If you run a café, a shop, a clinic, a cleaning company or a small care team, you probably know the Sunday-evening ritual: a spreadsheet, a group chat full of "can I swap Thursday?", and two hours of juggling availability, holidays and labour budget before the week even starts. For many owners we speak to, the rota is the single most hated admin task in the business.
It is also one of the places where AI earns its keep fastest. Not because it replaces your judgement about who works well together, but because it is very good at the tedious part: checking constraints, spotting gaps, forecasting how busy you will be and drafting a first version you can then adjust. This guide shows how to set that up in a small team, what to watch out for legally, and the prompts to use if you are not ready to pay for dedicated software yet.
What AI can and cannot do for your rota
It helps to be clear about the job before choosing a tool. A rota is really three problems stacked on top of each other:
- Demand: how many people, with which skills, do you need in each hour of the week?
- Supply: who is available, what are they contracted for, what holidays are booked, who cannot work with whom?
- Rules: rest breaks, maximum hours, minimum staffing for safety, fairness on unpopular shifts.
AI is strong on demand forecasting (it can read your sales or booking history and spot patterns you miss) and on rule-checking (it never forgets that someone already worked a late shift). It is weaker on the human side: knowing that two staff members have just fallen out, that a new starter needs a confident colleague alongside them, or that someone quietly needs Fridays off this month. Keep those decisions with a person. The realistic goal is AI drafts, manager decides.
Two routes: dedicated software or a general assistant
There are two sensible ways for a small business to start.
Route 1: Scheduling software with built-in AI
Workforce tools such as Deputy, Planday, Rotaready, When I Work and Homebase offer auto-scheduling: you set staff availability, skills and rules once, connect your till or booking system, and the software proposes a rota built around forecast demand. Most also handle shift swaps, clock-in and holiday requests in an app, which removes the group chat entirely.
This route makes sense once you have roughly 10 or more hourly staff, variable trading patterns (hospitality, retail, leisure) or several sites. Check three things before you buy: whether it integrates with your payroll, whether its forecasting uses your sales data rather than a generic template, and where the data is stored if you are in the EU or UK. Our guide to choosing an AI vendor has a fuller checklist.
Route 2: A general AI assistant and your existing spreadsheet
If you have five to ten people and a fairly stable week, ChatGPT, Claude, Gemini or Copilot can do a surprising amount with a well-structured spreadsheet and a good prompt. It costs nothing extra if you already pay for one of them, and it is a low-risk way to learn what you actually need before committing to software. The rest of this guide focuses on this route, because the same thinking applies when you move to dedicated tools.
Step by step: build an AI-drafted rota in an afternoon
Step 1: Put your rules in writing
Most rota problems come from rules that live only in the owner's head. Write them down once, in plain language, in a single document. For example:
- Opening hours: Tuesday–Sunday, 07:30–18:00. Closed Mondays.
- Minimum two people on site at all times; at least one key-holder at opening and closing.
- At least one barista-trained person on every shift before 11:00.
- No shift shorter than 4 hours or longer than 10 hours.
- At least 11 hours' rest between shifts; nobody works more than 6 days in a row.
- Weekend shifts shared fairly: everyone works at least one and no more than three weekend shifts per four-week period.
This document is also useful on its own. New managers can use it, and it is the core of the instructions you will give any tool later.
Step 2: Build a clean staff table
Create one sheet with a row per person: name (or initials), contracted hours, skills, days they cannot work, preferred shifts, and booked leave. Keep personal details out — the AI does not need phone numbers, addresses or anything about why someone is unavailable. "Not available Wednesdays" is enough.
Step 3: Estimate demand from your own data
Export eight to twelve weeks of hourly sales, covers or bookings from your till or booking system. Upload it and ask:
Prompt: "Here is 12 weeks of hourly sales data for a café. Identify the busiest and quietest hours for each day of the week, flag any weeks that look unusual (holidays, events, weather), and suggest how many staff I need per hour, assuming one person can comfortably handle up to €120 of sales per hour. Show the result as a table by day and hour."
Replace the €120 figure with your own rule of thumb — the point is to give the AI a ratio rather than letting it guess. Check the output against your gut feeling. If it says you need four people on a Tuesday afternoon when you know two are idle, find out why before you trust it.
