How to Build a Restaurant Staff Rota With AI Demand Forecasts

Forecast covers for each hour of each service, then convert them into staff needed by role using your own ratios, such as one server per 14 covers in the room. Let AI or a rota tool draft shifts around availability, skills, rest rules and a labour-cost target, and have a manager check it before publishing at least a week ahead.

The forecast decides the shape of the day: how many people each hour needs. People still decide the names. Keep those two jobs separate and forecast-driven rotas cut wasted hours without making staff feel run by an algorithm. Mix them up and you get cheap rotas nobody wants to work.

Follow me on Instagram@sagnikteaches

From a daily forecast to an hourly curve

A daily covers forecast (how to build one is in forecasting covers and cutting food waste) tells you Saturday will do about 170 covers. A rota needs to know when they'll arrive.

Connect on LinkedInSagnik Bhattacharya

1. Export time-stamped covers or orders from your till or booking system for the last six to eight weeks.
2. Group them by weekday and hour. For each weekday, work out what share of the day's covers arrive in each hour. An assistant can write the spreadsheet formulas if you describe your columns.
3. Apply the shares to the forecast. If 21% of Saturday covers usually arrive between 7pm and 8pm, a 170-cover forecast means about 36 arrivals in that hour.
4. Allow for tables still eating. Staff aren't only busy with new arrivals; tables seated in the previous hour are still being served. A simple way to capture this is "workload covers" = arrivals this hour plus half of last hour's arrivals. Adjust the half if your meals are unusually long or short.

Check which timestamp your export uses before trusting the curve. A common slip (illustrative): a restaurant grouped its till export by the time each bill was closed, because that was the default column. The resulting curve put Saturday's peak at 8-9pm and the rota loaded staff into that hour, while the kitchen was actually slammed from 7pm and half-empty by 9:30. Bills close about an hour after people sit down, so the whole curve had slid an hour late. Regrouping by the time of the first order (or the booking time, if the till doesn't record it) moved the peak back to 7-8pm, where the chef had always said it was. If your curve disagrees with what the team feels, suspect the timestamp before the team.

Subscribe on YouTube@codingliquids

Writing down your staffing ratios

Every restaurant runs on unwritten ratios: the head chef knows the line can handle so many covers an hour with two cooks. Write them down, because a forecast is useless without them. These are illustrative; replace them with your own after asking your chef and floor manager, then test them against a night that felt well staffed.

Testing a ratio takes two nights from your own records. Say last Friday felt well run: at 8pm there were 44 workload covers on the floor and three servers, which is about 15 per server. A Thursday that felt stretched had 36 workload covers and two servers, 18 each. The ratio you want sits just under the comfortable night, so 14 is a sensible starting figure, and anything above about 16 is where your team starts to struggle. Do the same for the kitchen, using the hour the chef remembers tickets backing up.

Also note the fixed jobs that don't scale with covers: kitchen prep before service, opening set-up, closing and cleaning. They go on the rota regardless of the forecast.

Worked example: Saturday at a 60-seat restaurant

Say a 60-seat restaurant (an illustration) forecasts 170 covers on Saturday, 60 at lunch and 110 at dinner, closed between 3pm and 5pm. Applying the ratios above:

Shaped into shifts that include prep and close, that comes to 26 cook hours, 30 server hours, 14.5 porter hours and 13 bar hours: 83.5 hours in total. At illustrative rates of $19 an hour for cooks, $16 for servers and bar, and $15 for porters, that's about $1,400, or 21.7% of forecast sales at an average spend of $38 a cover ($6,460).

Compare the owner's old habit of putting the full evening team on from open to close: five servers, three cooks, two porters and two bartenders all night. Same lunch, but 109 hours and about $1,820, or 28.2% of sales. The forecast-shaped rota uses 25.5 fewer hours on one Saturday, mostly by starting the peak-only shifts at 6pm or 7pm instead of 4:30pm.

Two cautions. The saving only holds if the forecast is close; a 200-cover night on the lean rota is a bad night for everyone. And short peak shifts only work if staff want them, and if they meet any minimum-shift rules that apply to you.

Setting a labour target that fits your week

A single labour percentage for the whole week hides the real picture. Set one per day instead:

1. Pull last quarter's hourly wage cost and sales by weekday. Decide whether salaried staff such as the head chef are in or out, and stay consistent; most owners leave them out of the rota target because their cost doesn't change with the rota.
2. Work out each weekday's actual percentage. Saturdays will be lowest, because the minimum crew is spread over more covers. A quiet Monday lunch may be two or three times the Saturday figure.
3. Set targets a little below today's figures, not at an industry number you read somewhere. A target you can't meet without breaking your own ratios just produces rotas that fall apart during service.

