How Can AI Help Hotels? Housekeeping Checks and Maintenance Requests

For a small hotel, AI works best on back-of-house paperwork: turning its own room standards into housekeeping checklists, sorting guest and staff fault reports into maintenance tickets with a category and an urgency, and drafting housekeeping rotas from the day's arrivals. The head housekeeper approves every checklist, and anything involving safety goes straight to a person.

None of these jobs faces a guest directly; guest messages and review replies have their own tutorials, as do pricing and upsells. The limit that matters behind the scenes is that the assistant has never seen your rooms. It works from the standards, notes and photos you give it and cannot inspect anything, so the gains are consistency (every room checked in the same order) and speed (a fault logged in a minute instead of remembered at five o'clock).

Follow me on Instagram@sagnikteaches

What can AI do back-of-house in a small hotel, and what can't it?

Five outputs are worth setting up, and each needs something from you first. The running example is a 24-room hotel with a head housekeeper, a part-time maintenance person and a manager who covers everything else.

Connect on LinkedInSagnik Bhattacharya

The sheet is the pattern throughout: the assistant drafts and sorts text, while counts, hours and dates are worked out where you can see the working.

Subscribe on YouTube@sagnikteaches

How do I turn my own standards into room-turn checklists?

Start from your own written standards, even if they are a single page, because the assistant will otherwise write a generic checklist from the internet's idea of a hotel room. Ask for one checklist per room state: departure, stay-over, arrival-ready check and deep clean. Keep each under about 25 lines, since longer lists get skimmed. A prompt that works:

You are helping the head housekeeper of a 24-room hotel.
Below are our written room standards. Turn them into a departure-turn
checklist for a standard double.

Rules:
- Use only our standards. Do not add services we do not list.
- One action per line, in the order a person walks the room.
- Maximum 25 lines.
- After the list, add a separate REPORT, DON'T FIX section: faults a
  housekeeper should log for maintenance, not repair.
- Use the words our team uses (duvet, bin, kettle).

OUR STANDARDS
(paste your one-page standards here)

An illustrative first draft, shortened to the main steps:

Four edits are typical. The assistant added "turn-down service with a chocolate on each pillow", which your hotel does not offer, so delete it. It wrote "comforter" and "trash can" where your team says duvet and bin. It slipped "wash windows" into a 30-minute turn; that belongs on the deep-clean list. And it left out the safe check until step 1 was reworded. The head housekeeper makes those edits, since that person knows what a real turn involves, and the approved list becomes the document that new staff are trained on.

The other room states are shorter versions of the same list. A stay-over service drops the bed strip and the deep dusting and keeps the bathroom, the bin, the restock and the step-9 test, which is why a stay-over is budgeted at 15 minutes against 30 for a departure. The arrival-ready check is a five-line walk-through for early or priority arrivals: smell, bed, bathroom, test, restock. Ask the assistant to derive both from the approved departure list and say which lines to keep, so the three versions stay consistent when a standard changes.

The REPORT, DON'T FIX column is the most useful part. Housekeepers see every room every day, so their checklist is a better source of maintenance requests than waiting for guests to complain. If the step-9 test fails in room 14 on Tuesday, the ticket exists before the next guest arrives. Hotel SOPs written with AI goes further on the standards documents these checklists come from.

How should inspection checks be sampled and scored?

A supervisor cannot inspect every room. A rule that is easy to run: inspect every room with an early arrival, one in three of the rest, and re-inspect any room that failed. Score ten items pass or fail, such as bathroom limescale, under the bed, bed made to standard, bin liners, kettle and cups, TV and remote, lights, curtains, smell and safe. Log each failed item in a sheet with the date, room, floor and a short note.

Take a week for the 24-room hotel (illustrative): 30 rooms inspected and 11 items failed. Paste the failed-item rows into a chat assistant, or add a column that categorises them, and ask: "Group these failures by item and by floor. List any item that failed three times or more. Do not name individuals." What comes back:

Limescale or grout in the bathroom accounts for 5 of the 11 failures, 4 of them on the second floor. Under-bed dust failed 3 times, missing kettle cups twice, and a TV remote once. Possible explanations to check: descaler stock or dilution on the second-floor trolley, and the order of steps 4 and 8.

That is a good summary because it points at process, and it also shows the limit. "Possible explanations" are guesses until someone opens the second-floor cupboard and finds the descaler empty. Ask for hypotheses, then test the cheapest one first. One more rule: never let a photo pass a room. Models can describe a bed in a photo, but OpenAI's help page says image descriptions can be wrong and object counts are approximate. A supervisor's eyes pass or fail the room, and a photo is only a record of a fail.

How do guest and staff fault reports become maintenance tickets?

