AI appointment agent for clinics: the receptionist who never misses a day

A day on the practice’s WhatsApp, without a system

Every message was a receptionist’s task: check the calendar and reply, find the deposit receipt, find another slot for whoever couldn’t make it, explain the time to the patient abroad. If Federico was running late, he told everyone one by one. And if someone didn’t show up, the slot was lost, because nobody found out in time to give it to someone else. In a traditional practice that work is done by a receptionist, on a salary. In a small one it’s done by the practitioner, between sessions: “if I have to cancel, my hands are tied”, Federico said when we started.

A traditional practice and Colella’s

This is how a practice with a receptionist compares with Federico’s, where the system does that job. The calendar is the same; what changes is who runs it, and when. At a traditional practice the receptionist works office hours: whatever comes in at night or over the weekend waits until Monday, and a slot dropped at the last minute is lost. Deposits are checked by hand, reminders depend on someone remembering and, if the practitioner runs late, someone has to call every patient. At Colella’s, patients book and pay on their own, whenever they like; the reminder goes out by itself, with a way to change the appointment; and a freed slot goes back on the calendar right away. Federico does the same as before: treat patients. But without a receptionist’s salary and without patients skipping without notice.

The practice’s rules, the same at 3 pm and at 3 am

Patients book on the website, leave a deposit and get their reminders on WhatsApp. And they reschedule or cancel on their own, without depending on anyone: if they can’t come, they don’t have to call or wait for an answer, they reply “can’t do Thursday” to the same reminder or go to “My account” on the website. The bot understands text and voice notes, offers the open slots, moves the appointment, carries the deposit over and frees the slot. If they cancel in time, the deposit goes back to their account on its own. The rules aren’t up to the AI. Dates, windows and refunds are enforced by the system, and the bot can only do what those rules allow. It doesn’t diagnose either, never gives out the practitioner’s personal number and, if a patient asks for a person, it steps aside.

Who does what: the patient, the bot, the physiotherapist and the system

The receptionist’s job got split up. Patients do on their own what they used to ask for by message, the bot answers and handles, and the system does what nobody has to remember to do. Federico sees it all in his panel and only steps in if he wants to. Each one sees what concerns them. Patients sign in to “My account” and see their appointments and files. The bot sees open times and the rules, but decides nothing outside them. Federico sees the day, the money and the records. And on the website, an assistant answers prices, hours and FAQs for anyone who drops by, without seeing any patient’s data. If a patient asks for a person, or the conversation gets stuck, the bot steps aside and hands it to Federico, who can reply himself from the panel.

Why the practice’s no-shows dropped to zero

It wasn’t the chatbot alone. It was two things that hold each other up. The deposit, which gives the patient skin in the game. And moving an appointment became easier than skipping it. Whoever can’t come lets them know, the slot goes back on the calendar and another patient takes it. No-shows turned into cancellations with notice: in the first three and a half months, 17 in 578 appointments.

The physiotherapist’s panel: the day laid out and every payment matched to its slot

At first Federico heard about everything through a Telegram bot. Since August 2026 he has his own panel, which installs on his phone. In “Today” he sees the day’s appointments, taps “Paid” or “No-show” and, if he’s running late, presses “Running late”: the system tells every patient that day. In the calendar he blocks a day or a time range, and the system cancels, refunds and notifies everyone affected. Every patient has a record, with their sessions, files, clinical notes and the conversation with the bot. Mercado Pago payments match themselves to their appointment and the balance stays on the patient’s account; if something is owed, it’s claimed over WhatsApp with one button. And in stats, month by month, no-shows, income, occupancy and how many patients come back.

What runs on its own, without anyone remembering

What depends on the receptionist’s memory at a traditional practice, the system does here on schedule. Messages only go out between 7 am and 9 pm: anything due overnight goes out in the morning.

• On paymentWhatsApp confirmation, the calendar event and, for online sessions, the video call link.
• After the sessionA survey every 5 sessions, without anyone having to ask.
• Standing slotsThe weekly slot books itself, with its exceptions, and syncs with the practitioner’s Google Calendar.
• RefundsIf the patient cancels in time or the practitioner blocks a day, the deposit goes back on its own through Mercado Pago.
• Last-minute bookingsThose booked late still get their reminder.
• ReviewsWhatsApp campaigns to ask for reviews, moderated before they show on the website.

The calendar filled right up

With the calendar open around the clock, the deposit at booking and slots that free up on their own, the practice filled up. In February, the first month, the system booked 14 appointments. In March, with the website open to patients, 112. Since June it’s over 200 a month, and today it has even more than at the start: 226 in September and 223 in October, with the month still running. In total, 1,539 confirmed appointments since February. And nobody else was needed to handle them: what would have been messages, calls and transfers now comes in on its own.

How it grew: every request from the practice, a feature

The first version shipped in February 2026 and by March the website was open to patients. What came next grew out of daily use: a patient living in New Zealand, online sessions, standing slots, the one-hour session. In August the system moved to its own platform, with the practitioner’s panel, and it keeps adding what the practice asks for.

Under the hood: one calendar, and the rules in the code

It started as automations on a server, and in August 2026 it moved to its own Next.js application, with its database on Supabase and deployed on Vercel. The patient’s website, the WhatsApp bot and the practitioner’s panel read from and write to the same calendar; Mercado Pago, Google Calendar and video calls connect from the panel. The bot uses an AI model with narrow tools: it can reschedule and cancel, but dates, the 24-hour rule and refunds are decided by the code, not the model. What used to be a chat, a calendar and someone taking notes is now an appointment with its status, its payment and its history.