From backlog to clickable prototype.

The starting point

Insurer Baloise is investing heavily in digitalisation and AI. It runs a broad portfolio of AI initiatives, and through internal AI Weeks and training days, employees across domains are brought into this new way of working, one business profile at a time.

That includes the people behind software analysis: functional analysts, product owners, product managers and digital experts. They bridge the gap between business input, which often arrives vague and messy, and the structured deliverables a developer can build from. Every software project at Baloise goes through a fixed analysis phase, with its own guidelines, templates and deliverables.

Lots of unstructured input on one side, a fixed output format on the other. That makes their process a strong candidate for a closer look through an AI lens. Baloise asked Baboon Labs to find out how far it could be optimised, and to build something around it for these profiles.

What we did

Baboon Labs designed a series of hands-on workshops for these profiles, spanning several business domains and spread over multiple days. Each edition opened with a crash course to get everyone on the same page: what AI-native analysis is, how the flow works, which tools it takes and where it usually goes wrong. An AI-native product owner then showed how he works today, side by side with how he worked a year ago. That made it concrete straight away.

Then the participants got to work themselves, with coaches on the floor to help where needed. They worked on real cases from their own portfolio, as close to their day-to-day job as possible. The cases ranged from greenfield to legacy. Greenfield means starting from scratch: going from an idea to written requirements and a clickable prototype, for example a new quote flow for brokers. At the other end was legacy analysis: building a functional understanding of an old database that has been running for years and whose documentation is no longer complete. In between sat the everyday work, like turning messy technical documentation into a clear release note, a user guide or a management presentation.

All of it happened in Claude, with Baloise’s context built into skills, templates and a design system matching the Baloise brand. That kept them within their own analysis guidelines, and they built their prototypes from that same context.

Results

By the end of each edition, every team had a complete analysis that followed Baloise’s guidelines, plus a working, clickable prototype in Baloise branding. No developer or designer involved. Analysis work that normally takes months was done in a few days.

Beyond prompting, participants learned how to work with context. Their own guidelines and templates were loaded into the model, so it worked from Baloise’s standards from the very first draft. They also saw how much their own domain knowledge matters, and what a real AI-native analysis flow looks like: the order you work in, where you let the model ask follow-up questions, where you as the analyst check the output yourself, and what you capture so a colleague can reuse it. Setting that up takes someone who knows the world of analysis and knows how to break it into pieces a model can handle reliably.

Participants rated the sessions 19.5 out of 25 on average. The number of aha moments scored highest, followed by the support from the coaches. One team started out openly sceptical: “AI in a domain as complex as insurance? That was never going to work.” By the end, that same team was asking how to bring this into their daily work.

After each edition, Baboon Labs delivered a debrief to management, with the coaches’ observations and a report recommending how to embed this way of working for the long term. What’s left is a group that has seen it work on their own cases, and a plan to make it stick.