Healthcare workflow automation: examples for clinics
Most clinic owners don't have good sales and marketing numbers.
Practice software is good at one thing: what happens once someone's already a patient. It goes blind everywhere before that point. And the reconciled financial side, what actually lands in Xero or MYOB, sits apart from all of it.
That gap is where a lot of workflow automation for healthcare pays off. It's not the only one either. Healthcare workflow automation is software doing a repeatable admin task for your clinic on its own, the same way every time. These are real examples of it, sized to a small AU or NZ clinic rather than a hospital. And they skip the invented case-study numbers a lot of vendor blogs lean on.
In brief: healthcare workflow automation is a fixed step that runs the same way every time. It shows up everywhere, from an after-hours enquiry logged at 9pm to the Xero and ACC reconciliation at month end. How much it saves varies a lot, from hours a week with a scribe to about five hours every fortnight from a synced system. None of it needs a hospital-scale build, just the steps written down properly.
What counts as automation, quickly
Read our full breakdown of AI vs automation if you want the whole argument. Short version: automation follows a rule, the same way, every time. AI reads something messy and makes a judgement call.
A US federal health-technology agency looked at where automation actually works across industries and boiled it down to a short test. A task is worth automating when it involves manual data entry, happens often, and has clearly defined inputs and outputs. It also needs clear roles for who does what (Zayas-Cabán, Haque and Kemper, 2021). Everything below gets checked against that test.
The referral system I built
The clearest example I've got is a referral system I built for my own clinic in New Zealand. It's the one that took the most hacking together.
It started because I wanted to improve retention and bring in new business. A referral-based incentive is a pretty standard recommendation for practices: someone has a good experience, refers a friend, and gets something for it.
The practice management software I used had no built-in way to do this. One tool I tried was too hard to use to bother with. So I built it myself.
The software had an open API. That's a door the software leaves open so other tools can read and write its data. I think I used Zapier at the time, connecting it to a spreadsheet in Google Workspace. It pulled in appointment data for some practitioners, not everyone. Each record got its own ID, and I logged the criteria that decided whether the automation should fire.
When those criteria were met, it triggered an email (I think it was Gmail), sent from the business account so it looked personal rather than automated. Attached was a referral card I'd designed myself. Whoever received it could show it in clinic as proof they'd been referred.
The practice software had a manual field for how someone found us. When a new patient came in through a referral, that field was updated by hand. That fed back into the spreadsheet, so the referred patient got tagged and received the same offer. I also built in a gap between emails, so referrers weren't getting one after every single session. They could also refer more than one person.
The bit I never solved
I ran it for six months and adjusted it three months in, based on what I was seeing. The benefit, I believe, was a discount following treatment. One thing I never worked out, and probably should have, was how much it cannibalised the referrals we were already getting through plain word of mouth.
Check this before you copy it
It ran in New Zealand.
AHPRA doesn't have a rule that names referral schemes. It treats a refer-a-friend discount as an ordinary gift, discount or inducement. That's allowed if the ad clearly states the terms and conditions (s.133(1)(b)). It's against the rules if it nudges someone toward treatment they don't clinically need (s.133(1)(e), AHPRA, Guidelines for advertising a regulated health service, § 4.5). AHPRA's own example is a discount that only pays off if someone books more treatment than they actually need.
In New Zealand, it's the Osteopathic Council that sets the rule. It allows an inducement only when the terms are clear and the patient's interests come first. It also strongly discourages coupons and gift certificates altogether (Osteopathic Council of New Zealand, Medical Advertisement Policy, September 2016).
So a scheme like this can work in Australia, on those two conditions. Check your own board's code too. AHPRA's isn't the last word on it. What I built worked for my clinic. Use it as a starting point, then shape your own version to fit your rules.
One more thing on that system: patient information was moving into a spreadsheet and an email account. That's exactly the kind of thing privacy law has something to say about. We've written separately about where privacy rules apply to any tool that touches patient data.
Where AI actually helps: notes and letters
Not everything on this list is a plain rule. Writing up a consult, or turning notes into a referral letter, needs judgement. Judgement is what makes something AI instead of automation. This is where ai automation in healthcare actually earns the name.
A May 2026 case study looked at two Australian general practices, published in the RACGP's own journal. One practice cut a correspondence task from three minutes to about ten seconds per letter. Across its whole AI system, the same practice estimated it had freed up about half a full-time reception role's worth of time each day. That time moved back into patient-facing work (Stevens, Mollenkopf and Foo, Australian Journal of General Practice, May 2026). Two of the three authors disclose ties to the AI vendors named in the study. One is a clinical advisor to Lyrebird Health. Another is Chief Medical Information Officer at Heidi. Two different vendors. Treat the numbers as real, from a real practice. Don't treat them as a ceiling every clinic should expect to hit.
