How to Read a SaaS P&L: What Founders Miss

All right, here is the direct answer. Read a SaaS P&L in four moves. Separate recurring revenue from services revenue. Look at what somebody put into cost of revenue. Split subscription gross margin out of the blended number. Then read the operating expense lines as percentages of revenue rather than as dollars.

All four moves exist because the statement in front of you was not built to answer your question. It was built to satisfy an accounting standard, and the standard is indifferent to whether you can tell a good software business from a mediocre one. So the numbers are right and the picture is still misleading, which is a harder problem than numbers being wrong.

Here is the reframe I would offer. Your P&L is not a scoreboard, it is a translation, and something gets lost in every translation. Signed contracts become recognized revenue on a delay. Two businesses become one gross margin. Cash you spent on engineers last year becomes an amortization line inside cost of revenue this year. Reading it well means knowing where the translation loses the plot.

So let me walk you through it in the order I actually do it. The top line, which is not one line. The discounts buried inside it. Cost of revenue, which nobody defines for you. The gross margin that is secretly two businesses. Capitalization. Then the operating lines, and the year the industry quietly cut them.

Your top line is not one line, and bookings is not any of them

Four numbers describe the same customer relationship at four different moments, and founders routinely quote whichever one is largest. Bookings is what got signed, which Wall Street Prep defines as the value of a contract signed with a prospective customer for a given period.¹ Billings is what got invoiced. Revenue is what got earned, and it is the only one of the four your accountant may put on the P&L. ARR is a projection of the recurring part, annualized.

That last one deserves a warning, because the most confident numbers in this industry come from it and no accounting standard governs it. In March 2023 the SEC staff looked at Alteryx and objected to the arithmetic. The company was annualizing contracts shorter than a year by dividing committed value by months in the term and multiplying by twelve. The staff's instruction was blunt: "please rename this measure to more appropriately reflect what it represents."²

Alteryx agreed, changed "Annual Recurring Revenue" to "Annualized Recurring Revenue," and disclosed that annualizing short contracts puts amounts into ARR exceeding total contract value. Read that twice. A public company, audited and well advised, was reporting a recurring revenue figure containing money nobody had contracted to pay. It came to about one percent of ARR, so this was a disclosure problem rather than a fraud problem. A private company doing the same thing has no staff reviewer to catch it.

The Commission's guidance on metrics like this is short and worth knowing, because your future investor will apply it. Release 33-10751 says a company should expect to disclose "A clear definition of the metric and how it is calculated," why it is useful, and how management actually uses it.³ A footnote adds that the metric should not deviate materially from what management uses to run the business. If your ARR definition lives only in somebody's spreadsheet, you cannot meet any of the three.

The discounts you already gave are hiding inside your revenue

Contra revenue is the least glamorous idea here and it moves the top line more than anything else on this page. The rule is that discounts, rebates, credits, refunds, price concessions and service level penalties are not expenses. They reduce revenue. So a company handing a month of free service to an unhappy customer is not incurring a cost of retention. It is reporting less revenue, in the same line as the sale.

The machinery sits in two places. Discounts and rebates contingent on volume or behavior are variable consideration, estimated up front and constrained where a significant reversal is possible.⁴ Payments to a customer for something other than a distinct good or service are consideration payable to a customer, reducing the transaction price directly.⁵ RevenueHub maps the four questions that decide which bucket applies, and notes the standard hardened what used to be a rebuttable presumption.

VMware's 2018 exchange with the SEC staff is the clearest real world illustration I know, because the company had to split its own channel programs across both buckets in writing. On the first it told the staff that "Rebates are variable consideration in accordance with ASC 606-10-32-5 through 32-9 and are accounted for as a reduction to the transaction price."⁶ On the second, marketing funds paid to partners, it said those were consideration payable to a customer. Two programs, two paragraphs, one identical effect on revenue.

The same filing settles something founders selling through partners get wrong constantly. Asked who the customer actually is, VMware answered that depending on the contract, it is either the channel partner or the end user. That decides whether the discount you gave a reseller is contra revenue or simply a lower price, and it opens onto the bigger version: whether you should report the gross amount at all.

