## TL;DR
AI is rewriting insurance -- from Lemonade issuing policies in 90 seconds to computer vision assessing car damage from photos to satellite AI evaluating property risk without inspection. Insurtech AI makes coverage faster, fairer, and more data-driven than traditional actuarial models.

## Core Explanation
Insurance AI: (1) Underwriting -- AI evaluates risk from alternative data: telematics (driving behavior), satellite imagery (property condition), IoT (smart home sensors). Replaces generalized demographic proxies (age, zip code) with individual risk signals; (2) Claims -- computer vision assesses damage (auto, property) from photos. NLP reads medical records and police reports. Fraud detection: ML identifies suspicious claim patterns (staged accidents, exaggerated damage); (3) Pricing -- generalized linear models (GLMs) extended with gradient boosting and neural networks. Telematics-based UBI (Usage-Based Insurance): pay-per-mile, behavior-adjusted premiums; (4) Customer -- AI chatbots handle policy questions, claims filing, and renewals.

## Detailed Analysis
Lemonade: full-stack AI insurer. AI chatbot handles everything from quote to claim. Claims paid in 3 seconds (world record for AI claims). Root: smartphone app measures 200+ driving variables over 2-3 week test period. Only insures good drivers, leading to favorable loss ratios. Zesty.ai: wildfire risk model (Z-FIRE) analyzes 100+ property features from satellite + aerial imagery with 10cm resolution. CNN assesses roof condition, defensible space, surrounding vegetation. Tractable: AI auto damage assessment used by 20+ global insurers. Computer vision identifies damaged parts, estimates repair cost, and flags potential fraud (damage inconsistent with reported accident). Key regulatory challenge: insurance is heavily regulated by state. AI pricing models must demonstrate they don't discriminate against protected classes. Explainability (SHAP, LIME) is essential for regulatory approval and consumer trust.