Half of the Interesting Things You Can Do With AI Don’t Require Your Data

Aug 12, 2026 · 3 min read

A StrideShift data monetization case study. Sector: pet insurance, Europe. Engagement: agentic research, synthetic data and ecosystem modelling.

The situation

A large European pet insurance business set out to find new commercial ground across itself and three partner and subsidiary companies. Between them, the four businesses already held a strong picture of the customer. The data was not missing. It was locked inside four separate companies with different shareholders, and pooling it meant access agreements, sharing arrangements and governance sign-off in all four.

That is months of legal work before a single question gets asked. Worse, nobody could say in advance whether the questions were the right ones. The group would have paid for the plumbing before deciding what it wanted to run through it.

What we did

We built the market instead.

We used our own agentic research methods to do it: coordinated AI agents that gather, cross-check and structure public evidence at a scale hand research cannot reach in the same time. From that we modelled the four businesses and the ecosystem around them, including the customers, the intermediaries, the service providers, and the flows of money and information between them. Then we generated a synthetic dataset that behaved like that structure.

Synthetic data is generated, not collected. It tells you nothing about any individual real customer. What it does is let you test how a proposition behaves inside a market before anyone signs a data-sharing agreement.

All of it — the research, the ecosystem model, the synthetic data — went into our AI Engine, which generated and scored the opportunities itself. The 120 ideas it returned were not the limit of what it could produce. That number was a setting. We tuned it to what the group could realistically read, understand, argue about and act on. An engine that hands over a thousand ideas has handed over nothing.

What came out of it

A catalogue of roughly 120 commercialisable opportunities, scored and sorted, and six of them built out as working demonstrations. The six were built to show the idea rather than to run in production. The point was to give people something concrete to react to instead of a slide describing what might one day exist.

The model does not replace the real data, and we said so from the start. It sets the order of the work. Instead of negotiating access across four companies in order to explore everything, the group could negotiate it for the few questions that had already survived a test. The governance work still has to happen. It happens second, and for a reason.