For the 60-seat restaurant, a filled-in version might read like this (illustrative figures, hourly staff only):

Saturday is already the lowest percentage and still has the largest gap, because that's where the old full-team habit cost most; Monday's target stays high because the minimum crew, not the rota, sets its cost.

An assistant is useful for the analysis if you paste the weekly figures in and ask it to lay out percentages by weekday and service; check the totals in a spreadsheet before relying on them.

Quiet services: when the minimum crew is the problem

On a quiet Tuesday lunch the ratios don't bite, because the minimums do: two cooks, two servers, a porter and a bartender, whether 12 people come in or 30. Forecasting helps here in a different way. If a service is forecast below the level where the minimum crew pays for itself, the decision isn't about the rota at all; it's whether to open, shorten the hours or run a smaller menu that one cook can handle. Put the forecast for your three quietest services next to their minimum crew cost for a month, and the answer is usually obvious.

The sum for that Tuesday lunch, with illustrative figures: 18 covers forecast at an average of $26 is $468 of sales. The minimum crew for a four-hour service is two cooks ($152), two servers ($128), a porter ($60) and a bartender ($64): $404, or 86% of sales before a single ingredient is bought. A shorter menu one cook can run, with one server who also pours drinks and one porter, costs $200, or 43%. Still expensive, but now the service at least covers its food. Closing Tuesday lunch altogether is the third option, and the forecast gives you the numbers to make that call calmly rather than on the back of one empty day.

Getting AI to put names on the shape

Once you know how many of each role you need each hour, the fiddly part is matching real people to it. An assistant does this well if you give it every constraint in writing:

Draft next week's rota as a table (rows = staff, columns = days).

Staffing needed by hour and role: [paste the table above for each day]

Staff: [name - role(s) they can cover - contracted hours - max hours
        - days/times unavailable - preferences]

Rules (never break these):
- Minimum shift length: [e.g. 3 hours]
- Minimum rest between shifts: [your rule]
- No closing shift followed by an opening shift
- Split shifts: only for staff marked "happy to split", max [n] a week
- Weekend evenings shared out so nobody works more than [n] of 4
- [Any rules for young workers that apply to you]
- Every shift needs at least one person trained on [till / bar / allergens]

Aim for total paid hours close to [target]. Where you can't meet a rule,
say which rule and why instead of breaking it quietly.

The last line is the one that saves you: an assistant asked for a "cheap rota" will quietly break a rest rule to hit the number. Asking it to name conflicts turns hidden problems into decisions for you. Here's the end of a typical reply (sample output, illustrative):

Total paid hours: 86 (target 83.5). Conflicts: 1. Saturday 7-10pm needs two bar staff, but only Bartender A is bar-trained and available after 6pm. I have left the second bar slot empty rather than assign an untrained server. 2. Meeting the hours target would require Cook B to close Friday and open Saturday, which breaks the close-then-open rule, so Saturday prep starts with one cook until 11am. 3. Server D has 6 shifts this week; their maximum is 5, but I found no one else available on Wednesday lunch.

Conflicts 1 and 2 are exactly what you want to see: problems named, not hidden. Conflict 3 is the kind to check, because the assistant broke a rule anyway and told you afterwards. Look at the availability you gave it; in this case a part-timer had marked herself free on Wednesdays and the assistant had missed it, so the fix is a swap, not a sixth shift. A skills matrix showing who can cover which station makes the "at least one person trained" rule easy to fill in; building a staff training matrix with AI covers that.

Then a manager checks it: does it feel right for the people, not just the numbers? Swap where needed and publish.

Rota tools that forecast for you

If building the curve each week takes more than an hour, a scheduling tool that reads your till data may be worth it. At the time of writing (September 2026):

• Homebase lists a free Basic plan for up to 10 employees; Plus at $70 a location a month adds an AI-powered scheduling assistant; All-in-One at $120 builds schedules from forecasted sales and labour targets (monthly billing; annual saves 20%).
• Lineup.ai lists $149 a location a month for forecasts plus scheduling, with a schedule builder and automated scheduling on top of its sales and item forecasts.
• 7shifts, built for restaurants, offers sales forecasting and labour budgeting on its paid plans; check its pricing page for which tier includes automatic scheduling.

