AI for Care Homes: A Registered Manager's 90-Day Implementation Plan

Phase AI in over three months: spend days 1 to 30 on rules, an approved-tool list and a staff briefing; run two small back-office pilots in days 31 to 60; measure, decide and write up what you found in days 61 to 90. Nothing touches care decisions or identifiable resident records until the rules have been tested.

The common failure is starting with the tool. A home that hands out logins in week one finds, by week four, that three staff are using personal accounts, nobody knows what has been pasted where, and the manager cannot explain any of it to an inspector. The order matters more than the software.

Follow me on Instagram@sagnikteaches

What stays out of scope for all 90 days

Write the exclusions first, because they are the part an inspector, a family member or a nervous night carer will ask about. For the whole of this plan, AI does not draft or alter care plans, risk assessments, incident records or medication documents, does not receive a resident's name, date of birth, diagnosis or photograph, and does not decide anything about a person's care. Those jobs may come later, as their own pilots, with a named reviewer and a decision recorded by the registered manager.

Connect on LinkedInSagnik Bhattacharya

Pharmacies draw the same line, and the pharmacy tasks never handed to AI are a useful model for drawing yours.

Subscribe on YouTube@sagnikteaches

That leaves a surprising amount of useful work: the home's rota messages, recruitment adverts, newsletters, activity calendars, policy drafts, meeting minutes from staff meetings with no resident detail, supplier emails, training quizzes and menu rewrites. If you want a longer menu to choose from, ten back-office uses in care homes lists them with the risks of each.

The worked example: a 40-bed home with 62 staff

Every figure below is illustrative. Picture a 40-bed residential home with a registered manager, a deputy, an administrator, two team leaders per shift, 62 staff on the payroll and a mix of day, night and bank cover. It has no IT department: a part-time contractor looks after the office computers and the broadband. The manager's week is eaten by rota gaps, recruitment, family emails, audits and paperwork. The goal for the 90 days is modest: give back roughly four to six hours a week across management and administration, without a single privacy incident, and finish with a short record the manager can put in front of an inspector.

Days 1 to 10: time three jobs before you change them

Pick the ten working days of a normal fortnight and have the manager, deputy and administrator note how long each of these takes: building and fixing the rota, writing recruitment adverts and replies to applicants, family and supplier emails, and the monthly newsletter. A paper tally sheet is enough. Without this baseline you will finish the 90 days unable to say whether anything improved, and a claim of "it saves time" is exactly what an inspector will ask you to evidence.

In the worked example the tally comes out like this: rota changes and agency or bank messages take the administrator about 5 hours a week; recruitment adverts, shortlisting emails and interview invitations take the manager and administrator about 3 hours a week between them; the newsletter and activity calendar take the activities lead about 4 hours a month; family and supplier emails take the manager 3 hours a week. That is about 12 hours a week of drafting-heavy work, which is where AI can help. It is also where no resident's personal information is needed.

Write the one-page scope

The scope has four lines: what AI may be used for, what it may never be used for, who may use it, and who to ask. Keep it to one page. A filled-in version for the example home reads: "AI may be used to draft rota messages with first names removed, recruitment adverts, newsletters, policy drafts for manager review and training quizzes. It may not be used with resident names, dates of birth, conditions, photographs or incident details. Only tools on the approved list may be used, on the home's accounts. Ask the deputy before trying anything else."

Days 11 to 20: build the approved-tool list

An approved-tool list is a short table the whole team can read in a minute. It names each tool, the account it must be used from, what it may be used for, what data may go in and who owns it. Start with one general assistant and, if the home already uses Microsoft 365 or Google Workspace, the assistant that comes with it. More than two tools in the first 90 days spreads attention thinly and makes the staff briefing longer.

