KCC says AI models could reset the pace of reinsurance pricing

Karen Clark & Company builds the catastrophe models that underpin reinsurance pricing and ILS structuring. A white paper the firm published this month, covered by Reinsurance News, argues that artificial intelligence could change how those models get built and refreshed. The paper describes an industry shifting away from statistical inference and toward physical models that AI augments, with effects on reinsurance pricing, underwriting, and capital management.

The perils in question are already expensive. Insurance Capital Daily's tracking puts 2026 US storm losses past $35 billion after the August derecho, using Aon and Gallagher Re estimates. KCC's paper targets exactly these events — severe convective storms, winter storms, wildfires — where historical data, the firm says, cannot capture how complex the risk has become.

First-generation catastrophe models were designed for rare, severe events: hurricanes and earthquakes. They generated hypothetical storms from statistical distributions and historical loss data, and the market made do with that architecture for decades. KCC argues that secondary perils, a shifting climate, and faster scientific discovery have outgrown it. For the perils where traditional statistics fail most, KCC has built high-resolution physical models.

These models work differently. Instead of relying on what has happened before, KCC's physics-based approach uses atmospheric equations and large volumes of environmental data to simulate weather systems as they develop. The severe convective storm model processes more than 30 gigabytes of satellite, radar, and weather data each day, generating hail, tornado, and wind-intensity footprints. LiveEvents, a KCC process, has compared modeled estimates against actual claims since 2018. The archive built from that work holds more than 100 terabytes of atmospheric data. Claims data alongside it runs to tens of billions of dollars. KCC describes this archive as the foundation for its next generation of AI-informed physical models, with output feeding catastrophe pricing, reinsurance decisions, claims management, and loss reserving.

For ILS investors, the update cycle matters most. KCC says AI-informed models could be refreshed in days, not months, putting pricing and underwriting on fresher information. A model that refreshes that quickly changes the rhythm of cat bond spreads and collateralized reinsurance terms. The effect runs both ways: more current models should price risk more accurately, yet faster revisions also mean faster repricing whenever the view of risk shifts. The speed carries particular value in a market where Insurance Capital Daily has covered cat bond sponsors testing softer terms while spreads hold firm.

Severe convective storms are KCC's focus — derechos in particular, because they trigger large losses and resist prediction. KCC says its researchers are training machine-learning systems on tens of thousands of radar images labeled by meteorologists, teaching the models to recognize different convective forms. The August derecho that pushed 2026 US storm losses past $35 billion is precisely the kind of event these models are meant to catch sooner. Other modelers are working toward the same goal: EigenRisk added hail footprints to its platform this month, Insurance Capital Daily reported.

Capital management is where the larger stakes sit. Insurance Capital Daily has reported AM Best's estimate that global reinsurance capital reached a record $705 billion — more money pursuing a smaller set of risk budgets. Faster model updates feed directly into allocation: which perils get written, how quickly limits are repriced, how much collateral stands behind an exposure. For collateralized reinsurers, an update touches both pricing and the capital they have to set aside. KCC's white paper does not spell out those collateral mechanics, but the logic runs through them.

None of this matters if the models cannot predict losses better. KCC has spent eight years comparing live estimates against actual claims, and that database is the evidence. A model that earns its speed by proving out against tens of billions of dollars in losses will find buyers in the ILS market. The next severe convective storm season will show whether the claims record backs the speed.