Ask five subscription teams to define churn and you will get one formula and five arguments. The arithmetic fits on an index card; the arguments are about who counts. Do failed payments count as churn when the subscriber never chose to leave? Does a paused plan count as retained? Is a customer who cancelled but still has twenty paid days remaining in the denominator this month or next? Until those questions have written answers, every dashboard number is a negotiation.
Pin the population before the percentage
Each core metric hides a population decision. Churn rate needs a rule for involuntary exits. Lifetime value needs a rule for recovered subscribers — does a customer saved by dunning re-enter with their original tenure or start a new clock? Failed-payment rate needs a decision about first-attempt versus eventual failures, because the gap between those two numbers is precisely the value of your recovery work. Teams that skip these decisions end up comparing months that were never measured the same way, then debating trends that are really definition drift.
A practical way through: keep a single internal sheet where every metric carries its formula, its population rule, and the date the rule last changed. When a rule changes, annotate the chart. It sounds bureaucratic until the first time it ends a meeting in five minutes instead of fifty.
A worked example, with the trap visible
Suppose 1,000 subscribers start the month, 40 cancel deliberately, 25 fail to pay and never recover, and 15 fail and then recover inside the grace window. A naive churn figure reads 6.5 percent. A voluntary-only figure reads 4 percent. The 2.5-point difference is not a rounding preference — it is the entire involuntary-churn problem, and reporting only the blended number hides the one part of churn your billing operation can actually fix. Report both, always, side by side.
For a worked example of that discipline applied to one number end to end, how we count recovered revenue states the population, the exclusions, and the edge cases in public — the same treatment every metric on your internal sheet deserves, so its caveats get decided on day one rather than discovered in a board meeting.
Keep a change log, or lose the trend line
Definitions drift quietly. Someone adds trials to the subscriber count, someone else starts excluding annual plans, a new dashboard tool rounds differently, and six months later the team is celebrating an improvement that is entirely definitional. The defence is boring and total: every metric definition lives in one document, every change is dated, and every chart that spans a change carries a marker on the date it happened. When the definition and the formula travel together, a new hire can reproduce any number the business has ever reported — and reproducibility, not precision, is what makes a metric trustworthy enough to act on.
Metrics should end in an action
The test of any metric is whether a change in it tells someone what to do next. Involuntary churn rising means look at decline mix and retry timing. Recovery rate falling means look at message deliverability and the card-update path. If a number can move without anyone knowing which lever it maps to, it is decoration. Choose fewer metrics, define them ruthlessly, and let each one own a lever.
That discipline is why Swipe Again reports recovery against the whole failed pool rather than a flattering subset — the denominator is the metric. However you instrument your store, write the definitions down where the formulas live. The formula was never the hard part; agreeing on what it counts is the whole job.
Top comments (0)