AI for medical practices · Sydney

AI for medical practices: a back office that keeps pace with the appointment book

In most practices the clinical side is well run and the back office is held together by the practice manager and two people on reception. Referral letters scanned and filed by hand, rejected claims worked through one at a time, recalls chased from a list. I help GP and specialist practices turning over one to ten million take that load off, privately, using the systems you already run.

Where a practice’s week actually goes

Ask a practice manager what fills the day and it is rarely the patients. It is the inbox of incoming letters and results that each need filing against the right record, the Medicare and health fund rejections that come back with a code nobody remembers, the recall list that is always a fortnight behind, and the roster that has to be rebuilt every time a doctor changes a session.

None of it is clinical, and none of it needs a doctor. That is exactly the work AI is now good at: reading a document, working out where it belongs, drafting the routine reply or the reminder, and putting a short list in front of a person to check. I work on the back office only. Diagnosis, treatment and anything a clinician signs stays with the clinician.

This is not a new practice system. You keep Best Practice or MedicalDirector, HotDoc or HealthEngine for bookings, Tyro for claiming and Xero for the books. The job is to connect AI carefully to what you already run, so the admin is drafted for your team and they check it instead of doing it from scratch.

What I’d take off your plate first

Usually Best Practice or MedicalDirector for clinical records and billing, HotDoc or HealthEngine for bookings, Tyro for Medicare and health fund claiming, Xero for the books, and Microsoft 365 or Google Workspace for everything else. I connect AI to those, rather than asking you to replace them.

Incoming correspondence, sorted

Specialist letters, discharge summaries and results that arrive by fax, email or secure messaging read for patient and type, then queued for filing against the right record. Reception confirms, rather than opening every attachment.

Recalls and reminders

Recall lists pulled from the practice system, reminders drafted in your wording, and a weekly list of patients who have not booked, so follow-up happens on time rather than when someone finds an hour.

Medicare and billing rejections

Rejected and unpaid claims gathered from Tyro and the practice system, each one explained in plain English with the likely fix, so the person reworking them starts from an answer rather than a code.

Supplier invoices and accounts payable

Medical supplies, pathology courier, cleaning and IT invoices read, coded and checked against what was ordered before they reach Xero. Duplicates and price changes flagged, with the query drafted.

Rostering and session changes

Doctor and nurse sessions, leave and locum cover tracked in one place, with the roster draft, the booking-system changes and the staff message prepared for the practice manager to approve.

Practice reporting and owner updates

Billings per doctor, appointment utilisation, debtors and recall rates turned into a one-page monthly report with a sentence on what moved, instead of an afternoon of exports and spreadsheets.

And the admin every office has

The same work turns up in every practice: email, spreadsheets, documents, and a Microsoft or Google setup that grew without a plan.

Invoicing and accounts payable

Supplier bills read, coded and matched to the order or the job before they reach Xero or MYOB. Mismatches drafted into an email for someone to approve. Overdue invoices chased politely, on a schedule.

Email that answers itself, nearly

The inbox triaged each morning, routine replies drafted in your tone, attachments filed to the right folder, and a short list of what actually needs you today.

Spreadsheets you stop rebuilding

The weekly report that someone copies and pastes together every Monday, produced before you arrive, with a sentence on why the number moved.

Documents and slides

Proposals, letters, reports and board slides drafted from your own templates and your own past work, so the first draft is eighty per cent there rather than a blank page.

Microsoft 365 or Google Workspace, set up properly

Shared drives that make sense, permissions that match who you are now, and AI connected to mail, calendar and files the same way for everyone instead of five people doing it five ways.

Project and job tracking

Status updates written from what actually happened in email and the job system, tasks pulled out of meeting notes, and a morning brief per person of what is due.

How it works: audit, roadmap, implementation, training, support

Sold one step at a time. Each step is quoted before it starts, stands on its own, and you decide after each one whether to take the next. Most businesses your size go from the audit to a working, trained setup in three to six months.

  1. Audit

    Two weeks, fixed price. A map of every system you run, a ten-minute staff survey, a few five-minute screen recordings of the work that eats the week, and short conversations with the people who run the business. It costs your team minutes, not hours.

  2. Roadmap

    A ninety-day plan written from what the audit found: which processes first, ranked by admin removed against cost and risk, which AI plan and accounts your team should be on, and a two-page policy on what may go in. Every item has its own price.

  3. Implementation

    I build the first one or two automations inside your own Microsoft 365 or Google Workspace, connected to the systems you already run. AI drafts, a person approves anything that leaves the building, and it runs beside the manual way until you trust it.

  4. Training

    Your team learns on their own work, not on demos. One standard setup for everyone, a short session per group, and one-to-one time with the two or three people who will carry it once I step back.

