Failure statistics are the most quoted and least examined numbers in subscription operations. A deck claims a category failure rate, a vendor page claims a recovery rate, a founder repeats both in a pitch, and somewhere along the chain the denominators fell off. This matters beyond pedantry, because these numbers steer real decisions: whether to build a recovery program at all, which vendor's promises to believe, and how much involuntary churn to blame for a growth problem that actually lives elsewhere. Quoting a number carefully is not academic caution. It is how you avoid buying the wrong fix with great confidence.
Three questions rescue most quotes. First: of what? A failure rate over all renewal attempts is a different animal from a failure rate over first attempts, and both differ from the share of subscribers who experience a failure in a year. Second: over what window, and whose population? A rate measured on enterprise annual contracts says little about five-dollar monthly plans, and a vendor's recovered-revenue figure measured across their happiest cohort says little about yours. Third: recovered by whom, counted how? A payment saved by an automatic card refresh, a payment saved by a subscriber clicking an update link, and a payment saved by a well-timed retry are three different mechanisms with three different lessons, and any statistic that blends them into one heroic percentage is marketing, not measurement. The discipline behind failed payment statistics worth quoting carefully is simply that every figure travels with its definition attached, or it does not travel.
It helps to keep a small set of honest ranges in your head rather than one dramatic point estimate. Renewal failure rates for consumer subscriptions commonly sit in the high single digits to mid-teens depending on price, card mix, and geography. A meaningful share of those failures are credential problems that no cadence can cure. Recovery, honestly counted against the full failed population rather than the convenient subset, is partial by nature, and any claim approaching totality should trigger the three questions immediately. Ranges feel less quotable than a single striking figure. They are also the only version that survives contact with your own billing data, which is the dataset your decisions actually run on.
For your own reporting, the standard to hold is reproducibility by a skeptic. Publish the population, the window, the definition of recovered, and the date the figures were pulled, alongside the number itself. When someone challenges the figure, and someone should, the argument becomes about definitions, which can be settled, instead of about honesty, which cannot. Internally, the same standard protects you from your own team: a monthly recovery figure that names its denominator cannot quietly improve because the denominator shrank.
None of this makes statistics useless. It makes them useful in the way instruments are useful, calibrated and read with their limits in view. The operators who quote carefully also tend to buy carefully, test carefully, and report carefully, and those habits compound into the only durable advantage in this corner of the business: believing true things about your own renewals slightly earlier than your competitors believe them about theirs. Swipe Again's published recovery benchmarks are maintained under the same rule, definitions attached, because a recovery vendor asking for trust on its numbers should be the first to show its denominators.
A final test, quick enough to run in a meeting: ask whoever quoted the number to state, in one sentence, what would make it go down next quarter. If the answer names a mechanism, a population, or a definition, the number is probably connected to something real. If the answer is a shrug or a larger number, you are holding a slogan. Slogans have their place in a pitch. They have none in an operating plan, and the discipline of telling the two apart, kindly and out loud, is most of what quoting carefully means in practice.
