%-to-111% quartile range, higher contract values correlating with higher NRR. The practical bar: mid-90s NRR is normal at an SMB-focused company, a six-figure-ACV enterprise company is expected materially above 110%, and the same 88% GRR is unremarkable in SMB and alarming in enterprise.

The most important finding for planning is not a retention number: expansion ARR contributes 40% of all new ARR at the median, unchanged between CY-2024 and CY-2025, rising to 58% between $50M and $100M ARR and 67% above $100M — the latter on a limited sample the survey flags as such. Past roughly $50M most growth comes from the installed base, which should change how you argue about S&M allocation in an AOP.

Acquisition efficiency: CAC ratio, CAC payback, LTV

Customer acquisition cost is usually expressed as a CAC ratio: S&M dollars per dollar of new ARR. Benchmarkit's CY-2024 medians are $2.00 for new customers — up 14% year over year, so new business got more expensive — and $1.00 for expansion. The blended ratio moved the other way, falling $0.19 (12%) in 2024 while remaining 10% above 2022, and the CY-2025 edition puts the blended level at $1.30. The combination is the useful part: new-logo acquisition got more expensive while blended acquisition got cheaper, which is a mix effect — the expansion shift above, arriving in the CAC line. When a CRO presents an improving blended ratio as a win, ask whether new-logo CAC improved or the mix simply moved.

CAC payback converts the ratio to months: [S&M ÷ (net new ARR × gross margin %)] × 12. For the example — $3.9M of quarterly S&M, $3.0M of net new ARR, 76.1% margin — that is 3.9 ÷ (3.0 × 0.761) = 1.71 years, or 20 months. That is a blended payback, since net new ARR includes expansion; on new-logo ARR against new-logo S&M it would be considerably longer.

LTV is commonly (monthly revenue per account × gross margin %) ÷ monthly churn rate, and LTV:CAC compares it to fully loaded acquisition cost, 3:1 being the conventional minimum. Treat it with more suspicion than anything else here: it divides by an estimated churn rate, so it is wildly sensitive at low churn, and the standard error is computing CAC from program spend alone, excluding sales salaries, commissions, tooling, and overhead. Know how to compute it, because it appears in early-stage decks and interviews, but do not propose it as the AOP efficiency metric: boards past Series B steer on CAC payback plus NRR, which carry the same information from observed data.

Magic number, burn multiple, Rule of 40

Magic number, developed by Scale Venture Partners to compare public SaaS companies on reported GAAP data, is [(current-quarter revenue − prior-quarter revenue) × 4] ÷ prior-quarter S&M. Below 0.75 is inefficient, 0.75 to 1.0 moderately efficient, above 1.0 very efficient and a signal to spend more. For our company — Q4 revenue $11.4M, Q3 $10.8M, prior-quarter S&M $3.9M — that is (0.6 × 4) ÷ 3.9 = 0.62, against a CY-2025 median above 1.0. Burn multiple, popularized by David Sacks of Craft Ventures, is net burn ÷ net new ARR: our example burns $2.44M and adds $12.0M of net new ARR, so 0.2x. Lower is better; the source gives a progression from 2.0x to 0.5x as a company scales but no numeric grade bands, so read under 1.0x as efficient rather than as a graded scale.

Rule of 40 is revenue growth % + profit margin % ≥ 40 — the most quoted and least specified metric in software, because "profit margin" is never pinned down. Three ways:

Margin definition used	Growth	Margin	Score
Adjusted EBITDA margin	25%	6.1%	31
Free cash flow margin	25%	(5.8%)	19
GAAP operating margin	25%	(9.9%)	15

One company, one year, scoring between 15 and 31 on a definition nobody standardized. Assume every Rule of 40 figure handed to you uses the most flattering available margin until proven otherwise. Benchmarkit's CY-2025 median score is 25, up from 15 the prior year, top quartile 43 — so 31 beats the private-market median and still fails the rule it is named after.

For public-market calibration, Bessemer reported the BVP Cloud Index averaging roughly 31% on Rule of 40 as of late 2023, with top-decile cloud companies near 48%. Date that figure whenever you use it. Bessemer also argues the rule is structurally wrong at scale, because "a margin increase has a linear impact on value, while a growth rate increase can have a compounding impact on value," and proposes the Rule of X: (growth × multiplier) + FCF margin, the multiplier around 2x private and 2-3x public. At 2x our company scores (25 × 2) − 5.8 = 44.2, against a Rule of X index average near 50 and a top decile near 80 on that same snapshot. The two scores are read against their own benchmarks, not against each other: the company is below average on both, but the gap narrows under Rule of X, which is Bessemer's point about growth being under-credited. Expect CEOs to prefer whichever one they win on.

The example sits at $40M ARR over 250 people, or $160K per FTE, below the $175K median — the kind of gap that gets a headcount plan questioned in an AOP review before anyone looks at a department.

9.5 Accrual, cash, and the commitment layer

Accrual accounting recognizes revenue when earned and expense when incurred, regardless of cash; that is the basis of every GAAP statement you will forecast against. Cash accounting records both when money moves, and in a private company it survives only in the cash forecast and the runway calculation. The gap is where the close lives: when a vendor delivers $60,000 of services in March and invoices in April, accounting books a $60,000 accrual in March. FP&A's job during close is largely catching the accruals accounting does not know about, since the budget owner knows work was delivered and accounting knows only what was invoiced (§8). Four mechanics generate most of the variance noise in your first few closes, and none is about anyone's spending:

• Accruals reverse. Booked in March, reversed April 1, real invoice landing in April. Over-accrue by $40,000 in March and April shows a $40,000 favorable variance nobody earned. Telling budget owners which of their variances are reversals is part of the job.
• Prepaids are the mirror image. Pay a $120,000 annual software contract in January and the expense amortizes at $10,000 a month. Annual vendor contracts live here, and the amortization schedule is a forecast input you own.
• Materiality thresholds are real and local. Most companies will not book an accrual below a floor — $5,000 and $10,000 are common at this scale — or the close would never finish. Accounting sets it; find out what it is, because below that line timing noise is simply accepted.
• The three-way match generates accruals for you. Matching purchase order to goods receipt to invoice means a receipt with no invoice creates a system accrual automatically. Where receiving discipline is poor that fails silently, and the manual accrual you chase in close week is all that stands between the P&L and a missing month of cost.

Which makes the close-week accrual confirmation a named deliverable: an email to every budget owner asking what was delivered but not yet invoiced, on a fixed day, deadline before the ledger closes. Accrued bonus and accrued commissions are the two largest recurring accrual lines you forecast yourself rather than collect.

Alongside accruals sits a third category that is neither cash nor expense: the commitment. Sign a $180,000 consulting contract and issue a purchase order and nothing is delivered, no invoice exists, nothing hits the general ledger. That $180,000 appears in the ERP as an open PO, and a good FP&A team pulls the open-PO report before closing a forecast, but it is tracked off-ledger for internal control only: no journal entry, no balance-sheet line, no effect on reported financials. That distinction is the next subsection, and the one place your government training is an outright advantage.

9.6 Public-sector translation: fund accounting to GAAP

This is where your existing fluency is most likely to mislead you, because the vocabulary overlaps and the mechanics do not.

Basis of accounting

WA state does not run on one basis. SAAM 80.20.50 is explicit: governmental funds use the flow of current financial resources focus and the modified accrual basis, recognizing revenues when "available and measurable" and expenditures "generally recognized when the fund liability is incurred, if measurable"; proprietary and fiduciary funds use the flow of economic resources focus and full accrual. Your DDA work sits on the governmental side, so your instincts are modified-accrual instincts.

Three consequences on full accrual. Long-term assets and liabilities live in the operating entity instead of being excluded from the fund, so depreciation is a real line on your P&L. The "available" criterion disappears: revenue is recognized when earned even if collection is a year out, which is what deferred revenue and DSO exist to manage. And expenditure becomes expense — an expenditure is a use of current financial resources, an expense is consumption of economic resources in the period it benefits. A three-year software license is one expenditure and three years of expense.

Lapse versus underspend: the same fact, opposite politics

In WA, "all appropriations shall lapse at the end of the fiscal period for which the appropriations are made to the extent that they have not been expended or lawfully obligated" (RCW 43.88.140) — which is why the encumbrance above matters, since a lawful obligation is what keeps authority alive. What expires is the authority, not a cash balance: for General Fund-State the money never left the treasury account, so nothing is sent back. You know better than I do what that does to behavior late in a biennium.

In a company the identical fact — budget approved, money not spent — is a favorable variance. Nothing expires, because no external authority was granted. The cash stays on the balance sheet, extends runway, and improves margin. Nobody sweeps it, and nobody has to be persuaded to let you keep it, because the question never arises.

Two cautions. A favorable variance is not automatically praised: "we underspent because we could not hire the three engineers in the plan" is a delivery problem wearing a favorable variance as a disguise. And next year's budget is built from somewhere: if the baseline comes off actuals rather than plan, an underspend quietly lowers your starting point — the same pressure that makes agencies spend down a lapse, arriving by a different mechanism. It is weaker, it is a budgeting convention rather than a legal reversion, and it is negotiable in a way a lapse never is (§3).

10. Long-range planning and strategic finance

Sections 3 through 8 are the operating rhythm: build a plan, lock it, reforecast, explain the variance. This section is the work above that rhythm — where the business needs to be in five years, what gets funded to get there, what a specific investment is worth, what you tell the board or the market, and what happens when you buy someone. Whether any of it is a separate department is a question of triggers, not headcount. Strategic finance splits off from FP&A when the recurring board, fundraise, and deal work fills a person, which at most software companies happens somewhere past $50-100M of ARR. Corporate development appears when the company has decided to buy things, which can be true at 300 people and false at 3,000. Investor relations appears at the IPO. Below those triggers all of this is the head of FP&A, the CFO, and maybe one person titled "strategic finance" wearing different hats in different meetings, and "strategic finance" on a job description often means the FP&A job with board decks attached (section 1).

The 3-5 year model

The long-range plan (LRP; also "the three-year plan," "the five-year model," "the strategic plan model") is a multi-year projection built at far lower granularity than the annual operating plan, tied to the company's stated strategy, and used to test whether the trajectory the company claims is arithmetically supportable. At a PE-backed company the LRP is built to the sponsor's underwriting case. Do not call it the LBO model: an LBO model is the sponsor's own transaction and returns model — purchase price assumptions, a sources and uses table, debt schedules, exit assumptions, and the resulting IRR and multiple of money to the equity. The two are separate documents that get held to each other, and mixing the names is a tell.

Know going in that this is standard vocabulary with no authoritative definition. Practitioner and vendor material treats the LRP functionally, as multi-year targets and guardrails set under uncertainty, without fixing a horizon or a structure; Vena's 2022 piece is typical in that it never defines the term or names a horizon, and its actual contrast is between an annual plan that "is set in stone and doesn't change often" and a forecast the team is "constantly updating." What follows is common practice, not a standard. Section 2 owns where the LRP sits in the calendar.

Dimension	Annual operating plan	Long-range plan
Horizon	12 months	3 years, or 5 if pre-IPO or PE-owned
Time grain	Monthly, weekly for cash	Annual
Revenue detail	By segment, rep, and product	By segment or product line only
Headcount detail	Position-level, by month, with start dates	By function, year-end totals
Opex detail	Cost center by GL account	Department totals, often as a percent of revenue
Built by	Every budget owner, coordinated by FP&A	Corporate FP&A or strategic finance, with the CEO
Status once approved	Budget of record; performance is measured against it	A direction at venture-backed companies, where nobody is graded on year 3. Not true under PE ownership, where the underwriting case is the exit math and management equity pays off against it (section 13)

The structural difference is granularity and who supplies it. The AOP is hybrid — a top-down envelope with a bottom-up build inside it and a reconciliation gap in between (section 3) — and every budget owner has a line in it. The LRP is top-down only. Nobody asks a marketing director what they will spend in 2029. A typical LRP runs off eight to fifteen drivers — ARR growth by segment, net revenue retention, gross margin trajectory, S&M / R&D / G&A as percentages of revenue, revenue per employee, a cash conversion assumption — and derives everything else.

The glideslope: what an LRP is actually for

The clearest published articulation of the LRP's real job is Dave Kellogg's Rule of 40 glideslope planning (2019). Rather than asking when you hit a Rule of 40 score of 40, you plan the path: what score you post each year on the way. His example is a $30M ARR company in year 5 planning toward an IPO in years 8-9, with a glideslope of -10%, then 0%, then +5%, which requires spending less for each incremental dollar of revenue every year. The base case has the company spending $1.38 to get an incremental $1 of revenue, against a median of $1.32 to acquire $1.00 of ARR.

The stress test is the valuable part. Hold expenses flat and miss revenue, and the score collapses to -32% then -42%, and the company ends up "needing to fund $42.4M more in operating losses than the original plan." Holding the planned glideslope through that miss would require efficiency to jump instantly from $1.38 to $0.49, which does not happen. The lesson generalizes: a multi-year plan is not a forecast, it is a set of constraints telling you in advance how much expense flexibility the trajectory needs to survive a bad year.

On the $40M ARR, 250-person company this guide uses as its default, an LRP for a Series D conversation starts from the mix that company actually posts — 76.1% gross margin, S&M 37% of revenue, R&D 34%, G&A 15%, operating margin -9.9% (section 9) — and slopes each line from there: 40% ARR growth decelerating to 30% and 25%, NRR at 112%, gross margin 76.1% to 80%, S&M 37% to 33%, R&D 34% to 28%, G&A 15% to 13%. That produces $40M to $56M to $72.8M to $91M of ARR and an operating margin moving from -9.9% to about +6%. Year zero has to be the P&L you closed, not a rounder number that makes the slope look better; an LRP whose first column does not tie to the actuals is the first thing a diligence reader catches. Note also where the room is. Operating expense falls twelve points of revenue across the plan, and G&A at 15% is already well under the 24% median, so it supplies two of them and the other ten have to come out of S&M and R&D. Say that yourself rather than letting a reader find it. Every number is a claim someone will interrogate, and the interrogation is the point. The model is not a prediction; it is the arithmetic behind the story.

Scenarios, and where the year-five number really comes from

Nothing above says where the drivers come from, and the honest answer is the thing nobody writes down. Two conventions structure the work, and neither is documented anywhere I could cite. Treat both as practitioner behavior you should expect rather than as sourced practice.

First, the scenario set. An LRP is normally cut as a base case (also "management case" — the trajectory you will actually run the company to), an upside case, and a downside case, with a separate investor case assembled at a fundraise. Under PE ownership the sponsor's underwriting case sits above all of them, and that is what management is measured on. Know which levers move between cases and which do not. Headcount timing, discretionary program spend, and gross margin usually flex. Net revenue retention and sales productivity are where cases quietly get fudged, because a point of NRR compounds through every later year and nobody can falsify it inside the model. If you want to find the soft spot in someone else's LRP in ten minutes, compare the NRR and the sales productivity assumption against the last eight quarters of actuals.

Second, and more important on your first cycle: you will frequently be handed the endpoint before you build the model. The CEO or the board arrives with an ARR number, an IPO window, or a valuation multiple, and the real assignment is to find a driver set that reaches it. Refusing is not the professional response, and neither is quietly solving for the answer. The professional response is explicit decomposition: "to reach $250M in year five we need NRR at 120% and sales productivity 30% above anything we have ever posted, so this plan is a bet on one of those two, and here is which." That sentence is the job. It converts a number someone wants into two assumptions someone has to own.

The related fact is that the board-deck LRP and the internal operating plan are often different files. The delta is the conservatism the CFO is holding, which is a governance choice rather than a deception, and it is the same instinct that produces a beatable annual plan and a guided range below the internal forecast. Find out which model you are being asked to build before you build it, and do not let the numbers cross-contaminate. Publishing the internal number in a board appendix is a mistake you make once.

The LRP is usually refreshed once a year alongside AOP kickoff and re-cut ad hoc for a board conversation, a fundraise, or a deal. Its link to the non-financial strategic plan — the mission, vision and measurable objectives that precede any strategy, then CFI's "Strategy Formulation," "Strategy Implementation" and "Strategy Evaluation" loop — runs through the drivers. "Move upmarket to enterprise" shows up as ASP rising, sales cycles lengthening, S&M efficiency worsening before it improves, gross margin holding. When the strategy narrative and the driver set disagree, one of them is wrong, and finding that is the highest-value thing FP&A does here.

Capital allocation and the investment case

Capital allocation is the decision about deploying finite resources across competing uses: organic growth, R&D, acquisitions, debt paydown, dividends, buybacks. Vena's framing is worth keeping — it "is not just about achieving higher returns with less spending. It's about measuring the return on internal investments and maintaining the cash flow that enables those investments." At a growth-stage private company the menu is short, nearly everything goes to growth, and the real decision is which bets in what order. An investment case (also "business case" or "investment request") justifies one incremental ask. Section 3 covers how these get raised and negotiated inside the AOP; here is the math they carry.

• NPV (net present value) — "the value of all future cash flows (positive and negative) over the entire life of an investment discounted to the present." Positive creates value, negative destroys it. CFI's illustration: $10,000 a year for 10 years at a 10% discount rate is worth $61,446 today.
• IRR (internal rate of return) — "the discount rate that makes the net present value (NPV) of a project zero." Accept above the hurdle, reject below. Two real weaknesses: it hides scale, and any project whose cash flows change sign more than once (a second investment tranche, an end-of-life cost) can have several mathematically valid IRRs or none. You will also hear that IRR assumes interim cash reinvests at the project's own rate; that framing is contested in academic finance and is not worth defending in a meeting. The practical objection at a growth company is simpler: a five-year IRR is dominated by terminal-year assumptions nobody believes.
• Payback period — how long it takes to recoup the investment. A screen, not a decision rule; CFI notes it "does not explicitly discount for the risk and opportunity costs associated with the project."
• Hurdle rate — the minimum acceptable rate of return. Most companies anchor it to WACC, on the logic that they could buy back their own shares and earn WACC instead, then risk-adjust upward per project.

A worked case

The $40M ARR company proposes an EMEA sales team: six enterprise AEs at roughly $310k fully loaded each (base, commission at target, benefits, payroll tax, T&E — load factors are in section 5), plus $740k for a UK entity, localization, regional marketing, and a solutions engineer. Year 1 costs $2.6M against almost no in-year revenue, because of ramp.

Year	Incremental FCF ($M)	Discounted at 10% ($M)	Cumulative undiscounted ($M)
1	(2.60)	(2.36)	(2.60)
2	(0.90)	(0.74)	(3.50)
3	1.80	1.35	(1.70)
4	3.40	2.32	1.70
5	4.60	2.86	6.30
Total	6.30	3.42

NPV at 10% is $3.42M, IRR about 42%, and payback lands mid-year-4 (cumulative cash is negative $1.70M entering year 4 against a $3.40M contribution), so 3.5 years. Against a hurdle near the cost of capital this clears easily, which is the practical problem with running growth investments through a WACC hurdle: almost everything clears. Growth companies therefore set project hurdles far above WACC, and payback does more work than NPV. At a cash-tight company — two years of runway, say — a 3.5-year payback is unfundable no matter what the NPV says. Our default company is not cash-tight (section 9) and can carry it, which is the useful lesson: the screen that decides here is payback against the cash position, not the hurdle rate.

Cost of capital and hurdle rates, benchmarked

The hard anchor for cost of capital is Aswath Damodaran's industry cost-of-capital dataset, January 2026 vintage, US firms.

Industry	Firms	WACC
Software (system & application)	309	9.34%
Software (internet)	29	10.66%
Computer services	64	7.83%
Total US market	5,994	6.96%

So a software company's cost of capital floor is roughly 9-11%. The hurdle rates companies actually apply sit far above that, and have for thirty years. Sharpe and Suarez, working from the Duke/CFO Magazine Global Business Outlook survey, report that in the second quarter of 2012 "the average respondent reported a hurdle rate of 14.1%, while the median value was 13.4%," against a BBB corporate bond yield near 4%; in the subsample of nonfinancial firms with sales above $100 million, "both the mean and median hurdle rates are even a bit higher, at 15%" (Federal Reserve FEDS 2014-02). Two findings from the same paper matter more than the level. Hurdles barely move: across five surveys from 1985 to 2011 the average shows "no evidence of a downward shift in concert with the general decline in interest rates," the mode is a constant 15% throughout, and the authors read that as evidence hurdle rates "are determined using rough rules of thumb, rather than fine-tuned calculations." And they rise with growth: sorting firms into quintiles by expected revenue growth gives group median hurdle rates of "12%, 12%, 14.5%, 15%, and 18.5%." A high-growth company demanding 18% on an internal project while its own cost of capital is 10% is behaving exactly like its peers. The gap is a rationing device, not a valuation.

What is genuinely unsettled is cadence rather than math. BCG's Ulrich Pidun and Sebastian Stange, relayed by Planful, recommend addressing cognitive biases by setting up a committee to review capital allocation proposals and bundling projects together to be discussed "quarterly, semi-annually, or annually," plus establishing accountability by testing alternatives in advance and instituting feedback loops from the people who implement. The same page relays an EY figure that 56% of CFOs worldwide said their capital allocation strategy needs to be completely rethought, and a McKinsey finding that companies with diversified investments "will be worth around 40% more than companies that don't." Both statistics are one hop removed from the primary studies; treat them as directional.

How the decision actually gets made

Ask "what's our hurdle rate" at a new employer and a common honest answer is that nobody set one. The mechanism in most growth companies is not a threshold test at all, and knowing that is worth more than knowing the formulas.

