How Many Staff Does a Restaurant Need? The Staffing Math

Ask ten operators how many staff a restaurant needs and you will get ten headcounts — and every one of them is answering the wrong question. A restaurant does not need a number of people; it needs a number of labour hours, placed where the covers actually land, and that number moves every week the forecast moves. Staffed by habit, labour quietly becomes the heaviest controllable line on the P&L. Staffed by arithmetic, it becomes the lever you control weekly.

The ranges below are typical, indicative teaching bands, not advice — your own sales history gives the exact figures, and formats differ by design. But the method is universal, and it is the same read we run at the start of every turnaround: forecast first, hours second, people third.

Headcount is the wrong unit

Contracts, visas and the org chart are written in people. Service is delivered in hours. The same 24-strong team can produce a lean, well-covered week or a bloated one, depending entirely on how its hours are deployed against demand — which is why two identical restaurants with identical headcounts can sit five points of labour apart.

The distinction matters more in this region, not less. In our experience across GCC operations, teams are sponsored, salaried and structurally fixed: you cannot flex headcount week to week the way a casual-labour market can. The roster is the only flex you have. That makes roster discipline the whole game — the payroll is largely committed the day the visas are stamped, and deployment is what decides whether that committed cost turns into served covers or into idle hours.

Covers per labour hour: the working gauge

The unit that converts a forecast into a roster is covers per labour hour: total covers served, divided by total labour hours deployed — all of them, front and back of house, prep to close, salaried and hourly alike. Its money twin is sales per labour hour, which weights the read by what those covers spent.

Do not import a target from someone else’s concept. A scratch-kitchen bistro and a delivery-first grill will run legitimately different figures, because the work behind each cover differs. Instead, mine your own history: take the last eight weeks, pick the shifts that ran well — service held, no meltdown, no ghost town — and calculate what covers per labour hour those shifts actually achieved. That figure, from your own best evidence, is your planning number. Then watch the trend: a gauge that drifts down over a quarter is telling you the roster is thickening faster than the trade.

Build the roster from the forecast, not the habit

Most rosters are last week photocopied. The habit roster feels safe because it is familiar, but it encodes every past mistake permanently — the Tuesday that was once busy, the double-cover Friday that once went wrong — and it never reads the forecast at all.

The forecast-built roster takes about thirty minutes a week with the numbers open:

• Forecast covers by daypart from your own history, adjusted for bookings, season and known events — not from optimism.
• Convert covers to hours using your covers-per-labour-hour figure.
• Place the hours on the demand curve: staggered starts, split coverage across the peak, short task-shifts for prep and close — rather than everyone in at ten and out at ten.
• Reconcile to money: price the built roster at the fully-loaded hourly cost and check it as a share of forecast sales, against the typical 25–30% band.

The habit roster asks “who worked last Friday?” The forecast roster asks “what does this Friday need?” They produce different weeks.

A worked staffing build

Take an illustrative 120-seat casual-dining room, with deliberately simple numbers — the arithmetic is the point, not the figures.

Friday forecast: 240 covers at an average spend of AED 75, so forecast revenue = 240 × 75 = AED 18,000. The restaurant’s own history shows 1.6 covers per total labour hour on well-run Fridays. Hours needed = 240 ÷ 1.6 = 150 labour hours. At a fully-loaded blended cost of AED 30 per hour — wages plus visas, insurance and housing spread across working hours — the day’s labour = 150 × 30 = AED 4,500, which is 4,500 ÷ 18,000 = 25.0% of sales. Healthy end of the typical band.

Now the two failure modes, same restaurant:

• The drift. Twenty unplanned hours creep in — an extra body “just in case”, an early start nobody questioned. 170 × 30 = AED 5,100, and labour is now 5,100 ÷ 18,000 = 28.3%. Still inside the band, but the drift has eaten the margin of safety in one week.
• The habit Tuesday. Forecast is 90 covers — revenue 90 × 75 = AED 6,750, hours needed 90 ÷ 1.6 ≈ 56. But the habit roster deploys a Friday-shaped 120 hours: 120 × 30 = AED 3,600, and the day runs at 3,600 ÷ 6,750 = 53.3% labour. Same team, same wages, same menu. The only difference is deployment.

