DEV Community

Swipe Again
Swipe Again

Posted on

Audit Your Retry Budget Before You Spend Another Attempt

Every failed payment gives you a small budget of retries, and most subscription businesses spend that budget blindly. The platform retries on its own schedule — a few attempts, spaced by a rule written for the average merchant — and when the budget runs out, the subscription cancels and the case closes. Nobody asks what each attempt was spent on. That is the question worth asking, because attempts are not interchangeable, and neither are the failures they chase.

Start the audit by sorting last month's failures by decline type, not by date. Expired and reissued cards form one pile: retries against a dead credential will fail every time, so the correct spend on that pile is zero attempts and one good card-update message. Insufficient-funds declines form a second pile: these are timing failures, and attempts spent before payday are wasted while attempts landed just after it succeed at a completely different rate. Generic and do-not-honor declines form a third, murkier pile, where a small number of carefully spaced attempts is reasonable and a rapid burst is not — issuers read bursts as noise, and some read them as risk.

Now count what the current setup actually spent on each pile. The usual finding is an inversion: the most attempts went to the dead-credential pile, because those failures retry cleanly and often, while the timing pile — the one with real recoverable money in it — got whatever schedule was left over. The platform was not careless. It was generic. It cannot know your subscribers' paydays, your price points, or which declines your issuer mix produces most. That knowledge is the entire case for a dedicated retry layer.

This is the lens the comparison in stripe dunning that allocates attempts by decline type instead of by habit is built around: attempts treated as a budget with a return, not a metronome. The evaluation questions change accordingly. Not "how many retries do you run?" but "what decides whether this failure gets another attempt, and when?" Not "do you send dunning emails?" but "which failures get a message instead of an attempt, because a message is the only thing that can fix them?"

Two audit numbers tell you whether the reallocation worked. The first is attempts per recovered payment: if it falls while recoveries hold or rise, the budget is being spent where it earns. The second is the share of dead-credential failures resolved by card update rather than exhausted by retries — the quiet metric that separates systems which read declines from systems which merely repeat charges. Track both monthly. They move before revenue does, and they explain the revenue when it arrives.

There is also a customer-experience line in the audit. Every unnecessary attempt is a small withdrawal from trust: a subscriber who sees three failed charges in a week reads carelessness into a business they otherwise like. A budget spent deliberately — fewer, better-timed attempts, paired with one clear message — is not just more efficient. It is quieter, and quiet billing is a retention feature in its own right.

One caution belongs in every audit report: do not judge a pile by a single week. Decline mixes shift with billing cycles, holidays, and the card portfolios your newest subscribers happen to carry, and a timing rule tuned on an unusual week will misfire on an ordinary one. A full month is the smallest honest sample, and two months is better where volumes allow it. The point of the audit is a durable allocation, not a clever one — a budget rule the team can leave running, re-check quarterly, and trust precisely because it was built on ordinary data rather than on the strangest week of the year.

Run the sort once, with your own decline mix. Most teams find the recoverable money was never in more attempts. It was in the allocation.

That allocation-first view is the working principle at Swipe Again: read the decline, spend the attempt where it can earn, and let the subscriber fix the failures only they can fix.

Top comments (0)