Pricing teams have never had more internal data. Yet many still struggle to answer a simple question: Are we pricing risk deliberately or drifting out of alignment with the market without realising it?  

We sat down with our Lead Data Scientist, Kalle Myllarniemi, to talk about how our new machine learning capabilities, Apollo and Atlas, came to life and what they now make possible.    

Q: Where did the idea behind Apollo and Atlas originate?    

Kalle:
We wanted to move beyond the headline stats.

For years, we have helped clients understand where they sit in the market. Above. Below. Moving up. Moving down. That is valuable, but it only gets you so far.

The bigger question is what you do with that information.

We kept coming back to the same thought. We collect daily pricing data across the market. We can see how risk is actually being priced in real decisions. Why are we only using that to describe position?

Apollo and Atlas came from that shift. Instead of helping clients understand where they sit, we wanted to help them understand why they sit there and whether that position is intentional.

That meant looking beneath the surface of price and understanding how rating factors and postcode risk are being treated across the market .

Q: What problem were you seeing in the market that made this necessary? 

Kalle:
Insurers are not lacking sophistication. Internal models are advanced.

But internal models only ever reflect your own book. They cannot show how competitors are pricing the same risks or which signals the wider market consistently weights heavily

That creates blind spots.

With rating factors, it can be hard to see whether you are deliberately differentiated or unintentionally misaligned. With postcode, strategies often drift not because modelling is poor, but because they are slightly out of step with broader market treatment

So the risk is not lack of capability. It is isolation.

Q: How did it feel seeing these propositions go from concept to reality?

Kalle:
It was not just seeing it become real that was exciting. It was the insights we began generating.

When you analyse premium outcomes at scale and isolate how individual factors influence pricing behaviour, patterns emerge.

Sometimes they confirm what you expect. Sometimes they challenge assumptions.

The first time we mapped a client’s structure against aggregated market behaviour, the differences were subtle but meaningful. A factor weighted slightly more heavily than the market. A postcode segment is treated slightly differently.

Those are not dramatic headlines. But over time, they shape portfolio mix and risk selection.

That was the moment it stopped being an idea and became something commercially powerful.

The Bigger Picture

Machine learning in insurance often focuses on internal optimisation.

Our approach focuses on market intelligence.

Because sustainable pricing performance is not just about building better models. It is about understanding how your pricing logic sits within a live, competitive market.