Reports arrive messily: a guest's message, a housekeeper's note, a front-desk call. A form with a few fixed fields (room, what is wrong, since when, whether a guest is in the room, a photo if possible, a tick-box for "safety issue") makes the sheet tidy. Then the assistant does the sorting. Here are three raw reports (illustrative):

1. Guest message: "hi the shower in 12 is only cold, been like that since last night, we're out til 6"
2. Housekeeper note: "r7 tv remote dead, also lightbulb by bed"
3. Front desk: "guest in 21 says the safe won't open, passport inside, flight tomorrow"

The instruction asks for four things per report: a category, an urgency (P1 now, P2 today, P3 this week, P4 planned), whether a guest is affected and how, and one question to ask. The second run produced:

The first run of the same prompt, before the "guest affected" field existed, rated the safe P3 under "furniture and fittings". Nothing in "won't open" says that a passport is locked inside. That is a realistic mistake, and it shows why impact on the guest belongs in the prompt: urgency is a function of what the fault does to a person, not what the fault is.

If you keep the log in Google Sheets, the AI function can do the sorting in a column: =AI("Classify this hotel maintenance report. Give a category (plumbing, electrical, heating or cooling, lock or safe, furniture, TV or Wi-Fi, other) and an urgency P1 now, P2 today, P3 this week or P4 planned. Format: category | urgency", C2). Google's help page lists some limits: it needs an eligible Workspace or Google AI plan, it cannot be nested inside other functions, only the first 350 selected cells generate at once, it cannot see the rest of your sheet unless you pass the range, and it does not refresh by itself when the data changes. If you work in Excel, the =COPILOT() worksheet function was retired on 14 September 2026, so paste the day's reports into a chat assistant or use Copilot's side pane instead. For the chasing and closing part of the loop, hotel maintenance software or AI: log, prioritise and chase is the longer read.

Closing the ticket matters as much as opening it. When the engineer has fixed room 12, a spoken note such as "swapped the shower cartridge, ran it three minutes and it ran hot, look again tomorrow" can go to the assistant with the instruction "Write a one-line log entry. Do not add anything I did not say." The entry should read: "Room 12: shower cartridge replaced; hot water confirmed after three minutes; recheck tomorrow." If it adds "tested to standard" or "guest informed", delete it. A log full of invented completions is worse than no log, because the next person trusts it.

Which faults must never wait for an AI triage?

Put a list on the form and on the wall, and make the safety tick-box ring the duty manager instead of waiting for any sorting. These faults go to a person immediately:

• A smell of gas, smoke or burning; a carbon monoxide alarm sounding.
• Sparking, exposed wiring or a socket that is hot to the touch.
• Water near electrics, a ceiling that is bulging, or any flooding.
• A door lock or window that will not secure.
• A lift with someone inside.
• Fire alarm panel faults or a blocked fire exit.
• Any guest injury.

The assistant is never asked about these, and it does not draft the response. Emergency services are called first when the situation calls for it, and the record is written afterwards. Equally, the rules for fire, water-system, gas and electrical checks follow schedules set by law and by your insurer, and they differ between places. An assistant can turn the schedule your contractor or adviser gives you into a tidy reminder list. It cannot tell you what the schedule should be.

Can AI draft the housekeeping rota?

Yes, with the sums done first. The workload comes from your numbers, not from the model. Monday (illustrative) brings 14 departures, 6 stay-overs and 4 empty rooms, with a standard of 30 minutes for a departure turn and 15 for a stay-over. That is 14 x 30 plus 6 x 15, which is 420 plus 90, or 510 minutes of room time. A housekeeper on a 6-hour shift with a 30-minute unpaid break is on duty for 5.5 hours, and after 45 minutes of trolley preparation and linen handling has about 285 productive minutes. Two housekeepers give 570, a margin of 60 minutes over the 510 needed. Check-out is at 11:00 and the first arrivals come at 15:00, so those two need to be on the floor by 09:30.

Friday has 20 departures and 2 stay-overs: 600 plus 30 is 630 minutes, which is more than two people can deliver (570), so Friday needs a third. The assistant did not work that out, and you should not ask it to. What you give it is the demand per day, the staff list with contracted hours and unavailable days, and the rules you already follow. Then ask for a draft rota in a table.

Draft a one-week housekeeping rota as a table (days across, names down).
Demand per day (housekeepers needed, start time, length):
Mon 2 x 09:30, 6 hours. Tue ... (paste from the sheet)
Staff: Ana 30 hours, not Sundays. Ben 24 hours. Cai 16 hours, not
Mon or Tue. (Replace with your own; use first names only.)
Rules: maximum 5 days in a row; at least one full day off a week;
no one under their contracted hours. If you cannot meet a rule, say
which one and do not break it silently.

Check three things before the rota is shared. Add up each person's hours against their contract. Count consecutive days, because "maximum 5 in a row" is the rule a draft most often breaks without telling you. And compare the number of people on each day with the demand row from your sheet. An illustrative slip: the draft gave one person six days running although the rules block said five, and showed Friday with two housekeepers instead of three. Both were caught by counting. Working-time and rest rules differ between places, so keep your own list of them and check the final rota against it. Rota tools can handle this later: Homebase's free Basic plan covers one location and up to 10 employees, and its AI-powered Scheduling Assistant sits on the Plus plan, listed at $70 a location per month (less on annual billing, and before any promotion). For the wider staffing picture, see hotel staff scheduling with AI and, for the order of rooms by arrival time, allocating rooms by arrival time.

