Automated wellbeing check-in calls

We make a tool in this category, so read this page with that in mind. We have tried to write it as we would write about a competitor, which means it is more critical than a sales page would be.

What it is

An automated voice service telephones a person on an agreed schedule, asks a short set of questions — how are you, have you eaten, is anything worrying you — and passes anything concerning to a named person on your team.

The case for it is straightforward. Many people receiving domiciliary care have long gaps between visits. A person supported on Monday and Thursday has three days where nobody from the service speaks to them. Something happening on Tuesday is found out about on Thursday.

Where it genuinely helps

Gaps, not visits. The honest use is covering intervals where the alternative is no contact at all, rather than replacing contact that already exists.

Escalation speed. A mention of a fall on Tuesday reaching the coordinator on Tuesday is a real improvement over finding out on Thursday.

Consistency. Everyone gets asked, on schedule, including the people who never ring in and never complain. Those are often the people services hear from last.

Where it fails, and this is the important part

It does not notice. A care worker at the door registers a great deal in ten seconds: whether the person seems muddled, whether the curtains are still shut, whether someone else is in the room. A phone call captures words. CQC’s own risk list includes that AI “can de-personalise health and social care”.

People say they are fine. They say it to humans as well. The difference is that a human hears the pause before it.

It cannot follow up properly. Real concern surfaces sideways, three questions in, once someone has decided you are worth telling. A script does not have a fourth question.

Some people find it upsetting. For someone with dementia, an automated voice on the telephone may be confusing or frightening. For someone with hearing loss it may be unusable. These are not edge cases in this population.

Someone may not be able to speak freely. If the reason for concern is in the room with them, a scheduled automated call is the worst possible way to ask.

The risk that matters most

Not that the technology fails. That it works well enough.

A service under commissioning pressure can show that automated calls are cheaper per contact than visits. The step from this covers the gaps between visits to this lets us reduce visits is short, and it is made in a spreadsheet by people who have never heard one of the calls.

CQC’s first principle is “AI to support, not to replace: AI can enhance, but not replace human decision making.” We would put it more plainly for this category: if introducing check-in calls results in fewer human contacts than before, the thing has been used to do harm, whatever the dashboard says.

Loneliness is not solved by a machine asking whether you have eaten.

If you are going to do it

1. Write down the commitment that this adds contact rather than replacing it, before you start, and keep it where a commissioner can see it.
2. Get real consent, from the person, with a simple way to stop. Someone who feels the calls arrived attached to their care package has not consented freely.
3. Tell them plainly it is automated, at the start of every call, and make sure they can reach a human.
4. Name who reads the alerts, and what happens out of hours. An alert nobody reads is worse than no alert, because it creates a record suggesting someone was told.
5. Agree what happens when nobody answers. This is the scenario the whole thing exists for, and it is the one most often left undefined.
6. Do the DPIA. See do you need a DPIA.
7. Ask people what they think after a month. Not whether the system worked. Whether they wanted the call.

What to measure

Not the number of calls made. Three things worth knowing: how many alerts led to an action that would not otherwise have happened; how many people asked to stop; and whether the number of human contacts went up or down.

If the third number went down, stop.

The honest summary

For services with genuine gaps between visits and a reliable person reading the alerts, this can add something real. For services that are already stretched and looking for a way to cover more people with less, it is the technology most likely to be used as a substitute for care while being described as an addition to it.

That is a judgement about how it gets deployed, not about whether it works. It is also the reason this page is rated “try with care” rather than “good to try”.