DEV Community

Swipe Again for Swipe Again

Posted on

A Census of One Month of Failed Stripe Renewals

Before changing anything about how a Stripe subscription business handles failed payments, try a cheaper experiment: take last month's failed renewals and sort them by hand. Not into succeeded and failed — that split tells you almost nothing — but into families, by what the decline is actually claiming about the world. An hour with the export, a spreadsheet, and the decline reasons Stripe already recorded will teach you more about your recovery problem than a quarter of dashboard watching, because the dashboard shows you a rate and the census shows you a population. Populations can be worked. Rates can only be worried about.

The first family is credentials: expired cards, replaced cards, closed accounts. These failures share a property that should shape everything you do about them — no retry schedule on earth fixes them, because the thing being presented to the bank is wrong. They heal in exactly two ways: silently, when a card updater refresh supplies the replacement, or by the subscriber acting. Every automated attempt spent on an unhealed credential case is not neutral; it teaches the issuer's systems that this merchant retries blindly, and it spends the subscriber's patience on notices about a problem described too vaguely to act on. In most censuses this family is a third or more of the whole, which means a third of the typical retry calendar is aimed at cases that were never going to answer it.

The second family is funds and timing: insufficient funds, daily or weekly limits, the temporary holds banks place and lift without telling anyone. These are the cases where a retry genuinely is the treatment, but timing is the dosage. A balance problem on the 24th of the month is a different case from the same problem on the 1st, and a limit problem clears on a cycle the subscriber's bank chose, not one your scheduler did. This family rewards patience structured as observation — attempts placed where pay cycles and limit resets actually fall — and punishes enthusiasm. Five attempts in five days against a monthly balance cycle is one real chance and four decorations.

The third family is the bank asking for the human: authentication required, issuer instructions to contact the cardholder, the declines where the reason is effectively "we would like the person, please." These cannot be retried into compliance. They need a message that says precisely what to do — open your banking app, approve this merchant — written like instructions from a person rather than an alarm from a system. In the census they are usually the smallest family and the most neglected, because they fit neither the retry tooling nor the standard dunning template, so they get whichever one is closer and fail quietly under both.

Sorting a month this way produces three numbers that reorganise the work: how much of your failure volume needs a person, how much needs a calendar, and how much needs a credential. That is the working content of stripe dunning — and once you have the three numbers, most tooling arguments settle themselves. You stop asking whether your retries are aggressive enough and start asking whether each family is being handled by the mechanism that can actually close it.

One more habit falls out of the census for free: re-run it quarterly. The mix drifts. A growing annual-plan share swells the credential family; a price rise swells the funds family; expansion into markets with stronger authentication norms swells the third. A recovery setup tuned to last year's census is a suit tailored for someone who has since changed shape. The export is already sitting in your Stripe account, updated daily, waiting to be read as a population instead of a rate.

That reading habit — reasons first, mechanisms matched to families, outcomes counted only when money lands — is the operating discipline Swipe Again was built around: Stripe's native retries keep their lane, and the work concentrates on the gaps the census exposes.

Top comments (1)

Collapse
 
omyvnss profile image
Om Yaduvanshi •

populations can be worked, rates can only be worried about is the line i would keep. the five attempts in five days point lands for the same reason, against a monthly balance cycle that is one real chance and four decorations. the third family is the one i would watch most, even though it is the smallest, because a vague alarm email is the worst fit for a case that needs a precise instruction like open your banking app and approve this merchant.