Food & beverage · 27 June 2026 · 2 min read
Two extra parties an hour shouldn't break a dining room
Most managers feel the cliff. Few can show it. So let me solve a real room, stage by stage, with the exact formula every call centre on earth uses to size its staff — Erlang C.
THE ROOM
10 tables. A one-hour average table, so each clears μ (mu) = 1 party per hour.
Capacity = tables × μ = 10 × 1 = 10 parties per hour. The ceiling. Fixed.
STAGE 1 — THE LOAD
Arrivals = λ (lambda) parties per hour.
Offered load a = λ ÷ μ, in Erlangs — the number of tables' worth of demand walking in.
Utilisation ρ (rho) = a ÷ c, where c = tables. How full the room runs.
STAGE 2 — THE WAIT FORMULA (Erlang C)
First, the chance a party finds every table full and must queue:
top = a^c ÷ ( c! × (1 − ρ) )
bottom = [ sum of a^n ÷ n! for n = 0 to c−1 ] + top
P(wait) = top ÷ bottom
Then the average wait once queued:
Wq = P(wait) ÷ ( c×μ − λ )
Two lines. Exact. No rule of thumb.
STAGE 3 — RUN IT
A normal night, λ = 6:
a = 6, ρ = 60%. P(wait) = 10%. Wq = 1.5 minutes. The room breathes.
Busier, λ = 7:
ρ = 70%. P(wait) = 22%. Wq = 4.4 minutes — already up threefold.
A good Friday — two more parties — λ = 8:
a = 8, ρ = 80%. P(wait) = 41%. Wq = 12.3 minutes.
Two extra parties an hour. The wait did not rise by a third. It rose EIGHTFOLD.
One more, λ = 9:
ρ = 90%. P(wait) = 67%. Wq = 40 minutes.
Nudge to 95% and it is 99 minutes — a guest waits longer than they will sit.
THE LAW
As ρ nears 100%, the wait runs to infinity. Not a slope — a cliff. The last sliver of capacity costs more than all of it before. This is why the trade caps near 85%: past it your queue stops tracking how busy you are and starts tracking how unpredictable your door is.
WHAT THE EXCELSIOR GM READS IN THIS
— Plan to the peak hour, never the daily average. The average hides the cliff; the cliff is where the reviews get written.
— A 40-minute wait is not a queue. It is a walk-out, a one-star, a table that turned once instead of twice.
— Pool the room. One shared waitlist beats four siloed sections; pooling buys wait-time you never put on payroll.
— Attack variability, not only headcount: staggered seating, reservation pacing, pre-order. Each flattens λ and adds capacity off the wage bill.
— The quiet one: luxury holds 60–70% on purpose. That headroom is not waste — it is the time to care, and it is exactly what the higher price buys.
Capacity is not a number. It is a threshold. Respect it, and the same room that drowns at 95% breathes at 80%.
The Excel that runs all of this — your own λ, μ and c — is built. Comment or DM and I will send it across.
Non multa, sed multum.