721 conversations at an aesthetic clinic: what the agent does and what it doesn’t

Since 1 July 2026, an aesthetic clinic has answered on WhatsApp and Instagram with an agent that I designed, built and operate. On 9 October I measured what has happened. These are the numbers, including the one that is not in my favour.

How many, and where from

From 1 July to the day of measurement, the agent has had 721 conversations. 562 came in on WhatsApp and 159 on Instagram. Test conversations are not counted.

Who answers

89% of those conversations are answered by the agent with nobody on the team stepping into the reply.

When a person is needed, the agent hands the conversation over: there is always someone in charge. And from the first message it says it is an AI, as Article 50 of the EU AI Act requires, in force since 2 August 2026.

When they write, and how long until they get a reply

Clients don’t write only during opening hours. And the first reply doesn’t wait for the clinic to open.

• 53% of conversations start outside the clinic’s opening hours.
• 8.4 s median time to first reply.

Reminders

The agent reminds clients of their appointments on WhatsApp. 75% of reminded appointments are confirmed by the client in the chat itself.

The number that is not in my favour

84%. The agent cannot create appointments. That is by design: it answers, proposes and reminds, but the owner creates the appointment, and she does so in 84% of cases.

I publish it like the rest. A custom system is also defined by what it doesn’t do, and this number shows it.

What is underneath

1. AI disclosure from the first message.
2. A person in charge to hand the conversation to.
3. Two independent reviewers on every delivery, one of them a different model in a clean session, and an executed test.
4. Operation: I operate it and measure it. This note comes from that measurement.