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 the published bands we run every diagnostic against — typical, indicative teaching bands, not advice; your own sales history gives the exact figures, and formats differ by design. The worked examples use illustrative round numbers chosen so the arithmetic is easy to check. 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.

Why is headcount 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.

What is covers per labour hour, and where do I get my number?

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.

How do I build the roster from the forecast instead of 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 published 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 week: forecast roster versus habit roster

Take an illustrative 120-seat casual-dining room, with deliberately simple numbers — the arithmetic is the point, not the figures. Average spend AED 75; the restaurant’s own history shows 1.6 covers per total labour hour on well-run shifts; the fully-loaded blended cost is AED 30 per hour — wages plus visas, insurance and housing spread across working hours.

With a constant productivity figure, the forecast-built roster costs the same share every day: 30 ÷ (1.6 × 75) = 30 ÷ 120 = 25.0% of sales, at the healthy end of the band. The habit roster deploys 130 hours every day — the “average day” photocopied seven times. Here is the week side by side:

Day	Forecast covers	Revenue (AED)	Hours needed (÷ 1.6)	Habit hours	Habit labour cost (AED)	Habit labour %
Mon	90	6,750	56.3	130	3,900	57.8%
Tue	90	6,750	56.3	130	3,900	57.8%
Wed	120	9,000	75.0	130	3,900	43.3%
Thu	160	12,000	100.0	130	3,900	32.5%
Fri	240	18,000	150.0	130	3,900	21.7%
Sat	260	19,500	162.5	130	3,900	20.0%
Sun	200	15,000	125.0	130	3,900	26.0%
Week	1,160	87,000	725	910	27,300	31.4%

Check the totals. Forecast roster: 1,160 covers ÷ 1.6 = 725 hours × AED 30 = AED 21,750, which is 21,750 ÷ 87,000 = 25.0%. Habit roster: 910 hours × 30 = AED 27,300, which is 27,300 ÷ 87,000 = 31.4% — through the 30% ceiling. The difference is AED 5,550 a week, from the same team on the same wages serving the same menu.

Now look at the shape, because the habit roster fails twice at once. Monday and Tuesday run at nearly 58% labour — 130 hours deployed against 56 needed, roughly AED 2,200 of avoidable cost each day. But Friday and Saturday are understaffed: 130 hours at 1.6 covers per hour can serve about 208 covers, against forecasts of 240 and 260. The habit roster’s flattering 20% weekend labour figure is not efficiency; it is a capped room — around 32 unserved covers on Friday alone, AED 2,400 of revenue at AED 75, turned away or served badly. Over-rostered on the quiet days, under-rostered on the busy ones, and the weekly total reads “31.4%” as if the problem were simply too many people. It is not. It is hours in the wrong places.

FOH and BOH: does the ratio follow 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:

Format	Typical shape	Why	What to hold accountable
Casual dining	Roughly balanced kitchen and floor	A full brigade behind a full service team	Total hours vs covers and revenue
Fine dining	More staff per cover, depth on both sides	Hosts, runners, senior floor roles; prep and pastry depth	Sales per labour hour, not covers alone
Counter-service and QSR	Kitchen-heavy	The guest does part of the service work	Throughput per station hour
Cloud kitchen	Almost entirely back-of-house	Front of house is a dispatch bench; no seats	Orders per station hour and packing capacity

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. A cloud kitchen’s staffing scales with order throughput and station capacity — not seats, because there are none — which is why its labour reads inside a different operating model altogether.

What does an hour of labour actually cost in the GCC?

The AED 30 in the worked week is a loaded rate, and the loading is where most labour budgets go wrong. The basic wage is the first line only. On top of it sit the visa and medical, mandatory health insurance, recruitment, often accommodation and transport, training before the first shift — and the end-of-service gratuity that accrues quietly across every contract. Under the UAE labour law, a full-time employee who completes a year of continuous service is entitled to gratuity at 21 days of basic wage for each of the first five years, and 30 days for each year after that. Illustratively: a basic wage of AED 3,000 a month is AED 100 a day on a 30-day month, so the entitlement accrues at roughly 21 × 100 = AED 2,100 a year — about 5.8% of basic — whether or not anyone books it monthly.

Price every rostered hour at the loaded rate, not the basic wage. A roster that reads 25% on basic wages and 31% fully loaded is the second roster, not the first — and it is the second one the bank sees. The cost-per-head read walks the full stack; the point here is that the covers-per-labour-hour arithmetic only produces a true labour percentage when the hourly rate in it is true.

Why is overtime a structural message, not a scheduling quirk?

Persistent overtime is the roster telling you something: 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, and it hides in the payroll run rather than on the roster, so nobody sees it until month-end.

The law sets the price. Under Federal Decree-Law No. 33 of 2021, overtime is paid at the normal rate plus at least 25% of the basic wage, rising to at least 50% for hours worked between 10pm and 4am (shift workers excepted), with additional hours capped at two a day and total working hours at 144 in any three weeks. Illustratively, on a basic wage of AED 20 an hour, 40 overtime hours a week cost 40 × 20 × 1.25 = AED 1,000 against AED 800 at the straight rate — a premium of AED 200 a week; the same 40 hours in the late-night band cost 40 × 20 × 1.5 = AED 1,200. On the payslip the premium looks manageable. It is not the real cost. The real cost is that premium-priced hours are worked by tired people delivering declining quality, that the two-hour daily cap means chronic overtime is quietly running through it, and that the overtime is almost always disguising an unfilled vacancy.

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 and a recruitment calendar that starts before the gap opens, not a permanently stretched team.

When does too few staff cost 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 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 worked week above shows the same thing at scale: the habit roster’s cheap-looking Saturday was 32 covers the room could not serve. The comparison that matters is always the loaded cost of the shift against the contribution of the covers it enables.

How do I read labour weekly, inside prime cost?

Labour is one half of prime cost, and prime cost — food and labour together, banded at 55–62% of revenue with the mid-60s as the hard ceiling — is the number that decides whether the model survives. Labour read alone flatters or panics depending on the week; a scratch kitchen legitimately runs labour high and food low, a prep-light concept the reverse, and only the sum tells you whether the trade is working. So the weekly labour read is short and non-negotiable:

Weekly gauge	What it tells you	Against
Labour as % of sales (fully loaded)	Whether the roster matched the trade	Your target inside the 25–30% band
Covers per labour hour	Whether productivity is holding	Your own eight-week trend
Overtime hours as % of total hours	Whether the base roster is too thin	A chronic floor vs event spikes
Rostered hours vs actual hours	Whether the plan survived the week	The forecast build
Prime cost (food + labour)	Whether the model survives	The 55–62% band

The red-line beyond which our rescue check reads labour as a structural emergency rather than a drift is 35% — deliberately above the 30% ceiling, because a business is urgent when it is past the line, not merely at it. Between the two is where a weekly reader fixes the roster for the price of thirty minutes, and a monthly reader discovers it four weeks late.

If you want the schedule-versus-covers picture in two minutes, the Labour Productivity read is built for exactly that; the Profit Leak Audit shows how labour sits beside food, rent and delivery commission as one of the four lines that decide margin; 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.