AI in a dental practice: missed calls, empty chairs, recall reminders

Half past ten in the morning and reception has one pair of hands: the patient in the chair is settling up, a new patient is standing at the desk with a form, and the phone is on its fourth ring.

That call was a new patient wanting a clean. It rang out at six and they rang the practice two streets away instead. That afternoon you look at the book and two gaps are sitting there from people who cancelled this morning.

Look back further and the patients who said they'd come back in six months are still sitting in the system. Nobody has made that call.

Jobs that don't need a person

Six admin jobs that happen every day. First line is now, second line is once it's running.

01The calls reception misses when it's flat out

NOWThe busiest part of the day is also when the phone rings most. Three rings with nobody on it and they call the next practice. Calls over lunch, after hours and on the weekend may as well not have happened.

AFTERA number that answers around the clock: gives the practice name, says up front that it's an AI assistant, asks what the call is about and takes a callback number, then sends you the summary. It doesn't assess anything clinical and it doesn't quote.

02Gaps left by last minute cancellations

NOWOne message in the morning saying they can't make it and that afternoon's chair sits empty. Reception wants to fill it, can't remember who was waiting, and then the day gets away from them.

AFTERThe moment a gap opens, the people on the waitlist who wanted that slot hear about it: "we've been keeping an eye out for you, Wednesday afternoon has come free". Patients who often cancel late get an extra confirmation step. A person still presses send.

03Recalls and routine check ups

NOWRoutine check ups, post treatment follow ups, ortho reviews. All of it runs on reception's memory and a sheet on the wall. Busy week and it stops, and once it stops nobody picks it back up.

AFTERBased on when each patient last came in and what was done, a who to contact list every week, with a real reason next to every name. Who's been contacted, and who has said don't message me again, is visible.

04The same twenty questions every day

NOWWhere to park, whether they need a referral, how long a first appointment takes, whether they can claim on their health fund, are you open Saturday. None of it needs clinical judgement, and three staff will still give three answers.

AFTERYou and reception agree one set of wording, and after that the phone, the website form, Google Business messages and social DMs all say the same thing. Change it once and it changes everywhere. Anything touching symptoms or fees goes to a person.

05Google reviews, the shopfront everyone sees

NOWYour happiest patients pay and walk out without leaving a word. One bad review sits there for three months with no reply, and it's the first thing a new patient sees when they search you.

AFTERAt the moment a patient is happy, a review request is drafted with a detail from that day in it, with no mention of discounts or giveaways and nothing about their treatment. A bad review comes in and a calm reply is drafted within 24 hours for you to look over.

06New patient enquiries spread across four inboxes

NOWWebsite form, Google messages, social DMs, chat apps, all on different phones and in different people's hands. Whoever sees it first replies, which often means nobody does.

AFTERThe four inboxes land in one place, and each one gets a decent first reply quickly, with a record kept. What you see at the end of the day is one list: who asked, what they asked, and which ones still have nobody on them.

What your week looks like once it's running

It isn't that everything runs itself. It's that everything finally has someone on it.

Mon

One list on your phone: 11 appointments today, two that need confirming early, and summaries of the four weekend calls nobody answered.

Tue

Someone cancels in the morning. Two people on the waitlist who wanted Tuesday morning hear about it, and the afternoon slot gets filled.

Wed

Nine names on the recall list, each message carrying what was done last time. Two you want to ring yourself, reception sends the rest.

Thu

A four star review mentions a twenty minute wait. A reply is drafted that doesn't argue and asks them to get in touch privately.

Fri

Next week's book, how many calls came in this week, and who never heard back. One page.

What AI can't do

This isn't a disclaimer. It's the condition for this lasting in a practice at all.

Any diagnosis, medication or treatment advice. A patient asks where it hurts and whether it needs pulling, or whether they should start on antibiotics first. The machine does not engage at all. It takes their own words down and passes them to the clinician. Not even half a sentence like "sounds like an infection".

How health information is stored and who can see it. In Australia health information is sensitive personal information and the rules are tighter than for ordinary customer records. What can go into a summary, how long it's kept and who can see it is set by you and your adviser under your existing privacy policy.

Calls on fees and treatment plans. Quotes, payment plans and what a health fund will cover are your judgement. Pushed for a price on the phone, the answer is always the same one line: someone will call back with an accurate quote.

Reschedules, cancellations and complaints that touch treatment. It builds the list and writes the draft, but nothing moves in the diary without a person saying yes. Book someone in twice and they're gone that week. Any complaint touching the treatment itself goes to a person immediately.

Don't guess which one to do first

You don't do all six. For most practices only one or two of them are actually costing money. The check asks 13 questions in practice language, takes 5 minutes, scores on the spot and tells you the three to do first.

Take the free check · result in 5 minutes Or have a look at what the academy actually installs, and what it costs

What practice owners ask most

A patient rings with a sore tooth asking if it needs pulling. Will AI just make something up?

No. It's boxed in to record and never assess: get their name, take down their own words, take a callback number, then say someone will ring them back shortly. For the first fortnight after it goes live, spot check the summaries, the same as you would with a new receptionist.

Reception already runs a booking system with reminder texts. Do we still need this?

The system's reminders treat everyone the same. Booked, so it sends. What this adds is three things the system doesn't do: an extra confirmation step for patients who often cancel late, filling a gap from the waitlist, and reaching people in their own language. The two run side by side without stepping on each other.

If patients can tell it's AI on the phone, will it put them off?

What puts people off has never been AI answering the phone. It's AI pretending to be a person and getting caught, which is why it says what it is straight away. The real choice in front of the patient is an assistant that answers on the second ring, or a phone that rings out.

Half our patients speak English and half don't. Can it cope?

One set of answers, spoken in either language, whichever the caller uses. Even when it can't follow, the floor is that the number and the reason go into the summary so a person can ring back.

Will patient information be used to train models?

Ask before you install anything: where it's stored, how long it's kept, whether it's used for training. Treat it as a hard requirement and rule out anything that can't answer. The bar in a practice is not the same as the bar in other industries.