How a Startup Financial Model Runs From Drivers to the Raise

A startup financial model turns a few assumptions about customers, prices and hires into a monthly forecast of cash, and from that the size of the raise and the months it buys. Its numbers will probably miss, and investors expect them to. The part worth building with care is the structure, which tends to survive.

The evidence on the numbers is plain. In a survey of 885 VCs across 681 firms, Paul Gompers and three co-authors found that “fewer than 30% of the companies meet projections”, and early-stage investors put the share at 26%. Among early-stage VCs, 31% do not forecast a company’s cash flows at all before they invest (Gompers et al., 2016). Investors still ask for a model, and founders still spend weeks on one. The parts that repay that time are the ones that carry the structure and the cash date, plus the growth rate, which decides the raise and can be ranked against real companies.

Exhibit 1. The parts of a first startup financial model, and the decision each one serves

Part	What goes in	What comes out	Decision it serves
Drivers	New customers a month, price, churn and hire dates, each with a source and a month by which actuals will test it	Every other line	Which assumption to test first
Revenue build	Customers × price, month by month	Revenue and its monthly growth rate	The plan: what the spending must buy
Headcount plan	Role, start month, loaded monthly cost	Payroll	The plan: when to hire
Other costs	Hosting, software, rent, marketing	Operating expenses	The plan: what can wait
Three linked statements	Income statement, balance sheet, cash flow statement	Month-end cash and a balance check	The runway: the month the cash runs out
Outputs	The milestone the next round needs, and a buffer for raising it	Raise size, runway in months, the growth rate’s rank among real companies	The raise

Source: CX Cash synthesis of the sources cited below.

A startup financial model is a phantom business, and investors discount its numbers

The electronic spreadsheet was invented for this kind of projection. Dan Bricklin had the idea in the spring of 1978 in a Harvard Business School classroom, where his finance class had to project the financial consequences of one company acquiring another on paper ledgers. VisiCalc went on sale for the Apple II on 17 October 1979 and sold more than 700,000 copies in six years (VisiCalc, Wikipedia). When Steven Levy surveyed its effects for Harper’s in 1984, one executive told him the tool let its user build “a phantom business within the computer” (Levy, 1984).

The phantom and the real business part company quickly, and not only on the downside. Lotus Development, maker of the rival 1-2-3, went public in 1983, less than a year after selling its first copy. Its senior product-design planner, Ezra Gottheil, told Levy that “Our own projections were violated on a daily basis”. Demand ran ahead of every formula in its plan.

Investors price this in. Harvard Business Review’s summary of William Sahlman’s 1997 article on business plans opens: “Every seasoned investor knows that detailed financial projections for a new company are an act of imagination” (Sahlman, 1997). Sequoia’s pitch guide asks a founder to explain the business model and gives the financials a single line: “If you have any, please include.” (Sequoia, 2019). In the Gompers survey, 47% of VC firms named the management team as the most important factor, more than any other. Firms that forecast at all use a median horizon of 3 to 4 years.

One study tested how much those projections weigh in a funding decision. David Kirsch, Brent Goldfarb and Azi Gera examined 722 funding requests sent to one American venture firm. Planning documents and some of their contents were only weakly associated with the firm’s decisions, and the information that mattered was “learned independently of its inclusion in the business planning documents” (Kirsch, Goldfarb & Gera, 2009). Hours spent polishing a fifth year of monthly revenue probably go into the part of the model its readers trust least.

But the structure of an early plan tends to survive to the IPO

Steven Kaplan, Berk Sensoy and Per Strömberg followed 49 venture-backed companies from an early business plan through the IPO to a later annual report, a median span of 63 months (Kaplan, Sensoy & Strömberg, 2005). At the business plan, the median company was 24 months old, employed 22 people and had no revenue in its last fiscal year. At the IPO, a median 34 months later, it had $7.2M of revenue and 124 employees; by the annual report, $35.1M and 378. Only 17%, 18% and 15% of the firms were profitable at the three dates. Every figure in those plans had to change by an order of magnitude or more.

The businesses themselves barely moved. The authors found business lines stable from plan to public company: “Only one firm changes its core line of business” among the 49. People changed more than the business did, and by the annual report half of the chief executives named in the plans were still in post. Thirty-one of the 49 IPOs took place in 1998, 1999 or 2000, so the sample leans toward the boom, and it contains only companies that reached an IPO: it shows what survives in a success, not how often plans survive in general. Within those limits, what a founder writes down about how the company earns money is the part most likely to last.