A principal that controls the good or service before transfer "recognizes revenue in the gross amount of consideration to which it expects to be entitled," while an agent reports only its fee.⁷ If your platform resells somebody else's product, processes payments, or fulfills through a partner, that call can move reported revenue by a multiple without moving your bank balance a penny. PwC is emphatic that it is a judgment you must support on the facts, not a choice.⁸ And it barely touches net income while transforming the top line and the gross margin percentage, which are the two numbers investors price off.⁹

Nobody will tell you what belongs in cost of revenue, and that is the point

Here is the fact that surprises founders most. SaaS Capital, which has reviewed thousands of these statements, opens its guidance by saying so directly: "Surprisingly, GAAP does not clearly define what should be included in a SaaS company's Cost of Sales."¹⁰ There is no authoritative list. Each company decides. Which means gross margin, the metric everybody treats as the objective measure of software quality, rests on a boundary somebody in your company drew, possibly without thinking about it.

The boundary that moves the number most runs through your customer facing team, and the useful test is behavioral rather than titular. If the team's focus is retention, satisfaction, engagement and enablement, it sits in cost of goods sold. If its focus is renewal events, bookings and persuading somebody to buy more, it sits in sales and marketing. The same firm calls the placement settled: "It is now established best practice to include the customer support and customer success department in Cost of Goods Sold."¹¹ Same person, same salary, two very different margins.

The other thing worth knowing is that the biggest line in cost of revenue is often not variable at all. Hosting reads like a usage cost and contracts like rent. Appian discloses a non-cancellable arrangement with Amazon Web Services where "purchase commitments under the agreement total $220.0 million over five years," with a minimum annual spend of $44.0 million running to October 2029.¹² That is a fixed obligation sitting inside the line every founder describes as scaling with revenue. If growth slows, it does not.

The one number on your P&L that is actually two businesses

This is the section I would keep if you only read one. When you say "our gross margin is 71 percent," you are almost never describing one business. You are describing a weighted average of a software business and a services business, and in SaaS the services business is frequently sold below cost on purpose.

Appian's second quarter of 2026 makes it concrete. Subscriptions brought in $157.7 million against $25.4 million of cost, an 83.9 percent margin. Professional services brought in $45.6 million against $33.1 million, which is 27.4 percent. The reported gross profit is $144.7 million on $203.3 million of revenue, or 71.2 percent.¹³ The software business runs thirteen points better than the number on the page.

Workday is the same story with a sharper edge. For the year ended January 2026, subscription revenue of $8,833 million carried $1,531 million of cost, an 82.7 percent margin, while professional services revenue of $719 million carried $790 million of cost, a margin of negative 9.9 percent. Blended, 75.7 percent.¹⁴ And here is the instructive part: Workday's income statement prints no gross profit line at all. It runs revenue straight to operating income. A nine and a half billion dollar company does not compute the number for you.

Zuora once printed both figures side by side. In its fourth quarter of fiscal 2021 the reconciliation table shows gross margin of 59 percent and subscription gross margin of 77 percent.¹⁵ Eighteen points. A founder reading 59 percent would conclude they had a weak software company. They had a 77 percent software company plus a services business sold below cost to win subscriptions.

Dave Kellogg built the cleanest thought experiment on this, constructing two companies identical below the revenue line except that one carries an extra hundred million of services revenue, and showing that the reader's judgment flips on percent of revenue optics even though the second company has more gross profit and more operating profit.¹⁶ He also puts the range plainly, describing typical SaaS services margins as negative ten to negative twenty percent and worse.¹⁷ That is practitioner judgment rather than survey output, and the three filings above sit inside it.

The population data agrees. Benchmarkit's 2026 study puts median gross margin on total revenue at 76 percent and median software gross margin at 80 percent, and explains the spread this way: "This gap reflects the dilutive impact of professional services and non-software revenue on the blended margin."¹⁸ The private company survey work David Skok ran with KBCM found median subscription margins of 78 percent, flat across four consecutive years, against services margins near 26 percent that had swung from about 11 percent the year before.¹⁹ That volatility is itself the argument for reporting the two separately.

