Stripe's Smart Retries are genuinely good at the job they were built for. Given a failed charge, the system picks a retry moment informed by signals from across the Stripe network, and it recovers a meaningful share of payments that a fixed schedule would have missed. Any honest teardown starts there: the feature works, it is included, and for a very small subscription business it may be all the recovery machinery the numbers justify. The interesting question is where its job description ends, because most renewal losses happen outside it.
A retry engine, however smart, only acts on the charge in front of it. It does not know that the card on file was replaced last week and the vault holds a token that will never succeed. It does not write to the customer, and it does not decide what the customer should hear, or when, or in whose voice. It does not distinguish the subscriber who would pay gladly if reminded from the one whose bank flagged the card as stolen, for whom another silent attempt is not neutral but actively unhelpful. Those distinctions are the substance of recovery work, and they live around the retry, not inside it.
The timing intelligence has a similar boundary. Network-level signals say a great deal about when cards in general tend to succeed. They say less about this subscriber: their payday, their timezone, the pattern of their previous renewals, the fact that their last two failures both cleared on the second attempt three days later. A system that treats every failure as an instance of a population will keep making population-level decisions for accounts whose individual history is sitting right there in the billing record, unused.
None of this is an argument for switching the feature off. It is an argument for reading it correctly, as one layer. Underneath it belongs card hygiene: updater coverage, pre-expiry outreach, a card-update path a customer can actually complete from a phone. Around it belongs communication: a first notice that arrives while a retry still has a chance, written like a person, carrying one clear action. Above it belongs measurement honest enough to say which recoveries the retries earned, which the emails earned, and which would have arrived anyway.
For the layer-by-layer version of that argument, stripe dunning starts with the feature's real strengths, then maps the work that still has to happen around it before a failed payment becomes a recovered customer rather than a quiet cancellation.
There is also a quieter cost to leaving the native feature as the whole system: silence. A subscriber whose card fails and who hears nothing interprets the silence in the worst available way — either the business did not notice, or it did not care. By the time a cancellation surfaces in a churn report weeks later, the moment when a short, decent message would have fixed everything has long passed. Recovery is partly a timing problem, but it is also a manners problem, and no retry schedule, however intelligent its timing, sends its regards.
The practical test for any team running on Stripe alone is a single question: take last month's failed renewals and sort them by what actually saved them. If the honest answer is that nobody can tell, the gap is not retries. The gap is everything the retries were never asked to do, and it will keep costing subscribers until something is assigned to do it.
That surrounding work is what Swipe Again is built to supply: retries woven into the gaps of the native schedule, customer communication in the merchant's name and colour, and counting that separates what each layer earned — so the native feature stays switched on, and stops being the whole plan.
Top comments (0)