Whichever you trial, test it on a week you've already rostered by hand and compare. AI shift scheduling against a rota spreadsheet goes through that comparison for small teams generally, and pubs and bars with events-driven peaks will find how pubs use AI for events, rotas and stock closer to their pattern.

Where forecast-driven rotas upset staff

• Hours that move every week. People have childcare and second jobs. Keep a stable core of regular shifts and flex only the edges.
• Sending people home early. If the forecast was high and managers routinely cut shifts mid-service, staff stop trusting the rota. Some places also have rules requiring notice or pay for late changes; check what applies to you.
• The same people always on Saturday nights. Tools optimise for cost and skills, which often means your best people get every peak. Write fairness rules in, as in the prompt above.
• Short shifts nobody asked for. A three-hour peak shift suits a student; it doesn't suit someone with a 45-minute commute.
• "The system did it." Managers blaming the tool for an unpopular rota is the fastest way to lose staff goodwill. The manager publishes it, so the manager owns it.

Checking the rota against the night

Every week, compare the plan with what happened:

1. Forecast against actual covers, by hour. Where the forecast missed, note why (a party booked late, rain, an event nobody flagged).
2. Labour cost as a share of sales, against your target.
3. Pressure points. Ask the chef and floor manager to jot down any time tickets backed up or sections were overwhelmed. Three notes at the same hour mean the ratio for that hour is too lean. A useful note is one line with a time and a cause: "Sat 7:40-8:10, tickets backed up, 45 arrivals against 36 forecast, walk-in party of nine." That one line tells you whether to fix the forecast (walk-ins weren't allowed for) or the ratio.
4. Overtime and late changes, by person. Rising numbers mean the rota isn't matching reality.

Adjust the ratios, not just the next rota. After six to eight weeks the ratios settle, and the weekly job shrinks to a 20-minute check before publishing. If the numbers keep showing you're short even on lean weeks, deciding between AI and a new hire helps you work out whether the answer is another person rather than a better tool.

Rota questions managers ask

Should I tell staff the rota is built from an AI forecast?

Yes. Explain that the forecast sets how many people each hour needs and that a manager still decides who works when. Staff accept a busier or quieter week more easily when they can see the reason, and they'll tell you when the forecast has missed something, such as a local event, which improves the next one. Hiding it tends to make every unpopular shift the algorithm's fault.

Do forecast-driven rotas mean more split shifts?

They can, because the demand curve for a lunch-and-dinner restaurant has a gap in the afternoon. Decide your policy before you let any tool draft shifts: many restaurants cap split shifts per person per week, or offer them only to staff who ask for them. Put that rule in the prompt or the tool's settings so it isn't overridden by the cheapest-looking option.

How far ahead should the rota go out?

At least a week is a common minimum and two weeks is kinder, but check whether any rules on scheduling notice apply where you operate. Forecast two weeks out using bookings and history, publish, then allow only small adjustments in the final days. If you're regularly cutting shifts at short notice, the forecast or your ratios need fixing, not the staff.

Further reads

• How to Train Front-of-House Staff to Work Alongside AI — Preparing your floor team to work alongside new AI tools.
• How to Survey Your Staff Before an AI Rollout (With Questions) — Asking staff what they think before rota changes land.
• AI Bias in Small Business Decisions: Hiring, Pricing and Credit — Keeping automated decisions about people fair.
• How to Use Booking Data and AI to Fix Your Class Timetable — The same demand-shaping idea applied to a class timetable.
• How Much Does AI Cost a Small Restaurant Each Month? — What a rota tool adds to a restaurant's monthly AI bill.
• What an AI Implementation Plan Looks Like for a Small Restaurant — One illustrative 48-cover restaurant's AI plan, start to finish: a week of tracking, three jobs chosen, four ruled out, and the day-90 numbers.
• Best AI Tools for Cafés and Coffee Shops, Grouped by Task — The AI tools a café can use, sorted by the job they do across the day, with what each costs and where each one lets you down.
• Where AI Actually Saves Time in a Small Restaurant — A task-by-task look at a small restaurant's admin week: where AI cuts real minutes, where it only moves them, and where it adds work.
• What AI Can and Cannot Do for an Independent Café — The desk jobs AI does well in a café, the ones it can't touch, and a 30-minute test on your own till data before you pay for anything.
• Can AI Help a Charity Forecast Demand for Its Services? — Use Excel's Forecast Sheet or a chat assistant to turn two years of service counts into a winter range, then add what the model can't see.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Homebase pricing page; Lineup.ai pricing page; 7shifts product information. Staffing ratios, wages and the worked example are illustrative. Checked September 2026.