Can AI keep a planned-maintenance list?

Yes, as a reminder list rather than a schedule. Planned maintenance is the unglamorous work that prevents the urgent calls: descaling shower heads, checking door closers, testing emergency lighting, servicing the boiler. The intervals come from the manufacturer's manual, your contractor and your insurer, and some are set by safety rules that differ between places. Give the assistant the equipment list with each interval written down, and ask for a month-by-month list:

Here is our equipment list with the service interval each supplier or
contractor gave us. Produce a month-by-month reminder list for the next
12 months. Do not add items or intervals that are not listed. Mark any
item with no interval as NEEDS INTERVAL and list those separately.

The NEEDS INTERVAL list is the useful output, because it shows what nobody has told you yet and gives you the questions to put to your contractor. Copy the reminders into a calendar that pings a named person, since a list left in a chat window reminds nobody. Rebuild it whenever something is added, such as a new boiler or a refurbished bathroom.

What stays with a person, and what about guest data?

• Safety and access decisions. Master keys, fire, gas, water and electrical issues, anything where someone could be hurt.
• Guests' valuables. Lost property that looks valuable, locked safes, medication. A person logs it and holds it.
• Charging a guest for damage. You decide, using your policy and the room's inspection record. The assistant can tidy the note afterwards.
• Staff performance. Patterns in the inspection log start a conversation; they do not settle one.
• Approving checklists and rotas. A person owns the final version every time.

On privacy, a room number and the fault are enough for a prompt. Leave out guest names, phone numbers, booking references and anything from a passport or payment card. For lost property, describe the item rather than pasting a photo of an ID. If you use a consumer chat plan, check that the model-training switch in its privacy settings is off; business plans do not train on your content by default. Give each staff member their own login, because the main providers' terms forbid sharing an account.

How will I know it is working after a month?

Choose four numbers before you start. Time from a fault being reported to the first action, by urgency level, using the timestamps in your sheet. Repeat faults, meaning the same room and same fault within 14 days, which tell you whether repairs are real or cosmetic. The share of inspected rooms that pass first time. And minutes spent drafting the weekly rota, which should fall once the demand sheet exists.

Keep a step if it makes one of those numbers better, or saves a manager at least 15 minutes a week, without adding errors. Stop a step if people are retyping the assistant's output into another system or skipping the checklist because it is too long. Then shorten the checklist before you blame the tool. In a 24-room hotel the whole set-up should take a few evenings: standards into checklists in the first, the form and the sheet in the second, and the first rota from the demand sheet in the third.

Questions hotel managers ask about back-of-house AI

Can AI look at a photo of a fault and say what is wrong?

It can describe what is visible, such as a stained ceiling tile or a loose tap, and that helps a ticket. It cannot reliably diagnose a cause. OpenAI's own help page warns that image descriptions can be wrong and counts approximate. Treat a photo as evidence attached to the ticket, and let the engineer decide what the problem is.

Do I need hotel software for this?

No. A shared sheet, a simple form and a chat assistant cover a small hotel. Your booking system probably tracks room status already, so use that for clean and dirty and keep the checklist and fault log in the sheet until volume justifies a paid maintenance module. Starting with software risks moving a broken process onto a nicer screen.

How often should the checklists be reviewed?

Whenever a standard changes and after any inspection pattern that keeps failing. The rule is that the checklist follows the written standards document, never the other way round. If a housekeeper suggests dropping a line, change the standard first, then regenerate the checklist from it.

Should the weekly summary name individual housekeepers?

Summarise by item and by floor first. Naming individuals in an AI-written summary invites over-reading small samples, such as two failures in six rooms. Look at individuals yourself, with the context of which rooms they were given, and talk to them before a pattern becomes a judgement.

Further reads

• Is AI Dynamic Pricing Worth It for a Small Hotel? — The front-of-house question this tutorial leaves out: room pricing.
• How to Upsell Room Upgrades and Late Checkout With AI — Upgrades and late checkout, written and timed with AI.
• Hotel Shift Handover Notes With ChatGPT: From Desk Logs to a Daily Report — Turn desk logs into a handover report between shifts.
• Where Should a Small Hotel Start With AI? — A wider order of jobs for a small property.
• AI for Hospitality: Which Jobs to Delegate and Which Stay With Staff — Which hospitality jobs to hand over and which to keep.
• How Property Managers Use AI to Triage Maintenance Requests — The same triage idea for rented properties.
• AI Consulting for Hospitality: What a Hotel Should Ask Before Hiring — Twelve questions to put to a hotel AI consultant, the month-one deliverables to expect, and a sample plan with numbers you can test.
• Airbnb Cleaning Schedule: Coordinate Cleaners and Changeovers — Build an Airbnb changeover schedule with AI: turnover windows, a cleaner handover sheet, short messages and checklist photos, with a worked week.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Sheets Help (the AI function), Homebase pricing page and OpenAI's ChatGPT image inputs FAQ, all read in October 2026. Checklist, sampling and rota figures are illustrations.