| By Gautam Naik Insurers are betting on a suite of new AI-driven techniques to better predict surging losses from climate-driven weather catastrophes ranging from unprecedented wildfires to hurricanes and floods. Less than three months into the year, natural disasters are already causing major economic disruption around the world, including recent fires across Los Angeles, an economy-denting cyclone in Australia, floods in Jakarta and a giant storm that left dozens dead in the US. According to a recent report by broker Gallagher Re, annual insured catastrophe losses of $150 billion have become "the new normal." While traditional models apply complex physics and elaborate computer simulations to estimate the probability of future losses, the results often can fall short. Flood models designed to measure the same risk have yielded conflicting outcomes. Wildfire models can struggle to accommodate the dizzying number of variables in play—everything from the role of human intervention to the possible flight path of a wind-borne ember. Some investors in catastrophe bonds expressly shun securities exposed to such perils because they don't trust the modeling. Every model "is an imperfect representation of a very complex phenomena," said Firas Saleh, director of product management at Moody's Corp. That's where artificial intelligence comes in. Its proponents contend it can provide a more accurate estimate of property-level risk for weather calamities.  Homes destroyed by the Palisades Fire in the Pacific Palisades area of Los Angeles, in January. Photographer: Jill Connelly/Bloomberg Once an AI model is trained on what it's supposed to look for, it becomes part of a modeling process that churns through huge volumes of data from aerial and satellite imagery to create a physical assessment of an individual property, eventually turning it into a risk report. Is the house made of brick or wood? Are there pine needles in the gutters, making combustion more likely? Humans would find it nearly impossible to perform such a detailed model and analysis, especially if a portfolio contains thousands or millions of homes. AI has now become invaluable to insurers seeking to price the risk, set premiums and have enough capital to pay claims from the worst catastrophe losses. "Weather and catastrophe losses are running ahead of the ability to manage them, and many insurers are having trouble sustaining their business because they're not getting the right rates," said Jay Guin, chief research officer of the extreme event solutions team at Verisk, a catastrophe modeling firm. "AI changes the equation." Zurich Insurance Group AG, one of the largest insurers in Europe, uses AI powered risk-modeling software to assess catastrophe risk and often tweaks it for its own purpose. "If there's fire hazard like vegetation, overhang or debris in your backyard that shouldn't be there, we can tell you to lower the risk otherwise we may not be able to underwrite you," said Ericson Chan, chief information and digital officer of the Swiss company. The demand for granular data has spurred risk modelers to boost investment in AI. Verisk's wildfire model has the usual ingredients, such as wind speed, vegetation growth and the impact of climate change. But the company says it also offers clients an extra layer of data: an AI-based analysis of homes, using images taken from satellites and low-flying aircraft. Its rival, Moody's Insurance Solutions, is pursuing a similar path. It recently acquired CAPE Analytics, which also uses AI techniques to provide "instant risk insights at the individual address level."  A driver waits with his tractor-trailer after it overturned during a possible tornado in Villa Ridge, Missouri, on March 14. Photographer: Robert Cohen/St. Louis Post-Dispatch Risk experts are increasingly using AI to crack an especially intractable challenge: modeling hailstorms, thunderstorms and tornadoes, collectively known as severe convective storms, or SCS. Insured losses from SCS events totaled $61 billion in 2024, the second-highest level on record, according to Aon Plc. Texas, which receives more hail claims than any other US state, recently approved an AI-driven model for SCS developed by ZestyAI, a San Francisco-based company. SCS losses are hard to model because insurance claims aren't always associated with a single, clearcut event like a hurricane or a wildfire. Instead, the damage often occurs from many small, even unnoticeable hail events that gradually impair a roof or home. "You don't get a cavity because you ate a large Mars bar," said Kumar Dhuvur, ZestyAI's co-founder and chief product officer. "You get it because you ate a lot of candy over a long period of time." The firm's SCS model uses a form of AI known as "machine learning," bringing together a complex set of variables: local geography, climate effects, a 3-D analysis of each building and roof and accumulated damage from historical storms. The model assembles the data into "hail scores" ranging from one to 10, which clients can use in the premium-setting process. "Insurers need a sharper tool," Dhuvur said. In the US, where there are about 100 million properties, "if you do it by hand, it won't compute—you need AI." Amica Mutual Insurance Co. says it uses AI-based models to get an edge in the Dallas-Ft. Worth area of Texas, where hail losses are high. The company uses a ZestyAI model to identify local homes that are at lower risk, based on individual features such as age, pitch and material of the roof. That allows it to price its cover more competitively. "We can keep our premiums more stable and that should enable us to grow in areas where other insurers can't grow as profitably," said William Pitts, an Amica managing vice president. Read and share this story on Bloomberg.com. | 
No comments:
Post a Comment