• The envelope comes first. The CEO and CFO set net headcount adds and a burn ceiling for the year, usually before any individual case is written. Everything downstream is competition for that envelope.
• Asks are ranked against each other, not against a hurdle. New money almost always comes from killing or shrinking something, so reallocation is the real allocation device. The question in the room is "instead of what," and a case that does not answer it loses to one that does.
• Most approvals are conditional. Stage-gated funding is close to universal: funded through the pilot, gate at the end of Q2 on two reference logos, named owner. Write the gate and the owner into the case yourself; if you do not, the approval is unconditional by default and you have lost your only leverage.
• The cost side is real and the benefit side is fiction. Every case understates integration and ramp and overstates the revenue. Your leverage is owning the assumptions and the tracking, not the verdict.
• Keep the register. Maintain a one-page list of funded cases with the promised metric, the gate date, and what actually happened. At most companies nobody else will, and being the person who has it is the fastest credibility a new director can buy.

When the asset is compute

At AI-adjacent companies the largest allocation decisions on the table are no longer sales teams. They are multi-year GPU or cloud reservations, model licensing versus self-hosting, and capacity bought ahead of demand. The same NPV and payback machinery applies, with three differences worth naming out loud. The commitment is take-or-pay and largely irreversible, so the downside case is the whole analysis rather than a sensitivity tab. The operating metric is utilization against committed capacity, not IRR, and it is the number the board will ask for. And because the spend lands in COGS rather than opex, the gross margin glideslope in your LRP becomes an output of this decision instead of an assumption you get to set (section 6 covers the COGS mechanics). Even asset-light software now guides capital expenditure explicitly: Salesforce's August 2026 results release carries full-year capex guidance of "approximately 1.5% of revenue" as a line item alongside revenue and margin. I have no citable source for the internal practice around compute commitments; this is description, not documentation.

Pricing decisions

Honest disclosure: this is the thinnest-sourced subtopic here. I could not find a citable practitioner source on FP&A's specific role in pricing — elasticity modeling, willingness-to-pay work, margin sign-off, or who formally owns the call. Available material treats pricing from a product and growth angle. Three useful data points. Paddle reports that among Fortune 500 companies "fewer than 5% have functions dedicated to setting the best price possible" and that most businesses "spend less than 10 hours per year thinking about pricing." Paddle also defines the value metric as "essentially what you charge for. For example: per seat, per 1,000 visits, per CPA, per GB used, per transaction," and claims companies that price on a value metric grow "at double the rate with half the churn and 2x the expansion revenue" against flat-fee pricing. ChartMogul supplies the standard corrective — price to customer-perceived value, not internal cost, because "your customers don't care how much you've invested in certain features of your product" — and lists the metrics companies actually scale on: internal users, active customers, API requests, storage.

What FP&A contributes: the breakeven, not the price

Nobody knows the elasticity and no model will produce it. The move is to invert the question so the argument lands on a number someone has an opinion about. Say the company writes $9M of new-logo ARR a year and product proposes a 10% list increase while sales predicts win rates will fall. With 15% of the increase given back in discount, realized price rises 8.5%, so revenue is unchanged at a 7.8% drop in win rate (1 minus 1 divided by 1.085). Call that the revenue-neutral breakeven and nothing more. On gross profit the tolerable drop is higher, because a lost deal also sheds commission, onboarding, and cost to serve, and that gap widens as gross margin falls, which is why this framing travels badly outside software. On lifetime value it is lower, because each lost logo also forfeits future expansion at your NRR. Present it as a range with those two bounds named, and you will not be corrected by the CRO.

The installed base is where the money is

The new-logo analysis is the one people run and the smaller half of the decision. On the default $40M ARR company, roughly $31M of base comes up for renewal over the year against $9M of new-logo ARR. A 7% renewal uplift on the base is worth about $2.17M; the 8.5% realized increase on new logos is worth about $0.77M. The base is worth roughly three times the new-business case, and it is where the churn risk sits. Run the same inversion: the uplift is revenue-neutral at 6.5% of incremental gross churn on the affected base (1 minus 1 divided by 1.07), so if account management thinks the increase costs you more than six and a half points of logo retention, it loses money in year one before anyone argues about elasticity. The uplift also moves net revenue retention, and the number you quote depends entirely on the denominator — get that right the first time you are asked, because quoting the flattering version is a known way to lose a room. Seven points is the move on the renewing slice only. NRR is measured against the full starting ARR of the cohort (section 9 defines it), so $2.17M of price-driven expansion against a $40M starting base is about five and a half points, not seven. Say which base you are on. Even at five and a half, price is the cheapest NRR lever available: no product ships, no rep sells anything, and the expansion arrives on renewal dates already on the calendar — which is also why it is the lever most likely to be over-pulled. Grandfathering, contractual escalators, and staged rollouts by cohort are the usual instruments, and each one is a different arithmetic problem you will be asked to run.

By convention rather than documented practice, FP&A owns the enforcement artifacts rather than the price: the discount approval matrix and floor price, the deal desk escalation thresholds, margin sign-off on non-standard terms, and the price realization reporting that says whether the increase actually landed — list versus realized ASP by cohort, which is the only way to find out that a 10% increase became a 2% increase in the field. Downstream, FP&A re-cuts the ARR waterfall assumptions the change implies (section 4).

Seats to consumption

The live pricing question at AI-adjacent companies is the shift from per-seat to usage-based or hybrid pricing, and it is a finance problem more than a product one. Four consequences to have ready. Usage revenue does not behave like contract revenue in the plan: a consumption forecast behaves much more like a caseload forecast, driven by volume and utilization rather than by signed contracts, which is territory you already know. ARR as a definition strains, because annualizing a variable usage month is a choice rather than a fact, and the definition you pick becomes the number the board tracks (section 4). Remaining performance obligation coverage falls, because uncommitted usage is not a performance obligation, so the forward visibility that cRPO gives a seat-based business partly disappears. And gross margin couples directly to price, because the priced unit is one you buy from a compute vendor, so a pricing decision is now a margin decision. The pattern most companies land on is hybrid: a committed platform floor that preserves forecastability plus metered overage that captures upside. That is convention as of 2026, not a sourced claim.

M&A modeling

Corporate development is "the group at a corporation responsible for strategic decisions to grow and restructure its business" — sourcing, valuation, negotiation, integration, plus partnerships and portfolio work. Mergers & Inquisitions frames it as buy-side M&A done in-house rather than through bankers. The split from FP&A is clean in theory (Corp Dev looks outward at deals, FP&A inward at the plan) and blurry in practice, because at any company that has not decided to be acquisitive there is no Corp Dev team and the head of FP&A builds the model.

The core artifact at scale is the merger model, which measures "the estimated accretion or dilution to an acquirer's earnings per share (EPS) from the impact of an M&A transaction." Build sequence: set offer value per share, structure the consideration (cash versus stock), estimate financing fees, interest expense, share issuance, synergies and transaction costs, run purchase price allocation for goodwill and incremental D&A, combine standalone into consolidated earnings before tax, then divide pro forma net income by pro forma diluted shares. Wall Street Prep's illustration has $4.00 standalone EPS becoming $4.25 pro forma, which the page reports as 6.4% accretion; the displayed EPS figures are rounded to the cent, so redoing it on those two numbers gives 6.25%. Synergies are "the incremental revenue generation or cost savings from the transaction, which we'll net against the costs of the synergies, i.e. the losses from the integration process and shutting down facilities" — and the netting is where cases go wrong, because integration cost is certain and revenue synergy is not.

What the deal actually looks like at your scale

Accretion/dilution is an EPS test, so it means something only at an acquirer with earnings. At an unprofitable growth-stage buyer the equivalent questions are what the deal does to ARR growth, burn multiple, and runway. The deal itself is usually an acqui-hire or a tuck-in, and the governing analysis is build versus buy: acquisition cost and integration time against the fully loaded cost and elapsed time of hiring that team yourself, at your actual offer-accept rate. Artifacts you should expect to touch, none of which appear in a merger-model tutorial:

• Sources and uses — where the money comes from and what it pays for, including fees and any cash left on the target's balance sheet.
• Quality of earnings (QofE) — third-party diligence on whether reported revenue and EBITDA are real. FP&A usually supplies the underlying data on one side or the other, and the QofE adjustments become your opening balance assumptions.
• Earnouts and retention pools — the earnout accrues and hits your P&L for two years, and the retention pool shows up as opex nobody budgeted. Both are recurring forecasting headaches and both get forgotten in the case.
• Escrow, holdback, and the working capital true-up — cash that leaves later, on a schedule the deal lawyers set.
• Integration cost — always understated in the case, and the line your post-mortem will be about.

On purchase accounting, one piece of common knowledge is out of date and worth unlearning. Step-up D&A on acquired intangibles still hits operating income and is the main reason a deal stays GAAP-dilutive for years. The deferred revenue haircut that used to suppress post-close reported revenue is largely gone. ASU 2021-08 "requires an acquirer to recognize and measure contract assets and contract liabilities (deferred revenue) acquired in a business combination in accordance with Revenue from Contracts with Customers (Topic 606)," under which "the acquirer applies the revenue model as if it had originated the contracts" — "a departure from the current requirement to measure contract assets and contract liabilities at fair value at the acquisition date." It took effect for "annual periods beginning after December 15, 2022 and interim periods within those annual periods," and applies "prospectively to business combinations occurring on or after the date of adoption" (USANA Health Sciences Form 10-Q, May 2023). Two practical consequences. If you read a pre-2023 merger model or textbook it will still show the write-down, and you should not rebuild it. And because adoption was prospective, a deal that closed before it will still carry the old haircut in the comparative periods, so a multi-year revenue bridge spanning both regimes has a discontinuity in it that is an accounting artifact rather than a business event. What did not change: your ARR bridge and your GAAP revenue bridge will still disagree after a deal, because ARR picks up the acquired book at its annualized run rate on day one while GAAP revenue only picks up the stub from the close date, and someone will ask you why. Assume the acquired P&L lands in your plan on day one whether or not anyone hands it to you cleanly, and that someone will eventually ask whether the synergies happened. One thing I could not source: who formally owns tracking realized synergies against the original case after close — Corp Dev, FP&A, or a standalone integration office.

Fundraising models and the board narrative

At a venture-backed company the LRP and the fundraising model are one spreadsheet viewed two ways: where the business goes, and how much cash that takes and when you must go ask.

CRV's August 2026 guidance is the most current numeric source available. Runway is cash on hand divided by monthly net burn ($600,000 against $75,000 a month is 8 months). Gross burn is total monthly cash expense; net burn subtracts revenue and is "the number investors watch most closely." Report both: $400,000 gross burn against $300,000 of revenue is healthier than $150,000 gross against $20,000. Use a trailing three-month average recomputed monthly rather than a single month, which insurance premiums and late collections distort (three months at $210k, $245k, $230k average to about $228k) — and weight recent months more heavily if burn is accelerating, because an average hides acceleration. Section 9 defines burn multiple and runway as metrics; the numbers here are the planning decisions built on them.

The 2026 targets: 18 to 24 months of runway at pre-seed and seed, and the same range has "hardened into the standard for later stage companies" too, because shorter targets no longer cover the interval between rounds. Open the round with 12 to 18 months left, well before cash drops under six, and budget 3 to 6 months from first contact to close — time that comes out of existing runway before the raise starts. Size the round to fund 18 to 24 months post-close, tied to a milestone rather than a round number of months. The caveat matters as much as the target: "two years of cash paired with inefficient burn means capital drains slowly without proportional return."

Sizing the round, with the arithmetic shown

Run that on a company that actually needs a round, which the guide's default company is not: at roughly $200k a month of net burn against $18M in the bank and a 0.2x burn multiple (section 9), it has years of cash and would raise opportunistically or not at all. Round sizing only bites when the burn is real, so the arithmetic below runs on a separate illustration — a Series C company at similar revenue scale that took the other road and spent into growth. Cash is $52M and the trailing three-month net burn is $2.1M a month, so 24.8 months at today's rate. Net new ARR of $16M against $25.2M of annualized net burn is a burn multiple of 1.58 (net burn divided by net new ARR; lower is better), nearly eight times the default company's. But today's rate is not the rate you will burn, because the hiring plan is already approved. Carry the cash forward on the burn path you actually committed to:

• Months 1-6 at $2.1M: $12.6M spent.
• Months 7-14 at $2.6M, as the hiring lands: $20.8M spent.
• Cash at a month-14 close: $52M minus $33.4M, so $18.6M. Not $52M minus a few months of today's burn, which is the mistake that makes every first fundraising model too optimistic.

Post-close, burn averages roughly $2.8M as the plan continues, so 21 months costs $58.8M and 24 months costs $67.2M. Against $18.6M on hand, a $40M round funds about 21 months post-close and a $50M round funds about 24. The band is $40-50M, and the choice inside it is a runway-target choice, not a market question.

Two things follow, and they are the reason to do this arithmetic before anyone asks. First, the raise should open around month 8-10 and close by month 13-16, because a 3-6 month process spends runway you have already committed. Model the round closing late; nobody ever models it closing early and is punished for it.

Second, the hiring plan is the fundraising decision — and deferring a hire is not the same lever as cutting one, which is the distinction most often fudged in the room where the round gets sized. Take ten hires at roughly $260k fully loaded each: $2.6M a year, about $217k a month, landing at month 7 in the approved plan. Run both moves against the burn path above.

• Defer two quarters. The cohort starts at month 13 instead of month 7, so six months of its cost drops out of the pre-close window: about $1.3M. Post-close, the cohort is on payroll in both versions, so the 21-month need is unchanged at $58.8M. Cash at close rises from $18.6M to $19.9M and the round falls from about $40.2M to $38.9M. A deferral is a timing move, and $1.3M is the whole of it.
• Cut them. The cohort is off the payroll for the eight pre-close months it would have worked (about $1.7M) and for all 21 months after the close (about $4.6M). Post-close burn falls from $2.8M to roughly $2.58M, so 21 months costs $54.2M instead of $58.8M, against $20.3M on hand. The round falls from about $40.2M to $33.9M — a $6.3M swing, nearly five times what the deferral bought.

Internalize that ratio, because the instinct in a cash-tight room is always to defer, and a deferral buys roughly a fifth of what a cut buys. Deferral is the right answer when the cohort is load-bearing for the milestone the round is sized to — you still need those people, you need them later. It is the wrong answer when it is standing in for a decision nobody wants to make, and the tell is that the same ten roles get pushed a second quarter at the next reforecast. Either way the hiring plan is the input you actually control. A revenue assumption flexed upward to close the same gap is a hope, and a board that has seen a few of these will read it as one. That is why the fundraising model drives the hiring plan in section 5 more than anything else does.

Two framings you will hear constantly. Default alive versus default dead (CRV attributes it to Paul Graham): holding spending flat and extrapolating current growth, does the company reach profitability on cash already in hand? Founders miss the transition into default-dead while the company still looks busy. And CRV's ordered levers for extending runway: pull cash forward (expand existing customers, raise ARPU, discount 2-3% for annual upfront payment, cheaper than raising and faster than debt); cut non-essential spend before touching headcount; renegotiate vendor terms while runway is still comfortable, since it gets harder once cash is tight; pace hiring to proven milestones; and use non-dilutive capital carefully, including R&D tax credits that can offset payroll tax up to $500,000 a year with contemporaneous documentation.

The board deck

Internalize Kellogg's argument before your first board cycle: "Slide count is time allocation, and time allocation is the de facto meeting agenda." The deck decides what the meeting is about. He names six segment types needing different slide treatment: an ops review wants dense slides with history, targets, plan percent and growth rates; a discussion wants one context slide then three slides with one question per title; a presentation is presenter-led; a proposal ends in a vote, which is exactly what budget approval is (section 3); an update is often one slide; a working session is whiteboard-driven. His procedural advice: "Go one-on-one first. You'll get far more candid answers than you ever will in a group setting." ICONIQ Growth adds that board reporting should place "emphasis on different types of key questions and corresponding metrics at different stages of growth," and offers slide templates companies should "customize all views based on their business model and unique priorities." That a growth investor publishes a function-specific deck template at all is my evidence that these decks are assembled modularly by function rather than written as one document; ICONIQ does not say so.

Guidance at public companies

Guidance is a forward-looking range a public company publishes for revenue, earnings, or other measures, usually for the coming quarter and the full year, issued with the earnings release and repeated on the call. It is voluntary; nothing requires it and some companies decline.

Two rules explain why it takes the shape it does. Regulation FD (17 CFR 243.100, adopted August 2000, effective October 2000) bars an issuer from selectively disclosing material nonpublic information to brokers and dealers, investment advisers and institutional managers, investment companies, or security holders likely to trade on it, without simultaneously (if intentional) or promptly (if not) making the same information public, with exceptions for people owing a duty of confidence and for registered offerings. That is why a number which moves the stock cannot go to one analyst first, and so gets drafted, vetted, and released as a document rather than said on a call. The other rule is why anyone is willing to publish a forecast at all: the statutory safe harbor for forward-looking statements, which covers "a projection of revenues, income (including income loss), earnings (including earnings loss) per share, capital expenditures, dividends, capital structure, or other financial items" so long as the statement "is identified as a forward-looking statement, and is accompanied by meaningful cautionary statements identifying important factors that could cause actual results to differ materially." Reg FD governs who you can say it to; the safe harbor governs whether you can be sued for it. That pairing is the reason every release and every call opens with the same block of cautionary language.

How the range actually gets set

The internal mechanics are practitioner convention rather than published practice — I could not source them against SEC, NIRI, or consulting material, and you should treat what follows as a description of what to expect rather than a rule. It is describable with confidence anyway, because the shape barely varies.

• Three numbers exist, not one. The guide is external and set below the internal plan on purpose. The AOP is what budgets are actually set to. The commit or stretch number is what sales is held to, above both. Confusing them in a meeting is the fastest way to be identified as new.
• Cushion is a standing CFO decision, not a quarterly one. A company has a guidance philosophy, and it changes rarely. Typical software cushion on next-quarter revenue is low-to-mid single digits; the full year is usually raised by less than the quarterly beat early in the year, so cushion is preserved for the back half.
• Guidance is mostly non-GAAP. Salesforce's August 2026 release is a representative shape: quarterly revenue ($11.42-$11.5 billion), quarterly revenue growth in constant currency, GAAP and non-GAAP diluted EPS, and cRPO growth of "approximately 14%"; for the full year, revenue, subscription revenue growth, GAAP operating margin of 20.1% against non-GAAP of 34.3%, EPS on both bases, operating and free cash flow growth, and capex as a percent of revenue. Note the range widths: the quarterly revenue range is about 0.7% wide and the full-year range about 0.65%. That precision is the point of the whole apparatus.
• The beat-and-raise is visible in the arithmetic. The same release raises full-year revenue guidance by $200M and decomposes it: "$100 million of organic growth, $200 million from the pending Contentful and Fin acquisitions, and $100M FX headwind." Every raise gets taken apart like that by analysts, which is why the raise is smaller than the beat.
• Who owns what. FP&A owns the bottoms-up forecast underneath the number. IR owns consensus — analyst model reviews, where the sell-side sits against your forecast, and the whisper number above published consensus. The CFO and CEO set the cushion. The disclosure committee and audit committee are briefed before release.
• Quarter end changes shape. A flash forecast within days of close, a weekly landing drill, and real pressure on revenue linearity, because a back-loaded quarter makes the guide unfalsifiable until the last week and nobody enjoys that.

The critique is better sourced than the practice. FCLTGlobal (2017) argues quarterly guidance attracts transient shareholders and pushes managers toward decisions that flatter a quarter at the expense of strategy, and should give way to long-term roadmaps tied to the fundamental economic drivers of the business. The campaign did not change behavior: the Salesforce release above is quarterly revenue, quarterly EPS, and quarterly cRPO guidance issued in August 2026. Section 13 covers the public-company cadence more broadly.

One note on the guide's core questions: section 3 answers Q3, the carry-forward-level analog and how companies separate baseline from new asks. This section supplements it from the other end. The DP-as-business-case material above shows what an incremental ask looks like once it must carry a financial return, which is the part the state process does not require of you and the private process does.

11. Governance, controls, and approvals

Here is the single largest structural difference between the budget you run today and the budget you will run at a company: a private-sector budget does not authorize spending. It plans it and it holds people accountable for it, but the legal act of committing company money happens somewhere else — in an approval matrix, on a purchase order, under a signed contract. In Washington the appropriation is the authority, and the statute proves it: RCW 43.88.130 bars any agency from expending or contracting to expend in excess of the amounts appropriated, and "any contract made in violation of this section shall be null and void." No company contract is void for exceeding a budget line. In a company the annual operating plan is a promise; the delegation of authority document is the authority.

Get that inversion straight and most of this section follows. Miss it and you will spend your first quarter telling a department head "you have budget for that" and being surprised when the purchase still does not go through.

What a budget does and does not authorize

The useful framing is the budget as a contract. Prince Oppong, a senior director of strategic finance at PayPal, put it to AFP this way: "A budget isn't just an aspirational exercise or an expression of strategy; it's literally a contract between multiple parties." It operates at three levels — management to the board, the CEO to the management team, and at a public company the enterprise to shareholder expectations. Consequences are commercial rather than legal: a missed number costs credibility, bonus, sometimes the job, and becomes legal exposure only where there is active misrepresentation.

