A recovery rate is a fraction. That sounds too obvious to say out loud, until you watch two honest reports describe the same month of billing and disagree by thirty points. Both did the arithmetic correctly. They simply were not counting the same thing — and because the counting rules were buried, or never stated, the larger number won the meeting.
This is a field guide to not being fooled, including by yourself.
One dataset, four defensible answers
Take a month with 1,000 renewals. Of these, 120 fail on the first attempt: 80 soft declines — insufficient funds, expired cards, temporary issues — and 40 hard declines, where the issuer has effectively closed the door. Over the following weeks, 48 of those subscriptions are eventually paid and active again. Six of the 48 are later refunded or charged back.
Now watch what the headline rate becomes, depending on four quiet choices.
1. All failures, eventually recovered, gross. 48 ÷ 120 = 40%. This is the most conservative common version, and the least flattering.
2. Soft declines only. Hard declines are excluded as “not really recoverable,” so 48 ÷ 80 = 60%. Same recoveries, same month — a twenty-point improvement achieved entirely by editing the denominator.
3. First retry only. Suppose 20 of the recoveries happened on the first retry. Reported as a first-attempt rate, the figure is 20 ÷ 120 = about 17%; reported the other way, a vendor might quietly switch to eventual recovery when their first-attempt number looks thin.
4. Net of refunds and chargebacks. The six clawed-back payments come out: 42 ÷ 120 = 35%. The business actually kept 35% of the failed cohort, not 40%, and certainly not 60%.
None of these calculations contains an error. Each contains a decision. That is why we insist on failed payment recovery benchmarks, with the counting rules stated before any comparison is allowed into a planning document — the rules are not footnotes, they are the result.
Why published benchmarks mislead even when nobody is lying
Benchmarks aggregate businesses that do not share a decline mix. A consumer app charging $9 a month has a failure pool heavy with insufficient funds. A B2B tool charging $400 a year has fewer, larger, stranger failures. A company that already runs good retries before a vendor arrives will show a lower vendor lift than one starting from nothing — not because the vendor worked worse, but because the easy recoveries were already being recovered. Comparing a raw rate across those pools is like comparing batting averages across different sports because both use a ball.
Vendor-quoted rates deserve a particular, polite skepticism. Ask which denominator was used, over what window, whether the pool was their customers (a self-selected, surviving group) or all trials, and whether recovered-then-refunded revenue was removed. A vendor who can answer those questions crisply is more credible than one whose rate is ten points higher and whose definitions are fog.
Build your internal benchmark first
Start with a cohort you can freeze. Take one month of first-attempt failures, close the books thirty days later, and do not let later recoveries wander back into the count. Record, for that cohort and nothing else: how many failed, how many were soft versus hard, how many were eventually paid, how much revenue those payments represented, and how much of that revenue was still yours after refunds and chargebacks.
Then stratify before you average. Break the cohort by decline category, by plan size, and by new versus established subscribers. Averages hide the only actionable information — which kind of failure you are actually good or bad at recovering. If your soft-decline recovery is strong and your expired-card recovery is weak, a single blended rate will keep that a secret from you for quarters.
Repeat the same frozen method monthly. The value is not the first number; it is that the numbers become comparable to each other. Only then does an external benchmark become useful, as a question — “their soft-decline rate is higher; is their denominator like ours, and if so, what do they do differently?” — rather than as a verdict.
The rule to keep
Refuse any recovery rate, internal or external, that arrives without its denominator, its time window, and its treatment of refunds. Those three omissions can move a result further than most product changes ever will, and unlike a product change, they cost nothing and recover nobody.
Counting carefully will not save a single subscription by itself. It will stop you from spending a quarter chasing a number that was never counting your problem.
Top comments (0)