Use business plans, not personal ones. On the vendors' own pages, checked in November 2026, ChatGPT Business and Claude Team both promise no model training on your content by default, Microsoft says Copilot prompts and responses are not used to train its foundation models, and Google says Workspace content is not used for model training outside your organisation without permission. Personal ChatGPT accounts work the other way round: chats are shared for training until the user turns off the "Improve the model for everyone" setting, so a personal account depends on every carer remembering. ChatGPT Business and Claude Team both start at two seats, so a home budgets for two logins at minimum. Ask your contractor to check the privacy and data-sharing settings on each account and to screenshot them for your file. In the file, describe each setting in plain words beside its screenshot, because vendors rename menu labels.

A short check before a tool is approved

• Is the account owned by the home, with a password manager or two-factor sign-in, so a leaver cannot take it with them?
• Does the vendor say in writing that business content is not used for training, and have you saved that page?
• Can the deputy remove a user in under ten minutes?
• Does the tool keep history you could be asked to produce, and who can see it?

Days 21 to 30: the staff briefing

Staff are your biggest exposure and your best source of ideas, and they hold views about AI that range from keen to frightened for their jobs. A 20-minute briefing, repeated for each shift pattern including nights and bank staff, does more than a policy emailed to everyone. Say plainly that the aim is to take paperwork off people, not to replace carers, and that nothing is used with residents' information.

Briefing outline (20 minutes, each shift)
1. Why: management paperwork is taking time away from the floor (2 min)
2. What we will try: rota messages, adverts, newsletter, policy drafts (3 min)
3. Three rules: no resident or staff names or details; approved tools only,
   on home accounts; a person checks everything before it is sent (5 min)
4. Live demo: turn a rough note into a newsletter item (5 min)
5. What to do if you slip: tell the deputy the same day, no blame (3 min)
6. Questions and sign-off sheet (2 min)

The "no blame" line carries real weight. Mistakes will happen, usually a first name or a room number pasted into a prompt out of habit. If the first reaction is a warning, the second mistake will be hidden. Treat the report as the behaviour you want, record it in the log, and fix the cause, which is usually a template that invited too much detail.

Finish with a signed acknowledgement sheet and one page of rules in the staff room. Keep the sign-off to one sentence: "I have been briefed on the home's AI rules and know who to ask." Anything longer goes unread. Your staff policy on AI, consent and inspections is the document behind this page, and it is worth having it drafted before day 30.

Days 31 to 45: pilot A, rota and recruitment messages

The administrator runs the first pilot because the work is repetitive, the content has no resident information and the outcome is easy to time. For ten working days, they use the approved assistant to draft the standard messages: a bank-shift request, a thank-you after a shift swap, a recruitment advert and the replies to applicants.

A prompt for the example home, with the identifiers removed in advance:

Write a short, friendly message to our bank staff group asking for cover on
a late shift on Saturday, 14:00 to 21:00, on the ground-floor unit. Mention
that the shift is paid at the standard bank rate plus the weekend uplift and
that replies go to the administrator by 4pm on Thursday. Keep it under 70
words and do not use exclamation marks.

A realistic output is a tidy 60-word message that does the job. What you fix: the assistant often adds a closing line such as "Your support makes a real difference to our residents", which sounds like an advert, and sometimes invents a rate or a deadline if the prompt does not state them. Rule one for this pilot is that numbers and dates come from the person, never from the tool.

Recruitment adverts without the cliches

The second job in pilot A is the care assistant advert. A tool will give you "join our family" language unprompted. A better instruction is to paste the real shift pattern, pay range, training offered and the three things staff say they like about working there, and ask for a plain advert of 150 words with no superlatives. The administrator edits for accuracy and the manager approves it before it is posted. Expect to cut the first draft by a third.

Days 46 to 60: pilot B, newsletter and activities calendar

Pilot B moves to something families see. The activities lead gives the assistant a list of what happened and what is planned, for example "garden planting on Tuesday, a visiting singer on the 20th, quiz on Friday afternoons", and asks for a one-page newsletter in the home's usual voice. No resident is named. Photographs and consents stay in the existing consent process, entirely outside the AI tool.