Two terms you will hear constantly. A budget owner (also budget holder, or cost center manager) is the manager accountable for a department's or project's spending plan — the person whose name is on the variance. Pigment's glossary splits it cleanly: FP&A sets the timelines, guidelines, and structure of the planning process; budget owners make the day-to-day spending decisions inside their approved allocation. Approval authority is a different thing: the right to commit funds. As Stampli frames it, budget ownership is responsibility for a spending plan and approval authority is the right to commit. They overlap and are never identical. A budget owner routinely needs someone else's signature to spend money they own the plan for. So "is it in budget?" is a necessary question and not a sufficient one. It is a gate, not a grant.

The reverse surprise is how little the line items bind. Inside a cost center, fungibility is close to total: an owner can move planned spend from travel to software to contractors without anyone re-approving anything, because the control is the total and the gate is the DOA matrix, not the line. There is no proviso restricting money to a purpose. What does need finance and usually the CFO is moving budget between cost centers, reclassifying opex to capex, or converting non-headcount budget into headcount. Design your variance thresholds accordingly: report at line-item granularity and you will chase variances that are pure recomposition.

Who approves the plan, and what the plan then becomes

The board approves the AOP — at a sponsor-owned company, a board the sponsor controls. The approval package is short: plan P&L by quarter, ending headcount, cash runway or leverage, a separate capex envelope (an aggregate capital number the board authorizes, spent later through the DOA matrix project by project), and the three or four assumptions the board is being asked to accept. Boards approve envelopes and assumptions, not line items.

Then the compensation committee converts the approved plan into bonus targets, and this is the origin of nearly every political behavior in planning. Meridian Compensation Partners describes the standard shape: target bonus is paid for "typical, expected performance, often 'budget' or 'plan'"; a common threshold performance level is 80–90% of target performance, and the most common payout at threshold is 50% of target bonus; maximum is typically set at 200% of target (companies with more predictable earnings may use 150%) and a common maximum performance level is 110–120% of target. Payout between the stated points is almost always straight-line interpolation. Put a number on it. A director with a $50,000 target bonus on a plan of $40M revenue, threshold at 90% and cap at 115%, sees zero below $36M, $25,000 at $36M, $50,000 at $40M, and $100,000 once revenue reaches $46M — above which nothing more is earned. That $36M cliff is why the revenue leader argues the plan number down in October, and the $46M ceiling is why a blowout year gets partly banked into next year's pipeline. Every expense owner pads for the same reason from the other side.

The consequence is that a company carries several versions of the same year at once, and part of your job is knowing which one a given conversation is about: the board-approved plan (the bonus and covenant number), the commit that sales leadership holds internally (usually above plan, because quota capacity is built with cushion), a stretch case used for hiring triggers and upside planning, and at a public company guidance to the street, which is normally set below plan so the company can beat it. Sandbagging is structural, not a character flaw. See building the AOP for how the negotiation runs and the revenue plan for the quota-capacity math underneath the commit.

The delegation of authority matrix

The delegation of authority matrix (DOA, sometimes "approval matrix") is the formal policy, anchored in board or executive action, that grants approval authority to roles at defined dollar limits. Read it as a grid: role × transaction type × dollar band. A complete one specifies the spend category (opex, capex, professional services, software), the threshold for each tier, the required approver by title rather than by name, whether that tier may onboard a new vendor, the point at which a second approver is triggered, the documentation required at each level, and the escalation path for anything out of policy (Ken from Finance).

Thresholds are company-specific; the shared pattern is dollar-banded tiers plus a dual-approval kicker somewhere in the middle. The bands below come from that same vendor glossary. Treat them as a shape, not a benchmark — this is illustrative vendor content, not survey data.

Approver	~$5M revenue, 10–50 people	~$50M revenue, 100–300 people	~$500M revenue, 500–2,000 people
Manager / department lead	$2,500 opex / $0 capex	$5,000 / $0	$10,000 / $0
Controller / finance manager	$10,000 / $5,000	—	—
Director	—	$25,000 / $10,000	$50,000 / $25,000
VP / department head	—	$100,000 / $50,000	$250,000 / $100,000
SVP / GM	—	—	$1,000,000 / $500,000
CFO	$50,000 / $25,000	$500,000 / $250,000	$5,000,000 / $2,500,000
CEO / owner	Combined with CFO tier	$1,000,000 / $500,000	$10,000,000 / $5,000,000
Board	Above $50,000 opex / $25,000 capex	Above $1M opex / $500K capex	Above $10M opex / $5M capex
Multi-approver trigger	Dual at $10,000+	Dual at $25,000+, triple at $250,000+	Plus separate software and professional-services columns

Two things to notice. The board sits above the CEO, not above the CFO, once a company is large enough to have both tiers. And at scale the matrix stops being purely about money: software and professional services get their own lower thresholds, and certain categories — anything touching customer data, anything with an indemnity clause — route to legal or security review regardless of amount. A $12,000 tool that processes customer records can be harder to approve than a $200,000 renewal with an existing vendor. Where three signatures are required, the third is structural rather than hierarchical: originator, finance, executive. Stampli's rounder rule of thumb is that limits step up roughly an order of magnitude per level, calibrated so the large majority of routine invoices clear at the first level.

Contract signature authority is a separate and usually much shorter list. Approving a PO is not authority to sign the agreement behind it. Signature authority is typically owned by legal, often reserved to corporate officers, with counter-signature rules and carve-outs that route to legal regardless of dollar value: uncapped indemnity, non-standard limitation of liability, a data processing addendum, auto-renewal beyond a set term, anything granting IP rights. A director with a $100,000 approval limit who cannot sign the contract is the normal case, and the two lists diverging is a routine reason a "fully approved" deal sits still.

Maintenance matters more than design. Stampli puts ownership of the document with the controller and approval of changes with the CFO; Ken from Finance recommends comparing actual approval patterns against the matrix quarterly and having the board or audit committee re-approve the matrix annually. The quarterly comparison is the useful one, and it comes with its own diagnostic: if a large share of invoices are clearing by exception, the thresholds are wrong, not the approvers. Enforcement should be systemic — amount-based routing in the procurement tool, hard blocks on incomplete approval chains, permissions tied to identity-provider role groups rather than named individuals, every override logged with reason, approver, and timestamp. A matrix that lives only in a spreadsheet is a document, not a control.

Headcount approval is a separate authority

Payroll is the majority of operating expense at a software company — 60–75% is the usual rule of thumb — and none of it moves through the PO chain. The hiring approval workflow is the budget control you will operate most often, and it runs on its own track. The plan produces approved positions — numbered slots or requisition IDs that follow the role across the HRIS, the applicant tracking system, and accounting (Pin). A hiring manager opens a requisition in Greenhouse, Ashby, or Lever; it routes through the HR business partner and the department head to finance before recruiting starts; finance checks that it maps to a budgeted, unfilled position at the planned level, location, and compensation band, and rejects reqs with no cost code or a band above plan. Executive sign-off sits at the end of that chain for senior or net-new roles.

The distinction that governs everything is backfill versus incremental. A backfill replaces an existing approved slot — the spend and the job already exist. Pin puts backfills at 24 to 72 hours and net-new headcount at one to three weeks, through three to five approvers depending on which it is. Incremental headcount expands the plan and needs a written business case and usually the CEO or the exec staff as a group, not the DOA matrix. The trap: a backfill at a higher level, in a more expensive location, or with an equity refresh is an incremental ask wearing a backfill's clothes, and it is your job to say so. This is also the one place the plan comes closest to being an authorization, since the position ID is what recruiting checks before posting.

Know what a hiring freeze actually freezes. In most companies it stops incremental reqs and open unfilled reqs; it rarely touches signed offers, and it often exempts backfills in revenue roles. And expect plan headcount, approved reqs, and filled headcount never to agree — reconciling the three, monthly, is an FP&A deliverable, not an accident. See headcount and compensation planning for the planning side.

Requisition to payment, and the spend that skips it

The spend chain runs in eight steps (Zip): need identified, supplier research or RFP for anything complex, purchase requisition approved and routed, purchase order issued with terms — an offer that becomes binding when the vendor accepts it, which is why vendor confirmation is the next step and why an unaccepted PO is not yet a commitment — vendor confirmation, goods or services received, invoice processed against a three-way match (the reconciliation of PO, receipt, and invoice before payment releases, the standard control against paying for things never ordered or never delivered), and payment. Routing keys on more than the dollar figure: Stampli lists budget availability, amount, department or cost center, vendor, project, and item category as co-equal criteria, usually combined, and Procurify layers role-based responsibility on top — manager confirms business need, finance verifies budget, IT assesses security fit, executives authorize material commitments — each approver evaluating only what they own.

Segregation of duties is the control this chain exists to deliver, and it is the first thing a SOX walkthrough tests. Requester, approver, goods receiver, and payment releaser must be four different people. The vendor master is a controlled table with its own approval, because adding a vendor is functionally the ability to pay one — which is why the DOA matrix often carries a separate new-vendor column, and why the lowest tier cannot open one. And any change to vendor bank details requires an out-of-band callback to a number you already hold, never a number supplied in the request; the FBI's Internet Crime Complaint Center advises using secondary channels or two-factor authentication to verify requests for changes in account information. Vendor-impersonation fraud is why the control exists.

Spend that never touches a PO

Processing a PO costs roughly $50 to $100 by Zip's estimate, so companies raise the no-PO threshold as they grow — and everything below it flows through corporate and virtual cards (Ramp, Brex, Navan, BILL) governed by card limits, merchant-category rules, and after-the-fact receipt and coding policy rather than by the DOA matrix. A department can accumulate six figures of SaaS through recurring card charges that individually clear every threshold. Vendr calls the general problem maverick spend and describes the exact failure: a salesperson buys software on a card with no notice to finance or IT, leaves, and the contract "quietly auto-renews for multiple years undetected."

Auto-renewal is the dominant failure mode, and the operative control is a calendar, not an approval gate. Most SaaS agreements renew automatically unless notice is given 30, 60, or 90 days before term end; by the time a renewal quote reaches the requisition queue the notice window has often closed and the approval is theater. Own a renewal calendar jointly with procurement, keyed to notice dates rather than renewal dates. The other measured process metric is retro (or confirming) POs — work started before the PO existed, invoice arrives for a commitment nobody approved. Volume of retro POs is a standard procurement health metric and a standard audit finding; when one lands, accrue it in the month the service was delivered, escalate the pattern rather than the instance, and decide with legal whether you are disputing the invoice or absorbing the year.

SOX and the audit committee at public companies

At a public company governance stops being purely internal policy. Section 302 (15 U.S.C. § 7241) requires the CEO and CFO to personally certify each report. What they actually sign is the SEC's form language at 17 CFR 229.601(b)(31): that they reviewed the report, that it contains no untrue statement of material fact, that the financials fairly present financial condition in all material respects, that they are responsible for establishing and maintaining both disclosure controls and procedures and internal control over financial reporting, that they "evaluated the effectiveness of the registrant's disclosure controls and procedures ... as of the end of the period covered by this report," and that they disclosed any change in ICFR during the most recent fiscal quarter that materially affected or is reasonably likely to affect it. The statute's original "within 90 days" evaluation window was replaced by period-end evaluation in 2003; you will still see it repeated in secondary summaries, and repeating it in front of a controller will get you corrected. Note the two regimes: disclosure controls and procedures ask whether material information reaches the officers in time to be disclosed; ICFR asks whether the financial reporting is reliable. FP&A usually feeds the disclosure committee — the body that reviews the earnings release and MD&A variance narrative — which sits on the DCP side.

Section 404 (15 U.S.C. § 7262) requires management's annual assessment of ICFR effectiveness and an external auditor attestation to it, but the attestation half, 404(b), reaches only accelerated and large accelerated filers. Status turns on public float and revenue (17 CFR 240.12b-2):

Filer status	Public float (non-affiliate market value)	404(a) management assessment	404(b) auditor attestation
Non-accelerated / smaller reporting	Under $75M, or annual revenue under $100M with float under $700M	Yes	No
Accelerated filer	$75M to under $700M, and not SRC-eligible on the revenue test	Yes	Yes
Large accelerated filer	$700M or more, and not SRC-eligible on the revenue test	Yes	Yes

The revenue test matters: a company with $200M of float and $60M of revenue is a smaller reporting company, stays non-accelerated, and gets no 404(b) attestation. Exit thresholds are lower than entry thresholds ($60M and $560M), so a company that dips below does not immediately fall out.

Section 301 (15 U.S.C. § 78j-1) reorders the org chart. Every audit committee member must be a board member and otherwise independent — no consulting, advisory, or other compensatory fee from the issuer. The committee, not management, is "directly responsible for the appointment, compensation, and oversight" of the external audit firm, and that firm "shall report directly to the audit committee." The external auditor does not work for the CFO. Section 407 (15 U.S.C. § 7265) requires disclosure of whether the committee includes a financial expert, and why not if it does not. The function you will actually deal with is neither of those: internal audit reports functionally to the audit committee and administratively to the CFO, and it is internal audit that walks you through a control narrative, performs the walkthrough, and tests your evidence. At a pre-IPO company an outside SOX readiness firm plays that role for a year or two before the first 404 opinion, which is usually when an FP&A flux review first becomes a documented control.

Beyond SOX

Two more public-company regimes touch FP&A directly. Exchange Act Rule 10D-1 (17 CFR 240.10D-1) directs the exchanges to condition listing on a company maintaining a policy that recovers erroneously awarded incentive compensation from anyone who "served as an executive officer at any time during the performance period," triggered by an accounting restatement, reaching back three completed fiscal years, and not dependent on fault. It covers any compensation "granted, earned, or vested based wholly or in part upon the attainment of a financial reporting measure," and the rule states expressly that "stock price and total shareholder return are also financial reporting measures." That raises the stakes on every metric definition FP&A writes into a bonus plan. Second, the guidance layer: Regulation G reconciliation discipline for every non-GAAP measure you publish, Regulation FD limits on what may be said and to whom, and the quiet period and trading blackout calendar. See how it varies by company type for the public-company rhythm.

What is actually in SOX scope, and what is not

Internal control over financial reporting is defined by PCAOB Auditing Standard 2201 as a process providing reasonable assurance regarding the reliability of financial reporting and statement preparation in accordance with GAAP. The standard never mentions budgeting or forecasting. What is in scope is how actuals get recorded: the close, journal entries, revenue recognition, reconciliations, and the period-end financial reporting process — ¶.26 puts "procedures used to record recurring and nonrecurring adjustments" and "procedures for preparing annual and quarterly financial statements and related disclosures" in the auditor's evaluation, and ¶.14 names controls over significant management estimates. Two failure grades: a material weakness creates a reasonable possibility that a material misstatement will not be prevented or detected on a timely basis; a significant deficiency is less severe but important enough to merit oversight attention. SOX programs are built on the COSO Internal Control — Integrated Framework, which is the reference model behind all of this.

So the rule about your own work has two halves, and only the first is the comfortable one. The AOP and the routine quarterly reforecast are not ICFR. No auditor tests whether you hit your plan. But the moment a forecast becomes support for an accounting estimate, it is in scope and inherits evidence requirements. That happens more often than new FP&A directors expect: goodwill and long-lived asset impairment testing, where the long-range plan cash flows are the model input; deferred tax asset valuation allowances, which turn on projected future taxable income; capitalized internal-use software; variable consideration under ASC 606; restructuring and lease-asset impairment. PCAOB AS 2501 tells the auditor, where an assumption rests on management's intent and ability to carry out a course of action, to weigh "the company's past history of carrying out its stated intentions" and "the company's written plans or other relevant documentation, such as budgets or minutes." Your budget is named in the standard. Keep the impairment-model version of the long-range plan archived separately, with documented assumptions, approvals, and a version date, because the question will be what you forecast last year and what actually happened. See long-range planning.

The other place FP&A lands in scope is the management review control — a review performed by management that the company designates as the control expected to catch a misstatement. Analytical and flux reviews are settled practice at public companies, not an exotic design; PCAOB inspection commentary uses "management's monthly review of budget-to-actual financial results" as its standing illustration, and warns that if the auditor "only looks to see that management performed the review and does not understand what management looked for in the review, what matters were investigated, and how they were resolved, the auditor has failed to obtain evidence that the review could in fact prevent or detect a material misstatement" (PCAOB). What varies company to company is the design, and these are the questions to ask in week one:

• What is the expectation? Prior period, prior year, budget, or an independently derived expectation. Designating the budget as the benchmark invites the auditor to test whether the budget is precise enough to be relied on, which drags your planning process into evidence scope. Some companies deliberately build the control against prior period for exactly this reason.
• What is the investigation threshold? Nearly always dual — a dollar amount and a percentage, both documented in advance rather than chosen after seeing the variance.
• What counts as an explanation? A named driver with corroborating support. "Timing" is not an explanation.
• Who signs, and what support is retained? Dated sign-off and the working file, kept where an auditor can pull it a year later.

Covenant and lender reporting at leveraged companies

At a leveraged company an outside party gets a contractual claim on your planning output. This is the closest thing the private sector has to the external, enforceable constraint you are used to, and unlike SOX it reaches the plan itself. Which version you meet depends on the capital structure.

• Covenant-lite with a springing test. Sidley reports that competitive dynamics, "particularly in the competitive middle and upper market," have increased adoption of covenant-lite provisions derived from the syndicated markets; in some of those deals maintenance covenants "benefit only revolving lenders, with term lenders receiving only cross-acceleration rights." A springing financial covenant is "a maintenance ratio that is dormant unless a specified trigger occurs," most commonly higher utilization of the revolving credit facility, and it is usually a first-lien or senior secured net leverage ratio. (That glossary describes UK LMA-style practice; the structure is the same in the US private credit market.) Sidley supplies the levels: quarterly testing is triggered "only if revolving credit usage exceeds a negotiated threshold of available revolving commitments (typically 40% or higher)," and in sponsor-friendly deals "testing can be avoided if revolving credit usage is reduced below the testing threshold after the covenant test date but before delivery of the compliance certificate." That changes your job: the live question is not a quarterly calculation but whether a revolver draw trips the test, which makes cash forecasting and draw timing an FP&A decision with covenant consequences. Two named variants: a covenant ratchet lets borrowers "reset covenant levels upward following incremental borrowings," and a covenant holiday suspends or softens testing "for a defined period following a material acquisition." Cushions wider than the 25–35% range below produce what Sidley calls a "cov-loose" structure.
• Traditional maintenance covenants. Still common in smaller and non-sponsor deals. Maintenance covenants require the borrower to stay inside specified ratios on a scheduled test, typically quarterly, whether or not the company does anything; illustrative levels are total debt to EBITDA under 5.0x, senior debt to EBITDA under 3.0x, EBIT to interest above 3.0x. Incurrence covenants test only on a triggering action — new debt, an acquisition, a dividend.
• Recurring-revenue (ARR) facilities. At a pre-profitability software borrower there is no EBITDA to lever, so Sidley notes market practice has moved to "an annual recurring revenue (ARR) covenant requiring a minimum annualized recurring revenue stream (adjusted for subscriber churn and retention) as a financial maintenance covenant and/or a cap on borrowing availability" — metrics that "focus on revenue durability and liquidity runway rather than near-term profitability."

How the level gets set should interest you more than the level itself, because it is set off your own model. Sidley describes covenant levels as chosen with deliberate headroom against the sponsor's projections: "A typical leverage covenant in a direct lending transaction may be set with a 25-35% cushion to the EBITDA projected in a sponsor or borrower model that is prepared and delivered to the lender prior to closing." Too tight and the company amends constantly; too loose and the lender loses its early warning. The base case you build at the deal table becomes the benchmark you are measured against for years, which is a strong argument against your own optimism.

Separately, affirmative covenants impose obligations (deliver audited annuals, maintain GAAP compliance) and negative covenants impose restrictions (dividends, asset sales, additional borrowing). Breach is technical default, and consequences ladder up: waiver, renegotiation, additional collateral, repricing, acceleration.

The single most FP&A-owned artifact in the whole process is the EBITDA bridge, because Credit Agreement EBITDA is a negotiated contractual definition, not GAAP and not the company's own reported adjusted EBITDA. It has an enumerated add-back list — restructuring and integration costs, transaction expenses, pro forma synergies, run-rate adjustments — and Sidley notes lenders "often impose caps (including, frequently, shared caps expressed as a percentage of EBITDA on certain projection-based or nonrecurring adjustments), time-based restrictions, or other limitations." You build and defend the bridge from GAAP net income to that number every quarter, and a scrubbed add-back can move leverage more than a month of trading. One escape hatch worth knowing: an equity cure lets the sponsor contribute cash treated as a dollar-for-dollar increase to adjusted EBITDA for covenant purposes, subject to negotiated frequency limits — a typical formulation bars more than two consecutive cures or more than two in any four consecutive quarters, with an overall cap such as four over a five-year facility (Colorado Banker).

The reporting mechanics are where FP&A lives. Credit agreements typically require monthly financial statements within 30 days of month end (sometimes 45) — consolidated balance sheet, income statement, and cash flow with comparative prior-year figures, certified by a Responsible Officer as GAAP-compliant. A compliance certificate, executed by a Responsible Officer and delivered with the financials, carries the covenant calculations, the EBITDA adjustment worksheets, and supporting detail satisfactory to the agent. And the plan itself is a deliverable: annual budget clauses are standard, with timing that genuinely varies — some require delivery before fiscal year start (by November 1, or 30–60 days prior), others within 45 to 60 days after the fiscal year begins, often bundled with the annual financials. Expect a monthly breakout, projected income statement, balance sheet and cash flow, and sometimes lender approval "not to be unreasonably withheld," with the prior year's budget continuing in effect if the new one is not approved. Read your own credit agreement's delivery section in week one; it sets your calendar.

