AI & SaaSAI SaaS CAC Benchmarks: Why Yours Has to Be Built, Not Looked Up

Lumina Clippers runs short-form video clipping campaigns, meaning creator-distributed video clips, for AI and SaaS brands. Not press clippings, not audio clipping. This article is about the cost of acquiring one customer, not the size of the budget.

Publishers that do segment AI-native companies segment them on growth, on gross margin or on retention, and then stop before customer acquisition cost.

High Alpha and Kyle Poyar's 2025 SaaS Benchmarks report, published November 12, 2025 from 800 or more companies surveyed in August and September 2025, is the one release that splits AI-native companies out. At the $1M to $5M annual recurring revenue (ARR) band it shows them growing 110 percent against 40 percent for general SaaS, on gross margins 5 points lower. It publishes no CAC split.

Emergence Capital's Beyond Benchmarks also splits AI-native out, reporting 4x faster growth and 21 percent higher retention. It publishes no CAC figure for that cut, and the report page discloses neither a publication date nor a sample size.

Kyle Poyar's Growth Unhinged piece of March 15, 2026 puts it plainly: "LTV:CAC was always fun-with-numbers, and now we need to make it official." LTV:CAC is a lifetime return multiple resting on a retention assumption; CAC payback is the months of gross profit needed to recover acquisition cost. Poyar publishes neither, and a metrics critique with no CAC figures in it is itself the finding.

Three organizations publish B2B SaaS acquisition-cost data with any regularity. They measure different things, and the one that discloses the largest sample publishes no CAC figure at all.

Benchmarkit's 2025 SaaS Performance Metrics report puts the median New CAC Ratio at $2.00 of sales and marketing expense per $1.00 of new customer ARR for CY2024, up 14 percent during that year. Benchmarkit discloses a survey base of 583 participants, though not the sample behind every metric, and it sells benchmarking software. Its 2026 report is titled "SaaS & AI-Native Metrics", which a reader will raise against this article's thesis. The phrase appears in the title and the URL but not as a segment anywhere in the body, and that summary page shows neither a sample size nor a publication date.

SaaS Capital's survey completed March 2026, its fifteenth annual, covers more than 1,000 SaaS companies and puts median selling costs at 15 percent of ARR and median marketing at 8 percent of ARR. It publishes no CAC and no payback figure: the largest disclosed sample in the field does not produce the number everyone wants. Why those spend percentages disagree from publisher to publisher is worked through in AI SaaS marketing budget benchmarks.

First Page Sage's figures, published April 17, 2025 and updated June 18, 2025, cover 2022 to 2024 and are one agency's client book. A separate Benchmarkit study of 323 B2B companies found only 52 percent of marketing teams measure marketing cost per dollar of pipeline, so the benchmarks that do get published rest on the half that does.

Two things circulate as AI CAC benchmarks and neither is. One reports faster payback for companies adopting AI tools, a different population from yours. The other reprints general B2B SaaS CAC ranges under an AI headline. Both are population errors, and a benchmark on the wrong population is worse than none.

One scope note. Everything below assumes subscription pricing with a stable annual contract value. If you price on usage, substitute trailing twelve-month revenue per customer and accept the volatility. If your go-to-market motion is not settled, start with marketing an AI startup.

Maximum allowable CAC is monthly gross profit multiplied by target payback months. Hold contract value and payback target constant and that ceiling is directly proportional to gross margin. Change either and the proportion stops holding, which is why both get locked before any margin comparison.

An AI company sells a $24,000 annual contract, which is $2,000 of monthly revenue. At a 55 percent gross margin, the midpoint of the 50 to 60 percent range for AI companies in Bessemer's AI Pricing and Monetization Playbook of February 10, 2026, which gives its basis as dozens of AI teams and states no sample count, monthly gross profit is $1,100. Against Bessemer's mid-market CAC payback target of 18 months, maximum allowable CAC is 1,100 x 18 = $19,800.

The same $24,000 contract at an 85 percent gross margin, the midpoint of the 80 to 90 percent range Bessemer reports for classic SaaS in the same document, produces $1,700 of monthly gross profit. Against the same 18-month target, maximum allowable CAC is 1,700 x 18 = $30,600.

Same product, same $24,000 price, same 18-month target, and the AI company's ceiling is $10,800 lower, 35.3 percent below the classic SaaS ceiling. The ratio of the two is 19,800 / 30,600 = 0.647, exactly 55 / 85. The ceiling tracks the margin and nothing else.

