Real, disclosed benchmarks

Search "SaaS churn benchmark" or "average SaaS growth rate" and the majority of what returns is content-marketing writeups citing each other in a loop, with no disclosed sample or methodology beneath any of it. This lesson uses only sources that disclose how their numbers were produced — real aggregated data from real companies, or a real survey with a stated sample size — and says plainly, for each one, what commercial interest the source has in the number looking a certain way. None of these sources are neutral; disclosed-but-interested is still meaningfully more trustworthy than undisclosed, and the difference is worth tracking.

#Baremetrics Open Benchmarks — real aggregated billing data, thin methodology disclosure

Baremetrics aggregates anonymized subscription-billing data from over 800 companies using its analytics product, and publishes live benchmark views (churn, MRR growth, failed-payment rates) segmented by average revenue per user. [Directional] — this is genuinely real, aggregated operator data (not a survey where companies self-report a number from memory), which is a meaningfully stronger evidence type than most of what circulates. It's downgraded from [Established] here because Baremetrics' own public pages don't disclose the underlying sample's composition (industry mix, company age, geographic spread) or a time-period breakdown in enough detail to know exactly what population "800+ companies" represents — and Baremetrics sells the analytics product this data comes from, so has an obvious interest in the benchmark tool looking useful and the underlying numbers looking authoritative. One disclosed figure from this source: Baremetrics' own data on failed-payment recovery found customers using its dunning-recovery feature recovered a median 410% return relative to what they spent recovering failed payments, from a December 2024 cohort of 148 customers — a real, dated, disclosed figure, specific to companies already using that recovery tooling rather than a general population.

#SaaS Capital's annual survey — large disclosed sample, lender's commercial interest

SaaS Capital, a specialty lender to SaaS companies, has run an annual survey of private B2B SaaS companies for fifteen years running; its most recent survey covers over 1,000 companies. Disclosed figures from the most recent survey: median growth of 22% overall (20% for bootstrapped companies, 25% for equity-backed); for bootstrapped companies in the $3M–$20M ARR range specifically, median 103% net revenue retention and 91% gross revenue retention; and equity-backed companies spend substantially more than bootstrapped ones across every function (roughly 70% more on sales, 100% more on marketing, 56% more on R&D, 100% more on customer success) for their extra five points of median growth. [Directional] — the sample size and survey cadence (fifteen years, over 1,000 respondents) are genuinely disclosed, which is real credibility, but SaaS Capital lends money to SaaS companies and has a direct commercial interest in the SaaS-as-asset-class narrative this annual report reinforces; treat the specific figures as real and worth anchoring on, and any narrative framing layered on top of them (how the report characterizes what the numbers mean) as coming from an interested party.

#ChartMogul's bootstrapped-vs-VC report — real billing data, undisclosed sample size

ChartMogul, a subscription-analytics company, publishes a periodic report comparing growth trajectories of bootstrapped and VC-backed SaaS companies using its own aggregated, anonymized customer billing data, cross-referenced with Dealroom's funding-status classifications. Disclosed finding: the top quartile of bootstrapped companies in ChartMogul's dataset reached $1M ARR only a few months slower than their VC-backed peers on average, and bootstrapped companies showed more resilience (smaller growth-rate drops) during the 2022–2024 slowdown than VC-backed companies did. [Directional] — real aggregated billing data again, which is a strong evidence type, but ChartMogul's own report doesn't disclose the exact number of companies in the underlying sample or state a margin of error, and — like Baremetrics — ChartMogul sells the analytics tool this data is a byproduct of, giving it an interest in the report driving attention to the product.

#The High Alpha / OpenView SaaS Benchmarks survey — large sample, self-reported

Originally run by VC firm OpenView Partners and now stewarded by High Alpha (also a venture firm), this annual survey draws over 800 self-identified SaaS company respondents (47% CEOs or co-founders in the most recent published sample), skewed toward US and European companies. Disclosed figures include median net revenue retention holding around 110% across several recent quarters. [Directional], downgraded specifically for the self-reported nature of the data (companies report their own numbers into a survey form, with no independent verification described) combined with the running firm's status as a venture investor with an interest in the SaaS category looking healthy — self-selection bias (which companies choose to respond to a benchmarking survey at all) is also a real, unaddressed concern with any voluntary-response survey of this kind.

#Reading these together

Notice what these four sources agree on directionally even though none of them are neutral and none would pass as [Established]-tier alone: bootstrapped SaaS companies grow somewhat slower than equity-backed ones on average, but not by the dramatic margin the "you need VC to grow fast" narrative implies; net revenue retention in the 100–115% range shows up repeatedly as a marker of a healthy business across independently-collected datasets; and gross revenue retention in the high-80s to low-90s percent is common even among genuinely healthy companies, which is worth knowing before assuming any customer loss at all signals a broken business. Convergence across multiple independently-interested sources, each with a different specific bias, is meaningfully more trustworthy than any single one of them alone — which is exactly why this lesson used four rather than picking whichever one had the most flattering number.