It’s near midnight and a personal carer is sitting at her own kitchen table, still in her uniform, a mug of tea gone cold beside the laptop. Down the hall her family is asleep. She’s not on the clock. She’s finishing the notes from a shift that ended hours ago — the meals, the mobility, the resident in the Banksia wing who seemed off after tea and had to be helped up off his bedroom floor.

She does this most nights. Everybody on her floor does. The care happened during the day, in the rooms, with her hands. The record of it gets written here, later, for free.

The care and the record live in different rooms

This is the part the sector rarely says out loud. The thing you are being funded and regulated for is not only the care. It is the evidence of the care: timestamped, specific, auditable. And that evidence almost never gets written at the moment the care is given. It gets written afterwards, from memory, by someone who is tired and unpaid and would rather be asleep.

Studies of care documentation put the load at up to around 40% of a shift spent on records rather than people (indicative, drawn from a 2024 Nuance/myneva study and US Surgeon General figures across the health sector). Whatever the exact fraction on your floor, you already know where the overflow goes. It goes to the kitchen table at midnight.

Call it the second shift. Nobody rosters it. Nobody pays for it. And it is quietly where three of your biggest exposures are actually settled.

Why 7pm is where the money and the risk get decided

Since the new Aged Care Act commenced on 1 November 2025, the strengthened Quality Standards expect evidence of care that is timestamped and auditable, and directors now carry personal liability for it. That changes what a thin note costs you.

Start with funding. Homes must now deliver 215 total care minutes per resident per day, including 44 registered-nurse minutes, reported monthly through the Government Provider Management System. Roughly one in three homes was short at the end of 2025, and from 1 April 2026 a metro home that falls short loses up to $33.41 per resident per day (Department of Health, Disability and Ageing; StewartBrown FY25). That is real money against a sector where the June 2025 residential direct-care result was A$9.82 per bed per day, about 3.2%, and 301 of 1,165 homes were running their direct care at a deficit (StewartBrown FY25). There is no fat here to absorb a penalty.

Then the funding you leave on the table. Under AN-ACC, how a resident is classified follows what the note says. “Assisted with walking” and “two-person assist, 50 metres” describe the same event and fund it differently. Vague notes quietly under-classify a bed. An illustrative figure from Healthcare Professionals Association guidance puts it near $30 per bed per day in legitimate funding never claimed.

Then the risk. There are eight SIRS incident categories, and a Priority 1 incident is reportable within 24 hours, under a scheme that expanded to home care in November 2025. The 24-hour clock starts when the incident happens, not when someone finally reads the note about it. If the line that mattered is sitting in an overnight log, unwritten or unread, the clock is running against you either way.

All three of these are decided by what one tired person can capture at the end of a short-staffed shift. And the staffing does not have slack to fix it: the sector runs an annual workforce shortfall of roughly 30,000 to 35,000 people (CEDA). The bottleneck is too few carers, not too many.

The trap is in the software

This is where it gets awkward, and where most vendors go quiet. The record that carries all of this, the residential clinical record, sits in the one tier of your stack that software can least reach.

We went looking, system by system, at what an AI agent can actually connect to across the aged-care landscape. The pattern is stubborn. The reachable tier is your rostering, workforce and home-care platforms: AlayaCare, ShiftCare, Humanforce, Deputy, Epicor Senior Living, Civica Carelink. These expose real interfaces: you can read care-minute and roster data, create shifts and timesheets, subscribe to events. But the big residential clinical systems that hold the actual care record (Telstra Health Clinical Manager, Leecare Platinum6, AutumnCare, Health Metrics eCase, Manad Plus) expose no self-serve public interface at all. They exchange data through My Health Record, FHIR or curated vendor-to-vendor connectors, not an open door. The medication systems (Webstercare, MedSig) link vendor-to-vendor only. And the finance and government layers (Mirus, AIM, the GPMS reporting rails) mostly consume data: they call Medicare and Services Australia; they do not hand your own data back out.

One more thing worth knowing, because it sinks the “our software already has AI” objection: several of these vendors have added their own AI features. Every one of them is in-product only. None is reachable as a service you could point your own reasoning at. So the AI trapped inside the closed clinical record cannot help the carer at the kitchen table, and the tier that can be reached is not the one holding the note.

