AI Tools for Facility Management in a Care Home: Maintenance Logs

A care home can use AI to turn carers' messages into structured maintenance tickets, summarise weekly safety-check logs so failures surface first, and draft contractor emails and chasers. The AI prepares entries; a named person decides priority and signs off every safety item. It never records a check that nobody physically did.

Most of the value sits in the dull middle: the vague "light's gone in the lounge" text, the log nobody reads until an inspection, the contractor who needs a third email. None of that needs resident information, which is why this is one of the safer places for a care home to start with AI.

Follow me on Instagram@sagnikteaches

Three streams of work, and what AI does to each

Facility work in a care home arrives in three different shapes, and they need different handling. Mixing them up is the usual reason a "maintenance AI" experiment disappoints.

Connect on LinkedInSagnik Bhattacharya

• Requests. Free-text messages from carers, kitchen staff and visitors: a dripping tap, a sticky door, a wardrobe handle. AI is good at turning these into a ticket with a location, a description and a proposed priority.
• Safety checks. Scheduled, recorded tests such as water temperatures, fire-alarm points, emergency lighting, call-bell tests, hoist and sling inspections, window restrictors. The checks are physical, so AI cannot do them. It can summarise the results, spot gaps and prepare the reminders.
• Contractor tickets. Work you cannot do in-house: lift servicing, boiler repairs, electrical testing. AI drafts the job description, the access instructions and the chasers, and keeps a tidy history.

The sign-off boundary runs through all three. Anything where a failure could hurt a resident, or where an inspector would ask "who checked this?", needs a human name on it. AI helps with paperwork around that name; it never replaces it.

Subscribe on YouTube@sagnikteaches

Turning a carer's message into a ticket

The first job to automate is intake, because it removes retyping and makes priority visible. Carers send what they send, often from a phone at the end of a shift. The prompt below asks the assistant to produce the same five fields every time, and to say when it is unsure instead of guessing.

You are helping the maintenance lead of a residential care home.
Turn each message below into a ticket with these fields:
Location | Problem | Proposed priority | Reason for priority | Missing information
Priorities: URGENT (possible harm to a resident or loss of a safety system),
SAME WEEK, ROUTINE. If a message could affect water temperature, call bells,
fire safety, doors, flooring or lifting equipment, mark URGENT and say why.
If the location is unclear, put "unclear" and ask one question.
Never include resident names. Do not invent details.

Messages:
1. "Lounge light flickering again, been like it all week. Also the kettle in the small kitchen smells of burning"
2. "Room 14 bathroom floor a bit slippy by the shower, I put a mat down"
3. "Call bell in 9 not going off at the desk, resident says she pressed it twice"

An illustrative response looks like this:

1a. Location: Main lounge | Problem: Flickering light, recurring for a week | Priority: SAME WEEK | Reason: possible fitting fault; electrical, but no stated heat or smell | Missing: which fitting
1b. Location: Small kitchen | Problem: Kettle smells of burning | Priority: URGENT | Reason: possible electrical fault and fire risk | Missing: kettle unplugged and removed?
2. Location: Room 14 en-suite | Problem: Slippery floor by shower, mat placed as stopgap | Priority: URGENT | Reason: fall risk | Missing: leak, worn surface or product residue?
3. Location: Room 9 | Problem: Call bell not sounding at the desk | Priority: URGENT | Reason: loss of a safety system | Missing: does the bell work from other points in the room?

Look at what the assistant did well and what you still have to fix. It split message 1 into two tickets, which is correct: a flickering light and a burning-smelling kettle are different risks. It raised the kettle and the bell to urgent without being told the specific items, because the prompt listed the categories. What it cannot know is whether anyone has already unplugged the kettle. That is the "missing information" column, and the maintenance lead's first action is to phone the unit, not to wait for the ticket.

Spend ten minutes testing the prompt on a month of your real messages before you rely on it. If it under-rates anything, tighten the URGENT list. A prompt that over-rates is annoying; one that under-rates is a risk. Where you want the same ticket format to flow into a shared tracker, the pattern in how property managers use AI to triage maintenance requests shows the same intake idea with a form in front of it.

Safety-check logs: summarise, never certify

Safety checks produce the records an inspector asks for first. They are also where an AI summary does the most harm if it is sloppy, because a smooth paragraph can hide a single failed test. So the design rule is blunt: failures and gaps come first, quoted from the log, and "all passed" is never a sentence the assistant is allowed to write.

The frequencies above are examples. Use the schedule in your own safety policy and your adviser's advice, and never let an assistant propose a frequency. What matters is the shape: a person does the check, a person signs, and the AI works on the record afterwards.

