How a physician linked ad spend to booked appointments

Industry

Private medical practice

Business area

Patient acquisition, appointment booking, and attendance

Automation type

WhatsApp AI agent with calendar booking, reminders, and campaign attribution

Services

Workflow design, AI agent configuration, WhatsApp platform setup, calendar integration, and CRM reporting

Tools

WhatsApp Cloud API, n8n, Google Calendar, Pipedrive, Meta Ads, and Google Ads

Key result

Appointments booked in the conversation itself, confirmed by reminder, and traceable back to the campaign that produced them

The practice was already spending on advertising. Campaigns ran on Meta and on Google, and both sent people to the same place: a WhatsApp number.

That number worked, in the sense that patients wrote to it and appointments eventually happened. What it could not do was scale, and it could not answer the one question that decides an advertising budget.

• The consultation price was asked several times a day, and answered by hand every time.
• Appointments were agreed inside the chat and copied into the calendar afterwards, which is where double bookings come from.
• Messages that arrived in the evening or at the weekend waited until the practice reopened, by which point some enquiries had gone elsewhere.
• Patients forgot appointments they had agreed to days earlier.
• Nothing connected a patient to the campaign that brought them, so two campaigns that produced very different results looked identical in the ad platforms.

The last point is the one that mattered most. Clicks and replies were measurable; patients seen were not. Spend was being allocated on the wrong evidence.

Before the system was built, an enquiry travelled like this:

1. The patient saw an ad on Facebook or Google and opened a WhatsApp conversation.
2. Someone at the practice read the message, in working hours.
3. The price and availability were answered manually.
4. A time was agreed over several messages, back and forth.
5. The appointment was written into the calendar by hand.
6. The patient was expected to remember it.

Every step was small. Together they meant that the speed of the practice's response depended on somebody being free to look at a phone, and that the source of each patient was lost by the second message.

The number was moved onto the WhatsApp Cloud API, which is what makes the rest possible: a number on the free Business app cannot be connected to other systems or send scheduled messages, so the platform was a requirement rather than a preference. The tool at WhatsApp Business app vs Cloud API walks through the same decision.

On top of it, we built an agent in n8n that handles the conversation end to end:

• It qualifies the enquiry first — what the patient is coming for, and whether this is a first visit or a follow-up — because that decides how long the appointment needs to be.
• It answers the consultation price directly, which is the question that arrives most often.
• It reads the physician's real free slots from Google Calendar and offers what actually exists, rather than a fixed list someone has to keep up to date.
• It writes the appointment back to the calendar, so the slot is taken the moment the patient accepts it.
• It records the conversation and the appointment in Pipedrive, together with the campaign that produced them.

The agent works at any hour. A patient who writes at eleven at night gets a price and a booked appointment in the same conversation, which is where most of the recovered demand comes from — those enquiries were not slow before, they were often lost.

One boundary was drawn deliberately: the agent does not give medical advice. It handles price, availability and logistics. Anything clinical goes to the physician.

The two platforms needed different mechanisms, because only one of them can hand information to a WhatsApp conversation on its own.

Both paths end in the same place. Every conversation reaches Pipedrive with its origin attached, and when the agent books an appointment, that appointment is recorded against the same record. The chain from campaign to booked appointment stays intact without anyone maintaining it.

What that changes is the question the practice can ask. Not how many people clicked, but which campaigns produced patients who booked — and those are rarely the same ranking.

A reminder that only announces a date does very little. The sequence built here asks the patient to confirm, which turns a message into a signal the practice can act on: a confirmed appointment, or one that needs attention while there is still time to fill the slot.

Because reminders go out days after the last message, they fall outside the window in which WhatsApp allows free-form replies. They are sent as approved message templates — a constraint of the platform that has to be designed for from the start, since template wording is submitted for approval rather than written on the day.

1. 01

   Trigger

   A patient opens a WhatsApp conversation from a Meta ad, from a Google campaign through the landing page, or directly.

2. 02

   Source Capture

   The campaign reference travelling with the conversation is read and stored before the first reply is sent.

3. 03

   Qualification

   The agent establishes what the patient is coming for and whether it is a first visit or a follow-up.

4. 04

   Price

   The consultation price is answered directly, from the information the physician configured.

5. 05

   Availability

   Free slots are read from Google Calendar at that moment and offered to the patient.

6. 06

   Booking

   The chosen slot is written back to the calendar, which closes it to anyone else immediately.

7. 07

   CRM Record

   The conversation and the appointment are recorded in Pipedrive together with the campaign that produced them.

8. 08

   Reminders

   A sequence of approved templates asks the patient to confirm ahead of the appointment.

9. 09

   Escalation

   Anything clinical, or any conversation the agent cannot resolve, is handed to a person.

The agent runs unattended for what it was built to do: reading the calendar, answering a price, booking a slot, sending a reminder. None of those can do damage that cannot be undone.

Everything else escalates. A question the agent does not recognise, a patient who insists, anything that sounds clinical — the conversation goes to a person rather than being guessed at. That split between what may run on its own and what has to stop and ask is the same one described in which processes are worth automating, and it is what makes an agent on a public channel safe to leave running.

Enquiries are answered whenever they arrive, and the appointment is booked in the same conversation instead of over a day of messages. The calendar is written once, by the system that read it, so the slot a patient accepts is genuinely theirs.

Reminders now come back with an answer, which gives the practice time to react to an appointment that is about to be missed rather than discovering it in the waiting room. The same problem, approached with a different stack, is covered in centralizing bookings and cutting no-shows.

And the advertising can finally be judged on what it produces. Each booked appointment carries the campaign that brought it, so the budget follows patients rather than clicks.

Results here were qualitative rather than measured through formal performance indicators.