One honest gap. The contents of a sponsor's own monthly reporting package — the KPI dashboard, the budget-versus-actual format, the cadence of value-creation-plan reviews — are not standardized in any public source I could reach, and vary by sponsor. What is documented is the post-close 100-day plan, the integration roadmap covering finance and accounting policy alignment, HR retention, and systems (Divestopedia); expect FP&A to be pulled into it. Treat any specific claim about sponsor package contents as company-specific until you see the template.

Translation: appropriation, expenditure authority, and allotment control

The practical upshot for your first quarter: learn the DOA matrix and the requisition workflow before you learn the chart of accounts. Together they tell you who really decides things, which is a different and more useful org chart than the one HR publishes.

12. Systems and modeling craft

Start with the fact that surprises people who read vendor websites first. Companies buy expensive planning software and then build the plan in a spreadsheet anyway. The 2025 AFP FP&A Benchmarking Survey (362 practitioners worldwide) found 96% using spreadsheets for planning and 93% using them for reporting on a daily or weekly basis, and separately found that 71% of respondents use an enterprise performance management tool at least quarterly. Both are true at once. AFP's follow-up, "Why Aren't FP&A Teams Using the EPM Tools They Bought?" (May 2025), gives the shape of it: among EPM users, 82% use spreadsheets to prepare data for the tool, 85% use spreadsheets alongside it, and 57% bypass it entirely.

So the honest description of a finance stack is not "we plan in Anaplan." It is: systems of record generate transactions, a planning tool holds the official version, a BI layer publishes the numbers, and a spreadsheet sits in the middle doing the actual thinking. Your job is to know which layer holds which truth, keep the mapping between them intact, and build the spreadsheet part well enough that someone else can pick it up. Budgeting approaches — top-down, zero-based, driver-based, rolling — are answered in section 3; this section covers only how a driver model gets physically built and where it lives.

The stack, layer by layer

Five layers. Confusing a system of record with a planning system is the most common orientation error a new FP&A hire makes.

One note on that first row. FP&A does not own the ledger and never posts transactions, but it is not true that it never writes: loading the approved budget into the ERP so owners can run budget-vs-actual natively is routine. The trap is that a budget living in both the planning tool and the ERP has one stale copy the moment a reforecast lands. Decide up front which one owners are told to open.

A sixth category sits beside accounting: close and reconciliation automation. FloQast (3,500+ companies) and BlackLine run the close checklist and account reconciliations. Controller's tools, not yours, but the close calendar they enforce determines when your actuals land — see section 8.

On the ERP tier a rough size ordering holds: QuickBooks and Sage Intacct below roughly $10M in revenue; NetSuite as the venture-backed and growth-stage default; then Workday Financial Management, SAP S/4HANA, and Oracle Fusion. A "NetSuite to Workday migration" on a job description means a year of your life will involve mapping tables.

The chart of accounts, and the mapping nobody shows you

This is the plumbing that consumes a first quarter, and the piece a government budget background least prepares you for — not because it is harder, but because the coding structure differs in kind. A commercial chart of accounts is a list of natural accounts, typically four to six digits and grouped so 4xxx is revenue, 5xxx cost of revenue, 6xxx operating expense, crossed with a small set of dimensions. Two always exist — department (or cost center) and legal entity — and many companies add class, location, or project. The account says what was spent; the department says whose budget it hits. That pair does the work fund, appropriation, program, and object of expenditure do in a state ledger, and does less of it: no fund, because no restricted source of money needs segregating, and no appropriation dimension, because nothing is appropriated.

The plan is deliberately coarser than the ledger. Accounting posts to 300 natural accounts; you plan 50 to 70 lines. You budget one line called "Software & subscriptions" and accounting posts to eleven accounts under it. The mapping table that rolls GL accounts up to planning lines belongs to FP&A, lives in the planning tool's metadata or a tab of your workbook, and is the most common place a budget-vs-actual silently breaks. Three failure modes:

• An unmapped account. Accounting opens 6485 "AI tooling" in March. Nothing maps it. It lands in an "other" bucket or drops out of the report, and your opex forecast is short by $400k with no error message. The defense is a standing check that sums mapped accounts against the trial-balance total and must read zero.
• A re-org. Support moves from Customer Success to Engineering, prior-year actuals stop being comparable, and every trend chart lies. Restating history is about a week of work; most teams instead keep a legacy-department crosswalk and report both views for a year. Decide which before the first board meeting after the re-org, not during it.
• A cost center with no owner, left by a project or a departed executive. Orphaned spend accumulates there because nobody gets a variance question about it. Audit cost centers against named budget owners quarterly.

One design rule pays for itself: a department owner should never see accounts they cannot act on. Keep allocations in a block below the controllable lines, or every variance conversation opens with a VP explaining that the miss was not his (section 6).

How actuals arrive, and how you prove they are right

The integration. Pre-built connectors exist between the major ERPs and planning tools, and where one exists it is a scheduled pull you never touch. A meaningful minority of mid-market implementations run instead on a trial-balance extract someone exports and loads, and multi-entity groups and recent acquisitions almost always run on file loads because the acquired company is still on its own ledger. Two consequences: the load is a full-period restatement rather than a delta, so a re-opened period must be re-pulled rather than patched; and it happens after the close is locked. Pulling actuals on day three of a six-day close gets a number that changes under you, and the "why did the variance move" email that follows is self-inflicted.

The tie-out. Before anything is published, reconcile loaded actuals to the GL trial balance at total company and by department — total-only agreement is not agreement, since two departments can be wrong by offsetting amounts and net to zero. Keep it as a standing report that must read zero. Recurring causes of a break: a new account nobody mapped, intercompany eliminations applied in the ERP but not the extract, top-side journal entries booked after your load ran, FX rates differing between systems. None are your error; all are your problem.

Actualization. Each month you replace a forecast month with an actual. The convention is a period flag per column — Jan A, Feb A, Mar A, Apr F, May F — with actual and forecast months visibly distinguished in every report. The rule underneath it: never type actuals over a forecast month in place. Flip the flag and let the model read from the actuals block. Manual overwriting keeps no record of what you originally forecast, so you can never answer the most useful question in FP&A: what did we think would happen, and what did we get wrong?

When three systems disagree about one number

At almost every growth-stage company the same metric has three or four values. Finance's ARR comes off the revenue schedule, RevOps' off closed-won opportunities in the CRM, the data team's out of a dbt model with its own churn logic, and last quarter's board deck agrees with none of them. Headcount does the same across HRIS, payroll, and plan FTEs, which disagree about contractors, interns, and slipped start dates.

The rule: for each contested metric, write down which system is authoritative, what the definition is, and who may change it, then publish a standing reconciliation between the authoritative source and the loudest competing one so the gap is a known quantity rather than an ambush. Definitions themselves are section 9; what belongs here is that this is a systems problem and it lands on you. Handle it as a reconciliation, not as someone being wrong — the CEO or CRO usually has a preferred number and it is usually the highest one.

The planning layer: what an EPM tool actually is

Strip the marketing and an EPM tool (enterprise performance management; also CPM, corporate performance management, or just "the planning tool") is four things a spreadsheet cannot do well:

1. A multi-dimensional database. Data stored by intersections of dimensions — account, department, entity, month, version, scenario — instead of by cell address. Anaplan's Hyperblock is one implementation; Oracle's Essbase, released by Arbor Software in 1992 and still under Oracle EPM, is the ancestor of the category.
2. Versioning as a first-class object. "FY27 Budget v3 Approved" is a stored, lockable thing, not a filename.
3. Concurrent multi-user write with workflow. Forty department owners submit into one model at once, each seeing only their own cost centers, with submit and approve states.
4. An audit trail. Who changed which intersection, when, from what.

Who uses what, by company size

The segmentation below is directional, drawn from vendors' own customer descriptions rather than an independent ranking, and vendors move upmarket and down constantly. Gartner's Magic Quadrant for Financial Planning Software sits behind a paywall, so the only accessible evidence of its contents is vendor announcements such as SAP's December 2025 post claiming Leader placement. Treat "named a Leader" as a vendor's own claim unless you have read the report.

Band	Common tools	What drives the choice
Seed to Series B (under ~$20M revenue)	Google Sheets or Excel; Datarails, Cube	Speed to stand up; nobody to administer a real tool
Growth stage (~$20M-$200M)	Pigment, Planful, Vena, Workday Adaptive Planning, Prophix (plan and close in one), HiBob Finance Suite; the spreadsheet-native tier — Abacum, Aleph	Headcount plan too big for a sheet; board wants versioned scenarios
Large / multi-entity / public	Anaplan, OneStream, Oracle EPM, SAP Analytics Cloud, Adaptive at the upper end	Consolidation across legal entities and currencies; SOX-grade audit trail

Two of the biggest names are now private-equity owned, which predicts pricing and support behavior and is a fair diligence question on any tool you would inherit. Anaplan was taken private by Thoma Bravo in a 2022 deal that closed at $10.4B; OneStream, public for less than two years, closed a $6.4B go-private deal with Hg in April 2026.

On implementation time use a range, not a benchmark. Workday Adaptive Planning claims an average deployment of 4.5 months, "even for some of the world's largest companies" — a vendor's own number, and still two quarters. The spreadsheet-native tier claims weeks: Aleph tells buyers to "skip the multi-month onboarding timelines" and quotes a customer at under three weeks to first quarter-end reporting. Anaplan and OneStream run two to four quarters with a partner. The rule behind the spread: implementation time scales with how much of your logic has to be rewritten in the vendor's modeling language. Budget for the partner fee too — roughly the annual license in the mid-market, several multiples at enterprise scale. That fee is what gets cut when a project runs long, and a descoped implementation is the shelfware failure below.

Excel and Sheets: conventions people will judge you on

Excel remains the standard for anything model-shaped: faster on large ranges, richer function set, and every EPM vendor's add-in assumes it. Google Sheets wins where many non-finance people type into one file at once — collection templates, rosters, trackers — so at Google-Workspace-native companies expect to receive inputs in Sheets and rebuild them in Excel.

The color convention

Color	Means	Example
Blue	Hard-coded input or assumption	1234
Black	Formula calculated on this sheet	=A1*A2
Green	Link to another worksheet	=Assumptions!B4
Red	Link to another file	=[Actuals.xlsx]Sheet1!$A$1

The scheme comes from investment-banking practice and is documented by Wall Street Prep, which adds a fifth color (dark red) for live data-provider pulls such as Capital IQ; in-house FP&A rarely has those. The rule behind the rule: a blue cell is a decision somebody made and can change; a black cell is arithmetic. Anyone reviewing your forecast wants to find the decisions in under a minute.

Structural rules

• Inputs, then calculations, then outputs. Assumptions grouped in one place, not scattered where they were first needed.
• One row, one calculation. The formula identical across every column, so auditing one cell audits the row. A hard-coded override buried in month seven is the classic source of a forecast nobody can reconcile.
• Never re-enter the same input twice — reference the source cell, because two copies will diverge — and break long formulas up. CFI's guidelines frame the test as printability: print the model and trace each line without getting stuck.
• Nothing hidden — no white-font cells, no values parked in column BZ — and no mixing periodicities side by side; monthly columns and an annual roll-up belong on separate sheets.
• Prefer MIN/MAX to nested IFs, and put a circuit breaker on any deliberate circularity (revolver interest is the usual case).

Function-level conventions

Those rules are about layout, and a model can satisfy all of them and still be unusable. These are what a colleague inheriting your file, or an interviewer running a modeling test, actually checks — settled practice rather than a published standard.

• The flat file rule. Land the GL extract as one long table on its own tab — one row per period, entity, department, account, amount — never reshaped by hand, and drive every report off SUMIFS or a pivot against it. This is the largest single difference between a workbook that survives a re-org and one that has to be rebuilt.
• Lookups. XLOOKUP where supported, INDEX/MATCH as the portable fallback, never VLOOKUP with a hardcoded column number: inserting a column silently changes the answer.
• No volatile functions. OFFSET, INDIRECT, NOW, TODAY, RAND recalculate on every change anywhere in the workbook; one of them used a thousand times is the standard cause of "this file takes forty seconds to open."
• Power Query for repeatable data prep, and the data model (Power Pivot) past a few hundred thousand rows. A cleanup step you repeat monthly by hand is a defect, not a workflow.
• A Checks tab. Assertions that must all read zero: departments sum to total, actuals tie to the trial balance, roster count equals plan FTEs, the balance sheet balances, ending ARR equals beginning plus net new. Put the aggregate check on the front tab and look at it before you send anything.
• One scenario switch. A single cell driving CHOOSE or INDEX across a block of assumption rows, so base, upside, and downside live in one file. Three copies of the workbook is how the downside case ends up with last month's headcount plan.
• Real dates in header rows, not text labels, so period logic can compare them. No merged cells. Check calculation was not left on manual before circulating.

On a modeling test the assessed behaviors are narrow: build from a blank sheet, put a timeline row with period flags across the top, use a corkscrew (beginning balance, plus additions, less reductions, equals ending) for anything that rolls forward, keep hardcodes out of formulas and in labelled blue cells, and state assumptions out loud.

One genuinely named, published standard exists here and comes up in interviews: FAST — Flexible, Appropriate, Structured, Transparent — maintained by the FAST Standard Organisation, a UK-registered nonprofit. The underused pillar is Appropriate: model at a level of detail matching your actual confidence. FAST's warning is against spurious precision — a tax assumption to three decimals beside a revenue forecast that could be wrong by 30% implies equal confidence in both. Coming from state work, the distinction that matters is between the plan and the forecast under it. Budget execution demands exactness of the plan, because it ties to an appropriation that is a legal number, even when the caseload forecast underneath carries an explicit range. In a company nothing forces the plan to tie to anything, so the discipline moves: decide which few inputs deserve precision, and be explicit about the range on everything else.

The three-statement model, and when you actually need one

A three-statement model links income statement, balance sheet, and cash flow into one connected workbook: net income flows to retained earnings, the cash flow statement drives ending cash, supporting schedules (depreciation from PP&E, interest from debt) feed back into the income statement. CFI recommends building it on a single worksheet rather than three, which reduces cross-sheet errors. It is the base object under DCF, LBO, and merger models — and interviews test it, because it proves you understand how the statements connect.

Calibrate, though. Day-to-day FP&A at a growth-stage software company lives in a monthly P&L by department plus a separate cash and runway model, not a full articulating build; the full model appears in the long-range plan, in fundraising materials, and with whoever owns the cash forecast (section 10, section 9).

Driver trees: a worked example

A driver tree decomposes a financial output into the operational quantities producing it, so the model has a small number of arguable inputs instead of a large number of typed numbers. CFI frames driver-based planning as forecasting from business metrics rather than historical extrapolation. Take the same $40M ARR, 250-person company section 4 plans to $52M, and decompose the year the way the tree would.

Branch	Build	Value
New business ARR	22 quota-carrying AEs × $750K fully ramped annual quota × 0.97 blended capacity × 75% planning attainment	$12.0M
Expansion ARR	$40.0M beginning ARR × 13.0% expansion rate	$5.2M
Contraction and churn	$40.0M beginning ARR × 13.0% combined contraction and churn rate	($5.2M)
Net new ARR		$12.0M
Ending ARR	$40.0M + $12.0M	$52.0M (+30.0%)

Two labels in the first row do distinct work and a sales-finance interviewer will ask you to separate them. The $750K is the quota a rep carries once fully productive. The 0.97 blended capacity is a different haircut entirely: it converts a roster of 22 bodies into ramped-rep-equivalents, the full-year selling capacity a team mixing tenured reps with new hires still ramping actually delivers. Notice what that particular value is doing. At 0.97 the roster yields 21.3 rep-years, which is close to no ramp drag at all, and this is a simplified annual build. A real capacity model computes ramped-rep-equivalents month by month off hire dates and a ramp curve, which is what section 4 does with the same roster: twelve of the twenty-two start inside the plan year, so the honest figure is nearer 17 rep-years and about $9.6M. The tree ties to the $12.0M plan only because blended capacity was set where the plan needed it, and putting that in one visible cell instead of thirty rep rows is most of the argument for building the tree. Read implied net revenue retention off it too: 13.0% expansion against 13.0% contraction and churn is exactly 100% NRR, and that is the number the board asks about.

Six blue cells drive the whole thing: rep count, quota, blended capacity, attainment, expansion rate, combined contraction-and-churn rate. Not thirty rows of rep-level hardcodes. The value shows up the moment someone challenges one: drop planning attainment ten points to 65% and new business ARR falls to $10.4M — $1.6M resting on one assumption, now a specific conversation with the CRO instead of a vague worry. Rep count is not yours to set (it comes from the headcount plan, section 5), and quota, ramp, and attainment come from sales capacity planning. The driver tree is where the two plans have to agree arithmetically, which is most of why it exists.

The failure mode is a tree with forty drivers, each one a number someone must own, defend, and update monthly. Keep the ones a business partner can argue about; hardcode the rest and label them.

Versioning

No published standard here — house convention, and it varies. The clearest practitioner framing I found is Lumel's seven-state lifecycle: draft, submitted, under review, approved and baselined (locked), active, superseded, archived. Two rules follow. Name files so alphabetical order equals chronological order and status is visible — FY27_Budget_V1_Approved, FY27_Q2RFC_V2_Submitted — never "final" or "latest." And the approved version locks: in an EPM tool a permission, in a spreadsheet world a discipline, with the budget of record read-only in its own folder and the live forecast a separate file referencing it. If the two are tabs in one workbook eleven people can edit, the budget will change quietly (section 7).

AI in FP&A as of 2026

Two survey numbers set the boundaries, and the gap between them is a lesson in interrogating a denominator before quoting an adoption statistic. Protiviti's global survey of 902 finance leaders, reported by CFO Dive in August 2026, found 77% of finance organizations have deployed AI in some form and use for financial forecasting at 76%, up from 58% a year earlier. But only 35% say they can confidently gauge ROI and only 14% are deploying against a detailed strategy. Data security has been the top-ranked concern three years running. Protiviti's Christopher Wright names the measurement problem: "The 'I' involves time, third party expenses and cost tracking... the 'tokenomics' has become a factor."

Against that, the 2025 AFP FP&A Benchmarking Survey, reported by CFO.com in January 2025, found only 23% of FP&A practitioners using AI daily, weekly, or monthly, 40% testing with plans to implement within a year, and 36% neither using nor planning to. The two are not in conflict, and the reason is mostly not elapsed time: Protiviti asked finance leaders whether their organization has deployed AI anywhere in finance, AFP asked practitioners whether they personally use it monthly or more. Protiviti's own prior-year forecasting figure was already 58% against AFP's 23%, so population and question wording do nearly all the reconciling work. Together they describe organization-level deployment that has not reached the analyst's daily work.

Treat the 23% as a historical anchor: published January 2025, with no comparable practitioner-level AFP figure since that I could find, and most widely circulated 2026 agentic-adoption numbers vendor-sponsored. The honest current position: agent pilots are common, agent-produced numbers reaching a board deck without a human recomputing them are not, and nobody has published a defensible productivity figure for either.

What actually shipped

• Copilot in Excel, finance edition. In June 2026 Microsoft added "skills" — reusable procedures for recurring work such as building a model, refreshing a monthly report, or running variance analysis — plus financial data connectors (FactSet, S&P Global, Morningstar, PitchBook, Daloopa) and, for auditability, edit attribution: Copilot's changes appear as Copilot's in the Show Changes pane, and a "Plan with Copilot" step states which ranges and formulas it intends to touch first. CFO Dive's coverage notes partner-built skills from Vena, Ramp, and others.
• Named agents in every EPM tool. Pigment's Modeler Agent and Analyst Agent are generally available; its Planner Agent, the one aimed at forecasting and scenario planning, is still tagged "coming soon" on the product page. Planful, Cube, Anaplan, and OneStream have equivalents. Check what sits behind a vendor's number before you repeat it. Pigment's Modeler launch post reports Carta getting a model build down from a week to two days, and quotes Figma crediting the agent with more than 80% of the way to a solid starting point — coverage of a first draft, not a speed gain. Single-customer anecdotes, both.
• Anomaly detection and close automation in the ERP and close tools — the most mature and least glamorous application, because flagging a coding error or an unusual accrual is a well-shaped problem with a cheap failure mode.

What the honest practitioners say

The most useful calibration point is that the company building the models has not finished doing this to itself. OpenAI's CFO Sarah Friar described in August 2026 two ambitions for OpenAI's own finance function — a "zero-day close" and continuously updated automated forecasting — and said plainly: "We are still building toward both ambitions."

Two FP&A Trends essays from the same weeks converge from opposite temperaments. The skeptical one, Lana Ilchenko's, argues the labor moves rather than disappears — "Someone has to correct it when it hallucinates. Someone has to handle the edge cases it drops" — and reports organizations "rehiring staff they eliminated" when headcount-reduction business cases met operational reality. On cost it relays two separately sourced figures worth keeping separate: a 73% AI-cost-overrun rate attributed secondhand to the FinOps Foundation's 2026 State of FinOps report, and, from Gartner, an estimate that agentic models consume 5 to 30 times more tokens per task than a standard chatbot. Neither is verifiable from the article itself. The optimistic essay, WAGO CFO Vignesh Dumonceau's "From Excel-Based Drivers to Agentic Planning," sketches a three-stage path — unified data layer, machine-learning-enhanced drivers, then multi-agent planning under human orchestration — and lands on the line worth keeping: "The model does not own the plan. FP&A does."

