How to Build SaaS Cash Flow Projections From Billing Terms

Cash flow projections for a SaaS company work best when they start from a handful of operating drivers: new bookings, retention, billing terms, collection timing and headcount. Among them, the contract terms decide when cash arrives, and they are what last year’s line items cannot show.

Salesforce shows how far timing and size can drift apart. In its fiscal year to January 2026 it recorded about 45% of the year’s billings in the fourth quarter. It then collected 43% of the year’s operating cash flow in the first quarter, which carried 24% of revenue (our arithmetic; Salesforce, 2026). A projection that extrapolates last year’s line items cannot see that gap coming. One built from drivers can, provided it stays small.

The drivers of SaaS cash flow projections, and what each one moves

Driver	What it moves	Where the number comes from	How public companies describe it
New bookings	Billings, then receipts one payment term later	Pipeline, win rates and sales capacity, by month	CrowdStrike: net new ARR is typically greater in the second half, particularly the fourth quarter
Retention (churn and expansion)	The size of each renewal invoice	Cohort history	SaaS Capital: median gross revenue retention of 90% on annual contracts
Billing terms	How much of a contract is invoiced at once	The contract mix: monthly, annual, multi-year	Okta invoices subscriptions mainly annually in advance
Collection curve	The share of each invoice paid 0, 30, 60 or 90 days after issue	The company’s own receivables ageing	Okta and Salesforce: payment due within 30 days of invoice
Renewal calendar	Which months carry the large invoices	Contract start and renewal dates	Salesforce: Q4 its strongest quarter for new business and renewals, Q1 its largest for collections
Headcount and payroll calendar	Most of the outflow, and when it leaves	The hiring plan with start dates; bonus, commission and payroll-tax dates	CrowdStrike: payroll taxes step up in the first half; Workiva: bonus and commission payments in Q1 and Q4

Sources: Forms 10-K of CrowdStrike, Okta and Salesforce (fiscal years to January 2026) and Workiva (2025); SaaS Capital (2025). CX Cash synthesis.

Receipts in month t = Σ billings(t − k) × share collected k months after issue · billings = invoices falling due under each contract’s term, after churn and expansion

Cash flow projections built from drivers earn their keep when the terms change

The forecasting literature sorts methods into two families. Extrapolation needs only the history of the number being forecast; causal methods need knowledge of what moves it. Kesten Green and J. Scott Armstrong, summarising decades of comparative tests, put it plainly: “For situations where knowledge of relationships is good and large changes are unlikely, as is common in the short-term, use extrapolation. If large changes are likely, causal methods provide forecasts that are more accurate” (Green & Armstrong, 2012). Causal models are most useful, they add, when relationships are strong, their direction is known, large changes are expected and the causal variables can be forecast or controlled.

Cash flow projections built from last year’s line items plus a growth rate are extrapolations, and an old kind. Green and Armstrong call exponential smoothing the most widely used extrapolation method, “with the possible exception of using last year’s value”. SaaS cash meets three of the conditions by construction: the link between an invoice and a receipt is contractual, its direction is known, and the company sets its own billing terms. The fourth, an expected large change, holds whenever those terms move, and the company knows of the move before the cash does.

The idea has a direct test. Armstrong, Fred Collopy and J. Thomas Yokum applied decomposition by causal forces, which splits a series into parts pushed in different directions, to 12 annual series, from automobile accidents to cigarette production, and made 800 forecasts. For the 9 series where the conditions were completely or partially met, “the forecast error (MdAPE) was reduced by more than half”; for the other 3 it made little difference (Armstrong, Collopy & Yokum, 2005). None of the 12 was a company’s cash, so the study shows the method can work rather than that it will here. A SaaS cash line does split the same way, into a growth force (ARR) and a timing force (the terms).

Adobe supplied the large change. In May 2013 it announced that new creative features would go only to Creative Cloud subscribers, and that Creative Suite 6 would be the last major update for perpetual licensees. In fiscal 2013 its revenue fell 8% to $4.06 billion and its net income fell 65%, a decrease the company put “primarily due to the revenue model becoming more ratable”. Operating cash flow fell 23% to $1.15 billion, after a year in which it had “remained stable compared to fiscal 2011” (Adobe, 2014). Subscriptions rose from 15% of revenue to 28%, and to 50% in fiscal 2014, when operating cash flow recovered by 12% (our arithmetic; Adobe, 2015).

