Most recovery systems are described as a funnel, which is the wrong picture. A funnel implies gravity does the work. Recovery is closer to a small workshop with three stations, and a failed payment has to be carried through each one by hand, in order, or it quietly stops being recoverable at all.
The first station is detection, and it is less trivial than it sounds. A renewal fails at 3 a.m. The processor records a code. Somewhere between that record and your team's morning, the failure has to become a case: a named subscriber, an amount, a reason, a clock that started when the charge failed and not when someone noticed. Shops that lose money quietly usually lose it here. The failure sat in a report nobody opened, aged three days, and arrived at the second station already stale. Freshness is not a nice-to-have in this work. A card problem addressed on day one is a service moment. The same problem addressed on day nine is an apology.
The second station is the retry itself, and the craft is restraint. Each decline code is a different instruction. Insufficient funds says wait for money to exist. A replaced card says the old number is dead and no retry schedule will resurrect it. A generic decline says ask again, carefully, differently. Teams that retry everything on the same fixed calendar are not running a strategy; they are running a metronome, and metronomes eventually train issuers to say no out of habit. The walkthrough of how the detect, retry, and recover sequence fits together in practice is worth reading precisely because it treats the retry as a decision rather than a default.
The third station is the one most diagrams leave off: the subscriber's own action. A meaningful share of failures cannot be retried into health at all. The card was replaced, the account moved banks, the billing address changed. Those cases resolve only when a person updates their details, which means the message asking them to do it is not marketing garnish on the process. It is the process. It has to arrive while the failure is fresh, say plainly what happened, and lead to a page where fixing the card takes less than a minute and never asks for anything twice.
Between the stations there are handoffs, and handoffs are where cases die. Detection without a clock produces a pile. Retries without the subscriber path produce noise. Subscriber messages without retry discipline produce people who fix their card and then wait, uncharged, until they assume the service ended and cancel something they meant to keep. A saved subscriber at the end of the line is evidence that all three handoffs worked, not just the loudest one.
There is a measurement consequence to the workshop picture, too. A funnel gives you one number at the end and no diagnosis. Three stations give you three numbers: how many failures became cases, how many cases the retries healed on their own, and how many needed the subscriber's hand. When recovery dips, those numbers tell you which bench to stand at. When recovery rises, they tell you what actually improved, which is the only version of the story worth repeating next quarter.
We run Swipe Again as those three stations in that order, because the order is the insight. Detect early, retry like it is a decision, and treat the subscriber's fix as a first-class step instead of a footnote. The workshop is small. That is the point. Small enough to watch every case cross the room.
Top comments (0)