Price Benchmarking in a Split Market: What the "Hard Market Is Over" Headlines Get Wrong

s the hard market really over?

For 32 straight quarters, commercial rates went up every single quarter for the better part of the decade. Then the streak ended. The industry index posted its first broad decline since 2017, with an average reduction of about 1.2 percent, and that number is what has everyone declaring the hard market finished.

The index in question is the outside data from the Council of Insurance Agents and Brokers, and the softening shows up hardest on the property side. On the broker side, one of the largest US brokerages is projecting commercial property to fall by roughly 4 percent on average through the back half of the year. Overall commercial premium growth averaged about 0.2 percent in the first half, the flattest it has been since 2017. On paper, the entire market looks soft. The problem is that "on paper" and "in the account" are two very different things.

Why does liability keep firming while property softens?

Property has been easing. Liability never got the memo. Commercial auto, umbrella, and excess kept firming right through the property softening, with some industry increases landing in the 8 to 15 percent range, while carriers quietly cut how much limit they will put on a single policy.

The reason is not capital and it is not weather. It is the courtroom.

No amount of reinsurance capital showing up in a given quarter fixes a trend that lives in jury awards. So the tidy market-wide average, that flat 0.2 percent everyone quoted, is doing what averages do: blending two things moving in opposite directions and handing you a number that describes neither of them.

What is a "K-shaped" market, and why do averages hide it?

A prior guest on the show described the current environment as a K-shaped market, and the common mistake is to picture asset classes. That is not what the term means here. It refers to risk-quality cohorts. Inside the same line of coverage, the most risk-averse accounts and the highest-exposure accounts are pulling apart from each other, one arm of the K climbing while the other slides down.

When a market splits this way, an industry average is worse than useless, because it actively hides the spread that determines what any single account actually pays.

How does price benchmarking read a split market?

This is where a market-wide figure fails and account-level data takes over. The only question that matters for a specific policy is where it sits within its own market, and you cannot get that from an industry average. You get it from carrier-level [price benchmarking] and comparables.

The episode used a live example from the pricing comparables view in the Advocate app: California multifamily liability, one line, one market. The cheapest quartile came in around 0.47 on rate, the median around 1.23, and the top quartile all the way up at 3.59. That is close to eight times the price inside the exact same line of coverage, same year, completely different worlds depending on the account.

That spread is the argument for [insurance data analytics] built on structured, carrier-level inputs rather than blended indices. In a split [property and casualty insurance] market, benchmarking is not a nice-to-have report. It is the only way to see the picture the average is hiding.

Why are data centers straining insurance capacity?

The second headline is about scale. A single hyperscale data center can carry an internal value somewhere between 20 and 30 billion dollars. New premium from data centers is expected to reach around 10 billion dollars a year, roughly double the size of the entire global aviation insurance market.

There are more than 4,000 data centers in the United States as of July 2026, with the top ten states holding about 60 percent of them. The concern is not only the size of the market, though. It is where the building is happening.

The coverage does not fit the existing boxes, because a single outage can cascade across an interconnected network instead of staying inside one building. A guest who structured one of the largest placements in the market, reportedly around 4 billion dollars, made the point that the constraint was never demand. It was capacity. Carriers are wary, and durable coverage once a building comes online is genuinely hard to source. Put that against a hyperscale campus worth 20 to 30 billion dollars, and even a record placement covers only a fraction of a single campus. This is one of the sharpest current examples in [commercial real estate insurance] of risk outrunning the capital available to cover it.

Why is AI adoption not fixing the insurance data problem?

The third headline is, arguably, the most important. AI stopped being a pilot project this year. Something like eight in ten insurance executives now say AI is embedded in their workflows, and on the operations side, 70 percent report using it, up from 58 percent a year earlier. Adoption is effectively settled. Then you turn the page.

Roughly two in three executives say poor data quality is actively slowing their AI models. Only about three in ten say they can get information quickly out of their AI, and more than eight in ten worry their models are being trained on data that is incomplete or simply wrong.

That gap is the reason the [ai in insurance industry] conversation keeps circling back to data rather than models. In a regulated industry, pricing and underwriting decisions have to be explainable, defensible, and auditable, and the rules on high-risk AI are tightening this year rather than loosening.

The approach Advocate described is the inverse of bolting a model onto messy data: carrier-level pricing and benchmarking that is already structured and connected, feeding a deterministic engine that returns the same answer every time and can always show its work.