AI for Home Care Providers Writes the Roster. The Claim Dies on Evidence

The visit happened. Tuesday, 9am, Mrs Kovac in Belmont, an hour of personal care. The support worker was on time, the shower went fine, and the notes say so.

Then someone asked a different question: where is the evidence that Mrs Kovac agreed to that price for that service?

Picture a home care provider in Geelong. Call the owner Leanne, a composite rather than a client. A hundred and forty clients on Support at Home, sixty support workers, two coordinators and one person in finance who also answers the phone when it rings.

That month a vendor showed Leanne what AI for home care providers usually means: a rostering tool. It matched workers to clients by skills, travel and availability, and rebuilt the week in seconds when someone called in sick. It was a good demo.

It would have scheduled Mrs Kovac's visit beautifully. It had nothing to say about whether the claim for it would stand up.

A visit you cannot evidence is a visit you delivered for free, or worse, a finding.

Where AI for home care providers is being pointed

Rostering, travel planning and voice notes are the pitch right now, and I will be fair to them. I build these systems for Australian businesses. If your coordinators spend every morning rebuilding the roster around sick calls, fix that. It is real time back.

But look at what the Department of Health actually checked when it reviewed the first two months of Support at Home claims. Its Cycle 1 payment integrity review covered sixty providers and 2,448 claims from November and December 2025. It found sufficient evidence of the agreed price for 35% of claims, sufficient evidence of delivery for 67%, and only 21% of monthly statements carrying every required inclusion.

The Department was careful to say those findings are about documentation, not proof that services were not delivered. That is exactly the point. The care happened. The paperwork did not hold together.

Rostering is where the demo lives. Evidence is where the money lives.

Why the evidence trail breaks

Under Support at Home, a claim is not one record. It is a chain:

service agreement → agreed price → scheduled visit → delivered visit → worker's notes → claim → monthly statement

Each link lives somewhere different. The agreement is a PDF signed at intake. The price sits in a spreadsheet or the client management system. The visit is in the roster. The notes are in the worker's app, or a text to the coordinator. The claim goes through the provider portal. The statement comes out of the finance system at month end.

At Leanne's, nobody owns the chain. Coordinators own the roster. Workers own the notes. Finance owns the claim. When a client changes services in March, the new price makes it into the roster but not into the signed agreement, and nobody notices until a statement goes out that does not match what the client was told.

Nobody decided to cut corners. Matching six places by hand, every month, for a hundred and forty clients, is simply the job nobody has time to finish.

The care is delivered by people. The evidence is assembled by whoever is left at five o'clock.

Run the numbers on your own claims

Treat it as arithmetic you fill in yourself.

Say you lodge six hundred claims a month and one in twenty has a gap: a missing agreed price, a note that does not match the hours, a statement line that does not reconcile. That is thirty claims. At twenty minutes each to fix, it is ten hours of coordinator time every month.

And that is the good outcome, where someone catches the gap. The bad one is the claim that goes out with the gap still in it and sits there until a reviewer asks for the evidence.

Every one of those numbers is invented and yours to replace. Pull fifty claims from last month and check each one against its agreement, its visit record and its notes. The shape survives any honest figure.

What AI for home care providers should do first

Fair question: why AI, and not a better checklist? A checklist works when someone fills it in. The slow part is reading. A model can read a support worker's note that says "showered, dressed, client tired, left at 10.05" and check it against a booked hour of personal care. A checklist cannot.

So the system walks the chain for every claim before it is lodged:

• It reads the service agreement and pulls the agreed price for each service.
• It matches each delivered visit to the roster and the worker's notes.
• It flags any claim where the price, the hours or the notes do not line up.
• It checks each monthly statement against the required inclusions before it is sent.

Every morning the finance desk gets one list, worst first:

• Kovac: personal care claimed at a price not in the current agreement
• Nguyen: visit claimed for 60 minutes, notes show 35
• Patel: monthly statement missing care management hours

When a note is unreadable or missing, it says so and puts that claim at the top of the pile instead of guessing. Nothing is lodged on its own. A person checks, fixes and submits.

Measure two things each month: claims flagged before lodgement, and statements that go out complete.

Automate the paperwork, never the care

No system should decide what care a client needs, change a care plan, or judge whether a worker's observation means a client is deteriorating. Those calls sit with your registered staff and your clinical governance, and they carry your registration.

The software tells you which claim cannot be evidenced. Your people decide what the client needs.

What AI for home care providers costs

Work down this list and stop at the first rung that works.

• The tools already inside your client management system. Many have agreement, pricing and statement features that are half configured. Turn them on first.
• A specialist claims checking product, priced per client or per month. Trial it for a quarter against the same fifty-claim sample.
• A custom build that reads your agreements, notes and rosters, walks the chain and flags gaps inside the software you already run. A few thousand up to the mid teens of thousands, depending on how many systems it must talk to.

I broke down where the running costs hide in what AI actually costs a small business. If you also deliver NDIS supports, the same evidence problem shows up there, covered in AI for NDIS providers.

The Australian layer

Two cautions. Your client records are health information, among the most sensitive data you hold. Ask any vendor where it is stored, whether it trains someone else's model, and whether it ever leaves Australia. Get the answers in writing.

And the rules are still settling. Support at Home is new, and the Department is reviewing claims in rolling cycles. Any system you build has to change when the guidance changes, which is one more reason to keep it reading your records rather than replacing them.

Common questions about AI for home care providers

What should a small home care provider automate first?

Claim evidence checks. They run on records you already keep, and one number tells you inside a quarter whether it worked.

Will AI replace care coordinators?

No. Calming a family, sorting a sick call at 6am and knowing which client needs a familiar face is the job. Matching six hundred claims to six hundred notes is not.

Is an AI rostering tool worth it?

If your roster genuinely eats hours every day, yes. But check your claims first. A perfect roster does not help when the claim behind each visit cannot show the agreed price.

Go back to Mrs Kovac. The worker turned up. The care was good. The record just never caught up with the visit.

If you want a straight read on how your claims would hold up, that is what a first call is for. Run the fifty-claim check yourself first, then book a strategy call and bring what you found, de-identified. We will walk the chain together before we talk about building anything.