Care Home Software UK: An AI Features Checklist Before You Sign

Ask five things first: which features use a trained model rather than rules, where the data is stored and processed, whether you can export everything in a usable format, how pricing scales per bed or per hour, and how the system helps you evidence care for CQC. Then insist on a trial with your own records before you sign.

Most care management systems are strong on the things that make a care home run: care plans, daily notes, medication administration, incidents, rotas and reporting. AI is a recent addition to that list, and vendors describe it in widely different ways. A checklist helps because it forces each claim into a question with an answer you can verify. The thirty items below are grouped so you can send them as one document and compare replies side by side.

Follow me on Instagram@sagnikteaches

What vendors say about AI, and what to test

Four public vendor pages, read on 4 November 2026, show the range of claims. I summarise them to show what a test would need to establish, not to rank the products.

Connect on LinkedInSagnik Bhattacharya

• Nourish has a dedicated Nourish AI page describing a system that turns care data into actionable intelligence, "explainable and privacy safe by design", with principles that include every model being explainable and auditable. The one named feature is AI-assisted care plan reviews, described as running with no external data exposure.
• Person Centred Software (PCS) describes PCS Intelligence, which turns the data you already record into evidence for regulators, boards and families, benchmarks your homes against comparable homes, and uses AI that it says spots patterns early to help anticipate risks. Its site footer links to a responsible AI statement.
• Access describes Access Evo for Care as a single view of everything: one workspace connecting its care software, with AI insights, an intelligence layer that monitors data and automates workflows, a rota exception assistant that suggests cover, and plain-language questions it calls Quick Asks.
• Birdie, which is built for home care rather than residential homes, describes SmartPlans, which turn a recorded assessment conversation into a draft care plan with each suggestion cited to what the client said, and a Q-Score it says predicts your CQC rating in real time.

Each of those descriptions is a claim, not a specification. "Spots patterns early" could be a statistical rule on falls data or a trained model; "explainable" could mean a plain-language reason or a log; a predicted rating is only as good as the inspections it was built from. Where a vendor publishes a responsible AI statement, read it before the demo and bring two of its promises with you to test. The checklist below turns each phrase into something you can ask and test. Your answer sheet should record the vendor's reply, the date, and who gave it.

Subscribe on YouTube@sagnikteaches

Group 1: Which features are real AI

1. List every feature that uses a trained model. Why: marketing blurs rules and models. Verify: ask for a written list and ask for the others to be labelled "rules-based".
2. For each, what goes in and what comes out? Why: you need to know what the model reads (daily notes, voice, images). Verify: a one-sentence answer per feature.
3. Who approves the output before it enters the record? Why: a care plan or incident entry carries professional responsibility. Verify: watch the demo and see the approval step.
4. Can each AI feature be switched off per user, per unit or per home? Why: you may need to pilot on one wing. Verify: ask to see the setting.
5. Is the AI a vendor model, a third-party model or both? Why: it decides where data travels. Verify: named provider in the contract.
6. What are the known limits? Why: good vendors know where the feature fails, such as strong accents or unusual terminology. Verify: ask for a written list.

Item 3 deserves emphasis. A tool that drafts text for a resident's record is useful only if a named person reads and owns every line, so ask to see who is accountable in the audit trail.

Group 2: Data location and processing

7. In which country is resident data stored, and where is it processed by any AI model? Why: storage and processing can differ. Verify: contract schedule naming data centres or regions.
8. Which sub-processors handle data, and can you be told when they change? Why: a new sub-processor is a new risk. Verify: a current list and a notification clause.
9. Is any resident or staff data used to train or improve models? Why: care data is highly sensitive. Verify: written answer, preferably in the contract.
10. How long are prompts, drafts, transcripts and audio kept? Why: retention should match your policy. Verify: retention table.
11. Who at the vendor can see resident records, and under what controls? Why: support staff access. Verify: access policy.
12. What is the breach notification time frame? Why: you have your own duties to report. Verify: hours stated in the contract.

Your obligations as a data controller do not move to the vendor when you sign. For the policy side, a staff policy for AI in care homes covering GDPR, consent and CQC shows how to write the staff rules around whatever you buy, and what to check in an AI tool's privacy policy and terms lists the clauses to read.

