AI for allied health: the admin layer, handled — the clinical work, untouched

Not with client information in it. The OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools — and a client's health information is exactly that. Beyond privacy, the clinical content of a report is clinical work that sits with the clinician, outside back-office AI's scope entirely. What AI legitimately does is surface which reports are outstanding — because unbilled sessions usually trace back to unwritten reports.

It can process the volume and flag the drift; it cannot make the interpretive calls. Pay-point progression, reclassification when a therapist takes on supervision, non-billable time that is still work time, part-time agreements varied in writing — these are judgements a person makes against Fair Work's definitions. Many underpayments begin as interpretation errors, which is why the working model is AI-assisted variance checks with human sign-off, on records kept for the 7 years Fair Work requires.

Yes, in a specific and useful way. NDIS invoicing typically happens outside the practice management system against the current price guide, which makes it the highest-risk re-keying job in the practice. AI consistency checks compare invoice lines against the delivered-session records before anything is lodged, and flag mismatches for a person to resolve. The scheme's rules and the exceptions stay human work — this is about catching errors before they become rejections.

It can make the problem permanently visible, which is most of the fix. AI keeps the weekly list automatically: every delivered session either billed or flagged with a reason — report not written, cancellation unresolved. What usually remains is a policy decision (billing practice matching your stated cancellation policy) and a named person who owns the list. Visibility from the AI, accountability from a human — that combination is what actually moves the bank balance.

No. It replaces the worst hours of their week — matching lumped deposits, re-keying NDIS invoices, checking splits line by line — not the roles. Someone still onboards clinicians with provider numbers and health-fund registrations, tracks AHPRA renewals and supervision loads, and handles the human side of a busy front desk. The admin team gets more capable, not smaller; the grind is what goes.

The honest answer: the setup is human, the consistency is AI. Every discipline arrives with its own referral pathways, item logic and report obligations, and encoding those rules is judgement work done once by people who understand them. After that, AI applies the rules identically at volume — coding by discipline and payer, flagging exceptions — which is precisely what multi-discipline month-ends need and spreadsheets never deliver.

Two places, both far from the clinical work: deposit-to-claim matching, because netted and lumped settlements are where reconciliation dies by hand; and the unbilled-sessions list, because it is the single report with the most cash impact. Get the review rhythm working there, then extend to payroll variance checks and NDIS invoice checks from a pattern you already trust.