AI and shift planning: what it can draft and what a manager must decide

AI shift planning in a hotel means using an AI assistant to draft a rota from rules you type in: coverage, contract hours, rest, skills and availability. It produces a first draft in a table. A manager still checks hours, rest, cost and fairness, approves the plan, and answers staff requests. The tool arranges constraints; it does not decide.

The rota is due tomorrow, two people are on leave and the night audit still has no name against it. An assistant can hand you a draft in a minute. Whether that draft is publishable depends entirely on what you typed in before you asked. This article separates what the tool can do from what you must decide, gives you a rule sheet to write first, and lists the six checks to run before the plan goes on the wall.

What can AI actually draft in a hotel shift plan?#

It drafts the first version of the rota: names against posts, days and hours, built from rules you type in. It arranges the constraints you give it; it does not know your people or your week.

The input is a rule set, not intuition. You specify how many people each post needs in each block of hours, each person's contract hours, approved leave, declared availability, and the skills each post requires. The output is a table draft with the rules it used visible, so you can see why a name landed on a shift. A draft you cannot audit is a draft you cannot correct.

Understand what the tool is. It is a language model arranging text and numbers. It does not check that the result is lawful, affordable or fair, and it will not volunteer which rule it quietly dropped unless you ask. It is genuinely useful on the tedious parts: the first pass across thirty people, the what-if question such as what happens if two more people take leave that week, reformatting the plan for the notice board, and writing the covering note.

It cannot know who stays calm at 03:00, who is three weeks into training, who has a verbal arrangement you agreed last spring, or who is about to hand in notice. If you give it an expected occupancy figure, it arranges staff around that figure. The figure is your decision, not its output.

This split matters because the drafting work and the deciding work fail differently. The draft fails by missing a rule; the decision fails by being wrong about people. If you approach covering the desk, floors and nights fairly as a management task first, the draft becomes a typing aid rather than a substitute.

Which rules do you have to type in first?#

All of them that matter. An assistant drafts only what you have written down, and every rule you leave out becomes a silent assumption in the plan.

Start with the coverage matrix: for each post, front desk, housekeeping, food and beverage, night audit, security, maintenance, how many people are needed in each block of hours, on weekdays, weekends, public holidays and in low season. Then the contract frame: weekly hours per person, the balancing period your contracts use, minimum rest between shifts, how night work is rotated, and who may approve overtime. These interact with working hours, the balancing period and night shifts as your contracts define them, so the numbers must come from your documents, not from memory.

Add skills and authorizations: who may run the night audit, who may close the cash desk, who speaks which language, who is trained on which system. Add availability and leave: approved leave, requested days off, school hours, and any restriction a staff member has declared and agreed to have recorded.

Then write fairness rules as numbers: weekends per person per month, night shifts in rotation, a cap on split shifts, and which rule wins when two rules collide. Finally, set the failure rule: when the assistant cannot satisfy everything, it must list the gap at the top of the draft and leave the slot empty, never fill it with a name that breaks a rule. Keep a version line: date, who asked, which rule sheet was used. The rule sheet is the real document; the draft is only its output.

How does the drafting step work in practice?#

One rule sheet, one request, one draft, one round of corrections, and a saved copy of each. Treat it as a typing job with a review, not a conversation.

Paste the whole rule sheet every time. A new session starts blank, and a draft built on half the rules still looks finished. Ask for a fixed table layout, one row per person per day, so two drafts can be compared line by line. Ask for the unsatisfied rules and the empty slots first, before the table; that list is the most useful part of the output.

Change one thing per round, for example move the night audit into a rotation, and keep each version with its date. If the plan is challenged later, the versions answer the question. Use initials or role codes instead of full names where you can, and never paste the reason for a leave. "Unavailable" is enough for the draft; the reasons belong in HR, not in a chat window, and your staff should know the same boundary from what staff may share with AI chat tools.

Time-box it. If correcting the draft takes longer than writing the rota by hand, the team is too small or the rules are too vague for this to pay off.

What must a manager check before the plan is published?#

Six things: hours, rest, coverage at the edges, single points of failure, cost, and fairness. The assistant can produce the numbers for the first three; the judgement on the last three is yours.

Hours first: the total per person against contract and the balancing period, with overtime flagged and approved by whoever holds that authority. Then rest: the gap between a late shift and the next early one, consecutive days worked, and the night rotation across the month. Then coverage at the edges, where rotas usually fail: the handover, the breakfast peak, late checkout, and the night audit that has to finish before the morning.

Single points of failure come next: a task only one person on the plan can do, with no backup on shift. Then cost: hours multiplied by the rates you supplied, against the department budget. If you gave it no rates, it has costed nothing. Finally, the fairness read-through: who has the weekends, who has the nights, and whether the same person absorbs every gap. Ask two of them before you publish.

Run the legal and contractual check with HR against your own contracts and collective agreements, and confirm your case with an adviser rather than with a draft.

How do staff see the final plan, and what can they change?#

They see one approved version in a place they actually look, and they change it through a written rule, not through the tool.

Publish the approved version only. A draft on the wall becomes a rumour, and the correction lands on the manager anyway. Keep one version everywhere: the back-office board, the staff group and the department screen must show the same plan with the same date. Write the change rule down: who may swap, how many days before the shift, who approves, and where the swap is recorded.

Give a simple written way to report a mistake or request a day off, and answer it as a person. Staff should never have to argue with a schedule that nobody will explain. Personal data rules apply here: availability, leave reasons and any health restriction are personal data. Collect only what the plan needs, tell staff why it is held and who sees it, and confirm the detail with your adviser. Keep a change log: the published plan plus each change with its date and approver. It settles most disputes in a minute.

What are the limits of an AI shift plan?#

It verifies nothing. It produces a plan that looks finished whether or not it is correct, and the accountability stays with the manager who publishes it.

Fluency is not correctness. A neat table reads as authoritative even when a rest period is broken or a post is left unstaffed. The tool has no view of the law or of your contracts; it applies the rules you typed, so if a rule was wrong, the plan is wrong in the same way, consistently. It has no memory of last month unless you give it last month's plan, so fairness across weeks has to be checked by a person holding the history.

It does not know cost unless you supply rates, and it does not know your budget. It cannot be accountable: when a guest complains about a closed desk at 02:00, the answer comes from a named manager, not from a tool. And it cannot hold the conversation with a staff member about why a request was refused. That conversation is the job.

When should you not use AI for the rota at all?#

When the rules are not written down, when the team is very small, or when the week is changing faster than a draft can be corrected.

If there are no written rules yet, the draft will encode the manager's habits and make them look like policy. Write the rule sheet first, even if you keep drafting by hand. With very small teams, three or four people, the rota is a conversation, and a table adds a step without adding anything.

Stay off the tool on complex contractual ground: collective agreements, apprentices, foreign work permits, part-time arrangements with special terms. Take advice; do not let a draft interpret a contract. In crisis weeks, sudden sick leave, a burst pipe, a group booking at short notice, the manager decides in the room, and the tool can reformat the result afterwards. And never let the rota become the quiet punishment for performance or discipline: those decisions are made openly, by a person, with a record. Finally, stop when the correction time exceeds the drafting time. Measure it once; the answer is usually obvious.

What to do next#

Write the rule sheet this week, before any tool drafts anything: coverage per post and hour, contract hours, rest, skills, availability, fairness rules as numbers, and the failure rule. Then let a draft do the typing and keep the six checks for yourself. The plan is a management document, and the signature on it is yours.