Why do customers stop coming back — and how do you find out in time?

Ask an operator why customers leave and you get a list of things customers said. Ask the data and you get a different list, because the customers who said something are a small and unrepresentative minority. Roughly nine in ten people who stop coming never tell you why; they simply stop appearing, and the absence is invisible unless something is counting.

Cause one: habit disruption. Most F&B loyalty is really habit — a route, a break time, a colleague you go with. When the route changes, the customer does not decide to leave; the trigger that produced the visit stops firing. This is the largest single category and also the most recoverable, because there is no grievance to overcome. A well-timed message reinstates the trigger.

Cause two: the unreported bad visit. A slow order, a drink made wrong, a staff interaction that landed badly. The customer does not complain — complaining is socially expensive and they have alternatives — so they downgrade you quietly from default to occasional. You will see a lengthening gap rather than a review. Recovery here needs acknowledgement, not a discount: a message that reads like a person noticing works far better than a voucher that reads like a mailing list.

Cause three: convenience displacement. Something new opened closer, or a delivery app made a competitor one tap easier. This is the hardest to reverse with messaging alone, because the competitor changed the cost of the alternative rather than your quality. What works is giving the customer a reason that is specific to you — a saved order, a held table, a reward that is nearly complete.

What connects all three is that they are legible in the visit record before they are permanent. A customer who came every four days and is now at eleven days has told you something, without saying anything. The mistake is applying one threshold to everyone: a fixed 30-day rule treats the four-day commuter and the six-week brunch customer identically, which means you contact the first three weeks too late and the second three weeks too early.

The recoverable window is roughly two of the customer's own visit cycles. Inside it, a relevant message recovers a meaningful share — commonly 15–30% depending on category. Outside it, the same message converts at a fraction of that, because the customer has already built a new habit somewhere else. This is why cadence detection is the part worth automating: the value is not in sending more messages, it is in sending them before the substitution hardens.