In a subscription or membership business, the cancellation is the end of a process that started weeks earlier. By the time somebody clicks cancel, the decision was made a while ago.
The opportunity is in the weeks before.
What the signals look like
They vary by business, but the shape is consistent: a drop in engagement relative to that person's own normal:
- Attendance or usage that has fallen off from their established pattern.
- A gap longer than their typical gap.
- A failed payment that has not been resolved.
- A support issue that was never properly closed.
- Approaching a renewal date with no recent activity.
The important part is relative to their own pattern. Somebody who came twice a week and now comes once is a stronger signal than somebody who has always come once a month.
Why nobody catches it
Because it requires noticing an absence. Everything else in a business demands attention: the phone rings, the email arrives, the customer walks in. A member who has quietly stopped coming generates no event at all.
That is exactly the kind of thing a system is good at and a person is not.
What to do about it
Define the thresholds first, in plain terms. "Someone who normally attends weekly and has not been in for fourteen days" is a rule you can actually implement.
Then decide what happens. The intervention that works is almost always personal and low-pressure: a short message from a real person asking how they are getting on, not a discount offer. A discount tells somebody who was ambivalent that the thing is worth less than they were paying.
Save the harder cases for a person
An automated check-in handles the mild cases. Somebody with a genuine problem needs a conversation, and the system's job is to surface them early enough that the conversation is still worth having.
The failed-payment case
Worth separating out, because a meaningful share of churn in recurring businesses is not a decision at all: it is an expired card and a notification that went to spam. A proper dunning sequence with multiple channels and retries recovers a lot of it, and it is the easiest retention work available.
Measuring
Track churn rate, and track the recovery rate of members who were flagged. If a flagged member is no more likely to stay than an unflagged one, either the signal or the intervention is wrong.