Churn gets treated as a retention problem, which is why so much money goes into win-back emails and discount ladders that move the number for one quarter and then let it settle back exactly where it was. Churn is a diagnosis problem. Somebody stopped getting the thing they came for, and the work is finding out where that happened before deciding what to build.
Which kind of churn do I have?
There are four, and they need completely different work. Customers who never reached the point where the product paid off. Customers who activated and then hit a wall as they grew. Customers whose situation changed and who were never going to stay. And customers who were the wrong fit at the sale, which is a marketing problem wearing a product problem's clothes.
That last point is worth sitting with, because it invalidates the most common piece of churn research anyone does. The cancellation form asks why someone is leaving. The answers cluster on "too expensive" and "not using it enough". Both are descriptions of the outcome. Neither tells you what happened three weeks earlier, which is where the fix lives.
| Kind | What it looks like in the data | What fixes it |
|---|---|---|
| Never activated | Concentrated in the first 30 days. Low or no use of the one action that defines the product. | Onboarding and first-run experience. Almost never price. |
| Activated, then stalled | Healthy first month, then a decline over two or three cycles. | A missing second use case, or a workflow that breaks at real volume. |
| Situational | Scattered evenly across tenure. Often tied to a role change or a budget cycle. | Very little. Some churn is not a defect and chasing it wastes the budget. |
| Bad fit at the sale | Clusters by acquisition channel or by segment, not by tenure. | Qualification and positioning, upstream of the product entirely. |
If you cannot tell which row your churn sits in, that is the finding. Every remedy below is wrong for at least two of these rows, so guessing costs a quarter.
How do I find the pain point that is causing churn?
Watch people use the product against a real task, then compare what you saw against where the data says they drop. Analytics tells you which step people leave at. It never tells you why, and the why is the only part you can act on.
These five, in this order, because each one narrows what the next has to look at.
- Cohort the churn before reading anything Split by signup month, by plan, by acquisition channel and by tenure. Churn that is flat across all four is situational. Churn that spikes in one is a defect with an address.
- Find the action that predicts staying There is usually one: the invite sent, the second project created, the integration connected. Compare the people who did it in week one against the people who did not. The gap between those two curves is the size of your onboarding problem.
- Watch somebody attempt the job for the first time Moderated sessions with people matched to your buyer, running the real task rather than a demo script. Five to eight is enough to find the blocking issues. You are watching for hesitation and for the moment somebody guesses.
- Interview the ones who stayed, not only the ones who left Churned customers explain their exit. Retained customers explain what nearly made them leave and what stopped it, which is the fix already working.
- Test the change before shipping it to everyone A fix that reads well in a meeting can make the number worse. Put the new flow in front of the same kind of person, watch it again, then roll it out.
- Cause fixed
- Discount offered instead
Illustrative, not a client result. Both cohorts look identical until the work lands in month three, which is the point: a retention curve only tells you anything once you can see the slope on both sides of an intervention. A discount flattens the curve for one cycle and then it resumes.
Growth is a process. Nothing here happens overnight, and anybody promising you overnight results is lying to you.
Find the cause from $100What is the difference between a churn reason and a churn cause?
A reason is what a customer tells you at the moment they leave. A cause is the event, usually weeks earlier, that made leaving reasonable. Reasons are easy to collect and almost useless. Causes are harder to find and are the only thing worth fixing.
A worked example. The exit survey says "too expensive". The cohort data says these accounts churned at month four, not month one, so they activated fine. The session recordings say the workflow that was pleasant with twenty items is unusable with four hundred. Nobody wrote "your bulk editing is broken" on a cancellation form. That is the cause, and the price complaint was the symptom of a product that stopped being worth its price at the size their business had grown to.
Which fixes actually keep customers?
The ones aimed at the specific step where value stopped arriving. Onboarding changes for never-activated churn, workflow and capacity work for stalled churn, better qualification for bad-fit churn. Discounts fix none of the three and buy about one cycle of delay on all of them.
There is an order to it that is worth respecting. Fix acquisition fit first, because every bad-fit customer you add is churn you have already scheduled. Then onboarding, because it compounds: every future cohort benefits. Then the stall points, which are the most expensive to find and the most durable once fixed. Retention campaigns come last, when there is something worth being retained for.
Do win-back emails work?
They recover a slice of the people who left for situational reasons and were going to come back anyway, which makes the campaign look better than it is. They recover almost nobody who left because the product did not work for them, and re-engaging that group without a fix in place usually produces a second, faster churn.
Lifecycle email is genuinely useful in the other direction: reaching somebody in the first two weeks, at the step where the data says they stall, with the one thing that gets them past it. That is not a win-back, it is activation, and it is where the return is.
How much churn is normal?
It depends so heavily on price point, contract length and market that a benchmark from somebody else's business is not useful to you. Your own trend line is. Churn rising against your own past three cohorts is a signal; churn sitting above an industry average you read somewhere may be nothing at all.
What matters more than the rate is whether it is concentrated. Churn spread evenly across every cohort and segment is the cost of doing business. Churn clustered in one plan, one channel or one month of tenure is a defect with a location, and defects with locations can be fixed.