Somewhere in your last board deck there is a churn benchmark: an industry average, sourced from a respectable report, against which your number is pronounced fine or alarming. This note is not an argument against benchmarks — it is an argument for reading the one you have with the suspicion it deserves, because churn benchmarks fail in a specific and predictable way: they average together businesses whose churn is made of entirely different stuff, and then invite you to compare your stuff to the average.
The composition problem
Churn decomposes into voluntary exits (decisions) and involuntary ones (payment accidents), and the ratio between them varies wildly by model. A B2B tool with annual contracts and corporate cards might lose almost nobody involuntarily; a consumer subscription on monthly billing with debit cards can lose more people to failed payments than to every competitor combined. When a benchmark blends those worlds into "media subscriptions average 6%," it has produced a number that describes no actual business, and comparing yourself to it can mislead in both directions — complacency when your voluntary churn is terrible but your involuntary is low, panic when the reverse holds.
The definition problem, briefly
Benchmarks also blend definitions: customer churn against revenue churn, monthly against annualized, trials counted or not. Before any comparison, force the two questions — which churn, computed how — and refuse to proceed until the benchmark answers them or admits it cannot. A surprising share cannot, at which point the number is decoration and should be treated as such in the deck.
What benchmarks are actually good for
Two things. First, existence proofs: a benchmark table broken out by industry and model can reveal a loss category you have not been counting at all — if businesses like yours routinely lose 2% a month to payment failure and your reports show no such line, the gap is in your accounting, not your performance. Second, orders of magnitude: knowing whether a problem is a 0.2% problem or a 3% problem in your model class decides whether it deserves a meeting. A benchmark set organized by industry, with the sources named and the definitions stated is built for exactly those two uses; treat its finer distinctions as hypotheses about your business, never as grades.
The benchmark to actually manage against
Your own history, decomposed. Last year's involuntary share against this year's, voluntary exits by stated reason over time, recovery rates by class — these comparisons survive scrutiny because the definitions are constant and the composition is yours. External benchmarks can start that conversation once a year. The internal series should be running it every month.
One more failure mode: the borrowed target
The most expensive benchmark error is not analytical at all; it is managerial. A leader reads that top-quartile businesses in the category hold churn under some line, sets that line as the team's target, and thereby commits the roadmap to a number whose composition nobody examined. If the line was achieved in businesses with a fraction of your involuntary share — different cards, different geographies, annual billing — the team will spend the year chasing voluntary-churn programs for what is substantially a payments problem, or vice versa, and will be judged against a target that was never reachable by the route they were funded to take. Targets should be set from your decomposed ledger outward: this much voluntary reduction from these programs, this much involuntary reduction from that recovery work. The benchmark can witness the sum. It should never dictate it.
Compare to learn what exists. Measure yourself to learn what is true. The first is a library; the second is an instrument, and businesses are steered by instruments.
Benchmark literacy and retention data notes, from Swipe Again.
Top comments (0)