Public-sector translation: AFRS, ABS, TALS, Enterprise Reporting

One thing to hold onto: what you have already done is EPM work in substance, even though nobody in a tech interview will recognize the product names. You have built plans in a multi-user system with formal submission workflow, version locking, role-based access, and an external approval gate — more governed planning infrastructure than most growth-stage companies have ever operated. Only the vocabulary is new. A locked baseline, a versioned submission, a mapping between ledger detail and plan detail, and a reporting layer separate from the transaction system are not.

13. How it varies by company type

Sections 2 through 8 describe one machine: set a target, build a baseline, separate new asks from it, negotiate, lock, reforecast, close, explain the variance. That machine is close to universal. What changes by company type is who holds the clock, what the plan is graded on, and where the political friction sits. Get those three right and you will guess most of the rest correctly. Get them wrong and you will show up to an AOP kickoff optimizing something nobody in the room is measured on.

The two axes that explain most of the difference

The first axis is clock flexibility: can the company renegotiate its own plan mid-stream, and with whom? A venture-backed startup rewrites its plan in a board meeting; a public company cannot rewrite what it already told the market. The second is the binding constraint: the number that, if missed, triggers a consequence automatically rather than through someone's judgment — months of cash at a startup, a leverage ratio in a credit agreement at a PE portfolio company, the guidance range at a public company. A large enterprise often has no single one, which is its own problem.

Which budgeting approach goes with which type

This section carries the company-type half of Q4 (which budgeting approach is used, and when); section 3 has the mechanics. None of what follows is a benchmark. It is the pattern I would expect from each type's sourced pressures, and it is worth asking about directly in an interview rather than assuming. Venture-backed companies run driver-based plans with milestone-gated spend, because a calendar build assumes money that may not arrive. Growth-stage SaaS runs a hybrid: a top-down envelope set as a percentage of ARR, filled bottom-up. Public companies more often build incrementally off the prior-year exit rate, because guidance continuity beats theoretical purity — though plenty run driver-based models underneath the incremental frame. Sponsors frequently reach for zero-based budgeting on indirect cost in year one, where the fast margin is, but it is a firm-by-firm habit rather than a settled pattern. Large enterprises build incrementally with a negotiated allocation layer on top: rebuilding four business units from zero at once is not achievable in one cycle.

Two approaches on the Q4 list are missing from that paragraph because they cut across it. A rolling forecast is a model on a continuous add/drop cycle — as a period closes it falls off the front and a new one is appended to the back, against a static budget that holds its projections fixed all year. The only truly rolling artifact most people will ever run is the PE portfolio company's 13-week cash forecast, described below. Large enterprises increasingly bolt a rolling four-to-six-quarter view onto a fixed AOP rather than replacing it, so the annual plan survives as the accountability document while the rolling view carries decision-making. Public companies almost never replace the annual plan with a rolling one, because guidance needs a fixed baseline to be measured against. Beyond Budgeting — a set of leadership principles and management processes aimed at freeing organisations from command-and-control cultures, including reducing annual target setting — is worth knowing as vocabulary. Pure implementations that abandon the annual budget outright remain rare and concentrated in a few European firms. Nobody is going to ask you to install it.

How the number actually gets to the target

Before the type-by-type material, one thing no process diagram shows. The submitted numbers in an AOP are not estimates. They are opening positions, and everyone in the room knows it. The devices that move them are general rather than type-specific, so this guide covers them elsewhere: sandbagging and budgetary slack on both sides of the roll-up in section 3.5; the unallocated reserve the CFO holds above it, a low single-digit percentage of operating expense and unstandardized, in section 3.6; the commit-versus-stretch split that decides which number the bonus actually pays on in section 2.5 and section 3.2; and the vacancy factor, the share of budgeted salary assumed unspent because roles sit open, in section 5 — which calls that same lever start-date slippage, a hiring haircut, or a hiring attainment factor. Vacancy factor is the phrasing you will hear from corporate FP&A. It is one assumption under two vocabularies rather than two levers, so a plan claiming both has double-counted, and it is the most abused number in any expense build: moving it from 5% to 8% on a $60M salary base finds $1.8M with no decision made, no headcount cut, and nobody accountable for the outcome.

What is type-specific is how a gap actually closes at a multi-BU enterprise, so take the $67M reconciliation worked through later in this section. In practice perhaps $22M of it is real — a product launch deferred to the following year, two open requisitions eliminated, a marketing program cut. The other $45M is timing and factors: a hiring plan pushed one quarter to the right, a vacancy factor moved three points, a contingency line trimmed. Everyone in the room expects most of the $45M to unwind by Q3. It closes the gap on paper in October, which is what the meeting was for. That split is practitioner characterization rather than a sourced finding, and it is the thing to test against the people you meet.

Who actually decides varies by type, and it is rarely FP&A. At growth-stage SaaS the CEO and CRO settle the revenue number in a conversation FP&A is often not in, and the expense plan is fitted to it afterward. At a public company the CFO and investor relations settle guidance first and the internal plan is built to support it. At a PE portfolio company the sponsor's operating partner sets the EBITDA number and management negotiates only its composition — which levers, in what order. Your influence is over shape, sequencing and credibility, not over the headline. Coming from a system where the number arrives as law, that will feel familiar in form and completely different in mechanism. Section 14 covers how to work it.

Venture-backed startup: the plan is a function of runway

At a pre-Series-B company the annual plan is a cash schedule with a P&L attached. The organizing number is runway: cash on hand divided by monthly burn, in months. Two burn definitions matter and are not interchangeable. Kruze Consulting defines gross burn as total monthly cash spend, revenue-agnostic, and net burn as spend minus revenue actually collected. Plan against net burn once revenue is reliable; watch gross burn when churn is high, because net burn can look stable while the revenue base underneath it shrinks.

Worked example. A Series A company holds $2.5M, spends $450k a month, collects $180k. Net burn is $270k, so runway is 9.3 months; gross-burn runway, the number if revenue went to zero, is 5.6. Kruze's benchmark is to start fundraising at about 10 months of runway remaining and treat six months as urgent. This company is at the trigger the day the plan is written, which reshapes the AOP: it is a nine-month plan plus a financing event, and every hire in months 7–12 is conditional. Hence the defining habit of the stage, milestone-gated spend — increases tied to milestones rather than a calendar, so the second rep hires when the first clears quota twice, not in April. Note also that runway is computed on cash you hold. A tranched round whose second half is milestone-conditioned is not in the bank, and including it in the runway calculation is a mistake you make once.

Burn multiple: whether the burn is working

Runway tells you how long the money lasts. It says nothing about whether the spending is producing anything, which is why the governing efficiency metric in venture board decks since the 2022 correction is the burn multiple: net burn divided by net new ARR over the same period, a capital-efficiency measure of how many dollars the company burns to generate each dollar of net new recurring revenue, originated by David Sacks. The company above burns $270k a month, $3.24M a year. Against $4.0M of net new ARR that is 0.81 and the board relaxes; against $1.5M it is 2.16 and the next board meeting is about a bridge round. The same source gives rough guidance of 1.5x–3.0x as acceptable from seed to Series A, below 1.5x as the growth-stage target and below 1.0x as best in class, and cites Capchase 2022 medians of 1.7x at $1–3M ARR falling to 0.65x at $5–10M. Treat those as circulating convention rather than a standard; the bands soften the earlier the company.

Expect a board to also ask whether the company is default alive, Paul Graham's test: assuming expenses remain constant and revenue growth continues at the rate of the last several months, does the company reach profitability on the money it already has? Default dead is the inverse, and it is a statement about the current plan, not a prediction. Both questions get asked of the plan you built, which means you should have run them before the meeting.

Target-setting: probability, not precision

Bessemer is unusually explicit about how hard each budget line should be to hit, recommending the revenue target at roughly 50% probability, profit/EBITDA at 70% (protected by contingency inside the expense budget), an individual sales quota at 30%, and Wall Street guidance at 90%. The 30% quota is deliberate over-assignment, so the aggregate of optimistic individual plans nets down to the 50/50 company number. It shows up as quota coverage above plan, and the arithmetic is exact: assigned quota ÷ plan = 1 ÷ planned attainment, so the $12.0M new-ARR plan built at 75% planned attainment needs $16.0M of quota on the street. That is 1.33x coverage, about 21 fully ramped AEs at a $750k list quota against the 22 the capacity build in section 4 ends the year with. Sourced practice puts over-assignment at 20–30%, roughly 120% in SMB and 135% in enterprise, so 1.2x to 1.35x is the band to expect. Note that the 90% rung is the one that only switches on at IPO — it is about credibility with the Street. A private board normally gets the 50% plan plus a downside case rather than a separate 90% number, because it approved the plan and is not the audience that rung was built for.

Two more habits from the same source. First, an explicit mid-year rebuild trigger: materially off-plan by June means writing a new H2 budget at the same probability levels using current data, not limping the original forward. Second, a warning about the Q4 hockey stick — plans that keep Q1 and Q2 reasonable, sweeten Q3, then load a big jump into Q4 on the theory that early-year hiring and marketing investment will have paid off by then. Bessemer's objection is not only that the jump gets missed. The year-end revenue run rate is what establishes the base for the following year, so a Q4 you budgeted and did not earn propagates forward into next year's target as well.

Growth-stage SaaS: a planned loss is the strategy

The thing to internalize before your first growth-stage AOP is that the operating loss in the plan is intentional and investor-sanctioned. Coming from an appropriated environment, where a planned deficit is a defect, this is the single largest reflex change on the list.

The evidence is quantitative. SaaS Capital's 2026 survey of more than 1,000 private B2B SaaS companies found equity-backed firms spend 70% more on sales, 64% more on G&A, and 100% more on marketing than bootstrapped peers at comparable size. Net of everything, bootstrapped companies run near 96% of ARR in total spend — breakeven by design — and equity-backed companies near 101%. Same revenue, same category, deliberately different cost structure, because one is buying profit and the other growth rate.

The same survey supplies the envelope. Median spend as a share of ARR runs roughly R&D 22% (unchanged year over year at the median, and higher at small scale — 24% in the $3–5M ARR band), sales 15%, G&A 15%, customer success 9%. At $40M of ARR with 250 people that is $8.8M, $6.0M, $6.0M and $3.6M, or 61% of ARR before marketing programs or hosting COGS. Which is how a top-down envelope gets set: pick the target ratio per function, multiply by planned ARR, and let the bottom-up build fight over the shape inside it.

Two cautions on using it that way. Which ARR you multiply matters more than the ratios do. Opening, average and exit ARR are three different denominators: the $40M company planning to $52M (section 4) averages about $46M across the year, so the same 61% supports $24.4M of spend on the opening basis, about $28.1M on the average basis and about $31.7M on the exit basis. That is a 30% difference between the ends of the whole opex plan produced by a definitional choice, and it belongs in the kickoff rather than in the reconciliation. Most companies set the envelope on average ARR; some use exit ARR; ask. And these are medians across a wide size range, not size-invariant constants — R&D as a share of ARR generally compresses as revenue grows, which is why a well-built envelope declines as a percentage over the plan horizon rather than holding flat. Use the survey cut nearest your own scale.

Rule of 40, and the argument that it is the wrong screen

The familiar screen is the Rule of 40: growth rate plus profit margin above 40%. Bessemer argues it under-prices growth and proposes the Rule of X: (growth rate × a multiplier) + FCF margin, the multiplier around 2x for late-stage private and 2–3x for public cloud companies, on the reasoning that margin improvement is linear in value while growth compounds. Their late-2023 read put the cloud market near 31% on Rule of 40 and 50% on Rule of X, with top decile at roughly 48% and 80%.

The two framings rank the same company very differently against their own benchmarks, which is why it matters which one your board uses. Take the running example scored in section 9: 25% revenue growth on a −5.8% FCF margin. Rule of 40 scores it 19.2 against a market average near 31, roughly 38% short. Rule of X at 2x scores it 44.2 against a Rule of X average near 50, roughly 12% short. Neither score is read against the other — each goes against its own benchmark — but a company that looks badly off the pace on one rule looks close to market on the other, which is Bessemer's point about growth being under-credited. Argue for a leaner plan on Rule of 40 logic while the board underwrites on Rule of X and you are arguing against the thesis they funded.

Why the pressure shifted: Bessemer's Cloud 100 data shows multiples compressing from 34x ARR in 2021 to 20x in 2025, while average time to $100M ARR fell from roughly a decade to 7.5 years, against a public benchmark near 8x. Less credit per dollar of growth, less time to produce it. On reporting, the same firm finds internal KPI counts scale from 3–5 metrics early to 15–20 by Series C, while the board deck stays capped at five or six.

Public company: the clock is statutory, the audience external

Public companies must file quarterly (10-Q) and annual (10-K) reports along with other mandated disclosures; private companies carry no such obligation. Every other private-company reporting rhythm is negotiable with a board or a sponsor. This one is negotiable with no one.

Plan of record, guidance, and consensus

The internal triad — board plan, internal plan, quota on the street — belongs to section 4 and survives the listing unchanged. What going public adds is two more numbers, only one of which the company controls, and confusing any of them is the fastest way to look new. The internal plan of record is what the organization is managed and compensated against. Guidance is the range given publicly, and it does not exist in any private company — section 2.5 names it as the fourth number for exactly that reason. Consensus is the average of sell-side analysts' published models, which is what the stock actually trades against on results day and which FP&A typically maintains as a tracked model of its own, line by line, so that the company knows before it reports whether it is beating or missing the number the market holds. The gap between plan and guidance is deliberate: Bessemer's ladder puts the internal revenue plan at 50% probability and the Wall Street number at 90%, because internal targets are motivational and external numbers are credibility instruments. Illustratively: an internal plan of $412M against public guidance of $400–405M. That gap is the beat-and-raise room.

Missing guidance is not survivable. Beating it by a large margin is a different problem rather than a free win: an oversized beat resets the bar you are held to next quarter and reads to the buy side as sandbagging, which is why the beat is managed rather than maximized. FP&A's job in that management is supplying the CFO with an honest range early enough that the guidance can be struck correctly in the first place.

The quarter is the unit of account

At a private company an annual plan that lands is a good plan. At a public company, phasing inside the year matters as much as the annual total, because each quarter is separately reported and separately judged. Worked example: a company plans 120 hires at the start of Q2 at $16k a month fully loaded, $1.92M a month once they are all on the payroll. Recruiting slips an average of six weeks. Those heads are on the payroll for the whole of Q3 either way, so nothing lands in Q3 that was not already going to; what the slip does is delete 1.5 months of cost per head, essentially all of it inside Q2. That is 120 × $16k × 1.5 = $2.88M of planned expense never incurred. Against a full-year operating expense base of $1.8B it is 16 basis points, invisible in the annual plan. Inside the quarter it is a $2.88M favorable, and at a 21% tax rate on 250M diluted shares that is $0.0091 — about a penny of EPS, enough to move a one-cent miss to roughly break-even on a number nobody decided to improve.

Two consequences, and they are the reason anyone cares. First, the favorable has no run rate behind it. Every head arrives, so the exit rate is fully loaded and next year carries the whole cost while this year showed a favorable; the CFO will say so on the call before an analyst says it for him, which means FP&A has to have isolated the hiring timing from real cost performance in the flux (section 8) rather than letting it sit inside a clean-looking opex line. Second, the variant. If recruiting makes the slip up by compressing the rest of the year's hiring into H2, the full year lands close to plan but Q3 and Q4 absorb expense the original phasing had put in Q2 — then the money really does move right, into quarters that were already carrying their own ramp. Both readings start from the same slipped start dates, and which one is true depends on a recruiting decision that has not been made yet in April. That is why a public-company hiring forecast is built by month and defended by month, and why "it evens out over the year" is not an answer anyone will accept.

Non-GAAP and segments

The plan of record at most public technology companies is a non-GAAP plan. Non-GAAP earnings are measures not prepared under GAAP's standard calculations and not required for external reporting, and the standard technology adjustments are stock-based compensation, amortization of acquired intangibles, and restructuring or other non-recurring charges. FP&A owns the GAAP-to-non-GAAP bridge, plans and forecasts in non-GAAP, and reports both. Nothing in state budgeting prepares you for a company whose primary internal operating measure excludes a real, large, recurring cost such as equity compensation; ask early which measure the plan and the bonus run on.

Segment reporting is the other public-company-only workload, and it is covered structurally under the enterprise heading below. The practical note here is that current US requirements have pushed external segment disclosure closer to internal management reporting — public entities must now disclose significant segment expenses regularly provided to the chief operating decision maker, which means the internal management view and the external footnote are no longer independent artifacts. I could not fetch the primary standard text this session, so verify the exact current requirement before quoting the standard by number.

Two calendar effects round it out. Blackout and quiet periods constrain who may say what and when, which pushes internal planning milestones away from the reporting window. And guidance mechanics — how the range is struck, Reg FD, consensus management — belong to section 2, with the control environment formalizing the close in section 11. The comparative point: at a public company the plan is a promise with a legal perimeter around it, so FP&A spends materially more time on documentation and version control than at any private company of the same size. Speed is not the virtue being rewarded.

PE-backed portfolio company: cash first, and the clock is contractual

This is the largest change in operating rhythm on the list, and the one an incoming hire is least likely to anticipate. AFP is blunt: post-close, reporting volume "doubles literally overnight," with daily cash-focused reporting starting immediately. The compression is structural, not cultural. Bain's 2026 read explains the arithmetic behind it: where a 2015 deal could clear a 2.5x return on invested capital over a five-year hold with about 50% leverage at 6–7% rates and only 5% annual EBITDA growth, today borrowing costs sit in the 8%–9% range, leverage ratios are closer to 30%–40%, purchase multiples remain in record territory yet largely stagnant, and deals only pencil out on EBITDA increases "closer to 10%–12% to generate that 2.5x return over five years". Bain's shorthand is "12 is the new 5". Those are annual EBITDA growth rates, not a compressed clock — the hold period did not shrink, the operating improvement required inside it roughly doubled. On $40M of EBITDA, compounding at 12% for five years reaches $70.5M; at 5% it reaches $51.1M. That $19M gap is the work, and it is why the same report urges firms to move from full-potential diligence to "hitting the ground running on Day 1 of ownership". Two other things change at once: sponsors mandate standardized reporting formats so portfolio companies can be compared, so you inherit a template rather than design one; and emphasis shifts from P&L to cash, on AFP's reasoning that cash flow cannot be hidden.

One caveat on the five-year hold, offered as characterization rather than sourced finding: it is the underwriting assumption, not a promise. Through the current cycle actual holds have run longer because the exit window has been narrow, which is why continuation vehicles and repeat sell-side processes have become ordinary. Practically, that means you keep the company exit-ready for years rather than months, and you may be asked to rebuild the model against a fresh set of underwriting assumptions without the company ever changing hands.

The credit agreement is the second binding constraint

Debt covenants are restrictions lenders place on a borrower, split into positive covenants (what the borrower must do — maintain ratio thresholds, deliver audited statements) and negative covenants (what it cannot do — pay dividends, sell assets, borrow more, pursue M&A without consent). The ratio tests named most often are Debt/EBITDA, Debt/(EBITDA − capex), interest coverage, and fixed-charge coverage.

Before the ratio math, the definition trap. The EBITDA in a covenant is not a number you read off the P&L. It is a defined term in the credit agreement with a negotiated add-back schedule, and it is routinely different from both GAAP EBITDA and the adjusted EBITDA in the board deck. A typical bridge: GAAP EBITDA $33.0M, plus stock compensation $2.0M, plus non-recurring restructuring $1.5M, plus the sponsor monitoring fee $1.0M, plus run-rate savings from a completed headcount action $2.5M, equals $40.0M of covenant EBITDA. Adjusted EBITDA in the general sense is the removal of one-time, irregular and non-recurring items from EBITDA, with share-based compensation, litigation expense, one-time gains and losses and above-market owner compensation among the usual normalizations — but which of those your lender permits, whether the run-rate savings line is capped (commonly at a stated percentage of pro forma EBITDA), how far forward it may look, and whether the cash netted in "net debt" is capped are all written in the definitions section of the credit agreement, not in any accounting standard. The quality-of-earnings work done in diligence sets the opening bridge. FP&A signs the quarterly compliance certificate that proves the calculation to the lender. Read the credit agreement in week one; it is the single highest-value document in the building.

Now the math. A company carries $180M of net debt against $40M of LTM covenant EBITDA under a maximum net leverage covenant of 5.00x, tested quarterly. Current leverage is 4.50x, and the covenant holds down to $36M of EBITDA ($180M ÷ 5.00, which lands exactly on 5.00x and therefore still complies). So covenant headroom is $4M: anything worse than a 10% EBITDA miss breaches. And that overstates the cushion, because the numerator moves too — an EBITDA shortfall usually arrives alongside a working-capital drain and a revolver draw, so an honest downside case flexes both sides of the ratio at once. That reorders the job: the downside scenario is a compliance test rather than an analytical exercise, and the first question asked of any reforecast is whether it still clears. Waivers and cure rights vary by deal. All of it is tracked against what practitioners call the value creation plan, the deal thesis expressed as measurable initiatives. AFP uses the term as settled vocabulary but I found no definitional source, so treat my description of its components (organic growth, margin expansion, multiple expansion, bolt-on M&A) as characterization, not a sourced definition.

Large multi-BU enterprise: the plan is a negotiation between P&Ls

At a single-BU company the AOP is one build. At a multi-BU enterprise it is N parallel builds, rolled up, compared against a corporate target, then re-cut — which is why the same process shape takes far longer in elapsed time. I could not source a defensible figure for typical cycle length at this company type (the consultancies that publish on it were unreachable this session), so I will not invent one. The claim is "longer, with an extra reconciliation round"; ask about actual dates in any interview.

