Most people start wrong with AI - lessons from a real case

At Easter and last week, I continued to work on a concrete case through my group AI Value Lab Oslo, where we use a real dental clinic as a learning arena to test how AI can be used in practice.

We didn't start with technology. We started the business.

Through analysis, research, interviews and discussions, we worked our way to one core point: Where is friction experienced in everyday life?

What works well today?
Where are the bottlenecks?
Where does the clinic lose time, capacity or potential income?

We also looked at what the competitors do better, particularly in how they meet patients digitally and operationally.

The picture that emerged was clear:
The professional quality and patient experience are strong, but there is a gap in the digital and commercial areas around. So not in the core delivery, but in what happens before and after.

That is why we are now working to modernize the patient journey, and consider how AI and digital solutions can strengthen:
🦷 booking and availability
🦷 follow-up and repurchase
🦷 clear information and patient communication
🦷 more efficient work processes
🦷 better operational support in everyday life

Specifically, we are now working with:
- further development of the website (better structure and more relevant information)
– new chatbot (never had before)
- new online booking solution (not previously available)

Everything is developed based on the actual needs of the patients. The point is not the tools themselves, but the order.

Most businesses, on the other hand, start with the question:
"Which AI solution should we choose?"

In practice, one should start in a completely different place by asking:
"Which decisions and processes affect the outcome?"

I recognize this from commercial organisations:
Prioritization of customers is done differently from person to person.
Follow-up of offers is governed as much by capacity as by potential.
Pipeline and forecast are often characterized more by optimism than structure.

This does not mean that it is done poorly, but that there is considerable potential for improvement. This is where AI can help!

A model can analyze patterns and suggest the next best course of action. But the value only arises when it is used to make better decisions in practice.

Since these AI posts are meant as expertise sharing, I also briefly mention the tools we are testing. Solutions have been developed with Claude and Claude Code, with content from FAQs and today's website built on localhost, with further plans for deployment via Vercel. Website work in parallel in Lovable. Not because the tools are important, but because more people are asking what is actually used in practice.

👉 In the next post, I look at why many AI initiatives stall, even when they get off to a good start.

PS. Easter was otherwise spent in Røros and Svalbard with experiences that remind me of something important: Not everything should be optimized. Cross-country skiing, alpine skiing, snowmobiling, dog sledding, wild animals and time with family and friends still beat most - the real is not artificial.