Extrapolating fiscal 2012’s stable cash would have overstated fiscal 2013 by $347.9 million. Adobe put the fall in cash down mainly to lower net income, and the fall in net income mainly to the revenue model, along with higher costs.

The billing calendar, not the revenue line, sets when SaaS cash arrives

Salesforce invoices most customers once a year, in advance, and its “typical payment terms provide that our customers pay us within 30 days of invoice”. Its fourth quarter has historically been its strongest for new business and renewals, and the company spells out the consequence: “our fiscal first quarter has typically been our largest collections and operating cash flow quarter” (Salesforce, 2026). The table below rebuilds fiscal 2026 quarter by quarter from its 10-Q filings and the 10-K.

Salesforce, fiscal 2026: revenue arrives evenly, cash does not, $ millions

Quarter	Billings and other	Revenue	Operating cash flow	Receivables at quarter end	DSO on revenue (days)	Days of billings in receivables
Q1 (Feb to Apr 2025)	6,885	9,829	6,476	4,354	39	56
Q2 (May to Jul 2025)	8,992	10,236	740	5,596	50	57
Q3 (Aug to Oct 2025)	8,700	10,259	2,316	5,474	49	58
Q4 (Nov 2025 to Jan 2026)	20,522	11,201	5,464	14,339	118	64
Year	45,099	41,525	14,996

Source: Salesforce Forms 10-Q for the quarters to 30 April, 31 July and 31 October 2025 (Q1 10-Q) and Form 10-K for the year to 31 January 2026. Fourth-quarter flows are the year less three quarters. Billings and other is the unearned-revenue roll-forward line; Q4 includes $651 million acquired with Informatica. DSO on revenue = receivables ÷ quarterly revenue × days; days of billings uses billings. Derived figures are our arithmetic.

Revenue stayed between 24% and 27% of the year in every quarter. Cash did not. The first quarter produced 8.8 times the operating cash of the second (our arithmetic). A projection that spread the year’s $15.0 billion evenly would have been $2.7 billion short in the first quarter and $3.0 billion long in the second. Other companies describe the same mechanism. SentinelOne reports a concentration of annual billings in its fourth quarter, and “This makes our fiscal first quarter our strongest for collections and operating cash flow” (SentinelOne, 2026). Snowflake has seen higher operating cash flow in its first and fourth quarters, after higher order volumes in the fourth (Snowflake, 2026).

Signal processing has an exact name for the mechanism. Convolving an input signal with an impulse response gives the output of a linear time-invariant system, where “the output is an accumulated effect of all the prior values of the input function”. Each value is weighted by a factor that depends on the time elapsed since it occurred (Wikipedia, Convolution). Cash receipts are billings convolved with the collection curve, the share of an invoice paid in the month of issue, the month after, and so on.

Last year’s receipts record the output and lose the input, so a line-item projection cannot tell a change in billings from a change in the curve. Real receivables depart from a time-invariant system in two ways: the curve itself shifts, as large customers and year-end invoices pay on their own schedules, and a disputed invoice is held back whole rather than in proportion.

The same arithmetic shows why days sales outstanding (DSO) misleads as a driver. Measured against quarterly revenue, Salesforce’s receivables read 39 days at the end of the first quarter and 118 days at the end of the fourth, on 30-day terms throughout. Measured against billings, they stayed between 56 and 64 days. The filings do not say how much of the excess over 30 days is late payment and how much is invoices issued late in the quarter. A company can say, from its own receivables ageing, and that ageing by customer segment is where the collection curve should come from.

Billing terms drift, and the drift moves cash while revenue stands still

A calendar that repeats can be extrapolated. A change in the mix of terms cannot, and at some companies the mix is moving. Snowflake bills most capacity contracts annually in advance, but expects more quarterly-upfront and monthly-in-arrears billing “as our customers increasingly want to align consumption and timing of payments” (Snowflake, 2026). Zscaler warns investors: “We cannot predict the mix of invoicing schedules in any given period” (Zscaler, 2026).

What a projection has to capture is the change in cash collected per dollar of ARR when the billing mix moves and nothing else does. Private companies show how wide the mix can range. ChartMogul’s 2025 billing report covers more than 2,500 SaaS companies on its platform in 2024. Annual plans supplied a median 47% of ARR at companies with $3M to $8M of ARR, and 28% at $15M to $30M (ChartMogul, 2025). The data come from one analytics vendor’s customers, report medians only and exclude companies with a single billing model.

