Tuesday, 12:40. A two-surgery practice on a market-town high street. The nurse is chairside, gloved, suction in hand. The reception phone rings four times and goes to voicemail. At that hour nothing is audible over the handpiece.
By six o'clock there are six missed calls on the display. Nobody looks at them. By tomorrow they are gone.
That is not a failure of organisation. It is the ordinary design of a small practice, where the person who answers the phone also assists, takes payments and deals with whoever has just walked through the door. The failure is a different one: not knowing what it costs.
Nobody publishes that number
We looked properly and could not find it: there is no published UK figure for unanswered calls in dental practices. Not from the profession, not from any sector register.
What there is instead is a cloud of statistics circulating on supplier blogs that fall apart on inspection. "35% of calls go unanswered" traces back to an article resting on "our latest study" — no link, no sample, no method — and it is not about dental practices at all. "Up to 30% of annual revenue" sits on an AI-receptionist vendor's own website with nothing behind it.
And nearly all of them play the same trick: three different problems blended together so the number comes out bigger. Unanswered calls, patients who fail to attend, and new-patient consultations that do not convert have different causes and different fixes. A practice reading that has no idea which one it has.
It is worth the bother because the pressure all points one way. The National Audit Office found 483 fewer dentists doing any NHS work in England in 2023/24 than four years earlier, and NHS hours per dentist fell from 27.3 to 24.8 a week between 2018/19 and 2022/23 — roughly five fewer NHS appointments per dentist per week (Nuffield Trust, February 2025). Demand is being pushed towards private lists, and the bottleneck it meets is a phone nobody can reach.
How to measure it yourself, in a week
You do not need a study. You need four numbers that are already in your phone system.
1. How many calls come in. Export a full week's log. Any hosted phone system gives it in two clicks; if the practice number is a mobile, the handset's own call log will do.
2. How many go unanswered, split into opening hours and out of hours. Out of hours hurts less — nobody was there. In-hours is the uncomfortable column, because somebody was there and was chairside.
3. How many you rang back the same day. Not the next morning. The same day.
4. Your actual new-patient rate. Cross-reference the diary for that week: of the calls you did answer, how many were new patients and how many ended in a booking. There is a "40–60% first-consultation conversion" figure doing the rounds that no article can source. Yours, you can.
The sum is a multiplication of three numbers that are all yours: missed calls never returned × your new-patient rate × what a new patient is worth to you in the first year. That is not a sector estimate; that is your practice. In our experience with service businesses, most owners do not have even the first of those four numbers.
The patient is no longer ringing first
There is a second, quieter loss, visible in what people type into Google: "dental practice taking new patients", "no NHS dentist taking on new patients", "losing NHS dentist". Those are not people browsing. Those are people hunting for a list that is open, which means the first practice to answer wins. If what they find is a web form that replies in two days, they go back to the results page — the same asymmetry we wrote about with contact forms — and the peak lands where it always does: evenings and weekends.
There is an economics point vendors leave vague. Replying to somebody who messages you first costs nothing in Meta fees: non-template messages are free inside the 24-hour customer service window, and Meta only charges when an outbound template is delivered (Meta's pricing documentation). On Meta's UK rate card effective 1 July 2026, an outbound marketing template is £0.0458 and a utility template £0.0159. Chasing costs money; answering the person already looking for you does not, and website chat costs nothing per message at all. Re-check before you sign anything — Meta revises these rates every few months.
What an AI agent cannot do in a dental practice
This is the part you will not find on competitors' websites, and it decides whether this is usable at all.
It does not triage. It cannot tell pulpitis from an abscess with facial swelling, and in an avulsion the window for replantation is measured in minutes. On any urgency signal it escalates to a person and points to NHS 111 out of hours. It does not assess.
It does not diagnose or take consent. Valid consent in the UK rests on a dialogue with the treating clinician about material risks — the standard set in Montgomery v Lanarkshire Health Board [2015] UKSC 11. An implant is not consented to in a chat thread.
It does not price anything that needs an examination or a radiograph. Published NHS band charges are set nationally and it can quote them. Which band a patient will fall into is a clinical judgement made after an exam, and the agent should say so.
It does not take a clinical history. Health data is special category under Article 9 of the UK GDPR, and WhatsApp's own Business Messaging Policy asks, in section 3, that you not use WhatsApp "for telemedicine or to send or request any health related information" where applicable regulations restrict it. The design consequence is simple: the agent collects intent, contact details and an urgency signal. Not symptoms, not history, not medication. And when a patient sends an unsolicited intraoral photo — they do, constantly — it does not read it.
It does not handle complaints or adverse outcomes. Post-operative pain, a failed implant, a request for a refund: human, immediately.
It promises nothing. No "pain-free", no lifetime guarantees, no discount on treatment. Whatever a practice is comfortable putting on a poster, an automated agent puts in writing, to a named individual, with a timestamp. That is a far worse place to be wrong.
And the limitation no supplier admits: answering quickly does not create space in the diary. If the next free slot is three weeks out, replying in nine seconds only moves someone up the queue sooner. This is for not losing the patient who was already calling you, not for inventing capacity.
One point of law worth checking rather than assuming. Article 50 of the EU AI Act, which requires people to be told they are dealing with an AI system, applies from 2 August 2026, and it turns on where the patient is rather than where you are (more here). Whether a UK-only practice is caught by anything equivalent is a question for your own advisers. Ours was simpler: we disclose from the first message regardless, because a patient who thinks they are talking to the receptionist says things they would not put to software.
If the sum comes out ugly
Start by measuring it. Those four numbers take an afternoon.
Zatio is one way of closing that gap: an agent that answers on WhatsApp and on your website chat, identifies itself as AI from the first message, books when booking is the right move, and hands over to a person the moment the conversation leaves its lane. We have two clients and neither is a dental practice, so we are not going to show you dentistry results we do not have. If it looks relevant, the demo takes fifteen minutes. If your sum comes out at three calls a month, you do not need us.