A voice agent that keeps an allied health network's appointment book full
The outcome we're after.
A multi-clinic allied health network lives or dies by its appointment book. Every empty physio or podiatry slot is paid-for time that earns nothing, and a missed call is a patient who books somewhere else. The front desk cannot answer every line while also greeting patients and chasing rebates. A voice agent backed by retrieval-augmented generation answers the phone on every line at once, books and reschedules, confirms upcoming visits and sends reminders to cut no-shows. It speaks only from the network's approved information about hours, fees, rebates and referrals, so it never improvises policy, and it hands anything clinical, sensitive or distressing straight to a person rather than attempt an answer it has no business giving.
Book a discovery call 
The phone line every allied health clinic loses money on
An allied health network runs on a full appointment book, and the front desk is where it leaks. Across a handful of physiotherapy, podiatry and dietetics clinics, the phones ring all day. People want to book, move an appointment, ask what to bring, check whether their referral is current, or work out what they will pay after a Medicare or private health rebate. The morning rush and the hour after lunch are the worst. Reception is greeting patients, taking payments and chasing rebates, so calls go to voicemail, the box fills, and a caller who could not get through simply books with the clinic down the road.
The cost lands twice. A missed call is a missed booking. A forgotten appointment is worse, because the clinician’s hour is paid for whether or not the patient walks in. No-shows quietly drain a network. An empty 40-minute physio slot cannot be sold back, and a busy practice can lose several of them a week without anyone tallying the figure.
Most clinics have tried a phone tree or a generic chatbot and found neither fits. A menu cannot understand “I need to move Thursday’s session because my referral runs out”, and a generic bot will happily answer a clinical question it has no business touching. Healthcare raises the bar. The patient’s details are health information under the Privacy Act 1988 and the Australian Privacy Principles, and the one thing a booking line must never do is stray into clinical territory.
Why a RAG-grounded voice agent, not a generic bot or IVR
The version that works is a voice agent on the clinic line that speaks naturally and is grounded by retrieval-augmented generation (RAG), so every operational answer comes from the network’s own approved content rather than a model’s guess. A patient rings, asks in their own words, and the agent books, reschedules or confirms against the live calendar, or reads back a fee, a location or a “what to bring” note from the network’s published material.
RAG headlines the build because grounding is the whole point in healthcare admin. Fees, rebate rules and referral requirements change, and a model trained on last quarter’s policy would state the old number with total confidence. Instead the agent retrieves from a current index of approved sources, so updating the source updates the answer. An OpenAI GPT model turns the retrieved passage into a natural spoken reply inside tight instructions on scope and tone, never beyond them. The booking, reminder and identity steps run through the network’s existing systems and Microsoft 365, and the whole service sits in an Australian cloud region so health information stays onshore.
A generic chatbot or an IVR menu fails on the two things that matter here. It cannot hold a real rescheduling conversation, and it has no reliable way to refuse a clinical question. This agent does both. Its scope is deliberately narrow. It handles bookings, reminders and operational questions, and it routes anything clinical, sensitive or distressing to a person. It does not assess symptoms, it does not triage, and it gives no clinical advice of any kind.
Building it, and the line we had to draw
The model was rarely the hard part. The friction sat at the edges, and one issue defined the whole build. The clinical boundary.
People ring a clinic to book and start describing why. “My knee has been clicking, should I see someone sooner?” A booking agent must never answer that. The fix was a hard scope boundary, not a softer prompt. We taught the agent to recognise clinical content and signs of distress, and the moment either appears it stops, makes no attempt at an answer, and routes the call to a clinician or front-desk staff member with the context already gathered. The patient is not made to repeat themselves, and the agent never crosses the line it was built to hold. That boundary is the product as much as the booking flow is.
Identity was the second piece. Before the agent discusses or changes any appointment it confirms who it is speaking with, because a booking carries health information and the wrong change in the wrong record is a real harm, not an inconvenience. The agent verifies identity and reads the appointment details back before it commits a change, and on a second failed attempt it hands the call to a person rather than press on.
Two practical constraints shaped the rest. Operational answers about fees, rebates and referrals are grounded in the network’s approved content through RAG, so the agent never improvises policy, and where a source is missing or out of date it offers the right person instead of guessing. And the booking system enforced rate limits, so calendar lookups were cached and batched rather than called on every turn. None of this is glamorous. All of it is the difference between a phone line clinicians trust and one they switch off.
What changed
In a representative deployment the voice agent answered every line at once, so the engaged tone disappeared from the morning and post-lunch rushes and fewer would-be patients hung up. Timed confirmations and easy rescheduling cut no-shows by roughly a third, which turned paid-for but empty clinician hours back into billable time. About one booking in five came in outside clinic hours, when the only previous option had been a voicemail box the network rarely cleared in time.
These figures are illustrative. They describe the pattern we see rather than a published result for a named network. The shape is the point. The routine, after-hours and reminder load comes off reception, the book stays fuller, no-shows fall, and every clinical or sensitive call still reaches a person. The agent makes the clinic easier to reach without ever pretending to be a clinician.
Where this fits
A booking voice agent is one application of our AI Agents service, built on a retrieval-augmented generation core, for an Australian healthcare network. It suits allied health because the work is high volume, rules based and entirely operational, which is exactly where a voice agent is genuinely useful and where the clinical line can be drawn cleanly. If your phone line is costing you bookings and your no-show rate is eating clinician hours, the place to start is to map your busiest call types and decide where a person must stay in the loop.
Representative outcomes
No-shows down
Timed confirmations and reminders cut missed appointments by roughly a third in a representative deployment, which recovered billable clinician hours each week.
After-hours bookings
About one booking in five was made outside clinic hours, time the network had previously lost to a full voicemail box.
Calls answered first time
The agent picked up the phone on every line at once, so callers stopped hitting an engaged tone during the morning and post-lunch rushes.
This solution applies our AI Agents service, built primarily on Retrieval-augmented generation , for the Healthcare sector.
Supporting stack: OpenAI GPT model, Microsoft 365.
Related solutions.
Representative Solution. An illustrative scenario based on how we deliver, not a named client engagement. Outcome figures are representative, not published results.
Frequently asked.
What are examples of AI in healthcare?
Which administrative tasks suit a healthcare voice agent?
Does the voice agent give medical advice?
How is patient privacy protected?
How does it actually reduce no-shows?
Stop losing patients to a busy phone line
We will map your busiest call types and show you how a voice agent would book, confirm and remind safely, onshore, with your clinicians in the loop and the clinical line firmly drawn.
Book a discovery call