I said no to an AI symptom chatbot in our patient app. Here is the one-pager I used

A short one, because three people asked me about it after a meetup.

In August our CEO came back from a conference wanting a chatbot in our patient app that answers "should I see a doctor about this?" questions. A competitor had one in a demo. The board liked it. I wrote a one-pager recommending we do not build it this year, and we did not. Here is the structure, in case it is useful.

1. What problem would it solve, and for whom?

I pulled our support data: 6% of patient messages are symptom questions. Most of them already go to the clinic's nurse line, which is the right place. The problem was real but small, and it already had a safe owner.

2. What does it change about our regulatory position?

This was the big one. Today our app is administrative: bookings, referrals, messages. Software intended to give information used for decisions about diagnosis or treatment can qualify as a medical device under the EU MDR, and an AI system in that position would likely be high-risk under the AI Act as well. That means clinical evaluation, a notified body, post-market surveillance. I asked our regulatory consultant for a rough estimate. The answer was "more than a year and more than your Series A".

3. What is the cheapest version that tests the need?

We added a clear "Is this urgent?" link to the nurse line and the national health line, with opening hours. Two days of work. We measure how often it is used.

4. What would make us revisit?

• Symptom messages grow beyond 15% of volume.
• A partner clinic asks for it and is willing to be the clinical owner.
• A certified component exists that we could integrate instead of build.

The CEO was not happy for about a week. Then the competitor quietly removed theirs from the demo. I did not say anything. Lagom smug.

Happy to be proven wrong, as always. But not about the notified body.