Group 3: Export rights and exit

13. In what formats can you export everything? Why: a locked-in system is a price rise waiting to happen. Verify: request a sample export in the trial.
14. Does the export include care plans, risk assessments, daily notes, MAR charts, incidents, audits and attachments? Why: partial exports are common. Verify: open the file and check each.
15. Are historic AI drafts and audit logs included? Why: you may need them to show who approved what. Verify: ask.
16. How long after termination can you retrieve data, and what is the fee? Why: it affects a switch. Verify: contract clause.
17. Is data deleted on request, and is deletion confirmed in writing? Why: retention duties end at some point. Verify: process description.

Group 4: Price, per bed or per hour

18. Is the price per bed, per user, per hour of care, or a flat fee? Why: the model changes with occupancy and staffing. Verify: written quote.
19. Is AI in the base price, an add-on or metered? Why: metered use creates surprises. Verify: ask for a cap.
20. What are the charges for extra homes, devices, training and integrations? Why: they add up. Verify: itemised quote.
21. What is the minimum term, notice period and annual uplift? Why: a long term with an uncapped uplift weakens your negotiating position. Verify: contract.

Here is how the arithmetic can look for an illustrative 40-bed home with occupancy at 90 per cent and 12 staff users logging in on a typical day.

These figures are invented to show the arithmetic only. The cheapest looks like the per-user model, but if the home takes on eight agency workers who need access, that becomes 20 users and $800, so the per-bed model is cheaper. Rebuild the table with the vendor's actual numbers and your own staffing pattern, and ask each vendor what counts as a user.

Group 5: CQC evidence reports

CQC's assessment framework gathers evidence in six categories: people's experience of health and care services, feedback from staff and leaders, feedback from partners, observation, processes and outcomes. Software cannot create the first four for you, but it can make the last two easier to show. CQC also publishes a separate page on which categories it prioritises for care homes and supported living, and its guidance pages carry a "last updated" date (29 April 2024 on the evidence categories page when I read it), so check for changes before you rely on a vendor's mapping.

22. Which standard reports support the processes and outcomes categories? Why: you want evidence ready before an inspector asks. Verify: ask to see report samples using sample data.
23. Can you filter reports by date, resident, unit and type of event? Why: inspectors ask for specific periods. Verify: run one in the demo.
24. Does the system log who edited a record, when, and what the earlier version said? Why: a clear audit trail is evidence of control. Verify: view one record's history.
25. What does "inspection ready" mean in the product? Why: it is a marketing phrase. Verify: ask for the exact screens or exports.
26. If a feature benchmarks you against other homes, where does the comparison data come from, and how current is it? Why: benchmarking is only as good as the sample. Verify: ask for the source and date.

A caution: a report that looks reassuring is not the same as evidence of good care. Use the software to retrieve and organise, and keep your own audits, resident and family feedback and observation records alongside it.

Group 6: Trial, training and support

27. Can you trial the system with your own anonymised records for at least two weeks? Why: demos use clean data. Verify: written trial terms.
28. What training is included, how long does it take, and is it paid time for staff? Why: adoption decides value. Verify: training plan.
29. What support hours apply, and what is the target response time for a fault during a medication round? Why: some faults cannot wait. Verify: service levels.
30. Can you speak to two similar homes using the product? Why: references show real use. Verify: names and calls.

Reading vendor replies: weak and strong answers

Replies matter more than the questions, and weak ones tend to look alike. Here are three illustrative pairs, so you can recognise them when they arrive.

The weak replies are not lies. They are non-answers, and you should score them as zero until the vendor replies with something specific enough to be wrong. That discipline is the most useful habit in the whole process.

A message you can send to every vendor

Sending the same email to each vendor keeps the comparison fair and gives you a written record. Edit the bracketed parts.

Subject: Written responses for our care software review

Dear [first name],

We are a [40]-bed residential home reviewing care management software. Please reply in writing to the attached questions, grouped under AI features, data location, export rights, price, CQC evidence and trial terms. Where a feature uses a trained model, please say so and name the model provider. Please quote on a per-bed and a per-user basis, and confirm what counts as a user.

