Designing AI for Real Life

What to build when your users aren’t asking for a conversation

A chatbot relies on a clean, comfortable assumption: the user arrives with a clear question and wants to talk about it.

Now consider what PensionDanmark looks after: pensions, insurance and health cover. Nobody wakes up wanting to chat about any of these. Most people don’t know what they should ask, and many would rather avoid the answers altogether.

PensionDanmark’s 850,000 members include scaffolders, truck drivers, cleaners and electricians. They open the app in a van at 06:40, or during a fifteen-minute break with their gloves half off. Asking them to chat means asking for words they don’t have, about subjects they would rather not think about. Yet the team was weeks away from shipping a generic chatbot anyway.

They built something else.

Henrik Harsbo, Head of UX & Design at PensionDanmark, shows what the team delivered, how testing with members reshaped it, and what 12 months in production revealed.

What you will take away:

• How to know whether your users will ever start a conversation, before you build the UI that waits for one
• Where the model’s job ends and your team’s begins
• Why long conversations aren’t necessarily a sign of success, and what to measure instead

Come if your users have better things to do than talk to your product.