Industry View · Insurance
Insurance is structurally an information business, which makes it the cleanest target for AI in financial services. McKinsey pegs the gen-AI revenue opportunity at $50-70bn, Conning finds 90% of insurers already somewhere on the gen-AI journey, and Chubb's CEO has committed to automating 85% of underwriting — even as a UnitedHealth class action over an algorithm with an alleged 90% error rate shows where this goes wrong.
Lemonade pet-claims cost cut 68% and improving loss ratios, McKinsey ~30% cost cut, but mostly pilots and rising legal/governance risk (UnitedHealth nH Predict)
Conversational AI and embedded quoting collapse acquisition cost, but margins flow to whoever owns the customer surface.
Gen-AI assistants ingest unstructured submissions and surface risk-appetite calls in real time, straight-through processing climbing from ~10-15% to 70-90%.
Telematics and ML pricing engines tighten loss ratios — Root and Progressive's data moats are the proof case.
Computer-vision damage estimates and LLM-drafted communications cut claims cycle time sharply — and concentrate the legal and bias risk.
Pattern-detection models flag anomalous claims earlier, with vendors citing 30%+ improvements in detection.
01 · The thesis
Insurance runs on the cost of acquiring, pricing and adjudicating information, which is exactly the cost AI compresses. McKinsey estimates gen AI could unlock $50-70bn in revenue and lift underwriter productivity by up to 50% while cutting underwriting cost by up to 30%; in selected cases it has pulled quoting times from weeks to hours and decision times to roughly 12 minutes. Conning's 2025 survey finds 90% of insurers somewhere on the journey and 55% in early or full deployment, and software-and-data spend has grown about 20% a year over the five years to mid-2025. But adjudication is also where AI becomes a liability rather than an asset. UnitedHealth is in a class action over the nH Predict tool — an algorithm plaintiffs say carried a 90% error rate, with nine in ten appealed denials reversed — and a March 2026 order compels broad discovery. Regulators are converging fast: 24+ states have adopted the NAIC model bulletin, and Colorado's quantitative bias-testing regime expanded to auto and health in October 2025. The winners will be those who treat governance and explainability as the product, not the overhead.
Conversational AI and embedded quoting collapse acquisition cost, but margins flow to whoever owns the customer surface.
Gen-AI assistants ingest unstructured submissions and surface risk-appetite calls in real time, straight-through processing climbing from ~10-15% to 70-90%.
Telematics and ML pricing engines tighten loss ratios — Root and Progressive's data moats are the proof case.
Computer-vision damage estimates and LLM-drafted communications cut claims cycle time sharply — and concentrate the legal and bias risk.
Pattern-detection models flag anomalous claims earlier, with vendors citing 30%+ improvements in detection.
02 · The two clocks
The AI-in-insurance market was about $10.4bn in 2025 and is forecast toward $13-15bn in 2026, with software-and-data spend growing roughly 20% a year over the five years to mid-2025 (Fortune Business Insights, McKinsey). Payoff clusters where the data is structured: claims, underwriting, pricing and quoting already account for 58% of disclosed AI use cases, and vendors report straight-through processing rising from 10-15% to 70-90% (Vantage Point). The gen-AI slice specifically was about $1.1bn in 2025 and is projected toward $14bn by 2035 — fast growth off a small base, with most carriers still in pilots rather than full deployment (Yahoo Finance / market research).
03 · Public players & exposure
We plot the listed players on two editorial axes — how exposed each is to AI disruption, against how ready its data, brand and position are to be the answer. The figures in the table are sourced; the placement is our read.
04 · Private flagships
The companies attacking this industry AI-first, with disclosed funding where available:
AI-native fraud detection and claims-decisioning engine sold into incumbent carriers across P&C and health.
Vision models that estimate vehicle and property damage from photos to accelerate claims settlement.
Digitizes and triages commercial underwriting submissions so risks route straight to the right appetite.
Transparent ML pricing-and-reserving platform built for actuaries, emphasizing explainability for regulators.
Generative-AI assistant that summarizes submissions and surfaces risk signals inside the underwriting workflow.
05 · Signals
06 · The exposure read
AI rewards clean, structured advantage and punishes friction. The line runs through who owns the data, the brand and the customer — and who is merely a step the technology can route around.
Sources
AI-in-insurance market: 2025 base vs near-term forecast
Source: Fortune Business Insights, AI in Insurance Market (2025); gen-AI sub-segment per market-research estimates (US$ bn)
Verisk ships gen-AI underwriting + XactAI
A data incumbent moving from raw data to AI-native underwriting and claims tooling signals where the margin pool is consolidating.
Colorado bias testing expands to auto and health
Quantitative bias testing under C.R.S. 10-3-1104.9 is now an enforceable obligation, not a voluntary bulletin or future date.
Lemonade posts company-best loss ratios
A 52% Q4 gross loss ratio and 31% premium growth start to validate the AI-native cost structure thesis.
Court compels UnitedHealth AI discovery
Broad discovery into nH Predict turns algorithmic claims denial into a live litigation and disclosure risk for the sector.
Half of US states adopt NAIC AI bulletin
24+ states adopting the model bulletin creates a de facto national governance baseline insurers must build to.
AI-native fraud detection and claims-decisioning engine sold into incumbent carriers across P&C and health.
Vision models that estimate vehicle and property damage from photos to accelerate claims settlement.
Digitizes and triages commercial underwriting submissions so risks route straight to the right appetite.
Transparent ML pricing-and-reserving platform built for actuaries, emphasizing explainability for regulators.
Generative-AI assistant that summarizes submissions and surfaces risk signals inside the underwriting workflow.
| Company | Stance | The sourced fact |
|---|---|---|
| Progressive PGR | Data moat | Snapshot telematics feeds ML pricing across millions of drivers, plus AI photo-damage estimating in claims (Klover.ai). |
| Chubb CB | Aggressive adopter | CEO Evan Greenberg has committed to automating 85% of underwriting using AI (Insurance Business / Deloitte). |
| Allstate ALL | Workflow gen-AI | Uses LLMs trained on internal data to draft claim communications inside adjusters' workflow (Emerj). |
| UnitedHealth UNH | Legal exposure | Class action over nH Predict alleges a 90% error rate; March 2026 order compels broad discovery (CBS News, DLA Piper). |
| Lemonade LMND | AI-native, scaling | In-force premium up 31% YoY to $1.24B in Q4 2025; pet claims cost cut 68% from $44 to $14 (Lemonade 8-K). |
| Root ROOT | Telematics pricing | Gross loss ratio of 59% in Q3 2025, below its 60-65% long-term target (Root 8-K). |
| Verisk VRSK | Picks & shovels | Launched a Commercial GenAI Underwriting Assistant and XactAI claims suite in September 2025 (Verisk newsroom). |
| Guidewire GWRE | Core platform | Cloud core-system incumbent embedding gen-AI across underwriting and claims workflows (Appit comparison). |
| Duck Creek DCT | Agentic platform | Launched an insurance-native agentic AI platform with a Model Context Repository and AI Assurance layer (PRNewswire). |
| Berkshire/GEICO BRK | Late catch-up | GEICO is rebuilding its tech stack and telematics to close a pricing-sophistication gap on Progressive (industry coverage). |