Wildfire risk modeling has long leaned on historical loss data to predict future fires, an approach that left insurers blindsided by catastrophic blazes in areas with no recorded fire history. Andrew Notohamiprodjo (pictured), chief data officer at Delos Insurance Solutions, said the company correctly flagged risk in areas like the Palisades and the site of the Camp Fire precisely because it abandoned that historical dependency in favor of a science-based model built specifically for the wildfire peril.
“Traditionally, fire was treated like a general peril rather than a specialist one,” Notohamiprodjo said. “The majority of losses don't happen in that general 85% of behavior, it happens at the 1% of the 1%.” He said models trained on historical information alone marked the Palisades and the Camp Fire area as lower risk simply because no large fires had occurred there since 1985, while areas like Oakland were flagged as extremely high risk based on a single past event despite years of subsequent mitigation work.
Delos built its wildfire model in partnership with a research think tank, designing it to simulate the worst-case physical scenario at a given location rather than extrapolate from historical loss curves. Notohamiprodjo said the approach also accounts for the human dimension of fire risk, noting that 85% of all ignitions today are human-caused, along with suppression response, evacuation behavior and property protection efforts that fall outside a purely environmental model.
The company's machine learning ensemble draws on remote sensing, vegetation modeling, and data from local fire stations and CAL FIRE, but Notohamiprodjo said the science team reviews every component the model raises or lowers rather than letting the system optimize unsupervised. “Science is present at every step of the way,” he said. “Location matters significantly.”
Delos has operated a cloud-based data lakehouse and machine learning operations infrastructure since its founding in 2017, a system Notohamiprodjo said allows a single scientist to submit hundreds of simulation models at once and review the results in an organized way. The recent wave of large language models, he said, has not changed how Delos differentiates itself in wildfire catastrophe modeling and risk assessment innovation, but it has sped up cross-departmental communication and the manual review of log outputs that previously consumed scientists' time.
“What AI allows Delos to do is scale the busy work,” Notohamiprodjo said, pointing to large language models' ability to scan textual log data for the few meaningful signals among hundreds of data points and to translate numerical findings into context for both internal teams and external reinsurance and carrier partners.
As Delos looks to expand into other catastrophe-prone regions of the western United States, Notohamiprodjo said the wildfire model itself requires little retooling because a US-wide version has already been operating for roughly two years. The wildfire model is one component within Delos' broader insurance model, which also covers hazard assessment, aggregation, probable maximum loss, pricing and the final underwriting product, an approach detailed further in the company's recent coverage of insurtech expansion into high-risk property markets.
Notohamiprodjo said the wildfire model will not be retrofitted to cover flood, hurricane or earthquake risk. It is the underlying hypothesis, embedding scientific expertise directly into the modeling workflow at a hyperlocal level, that can scale to other perils, provided each is handled by peril specialists rather than generalists. “Flood people are modeling flood, not health actuaries being asked to pivot,” he said, a specialization he called central to broader industry trends in catastrophe modeling and reinsurance strategy.
Asked what separates companies that will succeed in this environment, Notohamiprodjo pointed to fundamentals over hype. “The future is bright with the advent of AI and LLMs,” he said. “But the organizations that will succeed are the ones that can ride the rails of firm fundamentals and a clear vision.”
