Not postcodes. Individual properties. Every one.
We built an AI risk-intelligence platform for property and casualty insurers — pricing catastrophe risk at the individual property level, across wildfire, hail, wind, storm and water perils, with five separately regulator-approved models.
The platform covers 99.7% of U.S. properties at 95%+ verified accuracy — delivering up to 62x better risk segmentation than traditional actuarial methods.
Insurers priced risk by postcode. Regulators demanded transparency the methods couldn't provide.
Property and casualty insurers were grouping thousands of properties into single actuarial buckets based on postcode — meaning a well-built, low-risk property paid the same rate as a poorly-maintained one next door. Portfolios were systematically mispriced.
Regulators were simultaneously tightening requirements for transparency in catastrophe risk pricing — demanding that insurers explain their methodology at the individual property level, not at the postcode. The blunt actuarial tables couldn't provide that.
200 billion data points. Computer vision. Five regulator-approved models.
200B+ Property-Level Data Points
Training data drawn from 200 billion+ individual property observations — roof geometry, materials, vegetation proximity, construction year and historical claims — at the parcel level.
Computer Vision over Aerial Imagery
Aerial imagery processed by computer vision models that read roof geometry, surface materials and surrounding vegetation — inputs no actuarial table could ever encode.
Gradient-Boosted Tree Models
Purpose-built gradient-boosted tree models for each peril — wildfire, hail, wind, storm and water — trained on 20+ years of claims data per peril type.
Five Regulator-Approved Peril Models
Each of the five peril models has been individually approved by state regulators — so insurers can use them in rate filings with full regulatory standing.
LLM-Assisted Regulatory Filing Analysis
An LLM layer reads and analyses regulatory filing requirements per state — ensuring model outputs align with each jurisdiction's disclosure and methodology standards.
62x better segmentation. 99.7% of U.S. properties. Regulator-approved.
Better Risk Segmentation
Up to 62 times better differentiation between high and low-risk properties than traditional actuarial postcode grouping.
U.S. Property Coverage
Near-complete coverage of the U.S. residential property stock — priced at the individual parcel level, not the postcode.
Verified Model Accuracy
Independently verified accuracy across all five peril models — with each model holding individual regulatory approval.