Annual terms do buy some retention, but for B2B companies the gain looks small next to the cash. SaaS Capital’s 2025 survey of more than 1,000 private B2B SaaS companies found median gross revenue retention of 94% on multi-year contracts, 90% on annual contracts and 89% on month-to-month (SaaS Capital, 2025). The gap widens at low prices: ChartMogul reports median customer retention of 62% on annual plans and 41% on monthly plans for products with average revenue per account under $25. Both samples are self-selected, and neither separates the effect of the term from the kind of customer that chooses it. The direction is consistent; the size depends on the price point.

A shift to monthly billing at a hypothetical $6.0M ARR company: the same revenue and less cash, $ thousands

The B2B SaaS company has $6.0M of ARR, flat for the year so that only the terms change. Half its ARR is billed annually in advance on 30-day terms and collected the month after invoice; the other half is billed monthly by card. Renewals are spread evenly across the year. In the projection year, 40% of annual customers switch to monthly billing at renewal, with price and churn unchanged.

Quarter	Revenue	Cash, line-item projection	Cash, driver projection	Gap
Q1	1,500	1,500	1,350	−150
Q2	1,500	1,500	1,325	−175
Q3	1,500	1,500	1,400	−100
Q4	1,500	1,500	1,475	−25
Year	6,000	6,000	5,550	−450

Hypothetical company; every figure is our calculation. The line-item projection repeats last year’s receipts at last year’s growth rate (zero). The driver projection bills each contract on its own term and collects annual invoices one month after issue.

The company loses $450k of cash in the year, 7.5% of revenue, without losing a customer or cutting a price. That equals the in-year cash from $831k of new monthly-billed ARR signed evenly over the year, about 14% growth for this company. By the last month the cash run-rate is back to $500k, so the loss is a one-off, the float that annual prepayment had provided, and it would have reached $1.125M, more than two months of revenue, had every annual customer switched. A line-item projection shows none of it, because last year’s receipts carry no record of the mix.

Outflows follow the hiring plan and the payroll calendar

Receipts are only half of the projection. Most of the other half is people. Kruze Consulting’s analysis of its clients’ spending put payroll-related costs, contractors included, at 76% of operating expenses for venture-backed startups (Kruze Consulting, 2024). SaaS Capital’s 2026 survey puts median revenue per employee at private SaaS companies at $141,125, so the hypothetical $6.0M company would employ about 43 people at the median (SaaS Capital, 2026). The driver is the hiring plan with start dates, and a payroll line grown at last year’s rate is a poor substitute.

Timing within the year can matter as much as the total. CrowdStrike reports a lower margin in the first half of its fiscal year “due to a step up in costs for payroll taxes and annual sales and marketing events”. The same step moves the timing of its operating cash flow (CrowdStrike, 2026). Workiva describes “operating cash flows affected by the timing of employee bonus and commission payments during the first and fourth quarters” (Workiva, 2026). A 12-month average spreads those payments into months that do not have them.

The drivers also sort into two kinds. Hiring dates, prices, the billing terms on offer and the payment terms are set by management, so they belong in the plan as decisions. Churn, late payments and the timing of large deals act on the company from outside; they are disturbances and belong in the scenarios, so a board that approves hires funded by an upside scenario has treated a disturbance as a decision.

A driver model needs fewer drivers than a spreadsheet allows

The opposite risk is a model that tries to capture everything. Green and Armstrong’s guidelines put the useful number of causal variables in a regression at three or fewer, provided their effects can be estimated from many reliable observations. A 2015 review by the same authors found that complexity raised forecast error by 27% on average across the 25 papers that measured it (Green & Armstrong, 2015).

On our reading, the six drivers in the first table are enough for a typical B2B SaaS company. A driver earns its place when someone owns it, its value comes from a system of record, and a change in it moves cash within the projection’s horizon. Everything else can sit at a run-rate until a miss shows that it matters.

Green and Armstrong favour extrapolation while nothing large is changing

The best case against a driver model comes from the same literature. Green and Armstrong recommend extrapolation for the short term when large changes are unlikely. They cite a 1984 study by Makridakis and colleagues in which seasonal adjustment alone “reduced forecast errors by 23 percent” across 68 monthly economic series. Salesforce’s calendar repeats: applying fiscal 2025’s quarterly shares of operating cash flow to fiscal 2026’s total would have missed by at most $0.9 billion in any quarter, against $3.0 billion for an even split. A driver model also takes time to build and maintain, and every added driver is one more assumption to defend.

