Overbooking

A room that seats 90 might take 96 on a Saturday because experience says six will not arrive. Whether that is prudent or reckless comes down to one thing: whether the six is measured or assumed. Overbooking is the clearest case in restaurant operations where a guest data problem shows up as a service failure at the door.

How much can I safely overbook?

There is no industry figure, and any number quoted as one should be ignored. The safe cushion is a restaurant’s own no-show and cancellation behavior for that service, that day, and that time slot, measured over enough services to mean something. A Saturday 8pm slot and a Tuesday 6pm behave nothing alike, so a single house-wide percentage will be too aggressive on one and too timid on the other.

The costs on either side of the bet are not symmetrical, and that asymmetry should set the size of it. An empty table loses one seating of revenue. A walked party loses the seating, the guest, and years of their future spend. So the sensible place to pad is the slot where an error becomes a wait rather than a walk, meaning early service, a room with a bar, or a night the waitlist is deep enough to absorb it.

Confirmations and deposits complicate the arithmetic usefully. Both lower no-shows and therefore shrink the cushion available, because a book that reliably shows up is a book you cannot pad.

Can I trust my no-show rate enough to overbook on it?

Usually less than operators assume, and the reasons are in how the number gets produced. Reservation platforms disagree on the underlying signal: Resy, OpenTable, and SevenRooms expose a guest confirmation, while Tock does not deliver one, so an unconfirmed booking on Tock means nobody asked rather than the guest declining to answer. Overbooking against an unconfirmed count on one platform and repeating the same rule on another is comparing a measurement to a blank.

Marking discipline is the second variable, since a no-show only exists once somebody records it, and rooms differ enormously in how consistently that happens. Party-size shrinkage is the third, and it is invisible in no-show reporting: a six that arrives as four consumed a six-sized cushion without registering anywhere as a miss.

Who gets walked when the bet goes wrong?

Whoever happens to be at the host stand when the room runs out, and that is the wrong basis for the decision. A first-time booker and a regular of five years are not equivalent losses, and neither is a guest who has already been walked once. The information needed to tell them apart is a guest record attached to the booking, available before the room fills rather than during the argument.

The same record matters afterward. A walked guest who gets a call, a note on their profile, and a held table next time is recoverable. One who becomes an unlabeled bad night is not.