Illustrative result: the lead used to spend about 4 hours a month writing and laying out the newsletter, and in the pilot spends about 90 minutes including checks. The check matters. In one realistic failure, the assistant turns "visiting singer" into "live music from a well-known local band" because it fills gaps with plausible detail. Families act on what a newsletter says, so every claim is checked against the activities diary before it goes out.

Also in this window, add a third small job if the first two are going well: turning the manager's meeting notes into a staff-meeting summary. Keep the notes free of resident detail. If a meeting needs resident detail, it is minuted the usual way, not with AI.

Budget for the 40-bed home

All figures are illustrative list prices; check the vendor pages on the day. The cash cost is small. The real cost is the staff time, so count it.

Cash comes to roughly $370 to $630 including the contingency, so $400 to $900 is a safe range to put in the home's plan; it leaves room for a third seat or a longer contractor visit. Against that, the baseline shows about 12 hours a week of drafting work. If AI trims a third of it, that is about 4 hours a week, or roughly 30 to 35 hours across the eight or so weeks the pilots run, against about 76 hours of staff time invested. The first quarter does not pay for itself in hours; the second quarter can, because the rules, briefings and account set-up are not repeated. Say so in the review; an honest number is easier to defend than an impressive one.

Days 61 to 75: a measures sheet that fits on one page

Choose five measures and no more. The deputy fills the sheet in weekly from the log, and the manager reads it monthly.

The "corrections" measure is the one managers forget, and it is the best early warning. If more than about one draft in four needs a factual fix, the prompts are not giving the tool enough real detail, or the job is the wrong one for AI.

Days 76 to 90: decide, record and prepare for an inspector

For each pilot, the registered manager makes one of three decisions and writes it in the log: continue as is, continue with changes, or stop. Put the reasons in one paragraph. For the example home: pilot A continues with a rule that the administrator, not the assistant, enters all dates and rates; pilot B continues with a checking step against the activities diary; the proposed third job, summarising staff meetings, continues for another month before a decision.

An inspector may ask how the home uses technology and how it protects residents' information. A folder with five items answers most of it: the one-page scope, the approved-tool list, the signed briefing sheets, the log of slips and fixes, and the 90-day review. Open the folder in the conversation rather than waiting to be asked; it shows that the home treated AI as something to govern, which fits the well-led part of any inspection framework. Keep what you say to what is true: do not describe features you have not turned on.

Questions an inspector or a family member may put to you

"Does the AI see my mother's records?" The answer is no, and the scope page says so. "Who checks what it writes?" The named person for each pilot. "What happens if staff use it wrongly?" The log shows how slips are reported and fixed. "Why use it at all?" To move paperwork time back to the floor, and the measures sheet shows whether that happened. Keep each answer to a sentence and back it with a document.

A realistic slip and how it surfaced

In the example, a team leader, working late, pastes a handover note into the assistant to tidy the wording. It contains two residents' first names and a room number. She notices as she hits enter, reports it to the deputy the next morning, and the deputy logs the entry, checks the account's history settings, deletes the chat and records that no further information left the home's account. The fix is a one-line reminder on the briefing sheet ("Handover notes never go into an AI tool") and a new template for handover wording that the team leader can use on paper. Nobody was disciplined, the log gained an entry, and the home can now show how an incident was handled.

That is the strongest argument for the 90-day shape. The slip happened in a limited setting, was reported because the rules were clear and the culture was safe, and cost the home nothing but an hour. The same slip in an unplanned rollout would have been invisible.

What comes after day 90

With the foundations tested, the next candidates are the jobs you deliberately left out. Each one gets its own mini-plan: a named reviewer, a narrower data rule, a test on made-up examples before real ones and a decision recorded by the registered manager. Take them one at a time, and leave at least a month between them. If you are unsure whether to bring in help, this guide to hiring a care home AI consultant lists what to ask. If you would rather map the order with someone outside the home, that is what my AI implementation consultation covers, but the plan above is complete enough to run yourself.

