The challenge
In every care home, some residents join almost everything and some join almost nothing. The quiet ones can slip toward isolation without anyone deciding it should happen. A UK care home group wanted an AI resident engagement audit that would catch this early, while staff still had time to act.
The group also needed a record. When anyone asks how engagement is tracked and improved, the answer should sit in the system, not in someone's memory.
And there was a line it would not cross. Activity records describe older people and their days, so the AI could help only if it never learned who any of them are.
What we built: an AI resident engagement audit
The audit sits on an activities module we had already built for the group, running on React, Node.js, Express.js and MongoDB.
Interest profiles. Each resident has a list of interests, and activities are scheduled with suggested residents based on them.
Attendance with reasons. After each session, staff mark attendance, and any absence needs a reason before it can be saved.
Enrichment entries. Residents who took part get a chart entry from staff.
The activities audit. The AI reads a resident's interests, attendance, absence reasons and enrichment entries. For residents with low participation, it writes audit notes that highlight opportunities to improve their engagement.
An interest report. A list of the residents interested in each activity helps planners fill sessions with the right people.
Absence reasons are stored in each resident's profile. That gives every note a trail back to the records it came from.
How it works day to day
The wellbeing coordinator opens the audit and finds notes on the residents who have taken part least. Each note draws on what that resident was offered, what they declined and the reasons staff recorded, and points to opportunities to improve their engagement.
The coordinator checks the note against the resident's record, talks it over with the home manager and decides what to try next. The AI suggests. Staff who know the resident decide.
Because the evidence comes from attendance marks staff already make, the audit asks nothing extra of the care team.
What the AI never sees
The model only ever receives text that has been redacted and de-identified. We designed the audit around that rule from day one.
Identifiers are stripped at the source. Inside the group's platform, and before anything is sent, names, dates of birth, NHS numbers, phone numbers and addresses are taken out or swapped for stand-ins. A resident becomes a neutral placeholder.
The model writes about placeholders. It reads participation patterns and drafts notes without knowing whose record it is reading.
Notes are matched back inside the platform. The link between placeholder and resident stays in the group's own system, which attaches each note to the right profile.
Staff review every note. Nothing in a resident's day changes until the wellbeing coordinator or home manager has read the note and agreed.
We designed this with UK GDPR in mind, sending the model only what the audit needs.
The result
The group now has a single record of interests, attendance, absence reasons and enrichment for every resident. On top of it, staff receive audit notes on the residents who rarely join in, each traceable to the records behind it.
The audit gives the home an early prompt and a written trail of how engagement is tracked, without exposing resident identities to the AI. We have no measured outcome figures, so we don't quote any.
What the research says
Health data is the costliest to lose. IBM's 2025 Cost of a Data Breach report put the average healthcare breach at $7.42 million, the highest of any industry, against a global average of $4.44 million (IBM, vendor research). That is why names and other identifiers are stripped before any text leaves the group's platform.
Send the AI less. The UK Information Commissioner's Office says personal data must be adequate, relevant and limited to what is necessary, and suggests de-identifying data before it is extracted from its source and shared (ICO guidance on AI and data protection). The audit sends the model only activity records, with placeholders in place of identities.
Providers want proof technology is safe. In the government's 2025 adult social care provider technology survey, 44% of respondents said assurance that care technology is safe would help them adopt it (DHSC technology survey). Notes that cite the record, and a person who reviews each one, make the tool easier to trust.
Where AI fits next
These ideas are not live yet. A monthly digest could list residents whose participation has dropped, for staff to review. Once staff approve a note, a version could also be shared with relatives through a family portal.
Planning something similar?
Before you let AI review resident records, work through these questions:
What counts as low participation in your homes, and who sets that line?
Which identifiers must be removed before any text reaches a model?
Who reads each note, and how do they record what they decided?
Can every note be traced back to the records it came from?
The audit is one view of the same activity records behind our AI breakdown of offered versus refused activities. For the privacy side, read how to use AI without giving away your data. Our AI automation and healthcare software pages show related work.