None of which makes services bad. Upfront implementation work correlates with better retention, and paying for it out of gross margin can be the right trade. What is not defensible is running two businesses through one line and then benchmarking against companies that do not. SaaS Capital's valuation index excludes companies where a major portion of revenue comes from services, which tells you what happens to your comparables once the mix gets heavy.²⁰

Capitalization, and the margin that is not really there

Your subscription gross margin also contains items that are not cash and not really about delivering the product. Zuora's reconciliation breaks out its cost of subscription revenue and shows stock compensation, amortization of acquired intangibles and amortization of internal use software all sitting inside it, together lifting non-GAAP subscription margin from 77 to 81 percent.²¹ Later filings confirm internal use software amortization runs primarily through general and administrative and cost of subscription revenue.²²

Which brings up capitalizing software development, the single largest lever a founder can pull on reported profitability without changing anything real. Capitalize the engineers and this year's cash outflow becomes an asset amortizing over several years. SaaS Capital argues against it for private companies, describing the practice as skipping the income statement and pretending the cash you paid your developers is still in the business.²³ Worth knowing too that Meritech defines free cash flow as operating cash flow minus capital expenditures and capitalized software, so the people valuing you strip it back out.²⁶

The rules just changed. In September 2025 the FASB modernized internal use software accounting, and RSM's summary notes that "The new guidance removes the need to differentiate between project stages, making the rules neutral to development methodology."²⁴ Capitalization now begins when management has authorized and committed funding and completion is probable, effective for annual periods beginning after December 15, 2027. Deloitte reads the Board as expecting more cost to be expensed for software sold as part of a cloud arrangement.²⁵

AI is now doing to cost of revenue what nothing else has managed in a decade. The a16z analysis found AI gross margins in the fifty to sixty percent range against a sixty to eighty percent software benchmark, driven by inference cost and human support in the loop, and its advice reads like a warning label: "Track down and measure your real variable costs."²⁷ Benchmarkit is more reassuring so far, reporting that AI infrastructure costs have not yet materially compressed software margins at the median, though usage-only pricing shows a median of 62 percent.¹⁸ If you have shipped an AI feature and cannot see its inference cost as its own line, you are flying without that instrument.

The operating lines, and the year the industry quietly cut

Below gross profit the reading rule changes. Dollars tell you almost nothing and percentages of revenue tell you almost everything, because these lines are how you find out what a company decided to be. Benchmarkit's 2026 medians give you the yardstick: sales and marketing at 35 percent of revenue, research and development at 27 percent, general and administrative at 17 percent.¹⁸ Scale moves them predictably, with G&A falling from 22 percent below $5 million of ARR to 12 percent above $100 million.

What happened in 2025 is the most interesting thing in that dataset. Median research and development fell eight points in a single year, from 35 percent to 27 percent, while sales and marketing came down two and G&A came down seven over three years. Benchmarkit reads the data as signalling that 2025 was the year profitability became the primary operating objective across the industry. Aleph, which sponsored the study, puts the finding in one line worth carrying into a board meeting: "Software companies got more profitable and less defensible in 2025."³⁰

One caution. An expense ratio is only meaningful against a growth rate, and Benchmarkit is explicit that high research and development can be an investment or an inefficiency depending on trajectory. Note too that Kellogg's paradox runs both ways here, because services revenue in the denominator makes every operating ratio look better at the same time as it makes gross margin look worse.

One last test, the cheapest on this page. Jason Lemkin's rule is to compare cash collections against monthly recurring revenue and treat collections below MRR as a quality of revenue problem rather than a collections problem.²⁸ Anita Kutlesa says she has heard as many definitions of bookings as there are flavors of ice cream, and that founders who look only at the revenue number are not doing themselves a favor.²⁹ The P&L is one of three statements, and the one most easily made to look pleasant.

The four checks I actually run

When a founder sends me a P&L, this is the order. It takes about twenty minutes and tells me more than an hour of conversation would.

1. Split the revenue. Recurring subscription revenue on one line, non-recurring services and one-time revenue on another, each with its own cost line beneath it. If services is more than roughly ten percent of revenue and it is not broken out, nothing below it can be trusted.
2. Audit cost of revenue line by line. Hosting, third party software embedded in the product, the people who keep production running, and the support and success work aimed at retention. Anything aimed at renewal, upsell or persuasion belongs in sales and marketing instead.
3. Compute subscription gross margin separately. That is the number that tells you whether you have a software business. Compare it against a median around 80 percent, and expect to be lower before roughly $20 million of ARR and higher after.
4. Read the operating lines as percentages, then against your growth rate. Sales and marketing, research and development, and general and administrative, each divided by revenue, each compared to the medians, each interpreted in light of how fast you are actually growing.

Do those four and you have not changed a single number. You have made the statement answer your question instead of the standard's, which is the whole job. The books still have to be squeaky clean underneath, because none of this survives revenue recognized in the wrong month. But once they are, a well organized P&L tells you where the business actually makes money, and that is not something you should have to guess at.