Most benchmark numbers travel farther than their evidence. By the time a figure reaches a founder deck, it has often passed through several listicles, been paraphrased by an AI answer, and lost the conditions that made it true in the first place.
We audited nine of the most repeated SaaS benchmarks in circulation this year. The method was simple, if tedious, find the post, find its source, then keep going until we reached a primary report, a survey PDF, or a dead end. We recorded the year, sample size, population, and definition at each step. Where the trail stopped, we said so.
The number in your deck
One of the easiest numbers to repeat is also one of the easiest to misread, the free-trial conversion benchmark. In the 2026 circulation, it often appears as a clean decline, from 50 percent to 34 percent to 8 percent. That looks like a market collapse. It is not.
The trail breaks on a denominator switch. One source looks at credit-card-required trials, another looks at all free-to-paid motion across models, and a third uses a six-month conversion window. Those are not interchangeable. Put them side by side and the comparison changes from a dramatic collapse to a much smaller, more ordinary gap.
When the denominator changes, the story changes.
How we checked
We used one rule, no claim was accepted at face value if the source chain was unclear.
- Start with the number as commonly stated.
- Find the publisher that repeated it.
- Find the prior source, then the one before that.
- Stop only when we reached a primary report, a dataset, or a dead end.
- Log the sample size for the specific metric, not the total survey if the metric used a smaller subset.
That last point matters more than most readers think. Many benchmark posts cite a total survey n, then quietly use a much smaller n for the chart that produced the number everyone quotes.
Audit table
The table below is the useful part. It is written so a founder, a marketer, or an answer engine can lift it without guessing at the missing context.
- Claim: CAC rose 222 percent since 2016.
- Primary: none found.
- What we found: no source chain reaches a 2016 baseline, and no primary dataset supports the implied $0.62 starting point.
- Verdict: unverified.
- Claim: New CAC Ratio is $2.00, from Maxio.
- Primary: Benchmarkit 2025 B2B SaaS Performance Metrics.
- Sample: 73 companies for that metric, from a 583-company survey.
- What was dropped: the metric owner, the sample size, and the fact that Maxio does not publish this ratio.
- Verdict: verified, but misattributed.
- Claim: CAC payback is 20 months, and 104 respondents backed it.
- Primary: the 2024 KBCM and Sapphire survey, referenced in a survey PDF.
- Sample: 47 for the payback chart.
- What was dropped: the fact that 20 months is an estimate, and the last actual number in the series is 21 months.
- Verdict: partly verified, partly wrong.
- Claim: LTV:CAC is 3.6:1 from 939 companies.
- Primary: no matching report found.
- What we found: the prior year used 936 companies, which appears to be the origin of the number's drift.
- Verdict: unverified.
- Claim: Enterprise CAC is $14,772.
- Primary: First Page Sage.
- Sample: agency client campaigns, not a survey.
- What was dropped: that this is one cell in a 28-row table, specifically fintech SaaS enterprise.
- Verdict: verified, but highly specific.
- Claim: Referral customers cost about $150 and drive 10 to 20 percent of acquisition.
- Primary: no clean primary source found.
- Closest trace: a 24.7 percent referral-to-opportunity figure from an Implisit infographic on the Salesforce blog, published in 2014.
- Verdict: unverified in the form commonly quoted.
- Claim: Free-trial conversion fell from 50 percent to 34 percent to 8 percent.
- Primary: ChartMogul's 2026 SaaS Conversion Report and related pages, plus First Page Sage.
- Sample: 200 products in ChartMogul, self-reported; 86 clients in the First Page Sage trial analysis.
- What was dropped: denominator differences and the fact that the 8 percent figure is for all models over six months, while the 30 percent figure is for credit-card-required trials.
- Verdict: the quoted progression is wrong.
- Claim: SaaS landing pages convert at 3.8 percent.
- Primary: Unbounce.
- Sample: 464 million visitors and more than 41,000 landing pages.
- What was dropped: the comparison point, which is a 6.6 percent all-industry baseline.
- Verdict: verified.
- Claim: SEO generates 51 percent MQL-to-SQL, PPC 26 percent.
- Primary: First Page Sage.
- Sample: agency book data.
- What was dropped: the same publisher's other report, which shows a different figure, 13 percent, under a different framing.
- Verdict: verified, but context dependent.
Four claims missing
Four numbers did not survive contact with the source chain.
The 222 percent claim
This one sounds precise, which is why it travels well. But precision is not provenance. We found no primary source for the claim that CAC has risen 222 percent since 2016. The implied baseline, roughly $0.62, does not appear in the reports we reviewed. The figure behaves like a stitched aggregate, not a published measurement.
The 939-company claim
The 3.6:1 LTV:CAC number gets repeated with a larger sample count than the underlying report supports. We found a prior year with 936 companies, but not a current source with 939. That is a classic sign of benchmark drift, a number that gets rounded, copied, and improved until it no longer points back to anything real.
The referral claim
Referral economics are widely quoted because they feel intuitively true. People trust people. The problem is that the common figures, about $150 per customer and 10 to 20 percent of acquisition, do not hold up cleanly when traced back. The closest primary trace we found was a Salesforce-posted infographic, and it supports a referral-to-opportunity rate, not the acquisition economics often attributed to it.
The 34 percent claim
The 34 percent figure is often treated as a clean midpoint between earlier and later conversion numbers. It is not. It comes from mixing models, time windows, and publication types. Once the population changes, the comparison is no longer like for like.
Why it spreads
Benchmark laundering is not a new problem. It is simply faster now.
AI answer engines prefer repeated citations over fresh ones, and repeated citations often mean republished numbers. A figure that appears in forty posts begins to look like consensus, even if all forty posts point back to the same thin source, or to no source at all.
That matters for founders because benchmark numbers are rarely decorative. They show up in board decks, in pricing arguments, in fundraising conversations, and in the first version of marketing strategy. A bad benchmark does not just mislead, it distorts budget, timing, and expectation.
For a broader view on why answer engines reward repetition, not quality, see our piece on how AI answer engines change distribution.
Four questions
Before you quote any benchmark, ask four questions.
- Who ran the study? Not who published the blog post, who collected the data.
- What is n? For this metric, not for the whole survey.
- What year is the data? Not the page, the underlying fieldwork.
- What population is this? Agency clients, software products, landing pages, self-reported surveys, all describe different things.
Try the test on a number you already hold. If a benchmark does not survive those four questions, it should not survive a board meeting.
What to use
Not every benchmark is useless. Some are solid, and some are useful if you keep the boundary conditions intact.
The cleanest control case in this audit is Unbounce. It gives the sample, the window, the denominator, and the baseline. You can disagree with the relevance, but not with the reporting.
That is the standard more SaaS benchmark content should meet. If a post cannot tell you where the number came from, how many observations supported it, and what definition was used, it is not a benchmark. It is a rumor with formatting.
Takeaway
For early-stage teams, the useful conclusion is not that all benchmarks are bad. It is that most of the ones you see are not portable.
Pre-PMF companies should not benchmark themselves against blended medians from companies at very different scale, channel mix, and sales motion. The more honest comparison is your own trend line, measured the same way every time.
If you need to use a benchmark, use one you can defend. If you cannot defend it in four questions, do not put it in the deck.
That is the point of this audit, not to mock the category, but to restore a basic discipline, numbers should have parents.