Booking clinic appointments on WhatsApp: what an AI agent solves and what it must not

A badly booked appointment does not show up on the day it is booked. It shows up when two patients arrive at the same hour and somebody has to decide which one gets sent home. That is why booking on WhatsApp does not start with writing a friendly message; it starts by deciding where the availability the agent offers is going to come from.

Booking on WhatsApp without double-booking

Double bookings are rarely the mistake of the person doing the booking. They are the result of keeping two calendars: the one in the calendar app and the one in the head of whoever is replying. When the agent asks for open slots before proposing an hour, and creates the booking on that exact slot, the second calendar stops existing.

That is the order the booking recipe follows: first the bookable services with their duration and practitioner, then the open slots between two dates, and only then the booking itself. It is not an implementation detail: it is what makes the answer verifiable against the calendar.

Where your calendar lives changes the answer

If the clinic already works on a calendar, the agent reads that calendar. If the schedule lives inside the site’s own software (the case of an optician running Softix), the lookup goes against that system and not against a copy.

The question to answer before configuring anything is which of the two wins when they disagree. If the answer is “it depends”, there is no calendar yet. There are two.

What it must not do, which is the important part

An agent handling patients will get clinical questions. “Is this urgent?”, “which specialist should I see?”, “can I take something in the meantime?”. The right answer to all three is the same: do not answer, hand the conversation over.

This is not decorative caution. A wrong time slot is fixed with a phone call; wrong clinical guidance is not fixed at all. That is why the boundary is declared when the Met is configured (the healthcare solution spells out where it sits) instead of being left to the model in the moment.

Reminders are the easy part

Reminding someone of tomorrow’s appointment is the fastest thing to notice and the least difficult to set up. What is worth looking at is what the agent does with the reply: a patient answering “I can’t make it” is releasing a slot, and that slot is only worth something if someone, or something, offers it again.

When a person has to step in

The booking Met exists to resolve the repetitive part: which services exist, when there is room, confirming, reminding, rescheduling within policy. Everything else (an exception, an upset patient, an authorisation that never arrives, anything that smells like a clinical decision) belongs to the front desk.

Whether it was configured well is not measured by how many conversations it closed on its own. It is measured by whether the ones it handed over arrived with the facts already gathered, so the person does not have to start from zero.