Salesforce’s 2004 registration statement shows the same pattern inside one company (Exhibit 2). The filing states the revenue structure: “Subscription revenues are driven primarily by the number of paying subscribers of our service and the subscription price of our service” (salesforce.com, 2004). Subscription and support already made up 92% of revenue in fiscal 2001, the first full year, and the structure carried the company to the IPO while revenue grew 17.7-fold (our arithmetic). In fiscal 2004 it priced out at about $845 of subscription revenue per average paying subscription, or $70 a month.

Exhibit 2. Salesforce’s first five fiscal years: the structure held while every number moved

$ millions unless stated; fiscal years end 31 January	FY2000*	FY2001	FY2002	FY2003	FY2004
Revenue	0.0	5.4	22.4	51.0	96.0
Growth on the prior year			4.1x	2.3x	1.9x
Marketing and sales	2.5	25.4	24.6	33.5	54.6
Operating income (loss)	(5.6)	(33.6)	(29.5)	(10.5)	3.7
Net cash from operations			(13.2)	5.2	21.8
Cash and short-term securities, year end	12.6	22.2	11.7	16.0	35.8
Customers, approximate			3,500	5,700	8,700
Paying subscriptions, approximate		30,000†	53,000	76,000	127,000

Source: salesforce.com, Form S-1/A, Amendment No. 8 (22 June 2004): selected financial data, cash flow statement, customer and subscriber data. *From inception on 3 February 1999. †At 1 February 2001. Growth multiples are our arithmetic; blank cells are not reported in the filing.

The exhibit carries three lessons for a first model. Growth decayed every year, from 4.1 times to 1.9 times. Marketing and sales ran at 4.7 times revenue in fiscal 2001, so the spending came years before the revenue it bought. And the four fiscal years before break-even produced $79.2M of operating losses (our arithmetic), against $61.1M of net proceeds from convertible preferred stock. A model with the right structure absorbs surprises like these by changing its inputs. One with the wrong structure has to be rebuilt.

Therefore build a small startup financial model around the drivers that set the cash date

One study of planning suggests where the value comes from. Frédéric Delmar and Scott Shane followed 223 new ventures started in the first nine months of 1998 by a random sample of Swedish founders. Planning, they argue, helps founders “to make decisions, to balance resource supply and demand, and to turn abstract goals into concrete operational steps”. In their data it reduced the likelihood that a venture disbanded and sped up product development (Delmar & Shane, 2003). In their account the benefit runs through decisions, which argues for building a first model to produce three of them: a hiring calendar, the month the cash runs out, and the point at which the plan changes.

That calls for a short list of drivers. For a subscription business, the Salesforce sentence is the template: subscribers times price, with new subscribers a month, churn and price as inputs and revenue as the result. Each driver should carry its source and a month by which actuals will confirm or refute it, such as trial-to-paid conversion after the first 90 days of selling. Headcount needs the same treatment, line by line, with a start month for every hire. As Paul Graham of Y Combinator puts it, “Hiring too fast is by far the biggest killer of startups that raise money” (Graham, 2015).

The three statements belong in even a small model, because profit and cash can tell different stories. In fiscal 2003 Salesforce reported a net loss of $9.7M and generated $5.2M of cash from operations, largely because customers paying in advance raised deferred revenue by $12.0M. A founder reading only the income statement would have been wrong about operating cash by $15.4M in fiscal 2002 and by $14.9M in fiscal 2003, both times on the gloomy side.

The balance sheet checks that the links hold. A model whose balance sheet does not balance has a broken link somewhere, and every output downstream of it, the cash date included, is suspect.

The outputs follow from the structure. One is the cash date: at the end of fiscal 2002, Salesforce held $11.7M against $13.2M of operating cash used that year, about 10.7 months at that pace. The other is a written fallback. Graham’s advice is to know “precisely when you’ll have to switch to plan B if plan A isn’t working”. In a model that becomes a month and a trigger: a revenue level which, if missed by a given date, starts the cuts.

