Every quarter, some team runs a retention audit. It usually looks like this: pull the churn number, slice it by plan and tenure, read a sample of exit surveys, argue about onboarding, and leave with a roadmap full of product work. Three months later the churn number has barely moved, and nobody can say which of the roadmap items mattered.
The problem is not effort. It is sequencing. Most audits start with the customers who chose to leave — the loudest, most legible group — and treat the customers who were pushed out by a failed card as a rounding error owned by "payments." In subscription businesses, that second group is routinely a quarter to a third of all losses, and it is the only group whose cause you can read directly off a ledger instead of inferring it from survey prose.
A billing-first audit takes one afternoon and four exports.
Export one: ninety days of failed first attempts
Not retries — first attempts. Retries are your recovery system's diary; first attempts are the raw leak. For each one, you want the date, the amount, the decline code, and whether that subscriber is still active today. The single most clarifying number in the whole audit is the share of failed first attempts that belong to people who never came back and never cancelled on purpose. That is your involuntary loss, priced.
Export two: the decline-code histogram
Sort the failures by code and read the top five like diagnoses, not statistics. A file heavy with insufficient-funds codes is a timing problem: renewals landing before payday, retries fired on the same dead day of the week. A file heavy with expired-card and do-not-honor codes is a card-file hygiene problem: no pre-renewal update prompts, no account-updater coverage, descriptors customers do not recognize. Generic declines dominating usually means the retry logic is resubmitting unchanged transactions and hoping. Each code family has a different fix, a different owner, and a different cost. An audit that ends with "improve dunning" has not done this step.
Export three: the recovery timeline
For the failures that did eventually pay, how long did recovery take, and which attempt collected? If most recoveries land on attempts four through six, your early attempts are decoration. If recoveries cluster right after a specific email, that email is carrying the system and deserves to be protected, tested, and never casually rewritten by a brand refresh. If recoveries cluster on particular days of the month, your cadence should be rebuilt around those days instead of around a fixed interval timer.
Export four: the silent list
Subscribers whose payment failed, who never recovered, and who also never opened a single message. This group decides whether your problem is reach or offer. If the silent list is large and your sender reputation is mediocre, the fix is deliverability and channel mix — a text message where email dies — before it is copy. If the silent list is small and most non-recoverers did open messages, the fix is the ask itself: the update page, the friction at the card form, the absence of a save offer for people whose real problem was price, not plastic.
Scoring it honestly
The audit is only useful if it ends in a short scored list: each finding with an estimated monthly value, an owner, and a cost to test. That discipline is why a structured worksheet beats a slide deck — if you want a worked format, a retention audit worksheet that starts with the billing queue lays out the four exports above as fill-in sections with the scoring table built in.
Run this audit before the exit-survey audit, not after. Voluntary churn work is slow, expensive product work. Involuntary churn work is mostly operational, measurable within one renewal cycle, and — this is the part that surprises people — it improves the quality of every voluntary-churn signal downstream, because the people left in your exit data are the ones who actually chose to leave.
What we do with audits like this
At Swipe Again, this billing-first sequence is the intake for every recovery engagement: read the failure file before touching a single email. The order matters. Fix the leak you can read before you fund the leak you have to guess at.
Top comments (0)