Step 4: Ask for a draft rota
Now combine the three pieces:
Prompt: "Using the staffing rules, the staff table and the demand table above, draft a rota for the week starting 16 November. Cover the required staff per hour, respect every rule, stay as close as possible to each person's contracted hours and honour preferences where you can. Then list (1) any rule you could not meet and why, (2) each person's total hours versus contract, and (3) the total paid hours for the week."
The second half of the prompt matters most. Asking the AI to report its own rule breaches and hour totals turns a black box into something you can check in two minutes.
Step 5: Check, adjust, publish
General assistants are good at drafting and less reliable at arithmetic across a big grid. Paste the draft back into your spreadsheet and let formulas total the hours — do not rely on the AI's own sums. Then apply the human layer: pairings, training needs, anything you know that the data does not. Most owners find they change 10–20% of the draft, which still saves most of the time.
If you want to go further, a custom AI assistant with your rules document and staff table saved as knowledge turns this into a five-minute weekly job: upload the new week's leave and demand, ask for the draft, review.
Handling swaps, sickness and last-minute gaps
The rota you publish on Sunday rarely survives until Wednesday. AI helps here too:
- Cover finder: "Sam has called in sick for Thursday 07:30–14:00. Using the current rota and rules, list who could cover without breaking the rest-period or maximum-hours rules, ranked by who is furthest below their contracted hours."
- Swap check: "Alex wants to swap Saturday's late shift with Priya's Tuesday early. Does this break any rule?"
- Message drafting: ask for a short, friendly message offering the shift to the top two candidates. You send it; the AI does not.
Dedicated scheduling apps automate most of this, including pushing open shifts to eligible staff. With a general assistant, it still turns a 20-minute scramble into a two-minute check.
The legal and fairness side: GDPR, working time and the EU AI Act
Scheduling is about people's livelihoods, so a few guardrails are worth setting from day one.
Working time rules still apply. Across the EU, the Working Time Directive sets minimum daily rest of 11 consecutive hours, a weekly rest period and an average maximum of 48 hours per week, and national law often goes further. An AI draft is not a compliance check. Put the rules in your prompt or software settings, then verify.
Data protection. Rotas involve personal data, and availability notes can reveal sensitive information (health, family or religious commitments). Record the constraint, not the reason. Use a business plan of your AI tool that does not train on your data, and tell staff how scheduling decisions are made. Under GDPR, people have rights around decisions made solely by automated means that significantly affect them — another reason to keep a manager signing off every rota.
The EU AI Act. AI systems used to allocate tasks or monitor and evaluate workers fall into the Act's high-risk category, with obligations that fall mostly on the software provider but include duties for the business using it, such as human oversight and informing staff. Simple forecasting and a manager-approved draft are at the light end of this, but scoring staff on "reliability" to decide who gets hours is not. Our EU AI Act guide for small businesses explains where the lines are, and our piece on AI and employee monitoring covers the trust side.
Fairness. If you optimise only for labour cost, the AI will happily give the same three flexible people every unpopular shift. Write fairness into the rules explicitly, and once a month ask: "Show the distribution of weekend, early and late shifts per person over the last four weeks." It is the quickest way to spot resentment before it becomes a resignation.
What results to expect
For a typical team of 8–15 hourly staff, the realistic gains are:
| Area | Before | With an AI-drafted rota |
|---|---|---|
| Weekly rota building | 2–3 hours | 30–45 minutes |
| Covering a sick call | 15–30 minutes of messages | 2–5 minutes |
| Over- or under-staffed hours | Spotted after the fact | Flagged from sales data before the week starts |
| Rule breaches (rest, max hours) | Occasional, found by staff | Listed in every draft |
The labour-cost saving depends on how far your current rota is from your real demand curve. Owners who already staff tightly see mostly time savings; those who roster "the same as last week" often find a few hours a week of overstaffing in quiet periods. Measure it: compare paid hours against sales for four weeks before and four weeks after.
Start small. Draft next week's rota with the prompts above, alongside your usual method, and compare. If the AI version is as good or better twice in a row, switch. If you outgrow the spreadsheet, you will know exactly which rules and integrations a dedicated tool needs to handle — which makes choosing one much easier. For the customer-facing side of the diary, see our guide on using AI for appointment scheduling.
Where does your business stand on AI?
Take the free 3-minute AI Readiness Quiz and get a personalised score with your next steps.
Take the Free Quiz →