But the growth rate decides the raise, so rank it against real companies

A book printed in Venice in 1494 contains the arithmetic for the model’s most consequential cell. In his Summa de arithmetica, Luca Pacioli gave, without deriving it, an early version of the rule of 72: divide 72 by the rate of growth per period, and the result is roughly the number of periods it takes to double. His example was capital at 6% a year, which doubles in 12 years (Rule of 72, Wikipedia). Applied to a monthly growth assumption, the rule turns a percentage into a date. At 6% a month, revenue doubles in about a year; at 2.5%, in about 28 months. Graham made the same point with weekly rates: “A company that grows at 1% a week will grow 1.7x a year, whereas a company that grows at 5% a week will grow 12.6x” (Graham, 2012).

Real companies supply the distribution against which to read that date. ChartMogul’s 2023 report draws on more than 2,200 SaaS businesses that use its subscription analytics. It puts the median at around 2–2.5% growth a month throughout a company’s life, the top quartile at around 5–7% when starting out and the top decile at 10–17% (ChartMogul, 2023). Its median company took about 2 years and 9 months from its first paying customer to $1M of annual recurring revenue (ARR). Kyle Poyar’s 2025 analysis of 6,525 companies in the same dataset found that 13.4% reached $1M of ARR within three years of first revenue, and one in ten reached $10M within ten years (ChartMogul, 2025). Both samples are one vendor’s customers, and both reports name the bias that creates: survivorship in the first, selection toward serious businesses in the second.

The plan’s required monthly growth rate can be placed in that distribution before anyone signs. A plan that needs 10% a month from a small base asks the company to perform like the top decile for as long as the plan runs; one that needs 2.5% asks it to be typical. Exhibit 3 shows what the difference does to the raise.

Exhibit 3. The growth assumption sets the size of the raise

The company is hypothetical: B2B software with $25k of monthly recurring revenue ($300k of ARR), an 80% gross margin and operating costs of $200k a month, held flat to isolate the growth assumption. Its founders want $1M of ARR before raising again, plus six months of runway for the raise itself.

Monthly growth	Where it ranks (ChartMogul, 2023)	Doubling time, months (rule of 72 / exact)	Months to $1M of ARR	Cash used to get there	Raise including buffer
2.5%	About the median	28.8 / 28.1	49	$7.87M	$8.67M
4%	Between the median and the top quartile	18.0 / 17.7	31	$4.97M	$5.76M
6%	Top quartile	12.0 / 11.9	21	$3.35M	$4.14M
10%	Top decile	7.2 / 7.3	13	$2.06M	$2.85M

Source: hypothetical company; growth bands from ChartMogul (2023), monthly growth of SaaS businesses when starting out. Buffer: six months of net burn in the month the milestone is reached. The remaining cells are calculated by CX Cash.

A plan at the median growth rate needs about three times the money of a plan at the top-decile rate to reach the same milestone. The costlier case is the mismatch. A $2.85M round sized on 10% a month runs out after month 16 if the company grows at the median rate, with about $445k of ARR, less than half the milestone (our arithmetic). The founders would then be raising with high burn and slow growth, and Graham’s account of how startups die describes the result: “because they have high expenses and slow growth, they’re now unappealing to investors”. The defence is to size the round on a growth rate the company has already shown, and to show the board where the plan’s rate ranks among real companies.

Do founders need a written plan at all?

The case against is strong and partly right. Amar Bhidé argued in Harvard Business Review in 1994 that “a comprehensive analytical approach to planning doesn’t suit most start-ups”. In his view, “too much analysis can be harmful; by the time an opportunity is investigated fully, it may no longer exist” (Bhidé, 1994). A study of 116 ventures started by Babson College alumni who graduated between 1985 and 2003 found “no difference between the performance of new businesses launched with or without written business plans” (Lange et al., 2007). A 2010 meta-analysis found planning beneficial on average, with an effect that varies with the age of the firm and its cultural setting (Brinckmann, Grichnik & Kapsa, 2010). Add the weak link to venture decisions, and the written plan as a document earns little.

