Every subscription business reports a recovery rate, and most recovery rates are not comparable to anything — including the same business's rate from last quarter. Definitions drift: one team's numerator includes platform-native retries, another's excludes card updates, a third quietly drops cancelled cases from the denominator. The number still gets quoted in meetings with two decimal places. If you want to know whether dunning is actually working, you need a small set of metrics with definitions fixed in advance, read together, because any one of them can look healthy while the operation rots.
First, the failure rate itself: failed renewals as a share of renewals attempted. This is the input everything else works on, and a rising failure rate with a steady recovery rate still means more subscribers lost. Second, the recovery rate — with its definition written down: which recoveries count, over what window, and whether cases that cancel mid-sequence count as losses. They should. A rate that improves by reclassifying its failures is the oldest trick in this category. Third, split that recovery by source: platform retries, customer card updates, and your recovery process's own attempts. The split tells you what you are actually paying for, and it is the metric vendors least volunteer.
Fourth, time to recovery. Two operations with identical recovery rates can differ enormously in cash terms if one recovers in three days and the other in twenty-five; subscribers also behave differently across a long, anxious window than a short one. Fifth, attempt yield by attempt number: which retry in your sequence actually recovers, and which attempts recover so little that they mostly generate bank noise and support tickets. This metric is how schedules get shortened honestly — by evidence, not by fashion. Sixth, the age of the open queue: the share of cases still open past the point where recovery is plausible. A queue can post a fine rate while its tail quietly becomes a write-off backlog nobody owns.
Seventh, the number that disciplines all the others: recovered revenue per case opened, trended monthly. It folds rate, speed, and case value into one figure a finance reader can hold against the tool's fee in a single line. The definitions, formulas, and worked examples for the full set are laid out in seven metrics that show whether dunning is actually working, including the denominator choices that make each one honest.
A word about cadence, because metrics without a reading rhythm are just decoration. The failure rate and the queue's age want a weekly look — they move fast enough that a month of drift costs real subscribers. Attempt yield and time to recovery settle more slowly and read honestly monthly, alongside the fee comparison. The full set wants a quarterly definition audit: not re-deciding what the words mean, but checking that the dashboards still implement the definitions on paper. Definitions rarely change by decision; they change by a thousand small implementation choices, and the audit is how you notice before the trend lines quietly stop meaning anything.
Read together, these seven catch the standard illusions: a recovery rate propped up by platform retries, a schedule padded with useless late attempts, a queue whose losses are hidden in its age. Read separately, each can be gamed, and eventually someone — a vendor, a dashboard, a hopeful quarter — will game it. Fix the definitions, publish them inside your own team, and the monthly review becomes a short meeting about evidence instead of a long one about whose number to believe.
Swipe Again publishes its counting rules for exactly this reason: total recovered is the headline, platform recoveries are separated from the tool's own work, and a cancellation mid-case counts as a loss — so the seven numbers above reconcile against the processor instead of floating free of it.
Top comments (0)