Here is a weekly digest prompt. Paste a week of log lines (room numbers, readings, initials, comments) and nothing else.

Below is one week of safety-check log entries from a care home.
Produce, in this order:
1. FAILED OR OUT OF RANGE: every entry marked fail, or with a reading outside
   the limits I give you, quoted exactly with date and initials.
2. NOT RECORDED: any check on the schedule with no entry this week.
3. COMMENTS WORTH A LOOK: entries with words like slow, weak, loose, stiff, intermittent.
4. Everything else: one line only.
Do not write "all checks passed". If a section is empty, write "none found in the text supplied".
Limits: [paste the temperature ranges from your own water-safety policy, per outlet type].
Schedule: call bells weekly; fire points weekly; water temperatures weekly.

[paste log lines]

Notice the instruction not to write "all checks passed". The assistant only sees the text you pasted, so it can truthfully say only that nothing was found in that text. If a page of the log went missing, a model that says "all passed" would be wrong; one that says "none found in the text supplied" is honest about its limits.

A worked example: a 42-bed home with one maintenance lead

Take an illustrative 42-bed home with one maintenance lead on 30 hours a week and an external contractor for lifts, boilers and electrical testing. Before AI, requests reach the lead through a paper book, WhatsApp messages and a whiteboard. The numbers below are invented for the illustration, so replace them with your own counts.

1. Measure the current load. For two weeks the lead tallies how long each stream takes. Say it comes to about 14 hours a week: 5 hours sorting and re-typing requests (around 35 requests), 4 hours writing up and chasing safety-check records, 5 hours on contractor emails and calls.
2. Intake. Staff send requests through one short form (location, what is wrong, photo optional). The lead pastes the batch into the assistant each morning with the intake prompt. Sorting time drops from roughly 8 minutes a request to about 3, because the lead now checks and corrects instead of typing. Across 35 requests that saves about 175 minutes, or just under 3 hours a week.
3. Safety digest. On Friday the lead pastes the week's log lines. The digest takes about 10 minutes to generate and review instead of 45 to cross-check by hand. The saving is about 35 minutes a week, but the larger gain is that missed checks show up on Friday, not at the next audit.
4. Contractor tickets. The assistant drafts job descriptions and chasers from the lead's notes. Say 6 contractor jobs a week at 15 minutes each drops to 8 minutes each: about 40 minutes saved.

The total is roughly 4 hours of 14 back each week, about 29 per cent. That is a modest, believable figure, and the lead still reads everything. Cost: either ChatGPT Business at $20 a user a month billed annually or $25 monthly (from two users), or the AI that comes with, or can be added to, the office suite the home already uses. Microsoft 365 Business Standard is listed at $14 a user a month paid yearly, and the Business Standard with Copilot bundle at $23.50; Google Workspace Business Standard is listed at $14 with Gemini in Gmail, Docs and Meet. Check each vendor's page before buying, since business plans change.

Contractor tickets that do not need chasing three times

Contractors delay when the job is unclear. A good ticket says where, what, what access is needed and what must not be disturbed. Care homes add one more constraint: work near residents, noise during mealtimes, and keys or codes that cannot go in an email.

Here is a before and after. The lead's raw note:

lift 2nd floor slow and jerky again, been 2 wks, last service in aug, need someone this week pls, residents using it for meals so not 11-2 or 5-6

Prompt the assistant to write a contractor ticket from it, and the result is something like:

Subject: Lift fault, second floor, recurring since mid-October: visit requested this week
Fault: Lift travel is slow and jerky between floors; reported twice in two weeks.
Last service: August (see attached report).
Access: Please avoid 11:00-14:00 and 17:00-18:00, when residents use the lift for meals.
Please call the maintenance lead on arrival; do not rely on the front-desk bell.
Please send: arrival window, engineer name and a written report with any parts needed.

Check it for what the assistant could not know. It wrote "mid-October" from "two weeks", so check the date. It said "reported twice", which the note never said: that is an invented detail, and the lead strikes it. Hallucinated precision of this kind is the main contractor-ticket failure. Add a line to your prompt: "Use only facts in my note. If you need a fact I did not give, write a bracketed question."

For chasing, a short prompt gives you a polite sequence you can reuse:

Write three chasers for the lift contractor, at 2 working days, 4 working days and 6.
Day 2: a friendly check. Day 4: state the risk (residents' mealtimes affected).
Day 6: ask for a named engineer and a date, and say I will arrange another
contractor if I do not hear by end of day. Keep each under 70 words. No threats.