Bhidé and Lange have the document and the detail right. Sahlman’s article opens by mocking the belief that success needs “a decade of month-by-month financial projections”, and nothing in the evidence above rewards one. Their case says less about the arithmetic of cash. Lange and his co-authors exempt founders who must “raise substantial start-up capital from institutional investors or business angels”, which covers any company raising a seed round. Those founders face two decisions that no one can make without a startup financial model: how much to raise and when to hire. The model that answers them is small, and it stays useful only while actuals keep testing it. Graham’s advice is to “pick a growth rate they think they can hit, and then just try to hit it every week” (Graham, 2012).

Drivers, a balance check and a ranked growth rate

Founders building a first model before a seed round, and the investors and directors who will read it, can work from one list.

• Structure first. Drivers, a revenue build, a headcount plan, three linked statements and a balance check, with the drivers kept to the handful that move the cash date. The 3-statement model template and scenario planner below is a starting point for the structure.
• A source and a test date beside every driver. Each driver carries its source and the month by which actuals will confirm or refute it. A director can then ask which assumptions have been tested and which will be by the next board meeting.
• A ranked growth rate. The plan’s required growth rate, its doubling time and its rank in ChartMogul’s distribution are settled before the size of the round. Investors can ask for that rank before the five-year revenue line.
• Two cash-out months. The raise is sized on a growth rate the company has shown, with a written plan B and the month it starts. The model shows the cash-out month at the median growth rate beside the one at the plan’s rate.
• Less time on years four and five, which lie beyond the median VC forecasting horizon of 3 to 4 years. A reader learns more from the model’s structure, the part of an early plan that tends to last.

A model can miss every figure and still do its job

A startup financial model is a phantom business, and its numbers will miss. What tends to last is the structure, how the company earns, spends and runs out of cash, so that is where the work belongs. The one input that deserves special care is the growth rate, which sets the size of the raise and can be ranked against real companies before anyone signs. Built that way, the model can be wrong in every figure and still do its job. That is the test most models face, since early-stage investors estimate that about one company in four meets its projections.

References

• Bhidé, A. (1994). How entrepreneurs craft strategies that work. Harvard Business Review, March–April.
• Brinckmann, J., Grichnik, D. and Kapsa, D. (2010). Should entrepreneurs plan or just storm the castle? A meta-analysis on contextual factors impacting the business planning–performance relationship in small firms. Journal of Business Venturing 25(1), 24–40 (abstract).
• ChartMogul (2023). SaaS Growth Report (S. Jain).
• ChartMogul (2025). Against the Odds: The 2025 SaaS Growth Report (K. Poyar).
• Delmar, F. and Shane, S. (2003). Does business planning facilitate the development of new ventures? Strategic Management Journal 24(12), 1165–1185 (abstract).
• Gompers, P., Gornall, W., Kaplan, S. N. and Strebulaev, I. A. (2016). How do venture capitalists make decisions? NBER Working Paper 22587; published in the Journal of Financial Economics, 2020.
• Graham, P. (2012). Startup = growth. paulgraham.com, September.
• Graham, P. (2015). Default alive or default dead? paulgraham.com, October.
• Kaplan, S. N., Sensoy, B. A. and Strömberg, P. (2005). What are firms? Evolution from birth to public companies. NBER Working Paper 11581; published in the Journal of Finance, 2009.
• Kirsch, D., Goldfarb, B. and Gera, A. (2009). Form or substance: the role of business plans in venture capital decision making. Strategic Management Journal 30(5), 487–515 (abstract).
• Lange, J. E., Mollov, A., Pearlmutter, M., Singh, S. and Bygrave, W. D. (2007). Pre-start-up formal business plans and post-start-up performance: a study of 116 new ventures. Venture Capital 9(4), 237–256 (abstract).
• Levy, S. (1984). A spreadsheet way of knowledge. Harper’s, November; republished by Wired, 24 October 2014.
• Sahlman, W. A. (1997). How to write a great business plan. Harvard Business Review, July–August.
• salesforce.com, inc. (2004). Form S-1/A, Amendment No. 8, filed 22 June 2004.
• Sequoia Capital (2019). Writing a Business Plan.
• Wikipedia. Rule of 72; VisiCalc.

Salesforce’s figures come from its 2004 registration statement, and the growth bands from ChartMogul’s samples of its own customers. For the four papers marked “abstract”, only the published abstract is relied on. Ratios, sums and runway months without a citation are our own calculations on the Salesforce filing and the other cited figures, and the software company in Exhibit 3 is hypothetical.