One quiet Tuesday like that every week is roughly AED 1,900 of avoidable labour — week after week, invisible inside a monthly total.

FOH and BOH: the ratio follows the format

Operators often ask for the “right” front-of-house to back-of-house split. There is not one — there are typical shapes, and the shape follows the service model:

• Casual dining typically runs roughly balanced between kitchen and floor: a full brigade behind a full service team.
• Fine dining carries more staff per cover overall, and the extra depth lands on both sides — hosts, runners and senior floor roles in front, prep and pastry depth behind.
• Counter-service and QSR formats tilt kitchen-heavy: the guest does part of the service work.
• A cloud kitchen is almost entirely back-of-house. Its front of house is a dispatch bench, and its staffing scales with order throughput and station capacity — not seats, because there are none.

The practical rule: never copy a ratio across formats, and never judge your own by someone else’s. Derive the split from how your service actually works, then hold the total accountable to covers and revenue.

The overtime trap

Overtime is rarely a scheduling quirk. Persistent overtime is a structural message: the base roster is too thin for the real demand curve, or the forecast is fiction. Either way it is the most expensive labour you buy — premium-priced hours, worked by tired people, delivering declining quality — and it hides in the payroll run rather than on the roster, so nobody sees it until month-end.

Read overtime as its own weekly line: hours of overtime as a share of total hours. Occasional spikes around genuine events are the tool working as designed. A chronic floor of overtime, week after week, is an understaffing signal — and in the GCC, where hiring runs through visas and quotas and takes weeks, that signal has to be acted on early. Structural overtime is usually the true cost of an unfilled vacancy; the honest fix is recruitment and training done properly, not a permanently stretched team.

When too few staff costs more than too many

Overstaffing wastes money visibly. Understaffing wastes it invisibly, and usually wastes more. A shift below its real requirement caps the revenue the room can produce: tables turn slower, ticket times stretch, the team stops selling and starts surviving, guests feel it, and the reviews say so. Then the compounding starts — an overworked team churns, and every departure buys you recruitment fees, visa costs and unproductive training weeks.

The arithmetic is rarely close. Cut one eight-hour runner and you save 8 × 30 = AED 240. If service slows and the room loses just 20 covers at AED 75, that is AED 1,500 of revenue — about AED 1,050 of contribution even after a typical 30% food cost (1,500 × 0.70 = 1,050). Saving 240 to lose roughly 1,050 is not a saving; it is a leak with good intentions. The comparison that matters is always the loaded cost of the shift against the contribution of the covers it enables.

Read it weekly, inside prime cost

Labour is one half of prime cost, and prime cost — food and labour together against the typical 60–65% ceiling — is the number that decides whether the model survives. So the weekly labour read is short and non-negotiable: labour as a share of sales against your target inside the 25–30 band, covers per labour hour against your own trend, overtime share, and the prime-cost sum with food. If you want the schedule-versus-covers picture in two minutes, the Labour Productivity read is built for exactly that; and if you are still sizing the model, break-even tells you the covers per day the whole structure needs before staffing is even the question.

The GGB read

We do not begin with headcount, and we never begin with the org chart. We size labour from the demand curve — forecast covers, converted to hours through the operation’s own productivity evidence, priced fully loaded — and we judge the result only inside prime cost, because labour read alone flatters or panics depending on the week. A roster is a weekly control, not an annual decision; the operators who hold labour do the thirty-minute forecast build every week, watch the trend gauges, and treat chronic overtime as a hiring signal rather than a habit. The question is never “how many staff?” It is “how many hours, where, and what did each hour produce?” — asked every single week.