We would like a two-week trial with anonymised records before we decide, and the contract wording on model training, retention and exit.

Kind regards,
[your name]

Ask for the reply by a fixed date. A vendor who cannot respond in writing before you sign is likely to be slower after you sign.

Who in the home should own which questions

• Registered manager: groups 1 and 5, because the AI features and the evidence reports touch professional responsibility and inspections.
• Deputy or senior carer: group 6, because they will run the trial and judge the day-to-day fit.
• Whoever handles data protection: groups 2 and 3, including checking the contract against your own policies.
• Owner or finance lead: group 4, with a three-year cost sum rather than a monthly figure.

Splitting the work this way stops the review landing on one overloaded person, and it makes sure the person who will be accountable reads the part they own.

Hidden costs that a monthly price hides

Do a three-year sum, not a monthly one. Illustrative lines for the 40-bed home: handheld devices for carers (a one-off if the vendor does not supply them), Wi-Fi coverage in every room and corridor where staff record care, a paid training period for each new starter, the fee for each new integration (for example, with a pharmacy supplier or a payroll tool), and the annual uplift, which compounds. A 5 per cent uplift on a $6,720 yearly licence adds $336 in year two and about $689 in year three, so a flat price for three years is worth asking for. These figures only show the method, and your actual lines will differ.

A note on AI features that arrive after you sign

Vendors add AI features through updates. Ask how a new AI feature is introduced: off by default, or on? Will you be told in advance? Can you decline it without losing other features? A feature that arrives switched on, and quietly sends resident information to a new processor, is the kind of change your policy should catch. Put the answer in the contract as a notice and consent clause.

A data-location surprise, as it tends to happen

An illustrative case. A home signs for a system whose sales material says data is "hosted in the cloud with a leading provider". Months later, while preparing a privacy impact assessment, the manager asks where the AI drafting feature processes text and learns that the drafting model is run by a third party in a different region from the main database. Nothing was hidden: it was in the sub-processor list, which nobody had read. The home then had to decide whether to switch the feature off, ask for regional processing, or accept the transfer with extra safeguards, all while staff were already relying on it.

Asking questions 7 and 8 before signing would have put the answer in the contract. Make a habit of reading the sub-processor list on the day it arrives, writing the processing locations on one page, and checking it again each time the vendor notifies a change.

A demo script for the AI features

When a vendor demonstrates an AI feature, use a script rather than watching the polished path. These steps take about thirty minutes.

1. Use a fictional resident we supply: mobility aid, two recent falls, diabetes.
2. Enter three days of messy daily notes with abbreviations and a typo.
3. Ask the AI to draft a care plan review. Read the draft line by line.
4. Find one thing it got wrong or invented. Show where we correct it.
5. Show who is recorded as having approved the final text.
6. Turn the AI feature off for this unit. Show the setting.
7. Export this resident's record. Open it.
8. Show the log of what the AI read and produced.

If the vendor cannot find anything wrong in step 4, push: every model makes mistakes, and a vendor who says theirs does not is either not testing or not telling you. Step 8 is the one that separates serious products from the rest, because a log of what the model read and produced is what you need when someone asks how a statement reached a record.

A filled-in scoring sheet

Send the thirty questions to two or three vendors and score the replies. A simple sheet is enough.

Scoring (illustrative)  0 = no answer, 1 = partial, 2 = full, written
Group                     Vendor A   Vendor B
1 Real AI (6 items)          9          6
2 Data (6 items)            10         11
3 Export (5 items)           8          4
4 Price (4 items)            6          7
5 CQC evidence (5 items)     7          6
6 Trial and support (4)      6          7
Total (max 60)              46         41
Red flags:  A: AI trial not offered.  B: export limited to PDF.

In that illustration Vendor A scores higher overall, but Vendor B's weak export would be a serious worry. Total scores matter less than a single red flag in data, export or approval. Decide beforehand which items are deal-breakers, for example "no written statement on model training" or "no full export", and apply them before you add up points.

What a trial should test

A trial is where the checklist becomes real. Run one over two weeks with a small group and measure, not impressions.

