The two-thirds problem
Talk to any UK home care coordinator and ask what their day actually looks like. The honest answer, more often than not: "I spent two-thirds of it filling cover." Not building rotas. Not coaching carers. Not improving care plans. Re-shuffling the rota every time a carer calls in sick, runs late, or has a personal emergency.
That work is genuinely hard. To find good cover for a 30-minute visit at 7:15am you need to know: who's free, who's nearby, who already knows this service user, who's got hours headroom this week, who's trained on this person's medication, and who the service user actually likes. In your head, while three other carers are also calling in. Most rota systems hand you a list and say "good luck".
This update changes that. The new cover finder does the legwork before the coordinator picks up the phone. The AI rota builder (Pro tier) takes it further, generating an entire week's rota in under 60 seconds, ready for review. Here's what's actually under the hood.
How the cover finder works
When a visit becomes uncovered, sickness, sudden cancellation, late running, anything, the system runs every available carer through a five-factor scoring model. The factors and default weights:
| Factor | Weight | What it measures |
|---|---|---|
| Distance | 30% | Travel time from carer's previous visit (or home, if start of day) to the uncovered visit, via Mapbox routing. |
| Continuity | 25% | How many visits this carer has done with this service user in the past 90 days. Familiar faces score higher. |
| Hours headroom | 20% | Capacity left in the carer's contracted hours this week. Avoids creating overtime / availability conflicts. |
| Preferred carer | 15% | Whether the service user has explicitly flagged this carer as preferred (or excluded them). |
| Schedule fit | 10% | How well the visit slots into the carer's existing day without breaking another visit's window. |
The output is a ranked top-three. The coordinator sees scores like 87 / 82 / 76 with one-line explanations: "12 mins away, 4 prior visits, 8h headroom this week" or "20 mins away, no prior history, but on shift already". They make the final call, the AI just compresses the search space from "everyone available" to "the three most likely matches".
Weights are configurable. An agency that prioritises dementia continuity over travel time can flip the order. The defaults work for most agencies, but the point is they're not hard-coded.
The AI rota builder (Pro tier)
If cover finding is reactive, rota building is proactive. Once a week, usually Wednesday, the AI rota builder takes existing visit patterns, run templates, and carer availability, and generates a complete rota for the upcoming week.
The algorithm does roughly this:
- Group visits into geographic clusters using K-means on visit location
- For each cluster, identify carers who live within 15 minutes and have hours capacity
- Run continuity weighting, visits with high prior history with a specific carer go to that carer first
- Solve the remaining visits as a constraint satisfaction problem: minimise total travel time, maximise continuity, respect carer preferences and contractual hours
- Surface anything that couldn't be solved (e.g. a visit no carer has capacity for) for coordinator review
For an 80-carer agency with 480 visits in a week, the builder typically returns a draft rota in 45-60 seconds. Around 80% of the visits are placed cleanly; the remaining 20% are flagged for coordinator decisions. The coordinator now spends 90 minutes reviewing instead of 6 hours assembling.
What is, and isn't, actually AI here
"AI" is being used by every home care vendor at the moment, often loosely. To be honest about what we mean:
- This is AI: the cover-finder scoring model, the rota builder's constraint solver, retention risk scoring (separate feature), and document categorisation. These all involve trained models or scoring algorithms.
- This is NOT AI (despite the marketing language some vendors use): drag-and-drop rota grids, basic alerts, calendar conflict checks, route optimisation (that's Mapbox routing), or visit time tracking. These are useful, but they're regular software.
The honest answer to "is your rota system AI" is "the cover finder and rota builder are; everything else is good engineering." We think that distinction matters because it sets correct expectations.
A note on AI in care work
Care decisions affect real people. We don't let the algorithm decide alone, every cover suggestion goes through a coordinator, every generated rota is reviewable, every carer can flag a service user they shouldn't be sent to (or vice versa). The AI is a tool that compresses search space; the human is still in charge of the actual allocation.
We also don't use the algorithm for anything punitive. Bradford Factor scoring is calculated separately and surfaced as information, not enforcement. (More on that approach in our Bradford Factor guide.) Carer retention risk scoring exists, but it's surfaced to managers as "have a chat with this person", not "performance manage them out".
What does this replace?
For agencies coming from manual rota systems (whiteboards, Excel, paper diaries), the cover finder typically saves 40-90 minutes per coordinator per day. For agencies coming from older home care platforms without cover scoring (Birdie, CarePlanner, Access), the saving is smaller but real, typically 15-30 minutes per coordinator per day, plus better continuity scores because the algorithm consistently surfaces familiar faces. Side-by-side comparisons of every major UK platform are here.
The honest test for any AI feature in home care: does it free your team for higher-judgement work, or does it just produce more dashboards? If the answer is "more dashboards", it's not earning its keep.
Trying it on your own data
The cover finder ships on every plan, and the AI rota builder is part of the Pro plan at £10 per carer per month (the office team is unmetered). There's also a free starter tier for agencies up to 5 carers. Both features work best on agencies with at least 30 days of visit history loaded, the algorithms get noticeably sharper once they have continuity patterns to learn from.
The fastest way to see this in practice is to book a 20-minute demo, we'll run cover-finder simulations against a handful of representative scenarios from your current rota. Or start an account and import a week of visit data to evaluate the cover finder on your own rota. No card is required to begin, so there's no commitment.