MSI’s Chief Product Officer on why AI has turned predictive modeling into a commodity and why the MGAs that win the next pricing arms race will do it on the strength of their data and distribution relationships, not their models.

Key takeaways:

• AI has turned predictive modeling into a commodity: The ability to build, test, and iterate a pricing model around the clock is no longer a differentiator, it’s table stakes. Once every firm has access to a world-class modeler that never sleeps, the analytical work itself stops being the competitive edge.
• The real arms race is now about what you feed the model: Data quality, distribution relationships, vendor partnerships, and the infrastructure to capture insight from every customer interaction. These are what separate the firms that see risk sharply from those that don’t. The model is only as good as what goes into it.
• Volume within a specialty is itself a data advantage: Every risk a firm quotes, wins, or loses generates training data. Scale within a particular line or segment compounds over time into a proprietary view of risk that a newer or smaller entrant simply can’t replicate, regardless of how sophisticated its modeling tools are.
• Adverse selection is accelerating: The firms that pull ahead will take the risks worth writing and walk away from the ones that aren’t. The firms that fall behind will feel it twice, losing the business they want and keeping the business the sharper firm declined. That dynamic has always existed; AI is making it run faster.
• The durable competitive advantage is relational, not technical: Distribution relationships that generate proprietary data, the ability to assess and onboard new data and modeling vendors quickly, and the organizational speed to turn data into segmentation decisions in days rather than quarters. These are what will separate winners from the rest over the next three years.

Most of the conversation about AI’s impact on insurance is operational: faster quotes, smoother service and claims, leaner processes. Those gains are real and they matter. Firms that use AI to run leaner will fund lower expense ratios, lower premiums, and better customer experience.

Less discussed is what AI is doing to the pricing and underwriting segmentation arms race. That race is not new, and it has always rewarded the firm that knows risk better than the next one. What AI changes is its speed, and over the next three years that acceleration will do just as much to decide which MGAs win as operational efficiency.

To understand why, let’s split the work behind every pricing and underwriting model into two kinds: there’s work AI is newly capable of, and then the work that remains beyond it.

Making a Better Predictive Model

The first kind is predicated on the analytical work itself, which includes the modeling, testing, and iteration that turn raw data into a pricing or underwriting view. Especially in the past year, with tools like Claude Code, AI can now run that loop on its own, building a model, testing it, seeing where it falls short, and trying other approaches without waiting for a person to direct each step.

The arc is a familiar one. Chess fell to Deep Blue in 1997, Go to AlphaGo nearly two decades later, and competitive coding is falling now: in 2025, a single human programmer held off OpenAI’s model at the AtCoder World Tour Finals, and this year none did. Given a defined dataset and target, predictive modeling is no different: having a world-class modeler that can work 24/7 is now effectively a commodity. At MSI, we used AI to build and refine 20 versions of a predictive model for one of our E&S homeowners programs in a single day. That work used to take our best people weeks or months, and the model is in production today.

None of this eliminates the modeler. That’s because someone still has to know the business well enough to frame the problem, judge the output, and decide what the machine should learn from. 

What it eliminates is the edge that came from being better at raw modeling than the next firm. Once every firm has a world-class modeler that never sleeps, the difference in how sharply two firms see risk comes down to what each one feeds it.

Volume is an Asset

When it comes to the second kind of AI work, the areas that remain beyond AI’s competency, is everything the model learns from. 

That means distribution relationships that generate data competitors don’t have, standing partnerships with the best data and modeling vendors, the ability to assess and onboard new ones faster than the next firm, and the infrastructure to capture the insight in every customer interaction a firm already has instead of letting it go unused. And scale matters, because volume is an asset in this kind of work: every risk a firm quotes, wins, or loses is training data, so share within a specialty is itself a data advantage.

The firms that pull ahead will be the ones with proprietary data, the relationships to keep generating it, and the speed to turn both into sharper segmentation in days instead of quarters. The firms behind them will feel it twice: they lose the risks they want, and they keep the ones the sharper firm walked away from. Adverse selection is not new, but the speed at which it compounds is.

That’s the inflection point. The game MGAs are playing has not changed, but it is about to run on an even faster clock, and the firms that treat their data and relationships as the durable competitive advantage, not the models themselves, will be the ones winning on risk three years from now.