Computer vision for claims
Using image analysis models to assess the extent of damage from photos or videos, motor, home, catastrophe, and speed up indemnity, with a new risk of fraud through generated images.
Computer vision covers the techniques that let a model interpret images: recognizing objects, locating damaged elements, estimating an area or degree of destruction. In claims handling it is used to estimate damage from photos sent by the insured or taken by an adjuster: identifying the vehicle parts hit and whether to repair or replace them, costing from parts and labor schedules, assessing damage to a roof or facade. After a catastrophe, it analyzes aerial or satellite images to estimate quickly which buildings are destroyed or damaged across an entire area, before adjusters can even get there. The gains are shorter indemnity times, lower adjustment costs for small claims and faster allocation of adjusters to complex files. The limits are known: hidden damage the photo does not show, variable image quality, and fraud through manipulated or generated images, which generative models make easier and against which the origin and integrity of files must be verified.
Textbook case: after a hailstorm, an insurer receives 15,000 motor claims in one week. Policyholders photograph their vehicle with the app; the model values 60 percent of files without an adjuster, within 48 hours, and routes to an adjuster the files where it detects lighting or metadata inconsistencies suggesting edited images.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.