Every subscription business eventually quotes a statistic about failed payments — in a board deck, a planning doc, a pitch for better tooling. The numbers travel well because they are dramatic: a striking share of churn is involuntary, failure rates sit in the high single digits, recovery benchmarks promise a third or more of failures back. The problem is not that these figures are invented. It is that they arrive without their denominators, and a rate without a denominator is a rumour with a percent sign.
Start with the failure rate itself. Failed first attempts divided by what — all charges, all renewals, all active subscriptions? A business that counts retries as fresh failures will report a terrifying number and panic correctly about the wrong thing. A business that counts only first attempts against renewals due gets a calmer, more useful figure that can be compared month to month. Neither is lying. They are measuring different events and using the same name, which is how two honest teams end up arguing about whose benchmark applies.
The involuntary share of churn needs the same discipline. The claim that payment failure drives a large slice of subscriber loss is directionally sound, but the slice depends entirely on how exits are classified. A subscriber who cancels during a dunning sequence — after three notices about a card that never worked — files themselves under voluntary in most systems, even though the root cause was a billing failure nobody fixed. Count root causes and the involuntary share grows; count final clicks and it shrinks. Knowing which convention a statistic used is the difference between citing evidence and citing decoration.
Recovery rates are the most abused of the three. Recovered out of what pool, over what window, counting which endings? A rate computed only over cases still open flatters itself. A rate that counts a pause as a loss punishes a good outcome. The honest version states its pool — every failure that entered the process — its window, and its treatment of pauses, cancellations, and late manual payments, in the same breath as the headline number. A failed payment statistics page that shows its denominators alongside its rates is rarer than it should be, and it is the standard worth holding any benchmark to before it enters a planning document.
There is a practical test an operator can run in one afternoon: take last quarter's failures and recompute the three headline numbers under two different conventions each. If the story changes — if the problem doubles or halves depending on definitions — then the business does not yet have a measurement problem solved; it has a vocabulary problem, and every tooling decision made on top of it is a guess. Fix the definitions first, in writing, where the team can argue about them once instead of re-arguing them every quarter.
Definitions also decide who owns the problem: a failure rate owned by nobody in particular tends to be quoted by everybody and improved by none.
Benchmarks still earn their keep after that. They tell a business whether its first-attempt failure rate is ordinary or an outlier worth investigating with its processor. They set a plausible ceiling for recovery so targets are ambitious without being fictional. What they cannot do is substitute for the business's own counts, because a benchmark describes an average of other people's denominators. The mature posture is to cite external figures as context and internal figures as truth — and to keep the method for the internal ones boring, stable, and written down. That is the counting discipline Swipe Again publishes for its own recovery numbers, and it is available to any operator willing to spend the afternoon.
Top comments (0)