A voicebot in a clinic: what a voice assistant actually does when it answers the phone

One picture keeps coming back in our conversations with clinics. The receptionist is talking to a patient at the window while two lines blink on hold beside her. She can see them. She cannot answer, because there is a person standing in front of her. In the afternoon somebody pulls the missed-call list off the phone system and starts calling back. Some of those people pick up and say they have already booked elsewhere.

This is not a story about poor organisation. It is a story about one person being able to talk to one person at a time.

I am writing this because for several months now we have been building voice assistants that answer calls like these, and we can see that the market sells them nowhere near where the problem actually is. So let me say plainly what such an assistant does, what it does not do, and when there is no point putting one in.

Most of the clinics we talk to have already bought some kind of telephony. A digital exchange with a missed-call log. A helpline with a menu where pressing one goes to public-payer bookings and two goes to private ones. Wait-time measurement. A callback list exported to a spreadsheet.

Every one of those works exactly as promised. And not one of them reduced the load on the front desk, because they all solve the same thing: recording the fact that somebody called. None of them solves answering the call at the moment the patient is calling.

The whole difference is time. A patient who does not get through does not sit down and wait for a callback. They have a list of clinics in the search results and they call the next one. When our receptionist dials their number two hours later, she is doing work and he is already booked.

Which is why the question "how many missed calls do we have" matters less than the question "how many of them did we answer within the first thirty seconds".

The easiest way to show it is a walkthrough, not a feature list.

The phone rings. It answers on the second ring, at any hour, including when every human line is busy. The opening sentence is always the same and says outright that the patient is speaking to the clinic's automated assistant and can ask for a person at any moment. That is not a courtesy, it is an obligation, and I have written about it separately.

Then it asks how it can help. The patient says, in an ordinary sentence, that they would like to see a GP, ideally next week in the afternoon. The assistant checks available slots in the clinic's calendar, offers two or three, takes the choice, confirms the details, books the appointment and sends a confirmation by text.

When the matter falls outside what it can handle, it transfers the call to the front desk or logs a callback with an actual reason, not just a number. The difference between "missed call, 601 xxx xxx" and "she is asking about Tuesday's test result" is enormous for whoever makes that callback.

After the call, a record stays in the system: who called, about what, what was agreed and what still needs finishing by hand.

That is all. No magic. The value is not that the assistant has a pleasant voice. It is that the first half of these calls never reaches a human at all, and the second half arrives described.

Here is the line we do not cross, and I would rather state it early than late, because it decides whether it is worth talking to us at all.

The assistant does not assess anyone's health. It does not ask about symptoms in order to draw conclusions from them. It does not tell the patient whether their case is urgent or can wait. It does not suggest which specialist they should see if they do not know themselves.

It does not decide urgency or order. It does not move anyone up or down a list based on what it heard.

The reason is simple and it is not excessive caution. The moment a system starts assessing health or ruling on urgency, it stops being an organisational tool and becomes something governed by an entirely different body of law and a different approval process. That is a different league of cost and a different league of liability. We deliberately stay out of it, and I think a clinic that is being promised this at the offer stage should ask how the supplier intends to document it.

The assistant books, informs and transfers. That is it. That is enough to take most of the traffic off the front desk.

It also does not replace the practice management system. Every clinic conversation brings up a different name: mMedica, Falkdent, SmartDental, some dental system we have never heard of. Nobody wants to swap them and they are right, because those systems do things that are invisible from outside, such as settlements with the public payer. A voice assistant stands in front of such a system, never instead of it. Integrating with what the clinic already has is the core of this work, not an add-on at the end of the rollout.

We have seen enough voice assistant demos to know that a smooth conversation on stage says nothing about what happens at the front desk on Monday morning. Four things decide.

Booking into a real calendar. If the assistant cannot see actual doctor availability and does not write the appointment where the front desk will later see it, then it is not an assistant, it is a secretary recording notes. Somebody will retype it by hand anyway, which means the work has been moved, not removed. This is the same choice I described when writing about booking portals versus your own system: a calendar you do not control limits everything you can build on top of it.

A way out to a human in one sentence. The patient has to be able to say "I want to speak to the front desk" and get there immediately. An assistant that resists handing over the call generates complaints faster than it saves time.

Honest hours. The assistant can answer around the clock, but the clinic does not operate around the clock. You have to decide up front what happens to a call at ten in the evening: what the assistant may close on its own and what it only logs for the morning.

A trace of every call. A transcript or a note in the system, available to whoever speaks to that patient next. Without it, the first complaint comes down to one word against another.

I am writing this because it is the most common worry we hear, and it is usually justified.

Some patients will not use it and there is no point talking them into it. The older lady who has been calling for fifteen years and knows the receptionist by name should call and speak to a person. Unusual matters, test results, prescriptions, complaints, situations where the patient is upset, all belong to people.

The real effect is not that the phone stops ringing. It is that the front desk stops running three conversations at once. You see it most in the evenings and at weekends, which is when nobody used to pick up at all.

A few situations where we honestly advise against it.

A single-practitioner clinic with a calendar full three months ahead does not have the problem this solves. The phone rings rarely there and every call is short.

A clinic that does not know how many calls it misses is not ready either, because it will have no way of checking whether the rollout changed anything. That one is fixable in a week, more on which below.

And the hardest case: a clinic whose calendar is kept in a way no machine can read. A paper book, a spreadsheet on a shared drive, arrangements in the head of the head receptionist. There the first step is not a voice assistant, it is putting the calendar in order. Doing it the other way round ends in an expensive disappointment.

One thing you can do yourself, worth more than any offer you will receive.

Pull two reports from your phone system for the last month: incoming calls broken down by hour, and the share of missed calls in those same hours. Almost every exchange records this, even if nobody has ever opened it.

You will probably see two peaks, one in the morning and one in the afternoon, and within them a missed-call share higher than anyone at the clinic believes. That is one number, but it is the number that says whether there is anything to discuss at all, and it is the one you come back to three months later to check whether anything changed.

If you have no way of getting at it, write to me and we will point you to where it hides in your particular exchange. That is not a service, that is ten minutes.

Plenty of people can stand up a voice assistant today. Far fewer will answer the question of where the recording of a patient call is kept, on what legal basis you process it, how long you hold it and who has to know they were talking to a machine. We have written about that separately, because for a medical practice it matters more than voice quality.

A voice assistant is also only one element of the layer that sits between the patient and the practice management system. The rest, meaning online booking, reminders and reviews, is described on our page for healthcare, and what the tiers of a clinic website actually contain is set out in what a clinic website really costs.

We build these systems and we can show you one working before anything is agreed. There is not one person on our side: I usually run the scoping conversation, the integration and rollout are done by the team, and the clinic gets contact details for actual people rather than a shared inbox. If you would like to hear what such a call sounds like from the other end, write to me.