What is sourceable is the structure that causes the negotiation. Public companies report by operating segment: a component whose operating results are regularly reviewed by the chief operating decision maker to assess performance and allocate resources to it, and for which discrete financial results exist. A reportable segment's revenue and a defined measure of its profit or loss sit in the disclosures attached to the audited financial statements, get restated when the segment structure changes so that prior periods stay comparable, and are tracked year over year by analysts. So the external structure constrains the internal one, and reorganizing segments is expensive rather than merely inconvenient. The corporate group itself is not one of those segments — the corporate group does not usually earn outside revenues, and so is not considered a segment. That is a statement about whether corporate is a reporting unit, not about where its cost ends up. Segment reporting is typically based on internal reporting reviewed by the chief operating decision maker, and reported measures may differ from consolidated totals because of intersegment transactions and differing measurement bases — which is why the disclosure carries a reconciliation of segment totals back to the consolidated amounts. Whether corporate cost is pushed down into the segments or parked in an unallocated corporate line is therefore a company-by-company choice, not a rule, and both are common. That choice is the whole reason the allocation driver is worth fighting over: it decides whose margin carries the corporate cost, and by extension whose bonus does. Inside the segments, units are classified by what their manager is accountable for: a profit center generates revenues and profits or losses and its manager has authority over both how to earn revenue and which expenses to incur, a cost center is responsible only for its costs, and an investment center for its return on assets. Reclassifying a cost center as a profit center is an organizational change, not a chart-of-accounts edit.

Where the money actually gets fought over

Two mechanisms carry the political weight, and both are FP&A design problems rather than accounting problems. The first is allocation of shared costs (mechanics in section 6). The second is transfer pricing — the method used to sell a product from one subsidiary to another within a company, with variants including market rate, negotiated, contribution margin, cost-plus, and cost-based. The same source is candid about the dysfunction: too low and the selling unit refuses internal orders; too high and the buying unit goes outside while internal capacity sits idle; at pure cost the seller loses any reason to drive its costs down. Across borders it carries a tax dimension on top of the incentive problem, since the transfer price determines which jurisdiction books the profit, which is why the policy is usually owned jointly with tax rather than by FP&A alone.

Worked example, because the numbers are the argument. A $3.2B enterprise with four segments sets a corporate target of 18.0% operating margin, up 150 basis points. The BU submissions arrive at an aggregate 15.9% — a 210 bps gap, roughly $67M, closed in the reconciliation round described earlier in this section. This company pushes corporate cost down rather than holding it unallocated, so corporate IT costs $84M and is allocated to the units on headcount. BU A has 30% of headcount and 45% of revenue, so it absorbs $25.2M. Switch the driver to revenue and it absorbs $37.8M — a $12.6M swing on $1.44B of BU revenue, about 87 basis points of that unit's operating margin. No customer notices, nothing about the business changed, and the BU president's bonus moved. Which is why allocation methodology is litigated rather than decided, and why a new corporate FP&A director who changes a driver in their first cycle without socializing it will regret it.

Plan rates and constant currency

At any enterprise with revenue outside its reporting currency, one of the two or three biggest decisions of the annual plan is the set of plan rates — the exchange rates the plan is struck at, usually a forward strip or a trailing average fixed in the fall and then frozen for the year so that operating performance can be read without currency noise. Actuals are reported both at actual rates and at plan rates, and the monthly bridge separates volume, price and FX. Worked: €400M of planned revenue at a 1.08 plan rate is $432M; the euro at 1.02 delivers $408M, a $24M miss with no operational cause whatsoever. Which is why segment leaders report growth on a constant currency basis, and also why constant currency is the most-used device for reframing a bad quarter. Both uses are legitimate; know which one you are looking at.

Distinguish the two exposures, because only one is real money. Transaction exposure is the risk of loss from a change in exchange rates during the course of a business transaction — payables and receivables — and it affects cash flow and realized profits, while translation exposure is the impact of currency changes on the reported statements of foreign subsidiaries when consolidated into the parent's currency, which affects accounting values without immediate cash consequences. Treasury hedges the first. FP&A explains the second.

The reconciliation round has a name in some organizations. AFP describes "interlock sessions" where every team sits together to align tactics, priorities and resource commitments, with finance sizing the trade-offs and saying plainly when the plan is not feasible without more investment. The same piece calls the resulting budget "literally a contract between multiple parties: between management and the board, and between the CEO and the management team." The framing will feel familiar from public budgeting and the enforcement will not. An appropriation is a contract only as metaphor: it is a unilateral statutory ceiling, and a contract made in excess of the amounts appropriated is null and void. AFP's budget-as-contract is enforced by performance management — you miss it, you explain it, it shows up in your review and in the CEO's next board conversation. Nothing is void. For tooling at this scale see section 12.

Non-tech industries: three different P&L engines

Tech is this guide's default frame, but the question underneath — what drives the P&L, and therefore what the model is built around — has different answers elsewhere.

Manufacturing: standard costs, absorption, and who owns the volume

Start with ownership, because it is the biggest structural difference from every tech example above. The manufacturing annual plan is driven off the sales and operations planning cycle: finance consumes a demand plan and a supply plan and converts them to money. It does not originate the volume assumptions. That is a different seat from the one FP&A holds at a software company, where the revenue plan is built inside finance with the go-to-market team (section 4).

Standard costing means substituting an expected cost for an actual cost in the accounting records, with variances recorded to show the gap. Setting the standard is itself a forecasting exercise, built from the average of the most recent actual cost for the past few months and then adjusted for equipment age and scrap rates, projected labor-efficiency gains, scheduled wage increases, learning-curve effects, and negotiated purchasing terms. The tie to planning is direct: budgets inherently use standard costs, so budget-versus-actual and standard-versus-actual are the same machinery pointed at two horizons.

The two variance families are the price (rate) variance, which the same source defines as (actual price − expected price) × actual quantity purchased, and the volume (efficiency) variance, (actual quantity consumed − budgeted quantity) × standard cost per unit. Worked, assuming purchases equal consumption in the period: standard usage 2.0 lbs per unit at $2.10/lb; actual production of 500,000 units consumed 2.15 lbs each at $2.02/lb, so 1,075,000 lbs. Price variance is (2.02 − 2.10) × 1,075,000 = $86,000 favorable; usage variance is (1,075,000 − 1,000,000) × $2.10 = $157,500 unfavorable. Net $71,500 unfavorable: purchasing did its job, the plant did not, and the headline number alone would have told you neither. Confirm which convention your plant runs before arguing about a number — many shops compute price variance on quantity purchased and others on quantity used, and in a period where purchases and consumption differ the two produce genuinely different results. Same instinct as the price/volume/mix bridges in section 8.

Two things that pair will never explain. The first is absorption. Absorption costing apportions fixed manufacturing overhead to units at a predetermined rate, and overhead is under-absorbed when the amount allocated is lower than the actual amount incurred. If $12M of fixed overhead is absorbed at $12 per unit on 1.0M planned units and the plant builds 850,000, $1.8M sits unabsorbed and lands on the P&L with nothing about demand, price or unit cost having changed. The second is the annual standards refresh, which revalues inventory on hand and therefore moves reported margin at the start of the year before a single additional unit is sold.

Retail: comps, calendar, and open-to-buy

The headline planning metric is same-store sales (comparable-store sales), [(sales at existing stores in period T+1 ÷ sales at the same stores in period T) − 1] × 100, because it separates organic demand at established locations from growth that is really just new store openings. Check the definition before quoting the number: companies differ on how long a store must have been open to enter the comp base, and on whether e-commerce sits inside it, and both choices move the result materially.

Retail also does not run on calendar months. The NRF 4-5-4 calendar divides the year into months of four, five and four weeks so that comparable months hold the same number of Saturdays and Sundays and like days are compared to like days, with a 53rd week added approximately every five to six years and the following year restated to keep holidays aligned. A period is four or five weeks, not a month, and every year-over-year comparison needs the calendar alignment done before it means anything.

The merchandise budget runs on open-to-buy, which calculates how much inventory the business needs to buy to hit planned sales: opening stock + purchases − sales = closing stock, at cost. That source's example — $100,000 opening stock, $80,000 of COGS on $200,000 of sales at 60% margin, $20,000 on order — leaves $40,000 closing. If the plan needs $70,000 for the next season, open-to-buy is $30,000: the dollar authority the buyer has left. Set twice a year for six-month seasons and revised as actuals land, it is the closest private analog to an allotment — a periodized ceiling on how much of the plan may be committed, revised on a schedule. The break is what sits behind the ceiling. An allotment paces authority the Legislature appropriated, and spending past the appropriation voids the contract. Open-to-buy paces working capital, and a buyer who exceeds it has made a bad inventory call, not committed a violation.

Professional services: capacity, not bookings

A services firm builds revenue from a capacity model rather than an ARR waterfall (section 4): billable heads × available hours × target utilization × realized bill rate. Utilization rate is the share of a consultant's available time spent on billable work. Worked: 60 consultants × 1,900 hours × 72% × $225 per hour = $18.47M. One point of utilization across that base is 1,140 hours, about $257k, which is why the whole firm is managed on it the way a SaaS company is managed on headcount ramp (section 5). Be careful with benchmarks: I could not source a defensible median. The industry's main study is paywalled and exposes only a comparison — firms at the highest process maturity level see 42% more billable utilization than Level-2 peers — and the only absolute numbers I found are a vendor's illustration of firms moving from 65% to 72%. The 70–80% range quoted in conversation fits that, but is not a number I would put in front of a CFO.

What each type expects from a manager or director hire

One caveat: only the PE row is directly sourced. AFP reports that sponsors screen portfolio-company finance hires for prior PE experience as a credibility signal, comfort in high-velocity environments, business partnering skills over spreadsheet-focused work, automation ability, and the maturity to set realistic capacity boundaries rather than silently absorbing an unsustainable ask. The other rows are my synthesis of what each type's sourced pressures imply, not independent findings.

The table collapses two different jobs, so separate them before you negotiate a title. At every type, the manager job is owning models and a partner set, and being graded on accuracy and turnaround. The director job is owning the cycle, the calendar, the narrative to the executive team and at least one direct report, and being graded on whether the plan lands and whether you are willing to tell a business leader no. Knowing a BU's P&L better than its president does is a month-six bar for a director, not a month-one one, and a director who spends thirty days in the data building no relationships has failed the actual test.

Then watch the scope inversion, which is the most common mis-set expectation in a title conversation. A director at a 300-person growth-stage company covers the entire company and reports to the CFO. A director at a $3B enterprise covers one business unit or one function and sits three levels below the CFO. Both say director. The first is broader and thinner; the second is narrower and much better resourced. Which one fits depends on whether you want range or depth, and it is a fair question to ask in a first screen.

The practical read for someone moving out of a $5.2B state program: your experience maps most naturally onto the large multi-BU enterprise and the public company, both of which run on process discipline, roll-up negotiation, allocation methodology, and a fixed external clock — the types where your habits are assets rather than things to unlearn. Growth-stage SaaS and PE require a real change in reflex, toward a deliberate planned loss in the first and weekly cash and covenant math in the second. Both are easy to get visibly wrong in month one if you arrive assuming the old grading metric still applies. The vocabulary map for all of it is in section 15, and section 14 closes the loop on the first AOP cycle.

14. Business partnering and the operating rhythm

Sections 3 through 8 describe machinery: templates, versions, cadences, bridges. This section covers the layer that decides whether the machinery produces a plan anyone believes. None of the four core questions are answered here — the AOP process is section 3, revenue ownership is section 4. What is here has no template: how you work with the people whose numbers you consolidate, what the private sector means when it says a budget is "owned," and how the annual plan gets negotiated rather than how it gets designed.

This is also the weakest-evidenced part of the guide. The business-partnering literature is mostly practitioner opinion and vendor content, and its hard numbers are often stale or uncorroborated. Where a claim below is settled practice it is stated flatly; where it is one firm's framing, or my inference, it says so.

14.1 What "business partner" actually means

Finance business partnering is collaboration between finance and the operating units "to enhance decision-making and optimize resource allocation," in the AFP glossary definition. Bland enough to be useless, so use the practitioner version. Bartosz Obojski's three stages are worth memorizing: knowing the numbers, understanding the numbers, agreeing with the numbers. Knowing is the P&L, the KPIs, the balance sheet. Understanding is the operational drivers behind them, and his claim is that most finance people stop there. Agreeing is applying your own judgment to the inputs and being willing to say, in front of the business, that these are your numbers too.

The corollary matters more than the framework. Obojski explicitly excludes month-end close, report generation, variance analysis, process optimization and data-accuracy work from "true" partnering. Those are the price of admission, not the job. A team that does all of them flawlessly and nothing else is a well-run reporting function, and its own performance reviews will call it "not strategic."

This is not only a vendor preoccupation. AICPA and CIMA published design principles for finance business partnering in 2020, casting the partner as a knowledge orchestrator and naming the barriers as lack of clear strategy, insufficient training and poor communication channels. And it is what employers screen on: the 2024 FP&A Trends Survey reports business partnering as the top skill sought when hiring, at 50% of organizations, up nine points year over year.

A four-role ladder appears in Pigment's guide: controller (understand the business, hold cost to budget), advisor (stakeholder relationships, what-if analysis on real decisions), influencer (driver-based planning, identify value drivers), and strategic partner (scenario planning against what could move results). Two cautions before repeating it. Pigment attributes only its one-line definition of a business partner to a PwC report; the four roles carry no attribution, so do not call this "PwC's model." And Pigment presents them as responsibilities a team collectively establishes, not a seniority ladder — reading them as levels (analyst does controller work, a director is expected to be influencer and strategic partner or is an expensive manager) is my gloss, though it is how the levels get staffed. Team structure belongs to section 1, where AFP's three-level structural model (2019) is the standard reference.

14.2 What "budget ownership" means

This is the largest vocabulary shift in the section, and its bluntest statement comes from a boutique advisory firm rather than an institution: "Who owns your company's budget? If the answer is 'finance,' then nobody owns it." The same piece supplies the line to carry into interviews: the finance partner does not own the budget, they own the analysis. The operating model is a triad:

• Budget owner (also budget holder) — the VP or department head who controls the spending decisions and answers for the result. The load-bearing condition is real authority over the spend they are accountable for; accountability without authority produces theater.
• Finance partner — you. Reporting, variance analysis, forecast, challenge. Owns none of the spend.
• Executive sponsor — the escalation authority who approves exceptions and resolves conflicts when two owners want the same money. The source says "typically a C-suite member or senior director," not specifically the CFO, and plenty of companies never name the role at all. Budget owner and budget holder are the two words you can count on hearing.

What makes ownership real is that the owner built the number from drivers rather than receiving it as an allocation. Glencoyne's version is a marketing team justifying spend as "4,000 MQLs × $50 CPL = $200,000" instead of asking for a lump sum. That formulation is what lets a budget survive a mid-year cut conversation, because the cut now has a stated consequence in leads rather than an unstated one.

Controllable, allocated, and fully burdened

An owner's number has three layers, and confusing them is a common first-quarter mistake. Controllable spend is what the owner decides: program spend, vendors, headcount. Allocated spend is pushed into their cost center by a rule they did not write — facilities, IT, corporate overhead, cloud infrastructure split across product lines. A fully burdened view stacks both and is what margin analysis needs. The norm is that owners are held to controllable spend, see allocations reported but not charged against their performance, and renegotiate the allocation methodology at AOP rather than mid-year. The mechanics of pushing shared costs are in section 6; the partnering consequence is the warning. Changing an allocation driver mid-year restates every owner's variance history at once, and is the fastest way for a new finance director to lose the room. If you must, restate the prior periods first and walk each owner through their own before-and-after in private.

The ask is a negotiating position

The triad describes accountability. It does not describe how the decision gets made, and that gap is where a new director gets surprised. Start with the envelope. In most companies the target descends before the asks ascend: the CEO and CFO already hold an opex number and a burn or margin outcome they intend to land, sized without reference to the submissions. Bottom-up asks compete for a pool fixed before anyone opened a template. Expense budgets are also downstream of the revenue commit, which gives the CRO structural leverage no other VP has (section 4). And the ranking is usually pre-wired: by the time an ask reaches exec review it has been socialized in one-on-ones, and a request that first surfaces in the room is a request that loses. The reliable signal of where power sits is who the CEO calls when the number moves, not the org chart. Find the envelope-setter and the target in week one, because everything you build is an allocation problem inside them.

Then the padding. Owners submit above what they expect to receive because they expect a cut; finance haircuts because it expects padding; both sides know. That is budgetary slack, defined with its causes in section 3.5. What that section does not cover is what you do about it in the room. Four counters that work:

• Ask for a ranked list, not a total. "If I fund 80% of this, what is the first thing that dies?" A padded submission has no answer; a real one has an ordered one.
• Force the marginal-dollar question. What does the $6.2M-to-$6.5M dollar buy versus the $6.5M-to-$6.8M dollar? Owners argue totals; make them argue increments.
• Ask for the driver decomposition. It surfaces padding without accusing anyone, because a padded line has no driver behind it. It is the only reason a placeholder is findable.
• Do not haircut uniformly. An owner who submits honestly and gets cut like everyone else learns to pad next year. A flat percentage manufactures the behavior it complains about.

Worked example: a budget owner conversation at a $40M ARR company

A 250-person SaaS company at $40M ARR, calendar fiscal year. The VP of Marketing owns $6.2M: $3.8M of program spend and $2.4M of fully loaded headcount for 14 people. At AOP kickoff she asks for $7.4M. You do not own that decision, but you are rarely a bystander either — at most growth-stage companies finance sits in the approval path for the transaction even when it does not own the choice: new requisitions, purchase orders above a threshold, non-standard vendor terms, off-cycle asks (section 11). The honest formulation is that you own the recommendation and the gate; the owner owns the choice and the consequence. Pretending you have no influence disclaims a role you actually hold. Cube's five questions to a budget owner are a decent script: has every expense and revenue effect been accounted for; how were the numbers validated against history, benchmarks or market data; what known changes will hit this budget; how are risk and uncertainty handled; and what assumptions is this resting on, with what confidence.

Run through them and the $1.2M increment decomposes: $400k of demand gen tied to a mid-market push, $310k for two marketing hires at roughly $155k fully loaded each, $290k of contracted price increases on existing tooling, and $200k with no driver behind it. The first three are decisions for the CEO. The fourth is where the naive version of this story ends, with the placeholder exposed and everyone better off. That is not what happens. It comes out of the demand-gen line and reappears inside a defensible driver at a slightly higher cost per lead, or it returns as a Q2 off-cycle ask, or she pads two lines next year instead of one. What you bought is a $200k reduction and a piece of information about how this owner submits. Both are worth having. Neither is gratitude.

One live disagreement, because you will hear "ownership" used both ways. The triad draws a clean line: department heads own spend, finance owns analysis. Vena's August 2026 piece pushes further, defining decision latency as "the gap between knowing what to do and doing it" and arguing FP&A should own outcomes rather than plans: recommend what happens next, define the expected result, name who owns it. The term long predates the article, which does not claim to have coined it. Its supporting statistic — 73% of leaders estimating up to 5% of annual revenue lost to slow decisions — is properly sourced, to West Monroe's "Speed Wins" research: 214 C-suite executives and 1,000 managers at US companies with at least $250M of revenue, fielded November 2025. Self-reported estimates of a hypothetical loss, so directional rather than measured, but it is a named study you can cite. The reconciliation that holds: the budget owner owns the spending decision, FP&A owns making sure it gets made on time.

14.3 The operating rhythm

The operating rhythm (also operating cadence) is the fixed calendar of recurring meetings a company runs itself on. Section 2 covers the planning layers and section 8 the reporting package; what matters here is who sits in each meeting and what they owe each other. The most useful concrete practice in current sources is unglamorous: Chris Ortega, quoted by AFP in June 2026, tells finance people to "get out of your spreadsheets" and "have a biweekly one-on-one with your sales or marketing lead." That standing thirty minutes is where partnering happens. The MBR is where it gets audited.

A vocabulary trap in that table: QBR here means the internal quarterly business review. In a SaaS company the unqualified acronym more often means the customer quarterly business review that Customer Success runs with an account. You will hear both, sometimes in the same week. Ask which one is on the invite.

A month at the $40M ARR company. Books close on business day 5, but you have preliminary actuals on day 3 and are already calling owners about anything over threshold, so by day 5 you have explanations rather than questions. The MBR deck goes out end of day 7 for a day-8 meeting, and the rule separating competent teams from the rest is that nothing in that deck is new to the person it concerns. Days 9 to 15 are forecast updates and the 1:1s. Days 16 to month-end are the analytical work — pricing, CAC payback, the hiring-plan rebuild — the only part that compounds and the first part eaten when close slips.

14.4 Partnering by function

The generic advice above is the same for everyone. The actual conversation is not: what a sales leader needs from you has almost nothing in common with what an engineering leader needs.

Sales and go-to-market

The most consequential number you will touch is quota coverage: total quota assigned across the team divided by the bookings target, usually 1.2x to 1.35x — roughly 120% in SMB and 135% in enterprise (section 4) — so the plan does not depend on every rep hitting. Under it sits pipeline coverage, which is 1 divided by the win rate rather than the folk 3x (section 4). The partnering point is who supplies the input: the CRO's assumed win rate is usually the plan's real load-bearing number, and it is the one to challenge, not the enthusiasm around it. The rest of the job: commission plan design and the monthly commission accrual; forecast categories (commit, best case, pipeline) and why finance's number and the CRO's differ by construction rather than by dishonesty; deal desk and non-standard pricing approvals. One boundary to establish early — RevOps or Sales Ops usually already owns pipeline reporting, and rebuilding their dashboards in month one is a common new-director own goal that costs you the relationship you most need.

