AI in Insurance and the Rise of the K-Shaped Market

What is the K-shaped insurance market?

Zuk's core idea is that the market has stopped pricing primarily by line of business or geography and started pricing by risk-quality cohort. The result is a K. The top branch and the bottom branch pull apart, and the middle goes flat.

At the top of the K sit well-capitalized sponsors, modern assets, and clean loss histories, and carriers compete heavily for them. The bottom is highly commoditized, with little differentiation and its own set of challenges. The middle, which is most of the market, is the hard part: not a national account, not part of a small-commercial strategy, and difficult to get serviced or priced properly. Importantly, this is not about asset size. A mid-sized risk can be excellent and still get treated as middle-of-the-K simply because of how its quality is perceived.

What forces are driving the K?

Zuk points to several forces converging at once rather than a single cause.

That third force is where AI enters. The split between firms that are deploying technology well and those that are not shows up not only in pricing but in risk selection and in the performance of the underlying portfolio. Climate and weather add a fourth pressure, which the team noted has shown up repeatedly in their own pricing analysis.

The two faces of AI in insurance

One reason the AI conversation gets muddy is that two very different things get lumped under the same label. Agarwal drew the distinction clearly.

On the tool side, AI is already changing underwriting. An underwriter can look at something like a crime score, carve out a bespoke exclusion, and then apply that carve-out across thousands of risks at once. Smaller and smaller accounts are now seeing manuscript policies and nuanced liability carve-outs they have never seen before. On the risk side, AI is generating a wave of emergent exposure: cyber, data centers, and a surge in AI-driven litigation, with firms able to file enormous volumes of claims.

Is the insurance industry ready for AI risk?

Mostly not yet. Zuk describes a market that is reacting slowly and quietly absorbing AI liability rather than addressing it head on, a "see no evil" approach. The industry is, in his words, poorly equipped on the coverage side, unsure whether to include or exclude these exposures in policy forms.

The closest precedent both men landed on is cyber insurance ten or fifteen years ago. We are roughly at the advent of AI as a distinct insurable category. A few MGAs are just starting to write or price AI liability off model-specific risk, but the efforts are early and rudimentary, and pricing is genuinely hard when the underlying models change every month or quarter.

Why the middle of the market gets ignored

The middle is not ignored because it performs badly. It is ignored because, in a multi-trillion-dollar market with little transparency, it is structurally not interesting. Carriers chase the commoditized bottom for a while, then pivot to human-augmented specialty business at the top, and the solid middle gets passed over in submission queues without ever seeing pricing that reflects its real risk profile.

This is also where the middle is most exposed to blunt instruments. A mid-cap real estate portfolio or a regional manufacturer can get swept up with whatever thematic exclusion or endorsement is in fashion, including a silent AI-liability exclusion that may not even be relevant to its operations, the way PFAS exclusions have spread. Good risks end up subsidizing bad ones, and the lack of differentiation cuts both ways.

How AI and data can make the middle legible

There is no silver bullet, but there is a clear direction: legibility. Give brokers, risk managers, and the CFOs and controllers who actually buy the insurance the tools and insights to see where the market is pricing like-for-like risk, what the spread is, and which carriers are playing in that space.

Used well, AI and data let a strong but overlooked risk make its case: here is why we do not belong in that bucket, for these specific reasons. That is exactly what price benchmarking and insurance data analytics are for. They turn a good middle-market risk into one carriers compete for instead of pass over, and they surface coverage gaps, or an AI exclusion that should not apply, before a buyer ever signs.