How wide the AI margin gap actually is depends on the publisher. High Alpha's survey of more than 800 companies puts AI-native gross margins 5 points lower than general SaaS; Bessemer's midpoints put the gap at 30 points. Both Bessemer figures come from one document on one undisclosed basis, so that gap is at least measured the same way, but the true gap is unsettled and only your own gap moves your ceiling. Benchmarkit's 80 percent is a median for software gross margin, which excludes services, so it is not directly subtractable from a whole-company AI margin. At 80 percent the ceiling is $28,800, so the conclusion does not turn on which you pick.

Bessemer's State of AI 2025 deck, which states no company count, puts the fastest-growing AI cohort at about 25 percent, footnoting that margins there are often negative, and the next cohort at 60 percent. On the same contract at 18 months, that 25 to 60 percent range gives a ceiling from $9,000 to $21,600. The $19,800 is a midpoint case, not a benchmark. Replace 55 with your own audited margin.

Benchmarkit's 2025 New CAC Ratio of $2.00 isolates new-customer acquisition, which is what a ceiling constrains, so this article uses it over the blended $1.30 in Benchmarkit's 2026 report, which includes expansion. That $2.00 median on a $24,000 contract is $48,000 of acquisition spend, 2.42 times the $19,800 ceiling, or 142 percent over. Margin structure is not the explanation. Run the same median against the 85 percent classic SaaS margin and $48,000 is still 1.57 times the $30,600 ceiling. In months, a $2.00 CAC ratio at an 80 percent gross margin implies 2.00 x 12 / 0.80 = 30 months of payback, and Benchmarkit's own report says about 12 months is good. That $2.00 is a CY2024 median and Benchmarkit's 2026 data has efficiency improving since, so treat it as a high-water mark. Even as one it clears the AI ceiling by more than double. Compute your own ceiling before judging any channel against it, including the work it takes to turn a product demo into demand.

Monthly gross profit times target payback months. Three inputs, one multiplication, every input off your own books.

This formula is not new. It is gross-margin-adjusted CAC payback and any seed-stage CFO has run it. What is new is the cohort: companies whose gross margin sits below the structure it was calibrated on, and for whom the benchmark that would say where to sit does not exist.

Define the numerator first. CAC means all sales and marketing expense in the period, fully loaded salaries, commissions, tooling, agency fees and paid media, divided by the new customers closed in that period. Customer success and account management sit outside it, because they serve existing revenue. Two founders using different numerators produce ceilings that differ by half, which is how comparability gets destroyed.

If your current CAC is $30,000 against $1,100 of monthly gross profit, your actual payback is 30,000 / 1,100 = 27.3 months, which fails an 18-month target. The gap is not "spend less". It is $10,200 per customer. The calculator below runs the same arithmetic on your numbers.

A ceiling is per customer; a budget is not. Multiply it by the new customers you need. Forty against a $19,800 ceiling is a $792,000 maximum acquisition budget. Sanity-check that against revenue: the fifteenth annual SaaS Capital survey, completed March 2026 with 1,000+ participants, puts median selling costs at 15 percent of ARR, with marketing at 8 percent. Land far outside 23 percent of your own revenue and one of your inputs is wrong, either the ceiling or the customer count.

Five ways this goes wrong in practice.

Once the ceiling is set, every channel is a question of whether it clears, the same test we apply to our own campaign pricing.

You have three levers: raise gross margin, shorten payback, or change cost structure. If your gross margin today is under 30 percent, or negative, the formula returns a ceiling far below your current CAC, and that is not a reason to stop acquiring. Compute the ceiling at your projected margin at the end of the payback window, write the assumption down, and treat the gap between today's margin and that projection as the number your roadmap has to close. ICONIQ's survey of roughly 300 executives, fielded April and December 2025, has AI companies projecting about 52 percent average gross margin for 2026, the only published projection to check yours against and a self-report rather than an observed result.

Each has a counterweight.

Raise gross margin. Real, and slow. Inference cost per token has been falling, but a ceiling budgeted off today's margin across an 18-month window is wrong by construction in one direction or the other. State the assumption out loud.

Shorten payback. Only defensible if the sales motion supports it. Moving from enterprise to mid-market changes the target and the contract value at once, which breaks the proportionality the ceiling arithmetic depends on.

Change cost structure, not spend level. Paid channels clear through an auction, so cost per click has a floor set by other bidders. Creator-distributed short-form is priced as a unit rate instead. That is a different cost structure, not automatically a cheaper one, and cost per view is not cost per customer.