Churn

Churn arrives in hospitality from subscription businesses, where a customer cancels and the date is recorded. Restaurants get no notice. A guest who came every three weeks for two years simply stops, and nothing in a point of sale or reservation book registers the absence. The metric only becomes available to a restaurant that holds identified guest records over time, so it usually appears on the day a group adopts a guest data platform.

How do you measure churn without a cancellation?

By cohort and by window. Take the guests who visited in a period, look forward a fixed length of time, and count how many never came back. Two hundred first-quarter guests with sixty absent since is a 30% churn rate for that cohort, and it only means something next to the cohorts before and after it.

Cohorts also protect against the trap in a single overall figure. A restaurant adding guests quickly can hold a stable churn rate while retention among new guests gets worse, because the growing denominator hides it. Following each quarter’s intake separately shows whether the guests won now stay longer than the ones won a year ago.

How long should a guest be gone before they count as churned?

Long enough to be abnormal for your room, so the window has to come from your own data rather than a benchmark. A neighbourhood restaurant whose regulars visit monthly can treat ninety days of silence as a problem. A destination tasting-menu room where guests come twice a year would flag its best guests as lost on that rule, and a special-occasion restaurant on an anniversary cycle needs a window measured in years.

The defensible method is to use the interval a venue’s own guests actually keep. Look at the typical gap between visits for guests with a history, then set the threshold at a multiple of it, commonly two or three times. Do it per venue rather than per group, because a group running a bar, a neighbourhood dining room, and a fine-dining room has three different rhythms and one shared threshold would be wrong for all three.

Why does churn look worse than it is?

For two reasons that are artefacts of the data rather than guest behaviour. The first is unidentified volume. If a venue attaches a name to only 60% of its covers, a guest who has been in twice since their last matched visit reads as absent, and churn measures a failure to recognise people rather than the loss of them. Reported churn falls when matching improves, without any guest changing behaviour.

The second only shows up in groups. A guest who stops coming to one restaurant and starts eating at a sister venue looks churned inside the first venue’s numbers and is worth more than before at group level. Without identity resolved across venues, that guest counts as a loss and lands in a win-back campaign for a restaurant they traded for another one you own.