Questions managers ask before starting

Can carers use ChatGPT to write care plans?

Not as a starting point. Care plans hold identifiable health information and drive decisions about a person's care. Use AI for back-office writing first: rotas, newsletters, policy drafts, job adverts. Care planning with AI belongs in a later, separately approved pilot with a named reviewer.

Do we have to tell residents and families we use AI?

Be open about it. If a tool touches anything that affects a resident, say so in your privacy notice and tell families in plain words. Back-office drafting that holds no personal data needs less, but a short line in the family handbook costs nothing.

What if a night carer already uses a personal AI app?

Treat it as a finding, not a disciplinary matter. Ask what they used it for, explain the approved list, and give them a sanctioned route for the same job. Staff who feel punished stop telling you what they use.

Who should own AI in a home with no IT person?

The registered manager is accountable, but name one deputy or administrator as the day-to-day owner. Their job is the tool list, the log of pilots and the monthly check. Without one named person, rollouts stall by week six.

How much should a 40-bed home budget?

For the plan below, roughly $400 to $900 in cash across the 90 days, plus staff time. The larger cost is the hours: about 60 to 90 in total across the manager, deputy and administrator.

Further reads

• Home Care Agency AI: A 90-Day Plan for the Owner — The same idea for a domiciliary care agency.
• Adult Day Care Software for a Small Centre: Billing and AI Notes Checklist — Software checks for a centre running attendance and billing.
• Daycare Incident Reports With AI: Factual, Name-Free Narratives — How factual, name-free incident wording works in practice.
• AI Receptionist for a Dental Office: Weave, Zo or a Generic Agent? — See how another sector compares AI front-desk options.
• Mental Health Services AI: Intake, Waitlist and Referral Admin Jobs — Intake and waitlist admin in another sensitive setting.
• AI for Mental Health Providers: A Roll-Out Plan for Group Practices — A seven-step roll-out sequence for a group therapy practice, with consent wording, an approved-tool list and a 90-day review.
• AI for Pharmacies: The First 90 Days for an Independent Owner — A month-by-month AI plan for an independent pharmacy owner: phones and reviews, then rota and marketing, with dispensing kept out.
• Care Home Software UK: An AI Features Checklist Before You Sign — Thirty questions grouped into seven areas, from what counts as real AI to data location, export rights, per-bed pricing, CQC evidence and a trial.
• Children's Home Software: Which Systems Add AI and What to Check First — Mentor, OVcare and Lindi compared on what their own pages say about AI, plus the checks a registered manager should run first.
• Domiciliary Care Software UK: What the AI Features Actually Do — Birdie, CareLineLive and Nourish compared on the AI each really has, what it costs per hour or per carer, and the inspection evidence it creates.
• Home Care Agency Software Pricing: What the AI Features Add — What home care software costs on its own, what the AI features add on top, and how to price a 15-person agency three ways.
• AI Answering Service for Home Care Agencies: After-Hours Call-Outs — How a home care agency can use an AI answering service for sickness call-outs and urgent calls: scripts, escalation timers, message log and monthly costs.
• AI Policy for Aged Care: What to Cover and What Staff Must Not Paste — A checklist for an Australian aged care provider's AI policy: approved tools, the never-paste list, privacy impact assessment, consent and an incident log.
• AI Tools for Facility Management in a Care Home: Maintenance Logs — Turn carer messages into maintenance tickets, summarise safety-check logs and chase contractors, with a named person signing off every safety item.
• AI Tools for Nurses Who Run a Team: Rotas, Handover and Audits — A nurse manager's guide to AI for rotas, handover templates and audit write-ups, with a clear list of what must never go into a prompt.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business pricing page, Claude pricing page, OpenAI data controls FAQ, Microsoft Learn privacy page for Microsoft 365 Copilot, Microsoft 365 Copilot Business plan page and the Google Workspace generative AI privacy hub (all checked 4 November 2026).