AI in cat modelling will support ILS portfolio optimisation

Analysis summary

Subscribe to get comprehensive access to our exclusive analysis in the full platform. 

AI-enhanced cat models promise more accurate pricing and faster post-event analysis.

AI in cat modelling is starting to support ILS portfolio optimisation, with improved modelled outputs already enabling more accurate pricing, sources canvassed by this publication said.

Machine-learning AI, which can detect patterns within datasets and formulate predictions without humans providing frameworks, is being used to discern if there are links between weather systems in different parts of the world that can improve event predictions.

When being used together with satellite imagery to capture granular data about properties across wide geographic areas much faster than would have been possible in the past. This is particularly helpful for modelling secondary perils, which tend to impact properties in a smaller area compared to peak perils.

Looking ahead, ILS managers also expect AI to make it easier to interact with cat models, enabling queries to be made in everyday language rather than needing technical skills.

AI enhances exposure insights

AI-driven model improvements are making exposure data more accurate at primary insurance level, which will "ultimately feed downstream into more accurately priced ILS", a source said.

One key advance involves identifying relationships between past weather patterns and insured loss events that were not previously known. This can help underwriters to analyse potential impacts on portfolio exposures from weather patterns as they are developing.

“Modelling of perils has been siloed, with no explicit consideration of the relationship because the science has been immature,” explained Jay Guin, executive VP and chief research officer at Verisk.

“Unifying the models under one umbrella helps insurers to understand the correlations between risks,” Guin added.

One source said their firm had used AI to analyse storm steering – the atmospheric pressure and winds in the middle and upper layers of the atmosphere, which can influence hurricane strength and path. The analysis identified storm steering conditions that would increase or decrease the probability of landfalling hurricanes in certain US states.

"These findings could be used to determine if a deal was under- or over-priced based on the modelled outputs, and we can flex our exposure to those states accordingly," the source said.

Improved secondary peril modelling

AI-driven advances in secondary peril modelling are also helping to enhance ILS managers’ confidence when investing in those risks.

Applying AI to modelling wildfire, severe convective storm and winter-storm has been a focus for modellers as they involve “big data problems,” noted Dan Ward, senior director, product development at Karen Clark and Company (KCC).

In the case of wildfire, Moody's RMS is using AI to gather satellite imagery on individual properties more frequently, with more granularity and across a wider geographic area.

In an analysis, based on an AI-enriched update to its offering, the firm said the average annual expected loss on wildfire increased by 30% in around 6% of properties in a simulated portfolio of around 766,000 buildings across the US.

While for most homes adjustments to expected losses fell in the range of minus 5% to 5%, there were changes that indicated significantly under- or over-priced risk beyond this range.

“At an aggregate level, we might not see too much of a change, but there are extremes at both ends that would indicate there is bad risk underpriced, and good risk overpriced,” said Firas Saleh, director of product management at Moody’s.

MCPs welcomed by managers

ILS managers also anticipate that model context protocol (MCP) will enhance model effectiveness and access. MCP allows AI to access proprietary data where it has permission to do so, to blend it with opensource data, and to let users query it in everyday language.

This opens the prospect of models being more powerful and able to be queried by a range of people within a firm, not only analytics specialists.

“It’s not impossible to think that the way we retrieve and access modelled loss data will become natural language-driven in the future," said Rhodri Morris, executive director and head of ILS analytics at Twelve Capital.

The ability to query models in everyday language will make producing expected loss intelligence faster post event, supporting managers to communicate more quickly to end-investors.

However, sources caution that AI-driven modelling is still in its infancy and outputs are currently being validated by comparing to those produced through existing methods.

“AI could quite easily identify a subset of similar stochastic events in a dataset, which that would allow you to come to possible loss conclusions more quickly, but these would still need to be validated," said George Cooper, head of non-life analytics at Leadenhall Capital Partners.

Moody's has introduced MCPs across its platforms, while Verisk aims to introduce MCPs next year. KCC said its scientists did “not see additional benefit to the models from MCPs at this time”.

At present, the signals are that AI use in cat modelling will improve managers’ understanding of exposure, support more accurate pricing and enable faster and more accurate post-event intelligence.

Got a tip for us? Reach out confidentially here.

Frequently asked questions

By Jai Singh
5 October, 2026

Request your free trial today to unlock our complete intelligence platform.