Every subscription business has a drawer. Not a physical one, usually — a filtered view, a spreadsheet tab, a mental pile of renewals that failed last quarter and were never looked at again. They were not written off after a careful review. They simply aged out while the team was busy with work that had a louder deadline. That drawer is the most honest evaluation set a buyer owns, and almost nobody uses it that way.
Start by pulling thirty cases from it at random. Not the biggest, not the most recent — random, because a vendor demo always chooses its examples and a random sample refuses to flatter anyone. For each case, write down three plain facts: why the payment failed, in the bank's own language; what your current process did next, and when; and what eventually happened to the customer. Most teams can answer the first question from their billing platform. The second is usually patchier. The third often lives nowhere at all.
That gap is the first finding, before any software is judged. If a business cannot reconstruct what happened to thirty of its own failures, its problem is not only recovery — it is record-keeping. A good recovery system fixes both at once, because every attempt, message, card update, and cancellation lands on a case timeline that a person can read later without detective work. Ask any vendor to show you that timeline for a messy case, not a clean one. The messy case is the product.
Now run the drawer test itself. For each of the thirty cases, ask what a candidate tool would have done differently at the first decision point. Sometimes the honest answer is nothing: the card was closed, the customer had already cancelled in their head, and no sequence of retries changes that. Count those cases and set them aside. They define the ceiling. The remaining cases are where tools actually differ — a soft decline retried on the customer's payday instead of the system's schedule, an expired card flagged weeks before the renewal instead of on the morning it failed, a customer who needed one clear email rather than four increasingly urgent ones.
This is the standard behind payment recovery software tested against a live failure drawer: judge the tool on the cases you already lost, where the right answer is known in hindsight, instead of on a staged flow where every answer is convenient. A tool that would have changed ten of thirty outcomes, at a flat monthly price you can state in one sentence, is a different proposition from a tool that promises a percentage with no denominator and a contract you cannot summarise.
Pay attention, too, to what the tool does when a customer replies. In the drawer, there is almost always a case where the customer wrote back — confused about the charge, asking for a few days, disputing the amount — and the reply sat in a shared inbox while the automated sequence kept running underneath it. Recovery software that cannot pause when a human enters the conversation will eventually damage a relationship that the billing team then has to repair by hand. The pause matters as much as the retry.
Price belongs in the drawer test as well. Take the recoverable pool the thirty cases imply for a full month, apply a cautious recovery share rather than a vendor's best case, and compare that cautious figure with the actual invoice the tool would send. Flat pricing makes this arithmetic boring, which is a virtue. Percentage pricing makes the invoice grow exactly when the tool works, which may still be worth paying — but the buyer should see that shape on paper before signing, not discover it on the first successful month's bill.
None of this requires a pilot, a sales call, or a migration. It requires one afternoon, a random sample, and the discipline to write down what actually happened. Teams that do it stop shopping for features and start shopping for judgement: does this system read a case before it acts, spend its attempts where the evidence points, stop when it should, and keep a record honest enough to learn from? The drawer already contains every answer a demo is designed to avoid asking about.
That discipline is the through-line at Swipe Again — build the record first, let the recovery follow from what the record shows, and count only what can be shown case by case.
Top comments (0)