Product and engineering

At a SaaS company R&D is normally the largest single cost center, and it is the function this literature ignores. The dynamic differs structurally because engineering carries no direct revenue accountability, so the conversation is capacity, headcount and vendor spend rather than P&L: how many engineers, at what level, in which geography, and what the cloud and tooling bill does as usage grows. Two things you will own that product leaders do not think about. First, the capitalize-versus-expense judgment on software development cost, which moves reported margin without moving cash and which finance decides (section 6) — expect to spend real time collecting the time-allocation evidence behind it. Second, translation: engineering leaders argue in roadmap and story points, and someone has to convert that into dollars per feature and a build-versus-buy comparison. That someone is you, and doing it well is the fastest route into product reviews you were not invited to.

G&A and the rest

The smallest budgets and the most contractual spend, which makes the work renewals, vendor consolidation, and knowing every auto-renew date before it passes. Unglamorous, and the first place a CFO looks when the target changes.

14.5 OKRs and KPIs

OKRs (Objectives and Key Results) are a goal-setting system common in tech though far from universal — adoption thinned during the 2022-24 efficiency period, and plenty of companies run annual goals plus a metric sheet and call it OKRs. Ask which you are dealing with. The objective is qualitative and directional; the key results are "specific, measurable, time-bound" indicators of progress toward it. The canonical design is bottom-up, and Google's guidance describes "a mix of top-down and bottom-up suggestions"; common practice is that company objectives descend and teams author the key results underneath them, so the negotiation is over targets rather than direction.

Now the convention that will trip you, where the popular summary is dangerously incomplete. Most systems run two grades of OKR, and Google's playbook is explicit about both. A committed OKR is a promise: "the expected score for a committed OKR is 1.0; a score of less than 1.0 requires explanation for the miss, as it shows errors in planning and/or execution." An aspirational or stretch OKR is a bet, with "an expected average score of 0.7, with high variance." The famous sweet spot — "the sweet spot for an OKR grade is 60% to 70%; if someone consistently fully attains their objectives, their OKRs aren't ambitious enough" — belongs to the aspirational half only. The playbook names conflating the two as the central trap. So the practical instruction is not "do not treat 0.65 as a miss." It is: find out which kind you are graded on before your first review, because the same 0.65 is a success on one and a real failure on the other. Ask one more question while you are there — whether OKRs feed compensation. Google's guidance is that "OKRs are not synonymous with employee evaluations," but companies do it anyway, and where they do, every KR behaves as committed and sandbagging follows within one cycle.

KPIs (key performance indicators) are the other thing, and the contrast is worth keeping crisp: an OKR sets goals that drive change while a KPI tracks ongoing performance. OKRs run quarterly; KPIs "track long-term performance trends and usually stay stable for years." They interlock in one direction: a KPI going soft is what prompts an OKR, and an achieved OKR gets retired into a maintained KPI. Note also that at many companies the metrics that actually govern are the board-facing efficiency set — net revenue retention, CAC payback, Rule of 40, burn multiple — rather than the OKR sheet; those are defined in section 9.

FP&A's documented role here is translation and funding, not authorship. Pigment has FP&A driving alignment by sharing the key financial KPIs with the teams that own them; CFI's four tests for a KPI are relevance, measurability, actionability and simplicity. What no source documents is a process by which FP&A and a department head jointly write the objectives. Authorship almost certainly sits with the business, with FP&A costing and challenging, but that is an inference from the bottom-up norm rather than something the literature states. Ask directly wherever you land.

Worked example: pricing an OKR

The CRO's objective is "establish mid-market as a real segment." KR1: 60 mid-market logos closed in the year. KR2: mid-market from 12% to 25% of new ACV. KR3: mid-market CAC payback under 18 months. None of that is a budget. Your contribution is to turn it into one and hand the trade-off back.

Start with demand. Sixty logos at a $34k average contract value is $2.0M of new ARR. At a 22% close rate on qualified opportunities and an 8% MQL-to-SQL rate, that needs roughly 3,400 mid-market MQLs, which at a $115 blended cost per lead in that segment is $391k of demand gen. It also implies about $9M of qualified mid-market pipeline, since 1 divided by a 22% win rate is 4.5x coverage — worth saying out loud, because that is a bigger number than anyone in the room has pictured.

Then capacity, where you have to name your benchmark and your departure from it. The Bridge Group's 2026 AE research (158 B2B companies, tenth edition) puts median AE OTE at $200k against a median quota of $960k, a 4.6x quota-to-OTE ratio, with median ramp at 6.2 months. I am deliberately setting the mid-market quota lower, at $700k, because the segment has no playbook and these reps are new to it — that is 3.5x rather than 4.6x, and if the CRO wants to plan at the benchmark the headcount ask drops. At $700k and a $34k ACV, a fully ramped rep closes about 21 logos a year. Hire in March against a 6.2-month ramp and roughly 55% of annual quota lands in year one: about 11 logos per rep. Sixty logos therefore needs six AEs, not the three a full-year benchmark calculation would suggest.

Cost: six AEs at $200k OTE, roughly $260k fully loaded once employer taxes, benefits, tooling and allocated overhead are on top (section 5). Six times $260k is $1.56M annualized; hired in March that is ten months, so $1.30M in-year. Ramp does not change the cost — it is why only about half the bookings land. Add demand gen and the total ask is $1.69M. The plan books about 66 logos and $2.24M of new ARR, of which roughly $470k is recognized in-year, since bookings cluster in the last five months and average about two and a half months of revenue each. Now grade it against the CRO's own KRs, which is the point of the exercise.

• KR1 clears with slack — 66 logos against 60.
• KR2 fails, and not for a reason the CRO controls. The KR is a ratio whose denominator is company-wide new ACV, which the plan carries at $10.5M. $2.24M of that is 21%, short of the 25% target even in the good case. That sentence belongs in the planning meeting: a KR that cannot be hit by the plan funding it is a drafting problem, not a performance problem.
• KR3 clears on paper. $1.69M over 66 logos is $25.6k of CAC per logo, and at 75% gross margin a $34k ACV throws off about $2,125 of monthly gross profit, so payback is roughly 12 months against an 18-month target.

The sensitivity that matters is not spend, it is productivity. If ramp runs long and reps deliver 8 logos instead of 11, you book 48: KR1 misses outright, KR2 drops to 16%, and because the cost is fixed the CAC per logo rises to $35k and payback goes to nearly 17 months, so KR3 survives by a month it does not deserve. The honest sentence to the CRO is: this is a segment-entry bet, not an in-year payback bet; in-year revenue less this spend is $1.2M underwater, KR2 is unreachable as written, and KR1 and KR3 both turn on one assumption, which is how fast six new reps ramp in a segment we have never sold to. Delivered in the planning meeting rather than the Q3 variance review, that is what a seat at the table buys.

14.6 The "no surprises" norm

"No surprises" is a named rule in CFO governance writing — Board Agenda lists it first among three trust behaviors between a CFO and a board, ahead of ongoing dialogue and two-way communication: "Chairs and committee leads should hear about material shifts before formal meetings; bad news should come early and with context." What it is not is a formal FP&A framework with a canonical citation. Treat it as an established norm with an informal label. The clearest statement of the same idea inside FP&A is Abacum's: "Bad news doesn't get better with age. Communicate openly and avoid surprises in high-stakes meetings."

Operationally it means three things. A number that will embarrass someone reaches that person privately before it reaches a deck. A forecast change reaches the CFO when you believe it, not when the cycle would have surfaced it. And pre-wiring — walking each stakeholder through the material one-on-one beforehand — is not maneuvering but the expected preparation; skipping it reads as unprofessional. The example: on day 4 of close, marketing is $210k unfavorable to a $950k quarterly program plan. The wrong sequence writes it into the MBR deck and lets the VP meet the number in the room. The right one is a call that afternoon, which produces the real explanation — a conference booth planned for Q4, signed in September to hold the space — and a joint decision to move $210k out of the Q4 forecast so the full year stays flat. By the MBR the variance is a resolved item with a named owner, and you spent credibility on nothing.

14.7 Partnering through a cost action

Every example so far adds money. Most of what a new FP&A director actually does in 2026 takes it away. Deloitte's Spring 2026 European CFO survey of 1,136 CFOs across 12 countries found "unprecedented consensus on cost reduction, which ranks among the top three priorities across all surveyed countries for the first time in the survey's history," with only 23% expecting to grow headcount over the next twelve months against 36% expecting decreases. European rather than US, so read it as direction, not a US benchmark — but the direction is not subtle.

The CFO signals a target, and your first question is which kind of number it is. Take a $4M target at the $40M ARR company. Nineteen open requisitions with no offer extended are $3.1M of annualized cost, 78% of the target on their own — but those roles were phased across the back half of the year anyway, so freezing them saves only about $1.3M in-year. If the CFO means run-rate, the req freeze nearly does it alone. If she means cash out the door this year, it barely starts, and the rest has to come from vendor renegotiation, program deferral and contractor terminations, which are slower, more contractual and more visible. Ask which before you build anything.

Then build an option set by owner with stated consequences, not a uniform haircut across cost centers, and pre-wire every owner individually before any list circulates. Decide the freeze mechanics explicitly rather than letting them be discovered: offers already extended are honored, backfill policy is stated (a "hiring freeze" usually leaves backfills flowing — see section 5), and contractor conversions are named. Those details are where most of the money and all of the trust damage live. The VP who learns from a recruiter that her reqs were pulled has lost confidence in you permanently, and that is the concrete price of breaking the symmetric norm above.

One constraint here has no analog anywhere else in this section. During a reduction in force you will hold information you cannot share with the people you have spent a year teaching to tell you everything first. There is no clever way to hold both. What you can do is refuse to lie — "I can't discuss that" is survivable and "nothing is happening" is not — and get back to full candor the day the news is public. How you handle that week determines whether the relationship survives it.

14.8 Commentary and storytelling

Storytelling is a named, evaluated skill in private-sector FP&A, not a soft extra. Pigment defines the finance business partner as "a storyteller who delivers financial and analytical information to decision-making teams." CFI puts the demand more bluntly: companies want "FP&A people that don't just dump the data on the business's lap, but can tell them what to do with it."

The craft rules are concrete. Tailor to audience — executives need headlines, department leads need the detail underneath. Set materiality thresholds proportional to line size rather than absolute; as Wall Street Prep puts it, whether a million-dollar line is $100 off does not matter. Write the explanation next to the number on the analysis, not in a separate memo, and go get it in person with data already in hand, using the trip to learn what the team is struggling with rather than only to collect a sentence. And police direction words: "higher" and "lower" are ambiguous, because an expense coming in higher than plan is a negative variance to profit. The vocabulary used instead is favorable and unfavorable, always relative to profit.

The same variance, written three ways

• Raw: "Marketing expense was $210K higher than budget in Q3." True, useless, and ambiguous in exactly the way just described.
• Process-shaped: "Q3 marketing expenditures exceeded the allotment by $210K due to timing of conference obligations." Accurate, complete, and written for a reader whose question is whether the rules were followed. This is the register to unlearn.
• What lands: "Marketing is $210K unfavorable in Q3, all of it the conference booth we planned for Q4 and signed in September to hold the space. Full-year marketing is unchanged at $3.8M and Q4 program spend drops to $740K. Priya and I moved it in the forecast today. No impact to the full-year EBITDA commit." Same facts, ordered by what the reader has to decide, with the decision already made and the owner named.

The spoken register

Written commentary is half of it. These are the idioms you will hear in your first forecast call, each committing the speaker to something specific:

• Commit — a number you will be held to. Upside or best case — one you will not. Never let the two blur in a written summary.
• Sandbag — to submit a number you are confident of beating. An accusation about someone else, a tactic about yourself.
• Run rate — annualizing a recent period. The most abused shortcut in the vocabulary; always ask which period and whether it contained anything one-time.
• Hold the number — absorb a miss inside your own budget rather than passing it up. A real obligation, not a figure of speech.
• Take it out of the forecast — a reallocation someone must actually land, with a line and an owner.
• Controllable — the part of a variance the owner could have influenced, and therefore the part they will be judged on.
• Above the line / below the line — which line varies by company (gross profit at some, operating income at others). Ask.
• Walk the bridge — narrate the step-by-step reconciliation from one number to another, in order, out loud.
• Back to plan — "what's the plan to get back to plan" is a demand for a written recovery path with dates, not a rhetorical question.
• Haircut — an unexplained percentage reduction applied to a submission.

The WA-to-private term map is in section 15 and does not repeat these.

The sentence-writing is being automated. A May 2025 practitioner account from Genpact describes AI "automatically generating commentary (e.g., 'Q2 revenue is 5% higher than expected due to increased enterprise sales in North America')." That is roughly the first bullet above. The same piece has AI doing the stakeholder chasing too, assigning "engaging stakeholders" to automated notification agents. What it does not automate, and what the third bullet is made of, is deciding which variance is worth a phone call and how to frame it to the person who owns it. That is my read, not the article's claim. See section 12 for what the tooling does today.

14.9 How FP&A earns a seat

The test is behavioral, and Abacum states it well: "do my business partners want me in the room?" The same piece quotes a general manager describing the finance leader who had earned it: he "not only reported on the P&L, but I helped shape the P&L." The most-cited single trait in CFI's account of FP&A leader surveys is humble curiosity, named by 25-30% of leaders: genuine interest in how the operation works with no problem-solving agenda attached. Its example is a finance person who walked the production floor until they understood operations well enough to spot problems as they happened rather than in next month's variance. The tech equivalents are sales calls, contract negotiations, campaign planning, and product reviews you were not required to attend.

A concrete first ninety days, since the posture is easier to state than to execute:

1. Weeks 1-3: meet every budget owner one-on-one and ask each the same three questions. What decision are you trying to make this quarter? What number would change it? What report do I send you that you already ignore, and why?
2. Weeks 3-6: find the one recurring report the business hates and fix it. A fixed report buys more standing than a good analysis, because it costs them nothing to notice.
3. By day 60: ship one analysis nobody asked for that changes an actual decision. One. A second before the first has landed reads as hobbyism.
4. Not before day 90: propose a process change. Run one full close and one full forecast cycle first, or you will redesign around a problem that was this month's only.

Two months in the business without an opinion buys the standing to have one in month three. Budget the calendar time in week one, because it is what your director evaluates and no form records it. Watch also for the reputation you inherit: if your predecessor was the department of no, someone will test you early with a request they expect refused, and how you handle that request sets your position for a year. The failure mode in the other direction is old and well documented. A February 2017 FP&A Trends board write-up quotes Samipendra Chaudhury, then CFO of Nielsen Emerging Markets: "The finance community is guilty of having an inward-looking approach... the less we come across as watchdogs and more as an enabler it creates a more collaborative approach and FP&A is seen as problem solvers." A budget analyst who arrives fluent in controls and short on business context gets sorted into the watchdog category fast, and it is sticky.

14.10 What this means for you

Most of this is not new work for you. It is the same work with different words, a faster clock, and no rule to fall back on. You already run the finance-partner side of the triad: you hold program staff accountable for numbers you did not build and cannot spend, you evaluate requests against constraints set above you, and you defend a consolidated position you negotiated line by line. Consolidating dozens of program submissions into one internally consistent request is the AOP consolidation problem in section 3. Running a $5.2B forecast means you have already learned the thing that cannot be taught quickly: how to hold a number you did not personally build and still answer for it. Caseload-driven forecasting is driver-based planning under a different name. And decision packages are business cases against a hard submission date. Do not make the same claim about fiscal notes: a fiscal note is "an objective statement of the fiscal impact of proposed legislation" that "must be factual and objective," on a due date OFM assigns per request (generally 72 hours). It is the neutral cost estimate supporting somebody else's decision, and that register is one of the things you have to stop writing in.

Your current vocabulary	Say this instead	What actually differs
Agency / program / division	Budget owner, budget holder, cost center owner	A named individual, not an organization. Their name is on the line.
Agency budget office, budget analyst	Finance business partner, FBP	Embedded and continuous rather than cyclical, with no approval authority attached — unlike OFM, which approves allotments.
Allotment monitoring, expenditure review	Budget-vs-actual, BvA, monthly variance review	Monthly and forward-looking. The point is the revised forecast, not the compliance finding.
Higher / lower than allotment	Favorable / unfavorable to plan	Always oriented to profit, so an expense overrun is unfavorable, never "higher."
Agency performance measures	OKRs and KPIs (distinct things)	OKRs run quarterly in two grades — committed, scored to 1.0, and aspirational, scored to about 0.7. KPIs stay stable for years.
Briefing, narrative, decision package writeup	Commentary, the story, the walk	Ordered by the decision the reader faces, not by process chronology.
Allotment revision	Reforecast, latest estimate	No external body approves it; your CFO decides, monthly or quarterly. It revises the expectation and moves no authorized number.
Supplemental request	Re-baseline, re-plan, board-approved plan revision	Triggered by an event, not a session calendar, and it moves the plan of record rather than the expectation.

What will genuinely be different in your first AOP cycle. Three things, in rough order of what they will cost you.

1. Authority is negotiated, not granted. You will spend your first quarter looking for the rule that settles a dispute and there will not be one. A VP reallocates $300k across categories in March because their VP agreed; no proviso was violated because none exists. What you have instead is the quality of your analysis and your standing with that person, which is why the presence-then-fluency-then-challenge sequence is a job requirement rather than a nicety.
2. Informal relationship-building is a stated deliverable. The biweekly one-on-one, the sales call you sit in on, the product review you attend uninvited — these are what your director evaluates, and no form records them. A review saying "not seen as a partner by the business" means something specific and career-limiting.
3. Three separate clocks speed up, and not by the same multiple. Your reforecast cadence is already quarterly and you should say so in an interview: ERFC publishes four official forecasts a year on statutory dates (RCW 82.33.020), caseload forecasts come "at least three times each year" (RCW 43.88C.020), and allotment revisions are quarterly. What changes is the layer underneath. Formal reporting goes from periodic to monthly with commentary due on a fixed business-day deadline, the forecast refreshes monthly rather than quarterly, and ad hoc analysis goes from weeks to same-day. Nothing waits on an external forecast body or an OFM approval before the number moves. Precision expectations drop to match: a directionally right answer on day 2 beats a defensible one on day 12, and the instinct to get the number exactly right before speaking is the single habit most likely to read as slow.

The term map for the rest of the WA vocabulary is in section 15.

15. Public-to-private translation glossary

Sections 1 through 14 each carried a translation callout. This section collects them into one lookup table, then isolates the six mappings that will cost you something if you carry the state habit across unexamined. The table is ordered by the machinery each term belongs to rather than alphabetically, because the terms that mislead travel in groups.

Almost every break below comes from one fact: Washington's budget vocabulary encodes legal authority, and private-sector budget vocabulary encodes management expectation. An appropriation is "a legal authorization to make expenditures and incur obligations for specific purposes from a specific account over a specific time period" (OFM glossary). A board-approved operating plan is a number people agreed to try to hit. When a mapping feels too good, check whether you have quietly imported the authority along with the concept. You usually have.

The term map

The translations that mislead most

Six rows above are not merely imprecise. They will make a habit that serves you at DSHS misfire inside a company, in ways nobody will correct out loud.

1. The plan is not authority, for money or for people

You have spent years where the budget document and the spending permission are one instrument. They are separate systems in a company, and they fail in both directions. A director with $2.4M in her plan for a data platform cannot sign a $400K contract on the strength of that line; she needs a requisition, a purchase order, and someone whose delegated limit covers $400K. Mechanically that now runs through an intake tool such as Coupa, Zip, Ramp or Navan, where one purchase request collects legal, security, procurement and finance approvals in sequence. Whether FP&A is a named approver in that workflow decides whether you see a surprise renewal in advance or at the accrual, so getting added is a first-month ask.

The reciprocal is softer than the folklore suggests. Published policies commonly condition a delegated limit on the item being budgeted, set separate lower thresholds for unbudgeted spend, and make procurement approval a precondition of signature authority rather than a parallel track. So an unbudgeted $180K renewal usually escalates a level rather than clearing the day a senior VP with a $250K limit signs it. Usually is not always, and which model you are in is a week-one question, not an assumption. Headcount behaves the same way: eight approved roles are a budgetary provision, not permission to hire, and companies differ sharply on when a plan row becomes an open requisition. Some release the full year at AOP approval and let requisitions open on planned start dates, some gate quarterly against performance, some run a weekly hiring committee. Find out which, and find out who holds the trigger, because that person controls your largest line and does not report to you. Delete the sentence "it's in your budget." The question is who signs.

2. Monthly phasing is not an allotment

You will be asked to spread the annual plan across twelve months and load it, and the muscle memory says allotment filing. Nobody receives it, it is not a ceiling, and overrunning it violates nothing. Spread $9.6M of marketing evenly at $800K a month, then watch the user conference consume $2.1M in February alone: $1.3M unfavorable in the month, a year-to-date gap that persists until the offsetting underspend catches up in the back half, and a quarterly deck showing marketing over plan. None of it was real, and all of it needs explaining.

The counterweight is that phasing is consequence-free only inside a quarter. Quarters are not free, because the board sees quarters and the cash forecast is built off the phasing, so a badly phased quarter produces a cash conversation nobody needed to have. Put the effort where timing carries information, meaning hiring start dates, renewal months, the events calendar and usage seasonality. Get the quarter right even when the months inside it are approximate, and stop optimizing the spread of everything else. Bad phasing has no formal consequence, only your credibility bleeding out one variance comment at a time.

