A confirmation call earns its money when the patient says they will not come. Everything else in the process, from the day the calls start to the wording of the first sentence, should be designed around hearing that answer early enough for the clinic to give the slot to someone else.
That sounds like a play on words, and in the first meetings with a clinic it often lands as one. It is not. A process built to collect yeses looks different from a process built to find the noes: it starts earlier, it tries harder to reach the people who do not pick up, it treats an uncertain answer as a risk rather than a success, and it reports cancellations captured in time rather than confirmations obtained. We have run appointment processes for healthcare providers and insurers for years, and this one reframing does more for the result than any choice of voice or model.
We call it appointment confirmation. Internally we call it cancellation, because that is the business goal: learning early that a visit will not happen. – Dawid Wójcicki, Founder & CEO of Vee
The asset is the slot, not the appointment#
Think about what a clinic actually sells. Not appointments, but time: a specialist's hour, an imaging room, a procedure theatre with its staff and consumables already paid for. An appointment is a reservation on that time. When the patient turns up, the reservation turns into revenue. When the patient does not, the time is gone and cannot be stored for later.
The patient who did not come is only half of the loss. The other half is the patient who would have come, would have paid and was told there was nothing available. In private healthcare that second patient may well book with a competitor and stay there. So a no-show costs the visit that did not happen, the visit that could have happened instead, and sometimes the relationship that walked away while waiting.
Seen this way, a patient who confirms is useful information but changes nothing: the slot was going to be used anyway. A patient who cancels three days in advance is worth a great deal, because three days is enough time to offer the slot to the waiting list. A patient who cancels at eight in the morning on the day is worth much less. A patient who simply does not appear is worth nothing at all. The value of the information decays with time, and that decay is what the process has to beat.
What clinics are actually up against#
The scheduling problem is not unique to any one country, and the clearest recent numbers come from the United States. MGMA, the association of medical group practice leaders, polls its members regularly through MGMA Stat. Two of those polls frame the problem well, with the caveat that they are short practice-leader polls, not studies of patient behaviour.
| Poll | Finding |
|---|---|
| Patient access priorities for 2026 (poll of 9 Dec 2025, 236 responses) | Top patient-access focus for 2026: no-shows 27%, online scheduling 24%, phone access 22%, wait times 21% |
| Phones are still a bottleneck (poll of 10 Mar 2026, 294 responses) | Most time-intensive staff phone tasks: eligibility and prior authorisation 45%, scheduling 31% |
MGMA Stat, 10 Dec 2025 and 11 Mar 2026. Single-question polls of US practices. The March 2026 poll was sponsored by a company that builds AI clinical agents. Priorities, not measured no-show rates.
Two things stand out. First, no-shows top the list of what practice leaders want to fix, ahead of the digital tools that get most of the attention. Second, the phone is where the work piles up. Almost a third of respondents named scheduling among the most time-intensive phone tasks for their staff. The same people who are supposed to chase confirmations are already busy answering the phone, checking eligibility and rebooking.
In our work with European clinics, public and private, the picture matches this. Many confirm by text message only, have staff call when they have a spare moment, use an automated line that patients learn to ignore, or do nothing at all. Calendars are often not integrated across sites. Some private practices penalise patients who miss visits; others quietly overbook, putting several patients on one slot and accepting that someone will wait or be turned away. None of this is a technology problem first. It is a process that nobody has had the time to design.
Why staff calls and text messages fall short#
The usual alternatives each fail in a specific, predictable way. It is worth naming them precisely, because the design of a better process follows from the failure modes.
- Text messages only. A text reaches the phone but not necessarily the person. Response rates are lower than for a conversation, and a patient who ignores the text is indistinguishable from one who intends to come. A reply of 'OK' at midnight tells you little about the morning.
- Staff calling when they can. Reception staff call between other tasks, usually the day before. A person will rarely make three or more attempts to reach a patient who does not pick up, and nobody should expect them to: the queue at the desk is real and the unanswered call is not.
- A legacy automated line. Press one to confirm, press two to cancel. Patients who want to reschedule, ask a question or say 'probably' have nowhere to go, so they hang up. The process records silence, which is worse than no data because it looks like data.
- Overbooking. It protects the clinician's time at the cost of the patient's. It also hides the real no-show rate, so nobody can measure whether anything else is working.
All four share one weakness: they find out late, or not at all. The design goal is the opposite. Find out early, find out from almost everyone, and treat every unclear answer as something to resolve rather than something to file.
Timing: earlier than the day before#
The single most important design decision is when the first call goes out. Calling the day before is a habit inherited from staff workload, not from the economics of the slot. By the afternoon before, there is little time left to offer the slot to a patient on the waiting list, contact them and get a yes. For an expensive procedure or a specialist with a long queue, that is exactly the slot the clinic can least afford to lose.
So we call earlier. The schedule is agreed with each provider and depends on the type of visit, the length of the waiting list and how far ahead slots can realistically be refilled. Whatever its exact shape, it has to do five things:
- Early first contact. Early enough that a cancellation leaves real time to refill. The patient is asked whether the visit still suits them, with an easy way to say no or to move it.
- Retries for everyone not reached. Patients who did not answer are tried again within the agreed windows. A patient who did not pick up is not a confirmed patient.
- A follow-up before the visit for anyone still unreached, and for anyone whose earlier answer was uncertain: 'I think so', 'I'll have to check with work', 'probably'.
- A text message as fallback where calls have not connected, so the patient still gets the reminder and a simple way to respond.
- Daily status for the clinic: who confirmed, who cancelled, who moved, who is still unreached, ready for the reception team or the booking system to act on.
