DEV Community

Swipe Again for Swipe Again

Posted on

Start With the Failures You Can Actually Move

Every Monday morning the failure queue looks the same size, and that sameness is the trap. A hundred failed renewals read like one problem with one fix: retry harder, email louder, wait longer. In practice the queue is at least four different problems wearing the same label, and only some of them can be moved this week. The operators who shrink their failure rate steadily are not the ones with the most attempts. They are the ones who sort first and spend their attempts where movement is still possible.

Start with the movable middle. These are subscribers whose card was fine last month and whose decline carries a temporary reason: a balance that was short on one particular morning, a bank that asked for authentication nobody completed, a daily limit brushed by a larger annual charge. Nothing about the relationship is broken. What these cases need is timing and a clear next step, in that order. Retry them when the underlying condition is likeliest to have changed, and make any message they receive name the exact action, not a general warning about an account problem.

Next, separate the credential cases and stop retrying them as if they were timing cases. An expired card will be expired tomorrow. A replaced card will stay replaced. Presenting the old credentials again does not test anything except the bank's patience, and each pointless attempt makes the eventual ask look less serious. For this group the only productive move is an early, specific request for updated details, sent while the subscriber still remembers signing up and still wants the thing they bought. That is the practical core of a workable order for reducing failed payments, beginning with the cases that can still move: sort by whether the condition can change on its own, and let the sort decide whether you wait, retry, or ask.

Then look honestly at the cases that will not move. Some subscribers have closed the account, frozen the card after a dispute, or quietly decided and never said so. A mature recovery operation names an endpoint for these and stops. Not because persistence is wrong, but because endless attempts against a closed door train everyone — the team, the bank, and the remaining subscribers who share that bank — to treat your charges as noise. A written endpoint, applied evenly, also makes the rest of the numbers readable. You cannot judge a retry schedule if a fifth of its attempts are being spent on the dead.

There is a fourth group most queues hide: failures caused by your own setup. A descriptor a subscriber does not recognise. A renewal date that drifted away from the payday pattern the subscriber chose at signup. A plan change that reset the amount without a word of warning. These do not belong in dunning at all. Each one is a product or communication fix that prevents a whole class of future failures, which is worth more than recovering any single charge inside the class. Pull them out of the queue and into a fix list with an owner and a date.

A simple Monday discipline follows from the sort. Count the queue by reason before anyone touches it. Spend the first effort on the movable middle, send the credential asks the same day, close the dead cases at the agreed endpoint, and escalate the self-caused cases out of billing entirely. Then measure the week not by attempts made but by how the composition changes: the movable share should clear, the credential share should answer or age out, and the self-caused share should shrink toward zero over a quarter.

That discipline is the whole method at Swipe Again: read the reason before spending the attempt, ask early for the one thing only the subscriber can give, and fix the failures that were never really billing failures at all.

Top comments (0)