  5. Support

    A light monthly retainer: one call, a one-page summary, fixes, and the next routine your team worked out promoted into something the business runs. I read the AI news so your people do not have to.

Use AI without handing over your clients’ data

Health information is the most sensitive kind of personal information under the Privacy Act, and a practice holds a great deal of it: records, results, referral letters, Medicare numbers and billing history. None of that should go anywhere near a consumer chatbot, and none of it needs to for AI to help with the back office. Where a task can be done on de-identified data, such as reporting, it is.

The work happens inside your own practice systems and your Microsoft or Google tenant, on a business AI plan with company accounts for every person who uses it. If you already use an AI scribe, I make sure it sits inside the same policy, with patient consent handled the way your doctors already agree. Anything that goes out to a patient, a specialist or a fund waits for a person to say yes. I am not a lawyer and I do not certify compliance, but I will make sure the setup is one your own adviser can review easily.

The basics, every time

  • Company accounts on a business AI plan, never personal logins. Business plans do not train on your data.
  • AI connected to your own Microsoft or Google tenant, so files stay where they already live and access is revoked in one place.
  • A two-page policy your team can actually remember: which accounts, what may go in, what never does.
  • Everything I build lives in your accounts and is logged. I work from a staff login, not the master key.

Where most start

A typical first project: incoming correspondence sorted for filing

Most practice audits find the same quick win. Every incoming letter and result is read, matched to a patient and a document type, and queued for reception to confirm and file. Clinical review of what arrives stays exactly as it is now. It runs alongside the manual process for a fortnight until you trust it. Recalls and billing rejections are usually next.

Frequently asked questions

Will AI be making clinical decisions?

No. My work is the back office: correspondence, recalls, billing, rostering, suppliers and reporting. Anything clinical, including what a result means or what a patient needs, stays with your doctors and nurses. Where AI touches a clinical document, it is only to work out where it should be filed, and a person confirms it.

We use Best Practice (or MedicalDirector). Do we need to change?

No. I work with the practice system you have. Where it offers a proper way in, I use it. Where it does not, I start from the reports and exports your team already pulls, so nothing about your clinical setup has to change.

Is patient information safe with AI?

It can be, if it is set up properly, and that is most of the first conversation. Business plans that do not train on your data, accounts tied to your own tenant, de-identified data wherever the task allows, and a person approving anything that leaves the practice. Free personal chatbots are off the table entirely.

Is this worth it for a practice our size?

If your practice manager or senior receptionist spends hours a week on filing, rejections and recalls, usually yes. The audit will tell you honestly, and if a tidier template in your practice system would do the job, I will say so.

How much of my team’s time does this take?

Minutes, by design. Ten for the survey, ten for an optional screen recording, an hour with the key people. Whoever runs operations spends two or three hours across the first fortnight.

Do we have to commit to the whole thing?

No. Every step is quoted on its own and you decide after each one. Plenty of businesses take the audit and one project, and stop there until they see it working.

Are you tied to a particular AI vendor?

No. I lead with Claude because I think it is the strongest all-rounder for business work today, and I work just as happily with Microsoft Copilot, ChatGPT or Gemini if that is what your team already uses. Everything is built so the model is one part you can swap.

Questions medical practices ask me

Do I need an AI policy for my small business?

Yes. If your team uses AI at work, and most already do, you need a short written policy. Two pages is enough: which accounts to use, what may and may not go in, who checks AI output before it leaves the building, and who to ask.

Does ChatGPT or Claude train on my business data?

Not on their business plans: ChatGPT Business and Enterprise and Claude Team and Enterprise do not train on your data by default. On free and personal plans they can, unless each person turns it off, which is why business use belongs on a business plan.

Is it safe to put client data into ChatGPT or Claude?

On a business plan with company accounts, it can be: ChatGPT Business and Enterprise and Claude Team and Enterprise do not train on your data by default. On a free or personal account, treat it as unsafe for client information, which is also what Australia’s privacy regulator recommends.

Can my boss read my AI chats?

On a work account, assume yes if the business has a reason. Day to day there is usually no screen where a manager reads your chats, but on Claude Team the organisation’s primary owner can export its data, conversations included, and larger plans add compliance tools built for exactly that.

My staff are using their own AI accounts. Should I ban it?

No. A ban rarely stops it and always hides it. Give your team one approved AI tool on a company account, write down what may go into it, and ask people to move across. Keep one firm rule: company logins and client information never go into a tool the business has not approved.

Running a practice and doing the admin at night?

Tell me what you run, how many people, and what eats the most time. I will tell you honestly whether an audit is worth it.

Tell me about your business

Tell me what you're dealing with

Three lines is plenty. I reply personally, usually within a day, and the first call is free.