The schedule is a trade-off, and we say so openly. Call too early and patients have not yet thought about their week; some will confirm and later change their minds. Call too late and the slot cannot be refilled. That is why the follow-up for uncertain and unreached patients matters so much: it catches the people the early call could not settle, while the early call has already captured the clear cancellations with time to spare.
Hours, not days: running the whole campaign at once#
The second design decision is speed. A reception team works through a confirmation list one call at a time, between patients. An AI call centre does not have that constraint. Because the conversations run in parallel, a confirmation campaign for the coming days can be completed within a few hours, including repeat attempts for everyone who did not answer the first time.
Speed matters for a reason beyond efficiency. The sooner the whole list has been worked, the sooner the clinic has a complete map of risk for the period ahead, and the more time it has to act on it. A cancellation discovered at ten in the morning, three days out, can be offered to the waiting list before lunch. The same cancellation discovered by a receptionist on the afternoon of the day before is usually lost.
Parallel calling also changes what persistence costs. For a person, the third and fourth attempt on a patient who never picks up is the least rewarding work in the building. For us it is part of the schedule. Retry intervals depend on the outcome of the previous attempt and on what the patient asked for. All of this sits inside agreed calling windows, so persistence never turns into pestering.
Designing a conversation that makes 'no' easy#
If the goal is to find cancellations, the conversation must make it comfortable to cancel. That sounds obvious and is routinely ignored. Many confirmation scripts are written to elicit a yes: 'We look forward to seeing you tomorrow at ten, is that right?' A patient who is unsure, embarrassed or simply polite will say yes and then not come.
A cancellation-oriented conversation does a few things differently. It introduces itself clearly, including that the caller is an AI assistant calling on behalf of the clinic; Article 50 of the AI Act has required that kind of transparency since 2 August 2026, and we have always worked that way. It states the visit, the date and the place, and then asks an open question about whether the time still works, with rescheduling offered as a normal option rather than a failure. It listens for hesitation and treats 'I think so' as a status of its own, not as a confirmation. And it is short: the patient should be able to cancel or move the visit in well under a minute.
- Confirmed: clear yes, no hesitation. Nothing more to do.
- Cancelled: slot released immediately for refilling.
- Rescheduled or wants to reschedule: moved within the agreed rules, or passed to the clinic with the patient's preferences.
- Uncertain: flagged for a follow-up before the visit, and visible to reception as a risk.
- Not reached: retried within the schedule, then covered by a text.
These are the business statuses the clinic sees, in its own language. Behind them sit technical statuses (connected, voicemail, busy, invalid number) and dialogue statuses (what was said). Keeping the layers separate is what lets a clinic act on the result the same morning without listening to recordings.
Serving both sides of the appointment#
Cancellations do not only come from patients. A doctor falls ill, a machine needs servicing, a session is moved. In those cases the clinic needs to reach a list of patients quickly, tell them the visit cannot go ahead and offer an alternative. That is the same process run in the other direction, and it benefits from the same speed: patients told early are far more forgiving than patients told at the door.
The process also extends naturally beyond confirmation. Once the scheduling logic and the integration with the calendar exist, the same service can book new visits, move existing ones and remind patients about recurring care that is easy to forget: dental hygiene, orthopaedic insoles, follow-up visits, screening such as mammography. These calls are not about filling a gap in tomorrow's calendar but about keeping patients in a care pathway they already chose. For the provider, they are also a quiet way of protecting future capacity from the same no-show problem.
One more dimension matters in some health systems: patient ratings. Where the payer relationship depends partly on how patients rate their care, the experience of being reminded, rescheduled and kept informed is not a side effect. A clinic that tells patients early about a cancelled session and offers them a new time is building exactly the kind of rating it needs.
What to measure, and what not to claim#
We use the same four-rung ladder for appointments as for every other process. The first rung is reach: did we get through to the patient at all? The second is continuation: did they stay in the conversation after hearing who was calling? The third is the key information: a clear status for the visit, with cancellations captured and the time at which they were captured. The fourth is the business effect: attendance, slots refilled, revenue retained. We influence the fourth rung only indirectly, because refilling a slot depends on the clinic's waiting list and booking workflow, not on us.
Two practical consequences follow. A dashboard that reports confirmation rates is measuring the least valuable outcome. The report that matters shows cancellations captured, how far ahead of the visit they were captured, how many slots the clinic managed to refill and how attendance moved against a comparable period or a comparable group of clinics. And because the fourth rung lives in the clinic's own systems, the comparison should be built there, on the same specialties and the same types of visit, rather than taken from a supplier's slide.
What to ask before you automate confirmations#
If you run scheduling for a clinic network, a diagnostics provider or a medical insurer, these are the questions that separate a process designed to find cancellations from a reminder service with a voice:
- How many days before the visit does the first call go out, and why that day for this type of visit?
- How many attempts are made for a patient who does not pick up, at what intervals, and inside which calling windows?
- How is an uncertain answer handled? Is it a separate status, and does it trigger a second call?
- How quickly can the whole list for the coming days be worked, including retries?
- Can the patient reschedule in the same call, and how does that reach our calendar?
- What does the daily report show: confirmations, or cancellations captured and how early?
- How will we measure the effect on attendance and refilled slots, and against what comparison?
- Who listens to the calls, how often, and what happens when they hear something the statuses missed?
The answers reveal whether a supplier understands the business of a clinic or only the mechanics of a call. In our experience the hard part is not making the phone ring. It is agreeing the schedule, the statuses and the release rules with the people who run the calendar, and then listening, week after week, to the calls that did not go to plan. That is the work we do for providers, and the reason we still call the process by its internal name.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS