Metrics, retention and the organization · Answered by Golden Section from more than 400 B2B software companies observed
Split the number before you try to explain it. Separate logo churn from revenue churn, gross retention from net, seasonal pauses from real departures, and unproven accounts from the core book, then look for the trait the lost customers share. Most churn diagnoses go wrong because the founder is reasoning from one blended figure that mixes four different problems. We treat gross revenue retention of 95% or better as the level worth underwriting; the 2025 private median was 84%, and net revenue retention under 90% is a fix-first condition rather than something to fund. The next step is concrete: rebuild the last twelve months of cancellations on the ARR schedule, tag each one, and test at least ten churn predictor hypotheses against that list.
Diagnose churn on gross revenue retention for the core book, with seasonal pauses and provisional accounts reported separately, and only then decide whether the fix is product, onboarding, qualification or customer fit. Do not buy growth into a book whose gross retention you cannot explain.
| Metric | Value | What it means | Source |
|---|---|---|---|
| Gross revenue retention to underwrite | 95%+ | revenue retained from the starting cohort after contraction and churn, before expansionGolden Section's revised underwriting standard; below 80% is treated as disqualifying | Golden Section, publishedInvesting in Software, 2026 addendum |
| Private B2B SaaS gross revenue retention, median | 84% (2025), down from 88% | gross revenue retention, annualBenchmarkit annual benchmarks, n of roughly 340 to 580; the fall appeared at every quartile | External benchmarkBenchmarkit B2B SaaS benchmarks, as cited in Investing in Software |
| Net revenue retention floor | below 85% we pass; below 90% fix first | revenue from the starting cohort after churn, contraction and expansionequity pass line and lending minimum respectively | Golden Section, publishedGolden Section Equity and Lending pages |
| Churn predictor hypotheses | at least 10 | candidate factors across subscription details, use, support and satisfactionchurned customers must significantly outnumber the hypotheses for any relationship to mean something | Golden Section playbookChurn Identification Process |
A blended churn number hides its own causes. A customer who closes every January, a pilot account that was never going to convert, and a flagship customer who left because onboarding failed all produce the same canceled subscription in most billing systems. Averaged together, they point you at the wrong fix. The seasonal churn and core and provisional plays exist because boards, banks and acquirers price the whole book to the weakest cohort inside it.
Once the number is clean, the cause usually shows up as a cluster rather than a trend: every lost account bought the same tier, lost the same champion, or never reached the adoption metric. The churn identification process turns those clusters into leading indicators, which is how you move from explaining last year's churn to predicting next quarter's. Getting this wrong is expensive in a specific way. Churn you have to replace is not growth spend. At 84% gross retention, standing still consumes a large share of revenue before any real growth begins, which is why retention sets the achievable margin more than the cost structure does.
A vertical software company at $4M in annual revenue reports 86% gross retention and assumes the product is slipping. Rebuilding twelve months of cancellations on the ARR schedule shows three groups: seasonal accounts that return every spring, four provisional accounts signed on a champion's enthusiasm, and six core customers lost after implementations that ran long. With seasonal pauses pulled out and provisional accounts reported separately, core gross retention is 93%, and every core loss shares one trait, an onboarding that never hit its adoption metric. The fix is onboarding, not the roadmap. All figures are illustrative.
Very small books, roughly under 30 customers or fewer than 10 churn events, do not support pattern analysis; call every lost customer instead. Usage-priced products need retention measured on committed revenue, since normal usage swings can look like contraction.
From the Golden Section mistakes list, each paired with the play that prevents it.
Seasonal pauses logged as churn make a healthy book look like it is leaking and send the diagnosis in the wrong direction.
Unproven accounts blended into the core line drag the retention number down and hide whether the core product is actually holding.
Without a benchmark you cannot tell whether your retention is a crisis or ordinary for your segment.
In the order we would run them. Each is on its own page, most with a free Excel template.
Puts every contract change on its own dated line so churn and contraction can be measured on revenue, not logos.
Tags seasonal pauses at cancellation with a graduation rule, so real churn is counted in full and nothing else is.
Separates accounts that have not earned core status so the board sees what the core book actually retains.
Turns the clusters in your churned accounts into predictors and a monthly at-risk process with owners.
Customer plays The customer plays cover the full account lifecycle, from the ARR schedule through onboarding, adoption, renewal and offboarding, where most churn is caused or prevented.
State it as retention, on revenue, gross of expansion. We underwrite gross revenue retention of 95% or better as healthy for vertical software; the 2025 private median was 84%. Below 80% gross is a structural problem rather than a churn rate.
Churn is too much when it changes what capital you can take. Net revenue retention below 90% is a fix-first situation for our lending and below 85% is where we pass on equity. Well before that, if replacing lost revenue eats most of your sales capacity, churn is setting your growth rate.
Usually for one of a small set of reasons: expectations set in the sale that onboarding never met, poor product fit for a segment, loss of a key user, budget or ownership changes, a competitor, or a bad support experience. Find out which by clustering lost accounts on those traits rather than asking customers for exit survey answers.
Test at least ten predictor hypotheses against your historical churned accounts, keep the ones that cluster, and turn them into a monthly at-risk list with an owner. Check whether the predictors are leading or lagging, and measure whether churn actually falls among the flagged cohort.
Retention decides what capital a company can use. Our lending starts at 90% net revenue retention and our equity looks for retention above 100%, so a quarter spent diagnosing churn usually changes the terms more than a quarter spent raising.
Talk to Golden Section →Reviewed by Dougal Cameron, CEO & Co-Founder on 2026-09-23. Golden Section observations are labeled separately from external benchmarks and illustrative arithmetic.