AI for care providers – disability support, home care and residential care
Less time at the computer and more with people – without compromising on secrecy
I build AI support for care providers that cuts the time spent at the computer: language support when notes are written, draft rotas, induction for temporary staff and procedures you can ask questions about.
Anything about the service users is assessed and written by the staff. AI helps with the form, not the content.
In a group home, a home care team or a residential service, time at the computer is time away from the people you are there for. Notes have to be written after the shift, the rota has to work when someone is ill, and the temp starting tomorrow needs to know where the procedures are and how the alarm works.
That is where AI can take some of the load. But the data about service users is among the most sensitive there is – health, disability, everyday life in someone's own home – and it often concerns people who find it hard to protect their own privacy. That is why every assignment begins with where the data may be kept. What the tool should do comes after that.
Where the benefit is
Administration that can be reduced
Language support in social documentation
Staff write the note in their own words, and the tool suggests clearer Swedish and points out words that judge rather than describe. The documentation regulations of Socialstyrelsen (the National Board of Health and Welfare), SOSFS 2014:5, require records to be well structured and clearly set out, and the general advice says journal entries should be concise. The tool is set up not to add information, which is why whoever wrote the note always reads and approves it.
Rotas and replanning
Draft rotas based on staffing, delegations, language skills and continuity – the same staff visiting the same service users as far as possible. When someone calls in sick, the tool can suggest how to fill the gap. The coordinator decides.
Induction for temporary staff
An assistant that answers from the service's own approved procedures: the fire routine, how incidents are reported, how the day is organised, whom to call. It is built to decline questions about individual service users, and their details are not in its material – that belongs in the handover between colleagues. The technique is described on the page about AI knowledge bases.
Procedures in more languages and easy-to-read Swedish
Procedures and information material can be translated into the languages the staff group needs, and information for service users can be rewritten in easy-to-read Swedish. Someone who knows the language or the recipient reads it through before it is used.
Material for quality work
Anonymised incident reports can be compiled so that patterns become visible: times of day, types of event, procedures that are not followed. The service manager draws the conclusions.
The line
What AI should not do in care
Write implementation plans. The implementation plan (genomförandeplan) describes how a decided intervention is to be carried out for the individual, and according to Socialstyrelsen's general advice in SOSFS 2014:5 it should show whether the individual took part in the planning. According to the general advice, the plan should state whether the individual has taken part in the planning. A plan worded by a language model is built on what was in the source material, not on the conversation with the person. Language support for a text the staff have written can work. The content must come from the staff and the individual.
Assess service users. Interpret behaviour, suggest interventions or carry out risk assessments. That requires knowing the person, and a responsibility that cannot be handed to a system.
Replace the handover. What needs to be said between two shifts should be said between two people. An assistant can help the temp find the procedure, but not tell them how this particular day has been.
Be the only route to what is urgent. What applies in a fire, an alarm or a sudden illness is something staff should know without asking a screen.
Secrecy and privacy
Service users' data and the rules
Health data is a special category. Under GDPR (Article 9), processing data concerning health is prohibited unless one of the exceptions in the article applies. Anyone building an AI tool for care needs to start from that, not discover it afterwards. If such data is to be processed in a new system, a data protection impact assessment under Article 35 may be required before it is taken into use.
Secrecy follows the data. In municipal services, social services secrecy applies under the Public Access to Information and Secrecy Act (chapter 26, section 1). Anyone working for a private provider is instead bound by a duty of confidentiality under the LSS Act (section 29) or the Social Services Act respectively. Putting data into an external service means handing it to someone outside the organisation, and that has to be worked out beforehand, not afterwards.
The service and the contracts. A business service with a data processing agreement, processing within the EU, no training on your material and role-based access. The most sensitive material can run on your own server instead, and that trade-off is described on the page about local AI. If you are a provider for a municipality, the contract with the municipality may also have its own terms on personal data and subcontractors.
Supervision. IVO (the Health and Social Care Inspectorate) supervises social services and services under the LSS Act. An AI tool doesn't change the requirements for documentation, and that is one reason it should be possible to see how a text came about and who approved it.
The information describes the situation in November 2026. The new Social Services Act (2025:400) has applied since 1 July 2025, and Socialstyrelsen has amended its regulations accordingly, so always check the current wording. This is not legal advice. More on the basics is on the page about AI and GDPR.
The approach
Start where there is no service user data
A first meeting at no cost. With the service manager, and ideally with someone who works on the floor. We go through where the time disappears and whether AI is the right answer – sometimes it isn't.
A first workflow without sensitive data. Induction for temps, procedures in more languages or the framework of the rota. The service gets to try how AI works in everyday life before anything touches data about service users. The price is fixed and agreed in advance.
Assessment before the next step. If the tool is to handle data about individuals, the contract review and the impact assessment are done first, together with whoever is responsible for data protection in your organisation.
Training that fits the rota. Shift staff are rarely all in at the same time, so the walkthrough is arranged as shorter sessions for different teams.
Honestly
When it isn't a good fit
When the care management system already does it. If the system you use for documentation and rotas has the features you need, that is where the time should go: learning them properly.
When the procedures aren't written down. An assistant can only answer from what exists. The first step is then to write the procedures, and AI can help with the drafts.
When the real problem is staffing. AI can cut the time at the computer. It cannot be an extra person in the room when two service users need support at the same time.
When nobody can own the tool. Someone needs to be responsible for keeping its procedures up to date and making sure it is used as intended. Without such a person the tool quickly goes out of date.
Background
Who is behind it
I am an AI consultant and a speaker, not a social worker and not a lawyer. What I contribute is how AI systems are built so that they hold up in operation and can be reviewed. The knowledge of the service users and of the work is yours.
I also give talks to staff in group homes, daily activities and housing support about neurodevelopmental conditions such as ADHD and autism (NPF) and how to respond to them – there is more about that on the page about the talk for LSS staff. All the AI services are gathered under AI consulting.
Glänne & Söner is based in Lindesberg. I take assignments across Sweden, on site or remotely.
May we use AI with information about service users?
It can be done, but it is the last step, not the first. Data about health and disability are special categories of personal data under GDPR, and social services secrecy or a statutory duty of confidentiality applies on top of that.
It requires a service with a data processing agreement and a known place of processing, role-based permissions and an assessment before the tool is taken into use. Much of the benefit can be had without any service user data at all.
Can AI write implementation plans?
No, and I don't build tools that do. An implementation plan (genomförandeplan) describes how a decided intervention is to be carried out for an individual, and it is built on conversations with the person and on the staff's knowledge of them.
What AI can do is help someone who has written a text make it clearer. The content is the staff's.
Can AI help staff who write in their second language?
Yes, it is one of the areas where the benefit shows fastest. Whoever wrote a note can get suggestions for clearer wording and pointers to words that judge rather than describe.
The tool is set up not to add anything that wasn't there, and whoever wrote the note reads and approves it.
Does it work with our care management system?
That depends on the system and on what the supplier allows. Several care management systems have their own scheduling and documentation features, and if they are enough they should be used. I check what can be connected before you receive a quote.
How much does it cost?
It depends on which task is to be taken off your hands and how sensitive the data is. You receive a fixed quote after a first meeting, free of charge, before anything is built.
What takes the most time away from your service users?
Tell me where admin eats up time in your service, and we will look at what can be taken off your hands without touching service users' data first. The meeting is free of charge, and a quote has a fixed price.
