WhatsApp chatbot for a clinic: what to automate and what to leave with the front desk

A WhatsApp chatbot helps a clinic when it handles the repetitive administrative work: confirming appointments, rescheduling, sending the address and prep instructions, logging cancellations straight into the calendar. Anything about health, symptoms or treatment goes to a professional. A good system makes that split and shows the front desk what needs attention first.

Tuesday, 6:10 p.m. The front desk is closing out the register. On the clinic’s phone, 23 unread WhatsApp conversations (in Brazil, that’s where patients reach the clinic). One patient says she isn’t coming tomorrow. Another asks if he can take his medication before the test. A third wants to know if there’s parking. All three messages land in the same place, with the same weight, and whoever reads them decides on the spot what’s urgent.

The cancellation comes in through WhatsApp. The calendar only finds out later.

What you can automate without worry

• A confirmation the day before, with a simple reply: I confirm or I need to reschedule.
• Rescheduling within the slots the calendar opens up.
• Fixed information: address, parking, standard test prep, documents.
• Logging the cancellation in the calendar as soon as the patient replies, freeing up the slot.

Notice that almost none of this requires artificial intelligence. Much of it is scheduling system configuration and a well-written rule. Many scheduling systems already come with automatic confirmations nobody turned on. Start with what you already pay for.

Where AI comes in and where it stops

AI helps when a message comes in loose. “Hi, I’m running about 20 minutes late, is that still OK?” A rule doesn’t understand that. A language model understands it’s a late arrival, checks the clinic’s policy and drafts the reply or alerts the front desk.

The limit needs to be written down before the system goes live. Questions about symptoms, medication, test results or treatment go to a person. Always. The system replies that it will pass the question to the team and flags the conversation as a priority. It doesn’t try to help with the health question.

That “not trying” is the most important part of the project. And it’s the part that shows up least in demos.

Be wary of the “zero no-shows” promise

No-shows have causes a message can’t fix: traffic, a sick child, money being tight. A good confirmation reduces forgetting and, above all, makes cancellations show up early, while there’s still time to fit someone else into the slot.

Anyone who promises a percentage without looking at your calendar is guessing. The honest way is to measure first: how many no-shows per week, how many cancellations come in with less than 24 hours’ notice, how much time the front desk spends confirming. After the test, compare the same period.

Patient data and the LGPD

Under Brazil’s data protection law, the LGPD (Law 13,709/2018), health data is sensitive personal data. In practice, that calls for a few decisions before any system goes live:

• The system accesses only what the task requires: name, phone number, appointment time. Medical records stay out.
• It’s written down who accesses the conversations and how long they’re stored.
• The automation uses the official WhatsApp Business API. Unofficial tools can get the number banned and leave the data somewhere you don’t control.
• The vendor says where the data is stored and whether the messages are used to train any model.

If the answers come back vague, you already have your answer.

The human touch stays with the front desk

Taking the copying and confirming off the front desk is meant to free up attention for whoever is standing in front of them. The patient who is nervous about her test should get to talk to a person faster, precisely because the other 20 messages were already handled on their own.

If you want to see how your clinic does on these points, the AI-ization Diagnostic asks 10 questions about your routine and shows what you can configure, connect or automate, and where AI shouldn’t come in yet.

Of the messages that came in yesterday, how many really needed a person?