The number that moves first.

A restaurant losing regulars while still attracting first-timers shows flat sales. Underneath, the customer base is being replaced by people who cost more to acquire and spend less. Revenue looks stable right up until the new-customer flow slows, at which point the problem has been running for months.

Take all customers who ordered in the period. Count how many had ordered before. Divide. Do it monthly and plot it. The absolute level matters less than the direction, because the level depends heavily on format.

Almost entirely the experience rather than the marketing. Food consistency, timing accuracy, order correctness. When repeat rate falls, the cause is nearly always something operational that shows up in reviews at the same time.

More useful than a single rate: take everybody who first ordered in a given month and track what share ordered again within sixty days. Comparing cohorts month by month shows whether something changed and roughly when, which a blended rate cannot.

You can only measure it where you can identify customers. Marketplace orders generally do not carry identity, so a restaurant with most volume there is blind to its own retention. That is a strong practical argument for a direct channel beyond the commission saving.