Keep a running table of open tickets with date raised, contractor, chase count and the next action. The assistant can tidy it from your notes each Monday, but the table itself lives in your normal system.

Other jobs in the building that suit the same pattern

Once intake, digest and contractor tickets work, a few small jobs reuse the same prompts with little extra effort.

• Pattern spotting. Paste a month of closed tickets and ask which rooms and which fault types repeat. Say three tickets in six weeks name the same en-suite waste pipe: that is a fix, not three call-outs. The assistant suggests; the lead confirms from the originals.
• Planned-maintenance calendar. Paste the list of things with expiry or service dates (extinguishers, hoists, gas safety, electrical testing, legionella assessments) and ask for a calendar of what falls due in the next 90 days, with a reminder email draft for each. Dates still come from your certificates; the assistant only arranges them.
• Handover note for the lead's days off. A one-page "open tickets and urgent watch-list" note lets the on-call senior know what to do if the lift or a call bell fails at the weekend.
• Quote comparison. Paste two contractor quotes and ask for a side-by-side of what each includes, excludes and assumes. The decision stays with the manager.

If you want a broader view of how a home sequences this alongside care-planning tools, a registered manager's 90-day implementation plan puts facilities in context with the rest of the work.

Where this goes wrong in a care home

Four realistic failures, and how each shows up.

The tidy summary that hid a failure. A digest says "water temperatures looked fine" while one outlet's Tuesday reading sat outside the range in the home's written policy. It shows up when someone spot-checks the raw log. The fix is the prompt rule that failures come first and are quoted, plus the weekly five-entry spot-check.

Resident details slipping into a prompt. A carer's message says "Mrs [name] in 12 slipped by the shower". If that is pasted into a consumer chat tool, it has left your control. Tell staff the form has no field for names, remove names before pasting, and use a business plan where content is not used for training by default. How to stop AI tools training on your business data covers the switches.

A priority label treated as a decision. The assistant marks a sticky door "ROUTINE" and the door is a fire door that must close fully. Add a standing rule: fire doors, escape routes and anything involving a resident's mobility are always urgent, whatever the label says. Put that rule into the prompt and into the lead's own checklist.

Logging a check that was not done. The riskiest misuse is asking AI to "fill in this week's log". Never do it. The log records what a person observed, and an inspector, an insurer or a coroner will treat a false entry as far worse than a missed test. Make the rule explicit in your staff note: AI prepares and summarises; people test and record.

What the team needs to hear before the first paste

A ten-minute briefing at handover does more than a policy document nobody opens. Cover four points, in your own words. First, the assistant only ever sees maintenance text: room numbers, areas, equipment, never resident names, dates of birth or care details. Second, a ticket the assistant wrote is a draft until the maintenance lead has read it. Third, if a message sounds like an emergency (smoke, flooding, a bell that is dead, a resident at risk now), staff phone the lead or the senior on duty; they do not type it into a form and wait. Fourth, nobody asks the assistant to produce a record of a test: it summarises what people recorded.

The third point matters most. A form feels official, so staff sometimes treat it as having "reported" something. Write on the form itself, in one line: "Urgent or unsafe? Phone the maintenance lead or senior on duty now." That sentence costs nothing and closes the commonest gap between a digital tool and a live hazard.

A filled-in open-tickets table for Monday morning

Here is the kind of table the lead might ask the assistant to tidy from a week of notes. It is illustrative, with invented entries, and it shows the columns that make chasing automatic.

Ask the assistant to sort by priority, then by chase count, and to flag any urgent item with zero progress notes. It should not close a ticket. Closing needs the lead's initials and the retest result, which is the same sign-off rule again. One more small habit pays for itself: whenever a ticket closes, the lead adds a one-line cause ("worn seal", "tripped breaker"). A quarterly paste of those lines into the assistant gives you the pattern-spotting list above, and it is the cheapest way to decide whether a repeating fault deserves replacement instead of repair.

Photos, voice notes and the awkward request

Carers on shift often prefer a photo or a quick voice note to typing. Both can feed the same intake step, with care. A photo of a damaged skirting board or a stained ceiling tile is useful to the maintenance lead and rarely identifies anyone, but check each image for a resident, a name board or a medication trolley in the background before it goes into any tool. Voice notes should be transcribed by a tool your home has approved, and the transcript, not the audio, is what you paste into the assistant. If a transcript contains a resident's name, delete it before pasting.

