Every recovery system, whatever its brochure says, is three stations and two handoffs. A failure is detected. A decision is made about what kind of failure it is. Attempts and messages go out in some order, and either money lands or the case closes. The brochures photograph the stations. The losses hide in the handoffs — the places where a case waits, unowned, while one part of the system assumes another part is watching it.
Station one is detection, and it is less trivial than it sounds. A failed payment is not one event but a family: the charge that declined, the retry the platform scheduled, the card the bank quietly replaced, the subscriber who updated their details somewhere the billing system did not look. A system that detects only the first event works a partial list all month and reports its recovery rate against a denominator it chose itself. Honest detection means the queue contains every case that is actually failing, including the ones no webhook announced.
The handoff into station two — triage — is where reading matters. A decline code is a sentence about a specific card at a specific moment: this one is early, that one is final, this third one is a timing problem wearing a money problem's coat. Systems that treat the queue as uniform spend their attempts evenly and recover unevenly. A plain walkthrough of how recovery moves from detection through retry to a saved subscription lays the sequence out in the order it actually happens, which is the order worth auditing in any tool, including the one you already own.
Station three is the attempt itself, plus the message that travels with it — and the second handoff, between attempt and message, is the one most often dropped. A retry fired with no notice asks the bank to approve a charge the customer may no longer expect. A message sent with no working fix link asks the customer to do homework. The pair, timed together, is the unit of recovery; either half alone is a gesture. When a case closes unpaid, the useful autopsy question is rarely "how many times did we try?" It is "at which handoff did this case sit still, and for how long?"
None of this requires exotic machinery. It requires that the three stations agree on what a case is, that no case changes hands without a timestamp, and that somebody — or some system — reads the stillness report: the list of cases that have not moved in days. Stillness, in a recovery queue, is never neutral. It is a decision made by default, and it always decides against the merchant.
Judge any system, ours included, by its handoffs. The stations are easy to build. The handoffs are the product.
The same handoff test works in reverse, as a design tool rather than an audit. Lay your last twenty lost cases on a table and mark, for each one, the last station it demonstrably reached and the length of its longest stillness. Patterns appear quickly that no aggregate rate would show you: a detection gap on one platform's failures, a triage rule that parks a whole decline family indefinitely, a message sequence that ends two days before the final retry fires, so the last attempt arrives unannounced. Each pattern is a single, fixable seam. Recovery programs rarely fail everywhere at once. They fail at one seam, repeatedly, while the stations on either side report themselves healthy — which is why the seams, and only the seams, deserve the first hour of every review.
Published by Swipe Again — failed payment recovery at a flat $179/mo.
Top comments (0)