For a company whose terms, prices and customer mix are stable, and many are, Green and Armstrong’s advice holds: last year’s seasonal pattern grown at a sober rate is a respectable basis for cash flow projections, and a 13-week forecast covers the near term. Two things limit it. Terms are moving at some SaaS companies, Snowflake and Zscaler among them, and when they move an extrapolation fails in the year that matters, as it would have at Adobe. An extrapolation also cannot say why it missed. A driver model reports its miss by driver (billings, the collection curve, churn, hiring dates), and for a board that attribution is often the more useful product.

Build receipts from the billing schedule and read every miss by driver

A founder should build receipts from the billing schedule and a collection curve measured from the company’s own receivables ageing, and retire revenue times DSO. The share of ARR billed annually in advance deserves a monthly line, with its cash effect priced before anyone offers a prepayment discount or a switch to monthly billing. Payroll, bonuses, commissions and payroll taxes belong on the calendar month by month, from a hiring plan with start dates. Comparing each month’s projection with the actuals by driver gives every miss a named cause, and the free 13-week cash flow model below covers the near term.

Investors and boards should ask for last quarter’s miss broken down by driver, alongside the total, and for billings by quarter with days of billings in receivables, since DSO on revenue swings with the billing calendar. The share of ARR billed annually in advance, and how it has changed, is worth a standing question: a falling share is a cash event even when ARR is flat. So is the split between drivers that are decisions and drivers that are scenarios, with hiring approved only against the decisions.

A SaaS cash projection starts from the contract

Cash flow projections for a SaaS company should start from the contract. Revenue moves smoothly because the accounting spreads it; cash moves when invoices fall due and customers pay, on terms the company chose and can change. The forecasting evidence favours extrapolation while nothing large is changing and causal models when something is, and a change in billing terms is that kind of change. A small model with the terms first will show it. The supporting evidence comes from public companies’ filings and from forecasting tests on series other than cash, so a private company’s own misses, read by driver, are the better guide to how many drivers it needs.

Sources

• Adobe Systems Incorporated (2014). Form 10-K for the fiscal year ended 29 November 2013. (2015). Form 10-K for the fiscal year ended 28 November 2014.
• Armstrong, J. S., Collopy, F. and Yokum, J. T. (2005). Decomposition by causal forces: a procedure for forecasting complex time series. International Journal of Forecasting 21(1), 25–36.
• ChartMogul (2025). Billing is the invisible force behind your growth (SaaS Billing Report, S. Faustino).
• CrowdStrike Holdings, Inc. (2026). Form 10-K for the fiscal year ended 31 January 2026.
• Green, K. C. and Armstrong, J. S. (2012). Demand Forecasting: Evidence-Based Methods. Working paper, version 165, University of Pennsylvania ScholarlyCommons.
• Green, K. C. and Armstrong, J. S. (2015). Simple versus complex forecasting: the evidence. Journal of Business Research 68(8), 1678–1685.
• Kruze Consulting (2024). Startup Payroll Costs and Tax Budgeting for New Hires.
• Okta, Inc. (2026). Form 10-K for the fiscal year ended 31 January 2026.
• SaaS Capital (2025). 2025 B2B SaaS Retention Benchmarks (Research Brief 32). (2026). 2026 Revenue Per Employee Benchmarks for Private SaaS Companies.
• Salesforce, Inc. (2025). Forms 10-Q for the quarters ended 30 April, 31 July and 31 October 2025. (2026). Form 10-K for the fiscal year ended 31 January 2026.
• SentinelOne, Inc. (2026). Form 10-K for the fiscal year ended 31 January 2026.
• Snowflake Inc. (2026). Form 10-K for the fiscal year ended 31 January 2026.
• Workiva Inc. (2026). Form 10-K for the fiscal year ended 31 December 2025.
• Zscaler, Inc. (2026). Form 10-K for the fiscal year ended 31 July 2026.
• Wikipedia. Convolution.

The figures come from the filings, papers and reports above. Ratios, gaps and growth rates given without a source were computed from the filings themselves: Salesforce’s fourth-quarter flows are the annual totals less the first three quarters, and the DSO and days-of-billings columns use 89 days for the first quarter and 92 for the others. The $6.0M ARR company that moves customers to monthly billing is hypothetical.