That sounds like a dead end. It isn’t. It just tells you where the fix has to live. Not inside the locked clinical vendor, and not in waiting for it to open up. The reachable move is to redesign the evidence-generating process around that clinical core: draft the record at the point of care, grounded in the roster and care-minute data you can already read, and bridge into the clinical system through FHIR or My Health Record or a plain human export step. Never a silent back-door write.

What AI actually does here

At the point of care — dictated on the floor, typed at the station, not at midnight — a draft assistant turns rough words into a structured note: mapped to the resident, timestamped, with the specifics AN-ACC and the Standards want, and the missing fields flagged rather than invented. When a line looks like it might be SIRS-reportable, it does not decide. It surfaces the text, says how confident it is, and puts it in front of a named person against the 24-hour clock.

That is the whole design, and the discipline in it matters more than any feature: a reportable incident is not a place for a machine to be right on its own, and this never lets it be. The AI drafts and triages. A named clinician approves, edits or rejects every clinical record and every SIRS assessment before it becomes anything. The residential clinical record stays the source of truth. There is no parallel record and nothing is auto-submitted. Every step — draft, reviewer, edit, sign-off — is timestamped and attributed to the person who signed it, which under personal director liability is the point, not a footnote.

Be sober about the size of it, too. Rigorous studies of ambient documentation find a net saving nearer 15%, not the 90% you see on vendor slides (JMIR Medical Informatics, 2026; JAMA, 2026). We build to the honest floor and measure the real number, because a headline you can’t reproduce is worse than no promise at all.

The tools that do this already have names and shapes: a Shift Handover Summariser that turns an overnight into a one-page, RN-ready handover with flags raised; a SIRS Incident Triage Assistant that queues incidents into a likely category, priority and reporting deadline for human sign-off; a Care-Plan First-Drafter; a Care Policy concierge that answers policy questions with citations and routes anything outside the approved set to a senior person rather than guessing.

What it looks like when it works

The by-products are what pay for it. When care-minute evidence is captured cleanly as the shift runs, the monthly GPMS return stops being a scramble and the penalty exposure drops. When notes carry the right specifics, the AN-ACC funding that vague notes leak stays claimed. When the handover is drafted instead of retyped, the incident that mattered reaches a human while the 24 hours still mean something.

We’re new to aged care, and we say so plainly. What tells you this is real is happening elsewhere in the sector already — as independent industry results, not RMAI’s. ECH, working with CGI Australia, ran a seven-week proof of concept that took care-plan drafting from about 20 minutes to 36 seconds, saved a reported 3,000-plus hours a year across 2,300 clients, and lifted documentation accuracy from 35.3% to 92.5%. MercyCare, using Mirus, reported 75% less management time tracking care minutes and RN coverage. Australian Unity, with Biarri, lifted home-care visits delivered as planned from 60% to 90%. These are third-party results from comparable Australian providers, illustrative of what we build toward — with a clinician in the loop throughout — not claims about your numbers or ours.

But the result I actually care about is smaller and closer to home. It’s the carer who no longer drives the day home in her head to write it up in the dark. She captures it as she goes, the system does the shaping, a clinician signs it off, and the evening goes back to the person who gave the care. That’s the efficiency worth having. Not a leaner org chart. A carer who gets to be asleep down the hall with everyone else.

Where to start

You don’t need a transformation programme to find out if this is real for you. You need one honest look at where your notes actually get written, how long the evidence takes to become auditable, and which of your systems an AI could even reach.

Two low-risk ways in:

• Read the map first. We’ve written a plain-English brief on the aged-care software landscape — which of your systems can actually be connected, by what mechanism, and what “closed” is quietly costing you in penalties and underclaimed funding. → https://realmindsai.com.au/guides/aged-care/
• Book a free 30-minute discovery call. We’ll walk one real path — a shift note to its care-minute evidence, or an incident to its SIRS decision — show you where the gap sits in your stack, and name the three highest-value places to close it. Any build is quoted fixed-scope after that, on your numbers; focused builds typically ship in three to six weeks. → https://outlook.office.com/book/Consultation@realmindsai.au/?ismsaljsauthenabled

There’s a carer in your organisation finishing today’s records tonight, at her own table, for free. The only question worth asking is whether anything in your systems is going to give her that evening back.