The awkward requests are the ones that mix maintenance with care: "Mr [name] says his window won't stay shut and he's cold at night". The ticket should carry only the window, the room and the date. The comfort and care question goes to the care team through the care record. Teach staff to split such messages in two, and add a line to the intake prompt: "If a message mixes a building fault with a care concern, output the building fault only and add 'care concern for the senior' as a note without detail."

Checking it works after a month

Pick four numbers and look at them after four weeks. Time from request to ticket: it should fall to the same morning. Share of safety checks with a missing entry by the Friday digest: it should shrink week on week. Contractor jobs needing more than two chasers: count them against the previous month. And disagreements: how often the lead changed the AI's proposed priority, and in which direction. If the lead keeps raising priorities, widen the URGENT list in the prompt. If the lead never changes anything, either the prompt is excellent or the lead has stopped reading, so audit five tickets against the original messages to find out which.

Write one page for the team: what the assistant is used for, what it must never see, who signs what. Review it quarterly with the registered manager. If you would rather map the three streams with someone before you start, that is what my AI implementation consultation covers; get in touch if that is useful.

Care-home maintenance AI: the follow-up questions

Can AI decide whether a repair is urgent in a care home?

It can propose a priority, but a person must confirm it. Anything touching call bells, hot water, fire safety, hoists, doors or flooring that could cause a fall should be routed to the maintenance lead the same day, whatever the AI suggests. Treat the AI label as a first sort, not a decision.

Do I need a special maintenance system, or will ChatGPT do?

A chat assistant is enough to draft tickets, digests and contractor emails from text you paste in. It is not a record system. Keep the official log in your existing system or spreadsheet, and use the assistant only to prepare entries that a person then checks and saves.

Should resident names ever go into these prompts?

No. Maintenance work rarely needs them. Use room numbers or area names such as 'Room 14' or 'east corridor'. If a repair is linked to a resident's care, keep that detail in the care record and refer to it in the ticket only as 'see care plan'.

How do I stop staff trusting a tidy summary too much?

Keep the raw log entry one click away from every summary, and spot-check five entries a week against the originals. If a summary ever hides a failed check, change the prompt so failures are listed first and quoted word for word.

Further reads

• AI Consultant for a Care Home: Do You Need One, and What to Ask — Decide whether you need outside help before you buy anything.
• CQC Mock Inspection Template: Run One and Summarise Findings With AI — Use your maintenance records when you run a mock inspection.
• Care Home Software UK: An AI Features Checklist Before You Sign — Questions to put to a care-software vendor about built-in AI.
• Hotel Maintenance Software or AI? Log, Prioritise and Chase Requests — A comparable log-prioritise-chase loop in another setting.
• AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Cover for the maintenance lead's days off.
• How Much Time and Money Does AI Staff Training Take? — Plan the short staff briefing this workflow needs.
• AI in Care Homes UK: A Staff Policy Covering GDPR, Consent and CQC — A checklist-style AI policy for a care home: approved tools, data protection, consent, staff phones, oversight and the evidence to keep for CQC.
• Assisted Living Marketing Ideas: Eight Campaigns to Draft With AI — Eight practical campaigns for family decision-makers, each with a message angle, delivery channel, timing and human claims check.
• Care Home Audit Checklist: A Monthly Round Logged and Actioned With AI — A monthly care home audit checklist with scoring, an AI summary built from counts only, and an action log with owners and dates.
• Care Home Mandatory Training Tracker: Renewal Dates and AI Overdue Alerts — A spreadsheet tracker for care home mandatory training with renewal formulas, a weekly overdue list and a manager sign-off step.
• Care Home Medication Audit: A Monthly Template and AI Action Log — A monthly medication audit checklist for care homes: records and counts only, anonymous summaries and a signed action log.
• Care Home Residents' Meeting: Agenda, Minutes and You-Said-We-Did With AI — Plan the agenda, write accessible minutes and keep a you-said-we-did board current, with consent for recording and residents' names handled carefully.
• Nursing Home Survey Readiness Checklist: Evidence Binders With AI — How a nursing home can organise survey evidence: binder sections, a policy index, mock survey questions and a clear rule that AI summarises and staff verify.
• AI for Nursing Homes: What a Small Facility Can Run Without IT Staff — The office jobs a small nursing home can hand to AI with no IT department, which tools suit, and the privacy and vendor questions to settle first.
• Can AI Write a Care Home Fire Risk Assessment? What Stays With You — Where AI helps with a care home fire risk assessment (structure, action log, evacuation notes) and the judgements a qualified assessor must keep.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Microsoft 365 Business Standard and Copilot Business pricing pages, Google Workspace pricing page, ChatGPT Business pricing page (all checked 4 November 2026).