In this article
A table of two who stay half an hour longer than planned at 9pm feels like bad luck, not a pattern. But that same table may already be booked for the third time tonight — and a risk that looks small on any single booking stacks up over a service into something that stopped being a coincidence long ago.
Every reservation system does the same thing: it gives a table a fixed slot — 90 minutes, 2 hours — and books the next party the moment that slot ends. That is a plan on paper. What actually happens at the table follows no schedule: it depends on how many courses the guests order, how busy the kitchen is at that moment, whether someone is celebrating a birthday, and a hundred other things no system knows when the booking is made.
As long as the planned slot and the real duration line up, nothing goes wrong. The moment a table stays occupied even ten minutes longer than planned, a gap opens: the next guests can't sit down on time. On a table that turns over once a night, that's an outlier. On a table that turns a second or third time that same night — and on a popular table that's the rule, not the exception — that risk doesn't add up, it compounds: being safe once is easy, being safe three times in a row on the same night is a completely different sum.
This article isn't about serving faster or rushing guests. It's about the minutes you already plan in advance. With three numbers you already know — how many times you turn a table in one night, how often a booking runs late, and by how much — you can work out how likely it is that some table will free up late tonight, and how much buffer per booking you need to bring that down to something you're comfortable with.
Three assumptions that keep the delay invisible
"One late table is just bad luck"
On its own, that's true. The problem is that "once" is rarely the real stake. A popular table for four by the window doesn't turn once on a busy Friday — it turns two or three times: at 6pm, 8:30pm and 10:30pm. A 25% risk that a booking runs late is close to a non-event on a single booking. Repeat that same chance three times in a row on the same table, and the odds it happens somewhere that night climb to nearly 6 in 10. That's no longer bad luck — it's a property of how you plan that table, and it can be worked out before it happens.
"My average table time protects me"
An average of 100 minutes for a table of four says nothing about the spread behind it. Most tables sit close to that average or under it — quick meal, quick to leave. A small number sit well above it: a conversation that runs on after dessert, a second bottle of wine, a birthday whose cake still has to arrive. That's exactly the kind of distribution where an average misleads you: it's pulled up by a long tail, while your planning trusts that average as if every table sticks to it. The tables that run late don't do it by five minutes — they do it by fifteen to thirty, and those are the minutes your buffer has to absorb, not the average itself.
"Adding buffer just costs me revenue"
A minute of buffer you plan in now feels like a minute you're not renting out. But a minute you DON'T plan for, and that gets lost anyway because the previous table overran, disappears just the same — you just never see it on paper. It hides in a guest waiting twenty minutes at the bar for a table that was supposed to be "free", in a last sitting cut down to a quick bite, or in the very last booking of the night that you're forced to cancel because the table simply doesn't free up in time. No buffer isn't free revenue — it's revenue that spreads your risk of losing it across the whole evening instead of pricing it in beforehand.
The ultimate guide Restaurant Reservations Guide: 6 Steps to a Full House Restaurant reservations in 6 steps: cut no-shows, refill cancellations, pace the rush and fill quiet nights — the complete playbook for 2026. Open the guideThe math: how often, how likely, and how bad
Three numbers you already know from your own bookings — no assumptions from outside — together determine how big your risk is tonight and how much buffer stops it.
1. The risk compounds, it doesn't add up
If a booking has a chance f of running late, and you turn that same table n times in one night, the chance that a delay happens at least once is not f — that's what you'd think if you only considered the first booking — but 1 − (1 − f)n. At 25% risk per booking and three turns on one table, that's 1 − 0.75³ = 57.8%. That's the same math as "what's the chance it rains at least once in three days if the daily chance of rain is 25%" — and it's exactly why a table that turns three times a night has a completely different risk profile than a table booked only once, even though the risk per booking is identical.
At a fixed 25% risk per booking that a table runs late, the chance something goes wrong somewhere that night climbs fast — not in a straight line.
chance of at least one delay
number of times the same table turns that night
At one turn, 25% risk is a marginal case. At four turns — a busy terrace on a summer evening — it's already 68%: more likely than not.
2. The expected drag per turn is a sum, not a guess
When a booking does run late, it typically does so by a certain number of minutes — call that m. The expected delay you carry into the next booking on that same table is then f × m: the chance it happens, times how bad it is when it does. At 25% chance and a typical overrun of 20 minutes, that's 5 minutes of expected delay per turn — small on its own, but it's exactly this figure that stacks up over every further turn on that same table that night.
3. The buffer you need follows from how much delay you'll still tolerate on the last turn
Set a limit you find acceptable for the last booking of the night on that table — say, never sit down more than 10 minutes later than planned. Spread that tolerance across the number of turns, and subtract what you're already building up in expected delay per turn (formula 2): buffer = f × m − (tolerance ÷ n), never below zero. That way you're not working with a vague "add a bit of extra time", but with the exact number of minutes needed to hit your own tolerance, given how often your tables run late and by how much.
A table booked at 6pm, 8:30pm and 10:30pm. Without a buffer, every following turn drifts along with the delay of the one before it — the third guests of the night pay for the first table's overrun.
Without a buffer, the third table of the night is already 35 minutes behind before the guests have even sat down — and that's the table you're hoping will order dessert and coffee.
Work out your own buffer
Enter your own numbers — no estimates from us, only yours.
Check yesterday's or last Friday's booking list for your most popular table.
A rough estimate is fine — the tool shows how sensitive the outcome is to it.
—
This isn't a guarantee — it's the same logic airlines use to overbook and clinics use to buffer their appointment schedules: build the margin around how often and how badly things go wrong, not around the average.
What to do with this this week, this month and this season
One number at a time, in the order that pays off without forcing you to rewrite your whole reservation schedule at once.
This week — measure instead of guessing
- For your three most popular tables, note how often each booking actually ran late over the past two weeks, and by how many minutes.
- Count how often that same table gets booked in one night on your busiest day.
This month — build the buffer in where the risk is highest
- Start with the table that turns the most times per night — that's where the compounded risk is biggest.
- Increase that table's time slot by the recommended buffer, even if it feels like revenue you're giving up.
This season — revisit per busy period
- A summer terrace turns tables more often than a winter dining room — recalculate the buffer per season, not once for the whole year.
The delay was already in your planning, not in your service
A reservation system that treats every booking as an isolated block of time hides the real risk: that block repeats several times a night on your most popular tables, and a small risk that repeats stops being a small risk. The buffer that follows from that isn't extra service effort — it's a few minutes you already plan in advance, on exactly the tables where it matters most.