Every growth review has the same slide. Churn, one number, an arrow, a color. And almost every team treats that number the way a doctor would treat body temperature if it were the only reading they ever took — as a dial to push in the good direction. Lower is better. Make it lower.
The trouble is that churn is not a temperature. It’s closer to a symptom that a dozen unrelated illnesses all happen to produce. Two companies with an identical five-percent monthly churn can be dying of completely different things, and the number, by itself, cannot tell them apart. Worse, it can sit perfectly still while the thing underneath it changes entirely — improve one kind of churn, quietly grow another, and the headline holds steady while your business quietly swaps out its problem for a different one.
The number that averages away its own cause
A single churn rate collapses together people who left for reasons that have nothing to do with each other and demand opposite responses.
Some never got started — they signed up, poked around, never reached the point where the product did anything for them, and drifted off within days. That’s not a churn problem; it’s an activation problem wearing a churn costume. Some got real value for months and then it decayed, or they outgrew you, or a competitor got better — that’s a sustained-value problem, a completely different repair. Some didn’t decide to leave at all: their card expired and a failed payment quietly ejected them. And some were never a fit, got pulled in by a campaign that oversold, and were always going to go — arguably the healthiest churn on the list, and the most expensive to chase.
Averaging those four into one percentage is like averaging the temperature of a feverish patient and a corpse and reporting that the room is comfortable.
Timing is the cheapest diagnostic you own
Before you send a single survey, the shape of your churn already names most of its causes. Churn that clusters near signup means the product never landed its first punch — a first-run problem. Churn that rises slowly across the whole lifetime means value is decaying faster than it compounds. A sharp spike at one specific tenure — right after the annual renewal, right after a price change, right after you removed a feature — is practically pointing at its own cause. (This is a cousin of reading the cohort retention curve, but the question is different: there you ask whether the curve flattens; here you ask what the bumps and clusters are telling you.) You don’t need more instrumentation to read this. You need to stop flattening the shape into a mean.
The build decision
Stop asking what’s our churn and start asking which churn.
Split voluntary from involuntary before anything else — a startling share of what teams call “churn” is failed payments, and the fix there is dunning emails and a card-updater integration, not a single line of product work. Then separate never-activated from once-activated, because a fix aimed at one does nothing for the other: a slicker onboarding flow will not rescue someone who got value for a year and then watched it fade. And tag the wrong-fit segment honestly. It is tempting to chase it, because retaining anyone lowers the number — but a churn rate bought by clinging to users who never got value is a worse business hiding behind a better metric.
None of this is a plea for more dashboards. It’s the recognition that “reduce churn” is not one instruction. It’s four or five instructions wearing a single coat, and until you take the coat off, you’ll keep prescribing cough syrup for the patient with the broken leg — because both of them walked in with the same number on the chart.
Liked this? Get the next one in Working Theory.
Going weekly in August (it's in beta now). One genuinely interesting read on building, the brain, and the science most people missed.
Subscribe →