3. Nothing lapses, but the base still bites

Under RCW 43.88.140 unspent authority expires, so obligating it first is rational. Import that and you do real damage: a December spend-down reads to a CFO as waste. Do not swing to the opposite error either, which is treating an underrun as self-evidently good. An underrun is favorable in the P&L sense and is graded on what the money was supposed to buy. $600K of unspent demand generation next to a pipeline miss, or an R&D underrun that is really four unfilled engineering roles, reads as failure to execute rather than thrift, and the question after "why are you under" is always "what did we not get."

Underspend is not free in the next round, either. Most companies anchor next year's baseline on this year's actuals, so a cost center that spends $2.09M of $2.4M opens the next build near $2.09M unless somebody argues otherwise. That is the base-protection incentive you already know, relocated: fought in the AOP round with an argument rather than in Q4 with a purchase order, and negotiable because it is convention rather than statute. Say up front whether savings are swept or retained, because a company that has not decided has decided by default.

4. Nobody hands you the baseline, and nobody defends it for you

Carry-forward level arrives as control items in ABS the agency cannot change. The private run-rate baseline is a number you construct and then have to hold. At a $40M ARR, 250-person company with R&D at a $14.6M exit run-rate, the VP of Engineering will tell you her cloud baseline should be $1.9M rather than the $1.6M you carried, because a committed-use agreement steps up in March. She may be right. What differs is that no external authority settles it, no manual defines what qualifies, and "legally unavoidable" means nothing here. Expect real hours defending a layer handed to you free at DSHS, and keep the flat-continuation number visible under the adjustments so a moved baseline is visibly moved.

5. The revenue forecast is not ERFC, and treating it as adopted is a career error

ERFC and the Caseload Forecast Council are statutory bodies with no stake in your program, and you build to their numbers. The commercial equivalent looks structurally identical and is nothing like it, because the revenue plan is produced by people compensated on it. Sales submits $52M of new ARR; historical attainment has run 78 percent; three of the ten largest opportunities turn out to be the same customer counted at three stages. Adopting that is not deference, it is a failure to do the job. The state discipline that does transfer is RCW 43.88.030(1): adjustments to approved estimates are allowed but must be set forth in the budget document. Departures are fine, hidden departures are not, which is why a well-run company keeps the board plan, the internal plan and the quota carried on the street visibly distinct (section 4).

6. A business case is a decision package that carries a return and loses to someone in the room

Your decision packages are better written than most private investment requests, and the format maps closely: a scoped decision, a performance objective, a line of sight to strategy, fiscal detail. Two things change. The arithmetic: a DP argues merit against competing claims on a fixed appropriation and computes no NPV, IRR or payback, while a $1.2M customer-data-platform ask must state what it returns and in how many months, and that is the first gate rather than the last. And the politics: a DP loses to another agency in front of an appropriating body, impersonally, while a private request loses to the VP of Sales, who is at the table, whose ask was funded, and with whom you negotiate again in ninety days. The written case is necessary and not sufficient, which is why section 14 puts pre-wiring ahead of it.

16. Sources and further reading

Each link below was opened and described against the page that loads as of September 2026. Two categories go stale on a schedule: benchmark surveys are re-run annually and Big Four handbooks are re-issued, so a link that says 2026 today will quietly redirect to a 2027 edition. Where that matters the entry names the edition. On access, the full FASB Codification is free: the Financial Accounting Foundation eliminated the paid Professional View subscription on February 27, 2023 and opened advanced access to the public at no charge. SEC EDGAR, PCAOB standards and COSO summaries are free too. Some library guides still describe a paid tier.

Practitioner bodies and ongoing reading

The rough counterpart to GFOA resources and the budget-office listservs you already use.

• Association for Financial Professionals (AFP) — "A Better Budgeting Process". AFP is the main US practitioner and certifying body for corporate finance; it established and administers the CTP and FPAC credentials. The guide runs Pain Points, How to Budget: Key Decisions, How to Budget: Process, Budgeting Technology, Budgeting Case Studies, and Budgeting Wish List. "Key Decisions" and "Process" back section 2 and section 3; the case studies are the closest thing to watching another company sequence a cycle, behind section 14.
• FP&A Trends — "Why FP&A Needs to Break Free From the 100-Year-Old Budgeting Mindset". The traditional-annual-budget critique used in section 3 and section 10.
• FP&A Trends — "Reflections on Beyond Budgeting". Practitioner commentary on adopting and adapting the model.
• FP&A Trends — "88% Spreadsheets Have Errors" (July 26, 2017). Read it for where the statistic comes from, not the statistic: the article cites no study. It is the origin of a number in every planning-software pitch, and exactly the claim that gets a finance hire in trouble when someone asks for the source.
• Ray Panko — "What We Don't Know About Spreadsheet Errors Today". The underlying literature, by the researcher everyone is loosely quoting. His actual finding is narrower and more defensible: per-cell mistakes are uncommon, but a large model almost certainly holds at least one wrong final value, errors are hard to detect, and developers are systematically overconfident. Rates vary several-fold by methodology. Quote the direction, not the number, and the model-hygiene argument in section 12 survives cross-examination.
• Corporate Finance Institute (CFI) — "Types of Variances in FP&A". A commercial trainer's free reference page, behind the variance vocabulary in section 8.
• Mergers & Inquisitions — "The FP&A Director: First in Line for the Corporate Finance Crown, or Highly Paid Assistant to the CFO?". How the director and VP layer gets scoped and titled, for section 1. Both outcomes in the title are real, and the difference is mostly whether the role owns the plan or assembles it.
• Deloitte, CFO Insights — CFO Insights library. A topic collection (AI and digital transformation, operational efficiency, talent, value creation), worth a skim for how consultancies pitch CFO priorities. Not a budgeting-methodology source: nothing on ZBB or rolling forecasts, and its newest listed pieces date to early 2025.

Planning approaches: zero-based, driver-based, Beyond Budgeting, rolling forecasts

Behind section 3, section 7 and section 10. There is no correct method; most companies run a hybrid, and these cover the poles.

• CFI — "Zero-Based Budgeting". Baseline definition of ZBB: build every line from zero each cycle rather than from last year's actuals plus a growth factor.
• FP&A Trends — "How to Avoid the Common Pitfalls of Zero-Based Budgeting". Failure modes, mostly at large companies; the counterweight to the pure-ZBB pitch.
• Bain & Company — "What Is Zero-Based Budgeting?". ZBB as an organization-wide cost program rather than a line-item technique, the version most likely to appear in a PE-owned or turnaround job description. Bain claims consistent client savings of 15% to 25%; read that as a vendor's own number, since the firms selling ZBB programs also publish its results.
• CFI — "Driver-Based Planning". Driver-based forecasting: model a line as a function of an operational driver (headcount, seats, transactions) rather than a flat growth percentage. Underlies section 4 and section 6.
• CFI — "Activity-Based Budgeting". Separates driver-based (continuous driver) from activity-based (discrete activity or cost pool).
• Beyond Budgeting Institute — bbrt.org. The organization behind Jeremy Hope and Robin Fraser's work: replace the fixed annual budget with rolling forecasts and devolved, KPI-driven management.
• Harvard Business Review — Hope and Fraser, "Who Needs Budgets?" (2003). The article that put Beyond Budgeting in front of a mainstream management audience.
• Ramp — "Mastering the 6+6 Forecast". The midyear reforecast (six months actual plus six forecast, usually run in July) and the earlier 3+9. Interviewers test the distinction: a 6+6 is anchored to the fiscal year and stops at year-end, while a rolling forecast always extends a fixed horizon past it. The article treats them as complementary, not interchangeable. Used in section 7.
• Wall Street Prep — "Rolling Forecast Best Practices Guide". The build-it-yourself treatment of true rolling forecasts, and the better of the two on that mechanic.
• Protiviti — "Ten Steps to Elevate the Annual Budgeting Process" (2023). Modernizing an annual budget without abandoning it, which is where most companies actually land.

Revenue and SaaS metrics and benchmarks

Behind section 4, section 9 and section 13. Every publisher below re-surveys annually; check the current release before quoting a percentile out loud.

• Wall Street Prep — "SaaS Magic Number". The sales-efficiency ratio: the quarter-over-quarter increase in GAAP revenue, multiplied by 4 to annualize it, over the prior quarter's sales and marketing spend. The x4 exists because the numerator is one quarter's delta. Scale Venture Partners built it on GAAP revenue because public companies do not disclose net new ARR; private companies often substitute net new ARR, with no x4, because ARR is already annualized. Getting this backwards is a tell.
• Wall Street Prep — "Burn Multiple". Net cash burned divided by net new ARR added; the capital-efficiency metric running through growth-stage board decks.
• Kellblog (Dave Kellogg) — "Rule of 40 Glideslope Planning". A former SaaS CEO and board member treating the Rule of 40 (growth plus profit margin at or above 40) as a multi-year planning target rather than a scorecard line, which is what makes it relevant to section 10.
• Bessemer Venture Partners — "The Rule of X". Growth rate times a multiplier plus free-cash-flow margin, the multiplier roughly 2x private and 2x to 3x for efficient public companies. The direction is the opposite of what most people assume: Bessemer says the Rule of X is less relevant for early-stage companies burning cash, where burn multiple (target about 1x to 1.5x) is the better lens, and the growth weighting rises as a company matures toward public markets.
• Bessemer — Cloud 100 Benchmarks Report. Annual survey of top private cloud companies' growth, margin and efficiency.
• SaaS Capital — "Spending Benchmarks for Private B2B SaaS Companies" (2026 edition, surveyed March 2026, 1,000+ companies). Opex as a percent of revenue by ARR band; the fastest sanity check on a headcount or opex plan in section 5 and section 6.
• SaaS Capital — "What Is a Good Retention Rate for a Private SaaS Company?" (September 2025, data through December 2024). Gross and net revenue retention segmented by average contract value, not company size: the authors argue price point predicts retention better than age, revenue scale or industry, because companies selling at a similar ACV support customers alike. That cut matters when benchmarking the retention assumption in section 4.
• High Alpha — SaaS Benchmarks Report. Ninth annual edition (2025), previously operated by OpenView; 800+ respondents, about half CEOs or founders, weighted toward sub-$20M ARR. Growth, GRR and NRR, CAC, ARR per employee, go-to-market channel efficiency. The most commonly quoted percentiles in a growth-stage board deck.
• BenchmarkIt (Ray Rike) — 2025 SaaS Performance Metrics Benchmarks and 2026 SaaS and AI-Native Metrics. An annual private-SaaS operating-metrics survey. The 2026 edition benchmarks AI-native companies separately, because usage-based revenue and inference-heavy gross margins behave nothing like per-seat SaaS.
• Sapphire Ventures / KeyBanc Capital Markets — 2025 SaaS Survey. A long-running joint VC and investment-bank survey, and one of the most widely cited benchmark sources in the industry.
• Ordway — "What Is New ARR?". A billing vendor's glossary entry on ARR composition (new, expansion, contraction, churned), feeding section 4.
• Ordway — Cloud Usage-Based Pricing Report (2022). Of 150 public SaaS and cloud providers, only 10 with consumption pricing report ARR at all, computing it inconsistently (some 12x last month's revenue, some 4x trailing three months), while Snowflake, Twilio and others report no recurring-revenue metric. Useful for one thing: seeing how loose "ARR" gets outside per-seat SaaS. It does not measure how widespread usage-based pricing has become.
• PortfolioIQ — "Net New ARR". A second definitional source, cross-referenced in section 4 because vendors disagree on whether "new ARR" includes expansion.
• Mostly Metrics (CJ Gustafson) — the annual-planning index, a newsletter by a sitting tech CFO, roughly 27 planning pieces since 2022 covering long-range frameworks, board-deck templates and forecast-accuracy post-mortems. Part 1, The Kickoff covers participants, top-down versus bottom-up, and guardrails, the companion to section 2 and section 3. Part 2, Building Sales Capacity is the single most useful outside piece for section 5: quota-to-OTE ratios (minimum about 5x), rep ramp by segment (2 to 3 months SMB, 6 to 9 enterprise), 20% to 30% quota over-assignment for attrition and slippage, and reconciling quotas to a company revenue target.
• Pave — "The Compensation Cycle". A compensation-benchmarking vendor on how merit and equity-refresh cycles get planned and approved, behind the approval routing in section 5.

Accounting standards, controls, and reporting rules

FP&A does not close the books, but you need these well enough to explain why an accounting-driven number moved. The authoritative text is free at asc.fasb.org and at the PCAOB; the rest are commercial summaries and issuer handbooks.

• Baruch College Newman Library — "Public Access to Codification". How to reach the Codification, search it by topic, and find superseded pre-Codification standards. Caution: it still describes a paid Professional View tier the FAF eliminated in February 2023.
• Wall Street Prep — "ASC 606: Revenue Recognition". The five-step model, behind the recognized-versus-billed distinction in section 4 and section 9.
• KPMG — Software and Website Costs Handbook (February 2026). Covers ASC 350-40 (internal-use software), ASC 350-50 (website development) and ASC 985-20 (software for sale) — when internally developed software is capitalized rather than expensed, which drives the capex line at a growth-stage SaaS company in section 6. This edition incorporates ASU 2025-06, issued September 2025 and the first major change to internal-use software guidance in over two decades: it drops the stage-based capitalization model to fit agile development and adds interpretations for AI software and data costs. Effective for annual periods beginning after December 15, 2027, early adoption permitted, so both models stay live for two more years.
• BDO — "New ASU on Internal-Use Software Costs Guidance". A shorter free explainer on ASU 2025-06, for cross-checking without a download.
• KPMG — Segment Reporting Handbook (October 2025, reflecting ASU 2023-07's expanded segment-expense disclosure). Referenced in section 9 and section 13.
• EisnerAmper — "ASC 842's Effect on Equity Valuation". How operating-lease capitalization moves assets and liabilities onto the balance sheet.
• BillingPlatform — "ASC 340-40: Costs to Obtain a Contract". Capitalization and amortization of sales commissions, behind the commission-timing point in section 6.
• COSO — Internal Control — Integrated Framework (1992, refreshed 2013). Five components across operations, reporting and compliance objectives. Read it first for section 11: it is what a controller means by "the control framework," and it is management's document, not the auditor's.
• PCAOB — AS 2201, An Audit of Internal Control Over Financial Reporting. The auditor's counterpart to COSO: how the firm tests what management designed.
• PCAOB — AS 2501, Auditing Accounting Estimates. How auditors test judgment-heavy estimates, and therefore why FP&A gets asked to document the basis for one.
• SEC, Division of Corporation Finance — Compliance and Disclosure Interpretations: Non-GAAP Financial Measures (updated December 13, 2022). Staff Q&A under Regulation G and Item 10(e) of Regulation S-K: which adjustments the staff considers misleading, and the requirement that the GAAP measure get equal or greater prominence. Governs the adjusted-EBITDA and non-GAAP-margin numbers a public company's FP&A team helps prepare (section 13).

Systems, data, and tooling

Behind section 12. There is one recurring third-party map of this market, the Gartner Magic Quadrant for Financial Planning Software, current edition published December 1, 2025 and free through any vendor's reprint. Read that first, then treat every vendor link below as a primary source describing itself.

• Gartner Magic Quadrant for Financial Planning Software (2025) — reprints at Board and OneStream. Same report either way: quadrant graphic plus per-vendor write-ups. 2025 Leaders include Anaplan, Board, Oracle, OneStream, SAP and Workday, with Pigment a Visionary. Honor Gartner's disclaimer: it does not endorse vendors, and a reprint is published by a company that liked its placement.
• FAST Standard — fast-standard.org. An open modeling convention (Flexible, Appropriate, Structured, Transparent) covering layout, formula discipline and separating inputs from calculations. The model-hygiene reference for section 12, and the only free open convention here.
• dbt — documentation. The transformation layer between the warehouse and the planning tool, and increasingly where the numbers FP&A reports are actually defined. Worth reading even though you will not write it: it applies version control, automated tests, lineage and generated documentation to analytics logic, the discipline a planning model needs and rarely has.
• Wikipedia — Essbase. The OLAP-cube technology built by Arbor Software in 1992, merged into Hyperion in 1998, acquired by Oracle in 2007, still underneath much legacy enterprise planning. Also Anaplan for the connected-planning platform.
• Vendor pages, enterprise tier: Anaplan, Oracle EPM, OneStream, Board, Workday Adaptive Planning and its separate Financial Management ERP, and insightsoftware.
• Vendor pages, mid-market and spreadsheet-native: Planful, Vena, Prophix, Datarails, Cube, Abacum, Pigment, Aleph, and FloQast for close management.
• Pigment — "Introducing Modeler". Used in section 12 as an example of how newer vendors position AI-assisted model building.
• CFO.com — "Metric of the Month: Cycle Time for Monthly Close" (Perry D. Wiggins, CPA, reporting APQC data). The benchmark behind close-speed expectations in section 8: across roughly 2,300 organizations, top-quartile close is 4.8 calendar days or less from trial balance to consolidated statements, median 6.4, bottom quartile 10 or more. Note the date: these are 2018 figures, and automation has pulled the leading edge in since.
• CFO Dive — "OpenAI CFO touts move toward 'zero-day close'" (August 2026). The aspirational far end of the same curve, used in section 8.
• CFO Dive — "Microsoft beefs up Copilot in Excel for finance work". AI-assisted spreadsheet tooling reaching finance workflows, referenced in section 12.

Books

• Jack Alexander, Financial Planning & Analysis and Performance Management (Wiley, 2018), ISBN 978-1-119-49148-4. Publisher page. The closest thing to a comprehensive FP&A textbook in print, from forecasting technique to KPI design to M&A analysis. The one to finish if you finish one.
• Jeremy Hope and Robin Fraser, Beyond Budgeting (Harvard Business Review Press, 2003). Publisher page. The founding text behind section 3. Historically important, and a manifesto more than a method.
• Bjarte Bogsnes, Implementing Beyond Budgeting: Unlocking the Performance Potential (Wiley, 2nd edition, 2016), ISBN 978-1-119-15247-7. The implementation-oriented successor, by the practitioner who ran the change at Borealis and Statoil. If Beyond Budgeting comes up in an interview, this is what the people who actually run it work from, and it beats the 2003 text for saying what you would do differently on Monday.

Washington State references

The public-sector side of every translation callout, listed for precision rather than as new reading. The RCW chapter numbers are easy to transpose.

• OFM — "How the Budget Works", the counterpart to section 2, and the Glossary of Budget-Related Terms, authoritative for allotment, appropriation, decision package, maintenance level, proviso and fiscal year. Note what is not there: carry-forward level and policy level have no glossary entries and are defined in Chapter 5 below, which matters because carry-forward level is the term the baseline question in section 3 turns on.
• OFM — 2027-29 Budget Instructions (issued June 5, 2026; biennial requests due September 14, 2026), the direct analog to a company's annual planning kickoff package in section 3.
• OFM — 2027-29 Budget Instructions, Part 1: Operating and Transportation (June 2026), Chapter 5: Carry-Forward, Maintenance, and Policy Levels. The three budget layers, and the source for section 15. OFM calculates and issues the carry-forward level as a locked control item agencies cannot recalculate, which has no private-sector counterpart at all. The prior biennium's shorter standalone file is still posted as 2025-27 Chapter 5.
• OFM — 2025-27 Allotment Instructions. The allotment is an agency's plan of expenditures, revenues and FTEs for each month of the biennium, required by RCW 43.88.110 and revised by quarterly amendments. The nearest company analog is spreading an approved annual budget across twelve months and reforecasting quarterly (section 11). Where the analogy breaks hard: the appropriation is the legal authority, the allotment only the plan for spending it, and a company budget is neither. An approved AOP confers authority on no one, and nothing in a company corresponds to over-expenditure being unlawful.
• Caseload Forecast Council — cfc.wa.gov, created by RCW 43.88C.010, with forecasts governed by RCW 43.88C.020 (at least three a year). Loosely the counterpart to demand planning in section 7. The analogy breaks on authority: the statute makes the official forecast the required basis of the governor's budget document and the legislature's omnibus appropriations act. No company demand plan binds anyone that way.
• Economic and Revenue Forecast Council — About Us, created by RCW 82.33.010, with cadence fixed by RCW 82.33.020: four forecasts a year on statutory dates (November 20; February 20 in even years or March 20 in odd; June 27; September 27). Note the chapter: ERFC sits in Title 82, not the 43.88C chapter governing the CFC. Loosely the counterpart to top-down revenue forecasting in section 4 and section 7, and the analogy most likely to mislead you. The state's revenue number arrives from an independent body on a statutory date and is not negotiable. A company's top-down number is produced inside the building, owned by the CFO and the sales leader, and argued over until the CEO signs it.
• RCW — 43.88.030 (budget document contents and policy) and 43.88.110 (allotment of appropriations).
• fiscal.wa.gov — Budget Process and Budget FAQ. Plain-language overviews including supplemental sessions (section 2). Published by LEAP with OFM and the Legislature, not the State Treasurer.
• SEC EDGAR — Apple and Microsoft filing histories, cited in section 1 to confirm fiscal year-ends: Apple closes September 26, Microsoft June 30. An off-calendar fiscal year should feel familiar, since Washington's own runs July 1 to June 30, Microsoft's exact year-end. What differs is who decides: a company picks its year-end for business reasons and can change it, while the state's is fixed by statute.