• Time per daily note before and after, for five staff on different shifts.
• Error rate in AI drafts, by counting facts wrong or invented in a sample of twenty drafts. Compare with the safe way to draft a care plan review from daily notes.
• Report retrieval: time for the registered manager to produce a falls summary for a given month.
• Export: a full export of one resident's record, opened and checked.
• Staff view: one short, anonymous question to each user about whether it helped.

A realistic mistake from a trial: the vendor sets up AI drafting for a pilot unit, and a night carer finds that the drafts repeat "slept well" for a resident who had been up twice. The home had not defined who checks drafts on nights, so the error was noticed only a week later. The fix is cheap: write down who approves drafts on each shift before the trial starts.

Where staff time actually goes, and what AI should be judged on

The best test of an AI feature in a care home is whether it gives care staff time back for residents, without hiding anything from the registered manager. For a view of the back-office jobs where that can happen, see how AI can help in care homes: ten back-office uses. If you are considering several changes at once, a registered manager's 90-day implementation plan shows how to pace them so staff are not overwhelmed.

Finally, put every reply in writing and keep the replies with the contract. A promise made in a demo and not in the paperwork is not something you can rely on after you have signed.

Care home software and AI: questions from registered managers

Do I need AI in my care management system?

Not necessarily. Reliable care planning, medication records, daily notes and reporting come first. AI features are worth paying for only if they save measurable staff time or catch real risks, and if you can see how they reach their output. Treat them as an optional extra on top of a sound core system.

What does CQC want to see from care software?

CQC assesses services against its framework and gathers evidence in categories such as people's experience, staff feedback, observation, processes and outcomes. Software cannot supply that evidence for you, but it can make records easier to retrieve. Ask the vendor which reports map to those categories, and confirm with your own inspector's guidance.

Can I take my data with me if I leave a vendor?

You should be able to, but do not assume. Ask for the export formats, whether care plans, daily notes, medication records and incident logs are all included, how long you have after termination, and whether there is a fee. Test an export during the trial and open the file.

Is per-bed or per-user pricing better?

Per-bed pricing is easier to budget when occupancy is stable and many staff share logins. Per-user pricing can be cheaper for small teams but rises with agency staff. Ask for both quotes and model them against your occupancy and your real number of daily users before choosing.

Further reads

• AI Consultant for a Care Home: Do You Need One, and What to Ask — Whether outside help is worth it.
• Can AI Write Care Plans? What a Home Care Provider Must Check First — What to check on AI-drafted care plans.
• How to Evaluate an AI Implementation Proposal or Quote — Judge a vendor's proposal or quote.
• Questions to Ask an AI Vendor Before You Sign Anything — General questions for any AI vendor.
• AI Security Risks for Small Businesses and How to Close Them — Security risks to close before go-live.
• What Is an AI Scribe and How Does It Work in a Consultation? — How an AI scribe works in practice.
• 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.
• 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.
• NDIS Software for Small Providers: Per-User Costs and What You Get — ShiftCare's AUD per-user tiers, Brevity's per-client rates, worked monthly totals for a 12-person provider, and which AI features come with each plan.
• 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.
• CQC Mock Inspection Template: Run One and Summarise Findings With AI — Run a care home mock inspection by CQC key question, capture evidence, then use AI to summarise findings into a prioritised action plan.
• CQC Statement of Purpose Template: Draft Yours With AI and Check It — Draft a CQC statement of purpose with AI from facts you supply, then check every line against your registration, sign it off and set review dates.
• Care Home Daily Notes: Better Wording With AI, Checked by Staff — Before-and-after wording for care home daily notes, and the manager's sampling routine that keeps AI from adding anything that did not happen.
• Care Home Enquiry Form Template: Capture and Answer Leads With AI — A care home enquiry form template with field-by-field reasoning, fee and funding questions, tour booking and AI-drafted first replies.
• 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 Policies and Procedures UK: Keeping Them Current With AI — Use AI to compare each care home policy with current standards, keep tracked changes and a review log, and get staff sign-off before anything is published.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the Nourish AI page, the Person Centred Software home page, the Access Evo for Care page and the Birdie home page (read 4 November 2026); CQC's evidence categories page (read 4 November 2026); general contract-review practice. Check each vendor's current terms before signing.