Every host makes a version of the same bet constantly: will this table actually be free by 7:30, or will that table's current party linger past the estimate? Guess wrong in one direction and a guest waits unnecessarily; guess wrong in the other and a table sits empty that could have turned another booking.
Fixed assumptions versus real prediction
A common approach is a flat assumption ("90 minutes per table") applied regardless of party size, day, or what's actually been happening that night. Real turn times vary meaningfully by all three.
What live prediction is built from
Actual historical seat-to-close times for comparable party sizes and days, continuously updated against what's happening in real time that specific service, not a static rule applied uniformly.
What this changes at the host stand
A host can see which tables are predicted to turn next, not just which are currently occupied, enabling confident booking decisions instead of conservative guesswork that leaves covers on the table.
This connects directly to the Friday-night scenario in our "One Operating System" article: table management is one of the four pillars working together in that walkthrough, not a standalone feature.