From my own seat, speech-to-text tools help a lot here. That's a scribe used during a patient consult, or an internal one for writing up notes afterwards. I use Wispr Flow for the second kind myself. We sell no software and take no commission on anything we suggest, including that one. A letter built from notes is really just one text turned into another. That's exactly the kind of task AI is good at speeding up, a lot.
Run the same test on the rest of the list
Run everything else through the same test from earlier. If every step is fixed and repeats often, it's automation. If a step needs someone to read something messy and decide, that step is AI. None of it needs a hospital-scale build. This is clinic automation at the scale a small practice actually runs at.
| Workflow | What kicks it off | What it replaces |
|---|---|---|
| After-hours enquiry reply | An enquiry lands outside opening hours | Sitting in the inbox until morning |
| Appointment reminder | The day before an appointment, and again the morning of | Someone remembering to send it |
| Recall | A set time after someone's last visit | The same, forgotten |
| Booking-to-system entry | A patient fills in a booking form | Reception typing the same details twice |
| Contractor payout | The end of the month | Adding it up by hand |
| Xero and ACC reconciliation | Invoices and payments land in Xero | Reconciling the books by hand |
Front of house
An enquiry that lands at 9pm gets logged and gets a first reply straight away, instead of sitting in an inbox until morning. Speed matters more than people think. That's why we put an auto-responder on by default, the message that fires the moment someone gets in touch, before anyone's read it. Get the enquiry handed to someone fast. Aim to have that person on the phone within the hour. We've written more on this in our guide to AI receptionists for clinics.
Reminders and recalls run the same way, without anyone remembering to send them. The simplest version: an email and a text the day before an appointment, then again the morning of. Recalls fire at a set time after someone's last visit, on the same principle. Our guide to conversational AI in healthcare covers what a good reminder sequence looks like.
Behind the desk
Patient details captured once in a booking form can flow straight through to the practice management system. So reception doesn't type the same information in twice. The referral system above is the more involved version of the same idea. I built something similar to work out what I owed contractors each month, mostly automated.
Money and reporting
I connected Xero to my clinic and set up ACC invoicing (New Zealand's accident compensation scheme) alongside it. That gave me a complete, reconciled picture of the finances, and let me forecast off it too.
What this looks like outside the clinic
None of this is only a clinic thing. In sales work with Splose and HubSpot, I've built the same shape of automation. Meeting notes land in the CRM on their own, and a scheduled review flags gaps or risks on open deals. I've worked with both, and I'm saying so upfront.
The real numbers, and how Opal starts by mapping your time
It depends on the task, so I won't pretend there's one number for it. Take a scribe, for example: someone going from no scribe to using one saves hours and hours a week. Someone setting up a two-way sync between accounting software and practice software might save their team something more like five hours every two weeks. Both are real, and both are automation or AI doing very different amounts of work for very different reasons.
If you want help working out which task on your own list is worth tackling first, that's a different question. We answer it in which admin tasks to automate first.
Where we help: AI workflow automation starts with mapping where your team's time actually goes, then builds the automations that make sense for what we found there.
FAQ
What's the difference between automation and AI in a clinic?
Automation follows a rule the same way every time. AI reads something messy, like a consult note or a referral letter, and makes a judgement call instead. A task is worth automating when it involves manual data entry, happens often, has clearly defined inputs and outputs, and has clear roles for who does what (Zayas-Cabán, Haque and Kemper, 2021).
How did the referral system I built for my own clinic work?
My practice software had no built-in way to run a referral scheme, so I connected it through an open API, a door the software leaves open for other tools to read and write its data, using Zapier to link it to a spreadsheet in Google Workspace. When a record met the criteria I had logged, it triggered an email from the business account with a referral card I had designed myself. A manual field in the practice software flagged when a new patient came in through a referral, which fed back into the spreadsheet so they were tagged and received the same offer.
How much time does workflow automation actually save a clinic?
It depends entirely on the task. Going from no scribe to using one can save hours and hours a week. A two-way sync between accounting software and practice software is a smaller win, more like five hours every two weeks. Both are real, and both are automation or AI doing very different amounts of work.
What should happen when someone contacts the clinic after hours?
It should get logged and get a first reply straight away, instead of sitting in an inbox until morning. An auto-responder is the message that fires the moment someone gets in touch, before anyone's read it. From there, aim to have a person on the phone with them within the hour.