Suggesting cover for rota gaps

Twenty to seven in the morning, a carer rings in sick, and someone has to cover four calls starting at eight. This is one of the genuinely hard operational problems in domiciliary care, and it happens constantly.

The appeal of a tool that produces a shortlist in seconds is obvious. The complications are mostly about people rather than technology.

How it works

A tool holds availability, training records, location and contracted hours, and when a gap appears it produces a ranked shortlist of who could cover, with a reason for each.

The coordinator still rings round. The tool replaces the mental list, not the conversation.

Where it helps

Speed at the worst moment. Six-forty in the morning is when a coordinator is least able to hold twelve constraints in their head.

Not forgetting people. A coordinator under pressure rings the people they think of first. A system considers everyone who meets the criteria, including the person who joined six weeks ago and has never been asked.

Compliance constraints. Working time limits, required training, DBS status — these are exactly the checks a machine should do, because they are rules, and forgetting one has consequences.

Where it goes wrong

It learns your habits, including the bad ones. This is the main risk and it is quiet. If four people have always said yes, they have the best record, so they get suggested, so they get asked more. Over a year the tool has built you a burnout machine with an audit trail showing it was fair.

Fix: check monthly who is actually being suggested and asked. If it is the same names, that is a finding about the tool, not about those staff.

It disadvantages some people systematically. Staff with caring responsibilities of their own, or who work part-time, are less often available at short notice. A system that ranks on past availability will rank them lower, permanently. Most of them are women. That is a pattern worth watching for rather than discovering later.

The data is usually not good enough. These tools depend on accurate availability, travel times and up-to-date training records. Many services do not hold that accurately, and a confident shortlist built on stale data is worse than no shortlist.

The real question is not in the system. The coordinator’s actual thought is often “I could ask her, but she covered twice last week and her mum is ill”. That is not in any field, and it is usually the right call.

The bit people forget: this is staff data

Every other page on this site is about the data of people you support. This one is about your employees.

That matters in two ways. Your staff privacy notice probably needs updating to say that AI is used in allocating work. And if shift allocation becomes substantially automated, it is worth asking whether Article 22 is engaged, because who gets offered shifts affects people’s income — the ICO’s test is whether a decision has “legal or similarly significant effects”.

Keeping a human making the offers is the straightforward answer, and it is what the tool should be designed around anyway.

How to trial it

1. Check your data first. If availability and training records are not accurate, fix that before buying anything. You may find the data work solves most of the problem on its own.
2. Run it in parallel. For a month, let the tool produce a shortlist and let the coordinator work as normal. Compare.
3. Ask the coordinator what the tool missed. Their answers are the specification for whether this is worth it.
4. Count who gets suggested. Across the trial, how many different people appeared? If it is a small group, the tool is entrenching rather than widening.
5. Tell your staff you are trialling it, and what it does.

What good looks like

A tool that hands the coordinator five names they can choose between, including one they would not have thought of, while doing the training and working-time checks reliably.

What bad looks like: a tool that tells the coordinator who to ring, and is right often enough that they stop thinking about it.