Moody's has found that artificial intelligence-driven property data can materially change how US severe convective storm risk is priced and allocated across insurance portfolios.
According to a new white paper, Moody's said that adding enriched attributes led to a reduction of about 5% in modeled average annual loss for severe convective storm. At portfolio level, that suggests building stock that is somewhat stronger than the model’s default assumptions.
Moody’s stressed, however, that this small headline change masks much larger movements underneath. Nearly half of all properties recorded a change in modeled loss of more than 15%, either up or down, once enriched data were applied. More than 226,000 locations saw material increases in modeled loss, while over 266,000 saw decreases of similar magnitude.
For insurers, the finding is that a modest portfolio credit can conceal a major reshuffling of risk within the book. From an underwriting, pricing and accumulation-management standpoint, that redistribution may matter more than the top-line shift because it alters where loss is expected to concentrate in future events.
The study highlighted Texas as an example of how regional experience and building practices affect modeled results when secondary data are available. After enrichment, state-level modeled loss for severe convective storm fell by about 11%.
Moody’s linked this to the impact of repeated hail events on the building stock.
In recent years, Texas, particularly the Dallas–Fort Worth area, has suffered multiple large hailstorms, prompting widespread roof replacement and upgrades. In the sampled portfolio, more than two-thirds of Texas properties were assessed as having roofs in excellent condition, and many had brick veneer cladding, which performs better against hail than more fragile materials.
At county level, this translated into double-digit reductions in modeled loss for many locations. Within the same state, however, properties with older roofs and more vulnerable cladding experienced the opposite effect, with enriched data pushing modeled loss higher. The ability to distinguish systematically between recently upgraded homes and those where vulnerability persists is presented as a potential underwriting edge.
Moody's white paper also examined how enriched data interact with different sub-perils within the severe convective storm model.
According to the results, in hail-dominant regions, such as Texas and the Great Plains, dense tree cover can shield roofs and exteriors, reducing hail impact. In these areas, properties with high tree density tended to see decreases in modeled loss after enrichment.
In contrast, in regions where straight‑line wind is the main driver, such as parts of the Northeast and California, the same trees become a debris hazard, raising the risk of impact damage. There, enrichment led to increases in modeled loss for tree‑dense properties.
Meanwhile, to illustrate the potential underwriting impact, Moody’s compared two three‑story homes in Aurora, Colorado, a city in a high hail‑frequency corridor.
One property recorded a 62% increase in modeled loss, driven by wood cladding, a steep gable roof with more exposed surface area, a roof‑mounted solar array and limited shielding. The other saw a 23% reduction, reflecting brick veneer, a hip roof, some tree protection and no rooftop equipment.
The higher‑risk house is the newer of the two based on year built. In the enriched run, specific secondary modifiers outweigh age. Without this information, a carrier might treat the risks similarly or even favor the newer build. Enrichment allows the insurer to price the more resilient property more competitively and flag the other for closer review.
The analysis suggests that systematic enrichment of exposure files can support more granular underwriting rules and pricing, rather than relying solely on standard submission fields and broad territory relativities. Over time, portfolios assembled with this type of property‑level discipline may show different concentration and volatility characteristics than those built on coarser assumptions.
Reinsurers can use enriched data to benchmark cedant portfolios beyond aggregate metrics such as total insured value and average age. A book skewed toward well‑maintained, recently re‑roofed homes is not equivalent to one dominated by older, deteriorating stock, even if their headline figures are similar. Consistent secondary attributes across all locations could strengthen confidence in cedant submissions and inform treaty structures.
The study also raises questions about how often exposure data should be refreshed. Many of the attributes with the greatest modeled impact — such as roof age and condition, vegetation and cladding — change over time as properties are maintained, upgraded or allowed to deteriorate. If catastrophe models rely on static data captured at binding and rarely updated, modeled results may drift away from the current risk profile.
Moody’s argued that aerial imagery–based enrichment can provide a scalable way to keep key secondary modifiers current across large books, improving the explainability and stability of model output.
As the US property catastrophe market moves into what some view as a softer phase after several years of rate hardening, the ability to separate higher‑ and lower‑risk properties within seemingly homogeneous books may help determine which insurers grow profitably and which accumulate exposure on less informed terms.
