Ask any practice manager in a busy Sydney or Melbourne clinic where the day actually goes, and the honest answer is rarely patient care. It is the front desk: the phone that never stops ringing, the recall list that needs chasing every month, the referral letters waiting to be filed, and the Medicare rebate questions that eat into every gap between consults. Reception is the busiest job in most Australian practices, and it is also the job most short-staffed on a Monday morning. Practice owners who ask us where an AI assistant should start almost always assume it is clinical. It rarely is. The bigger opportunity, and the safer one, sits at the front desk.
Where the reception hours actually go
Before deciding what an AI assistant should do, it helps to see where the hours are already spent. In a mid-sized general practice or allied health clinic, the daily workload usually breaks down into a handful of repeating jobs, most of them text-heavy and predictable rather than clinical:
Inbound calls for bookings, rescheduling and results queries, often 40 to 60 calls a day in a mid-sized GP clinic
Recall and reminder campaigns for immunisations, chronic disease reviews and cancer screening
Referral chasing: confirming specialists received documents and following up on missing reports
Medicare and private health fund queries, including HICAPS claims and rebate explanations
Waitlist and cancellation management, filling appointment gaps left by no-shows
A full-time receptionist in a Sydney or Melbourne practice costs roughly $58,000 to $65,000 a year in wages once superannuation and other on-costs are added, and most practices run two or three people across the week just to keep the phones covered during opening hours. None of that spend goes toward anything clinical. It goes toward logistics: matching patients to appointment slots, chasing paperwork, and explaining the same Medicare rebate rules dozens of times a week. A practice with three GPs and one allied health clinician can easily spend $150,000 to $180,000 a year on reception and admin wages before a single extra clinical hour is added.
What Claude can safely absorb
The tasks above are mostly repetitive, text-based and low-risk if a human reviews the output before anything reaches a patient. That makes them a reasonable starting point for an AI assistant like Claude, used through a tool such as Claude Cowork that connects to the practice's inbox and calendar, rather than a stand-alone chatbot staff have to operate by hand. In practice, that looks like:
Drafting recall and reminder SMS or email copy for a nurse to approve before sending
Summarising incoming referral letters into a one-line handover note for the GP
Triaging a shared inbox so results queries and urgent messages surface above routine admin
Drafting replies to routine questions about opening hours, fees, parking and telehealth options
Preparing a daily reconciliation note for HICAPS and Medicare batch payments
One Brisbane allied health clinic we worked with cleared a backlog of around 200 referral emails in an afternoon using an AI-assisted triage pass, a job that had been queued for over a week. Based on the practice's own time estimates, that kind of ongoing triage is worth close to $15,000 a year in reclaimed reception capacity, without adding a single extra hire. The same setup drafts recall SMS copy overnight for a nurse to approve each morning, and prepares a short handover note on each incoming referral so the GP is not reading the full letter cold. Nothing sends automatically. A person still presses send on every message a patient receives.
What has to stay with a person
None of this means handing the front desk to an agent. Clinical judgement, anything requiring a nurse's or doctor's discretion, distressed or complaining patients, and any decision about disclosing health information under the Privacy Act still needs a person to make the call. Claude can draft a message; it should never be the one deciding whether that message is appropriate to send to a specific patient. The practices that get this right treat the assistant as a fast first draft, not an unsupervised decision-maker, and they say so plainly in their own policies and to their patients.
Never let an assistant make a clinical urgency call on its own
Keep a human review step on anything patient-facing for at least the first few months
Log what was drafted by AI versus written by staff, so an audit trail exists if a complaint is raised
Treat the Privacy Act and My Health Records obligations as the floor, not a box to tick once
What this is worth in a real practice
For a two-GP practice, absorbing even a third of front-desk admin into an AI-assisted workflow is typically worth about $20,000 a year in freed reception capacity, enough to add extra consulting hours, reduce receptionist overtime, or simply stop losing good staff to burnout from an impossible phone queue. The rollout that works best is staged rather than immediate. Start with low-risk recall messaging in week one. Add referral inbox triage after a couple of weeks once staff trust the tone of the drafts. Move to Medicare and HICAPS query drafting only once the review habit is properly established across the team. Review the error log monthly rather than assuming it will stay quiet on its own, and keep a standing agenda item at the practice meeting to ask staff what still feels slow.
If your practice in Sydney, Melbourne or anywhere else in Australia is trying to work out where reception hours actually go, that is usually the right place to start, not with a big software project. Book a short call and we will map your specific front-desk workload before recommending anything.
