ZestyAI for Property-Level Climate Risk Scoring

In-Depth Review

ZestyAI builds AI models that score individual properties for climate-related peril risk. The core products are Z-FIRE (wildfire), Z-FLOOD (inland flooding), and Z-WIND (hurricane, wind, and hail). Each model operates at the property level, meaning two houses on the same block can receive different scores based on their specific building characteristics, surrounding vegetation, terrain, and exposure.

What ZestyAI Does for Insurance Professionals

Traditional catastrophe models and hazard data providers typically operate at the ZIP code, census tract, or grid-cell level. A carrier writing in a wildfire-adjacent territory gets the same hazard score for every property in the zone, which forces binary decisions: write the entire territory or impose a moratorium. ZestyAI’s property-level approach lets carriers differentiate within territories, identifying the specific properties where risk is genuinely elevated versus those where defensible space, construction materials, or topography reduce actual exposure.

The practical implication is selective underwriting. Instead of declining every application in a fire-prone ZIP code, a carrier using Z-FIRE can identify which properties have adequate defensible space, non-combustible roofing, and favorable terrain positioning, and write those while declining or surcharging the genuinely high-risk addresses.

Key Capabilities

Z-FIRE is ZestyAI’s flagship model and the first AI-based wildfire risk model approved by a state insurance regulator (California Department of Insurance). It scores properties based on vegetation density and type, terrain slope, defensible space, building materials, distance to wildland-urban interface, and historical fire behavior in the surrounding area. The regulatory approval matters because it establishes a precedent for using AI model output in admitted market rate filings.

Z-FLOOD addresses a gap in traditional flood risk assessment: inland and pluvial flooding that FEMA flood maps do not capture. FEMA maps are designed for riverine flood zones, but a significant share of flood claims originate from properties outside designated flood zones due to surface water accumulation, inadequate drainage, and localized precipitation events. Z-FLOOD scores these risks using property-specific elevation, drainage patterns, and historical precipitation data.

Z-WIND covers hurricane, wind, and hail risk for Gulf Coast, Southeast, and Midwest territories. The model considers construction type, roof characteristics, building age, and exposure to prevailing storm tracks. For carriers where wind and hail are the dominant loss drivers, this provides a more granular risk view than territory-level wind pool assignments.

Pricing

Enterprise contracts only. No public pricing. Typical deal structures involve per-property scoring fees or annual portfolio licenses. Ask about: minimum volume commitments, per-peril pricing (whether you can license Z-FIRE without Z-FLOOD), data delivery format (API vs. batch), and refresh frequency (how often scores are updated as new data becomes available).

Honest Assessment

ZestyAI’s property-level approach is genuinely differentiated. The question is whether the practical lift over your current rating factors justifies the cost and integration effort. The California regulatory approval is a strong signal, but it does not automatically transfer to other states. Each state DOI has its own review process for AI-based rating models, and carriers should budget time for compliance work.

Model transparency is a consideration. ZestyAI provides scores, not the underlying model weights or feature importance rankings. If your actuarial team needs to explain the score to a regulator or to internal stakeholders, ask ZestyAI for model documentation that describes the inputs and methodology at a level sufficient for your filing requirements.

One Thing to Test Before Committing

Request Z-FIRE and Z-FLOOD scores for 500 properties from your current book in a high-peril territory, including properties with known claims in the last five years. Build a simple lift chart: do ZestyAI’s high-risk scores correlate with your actual loss experience? The separation between claimed and unclaimed properties in your own data, not in ZestyAI’s published validation statistics, is the metric that determines whether the model adds real value to your underwriting.