Insurance: the full guide

An insurer sells a promise. It collects a premium now and pays claims later. Its margin is what is left after claims and expenses, plus what it earns investing the money in between.

Five pools of money decide an insurer's year.

What sets insurance apart is oversight. A retailer can change a price without asking anyone. An insurer's rates, forms and claims practices answer to regulators in each state where it sells. So every place AI touches the money is also a place a regulator can ask how it was done.

Capacity is not cost. Faster files are worth money only if the insurer needs fewer adjusters, less overtime or fewer outside firms, or wins business it would have lost.

Approvals and denials are not the same. AI that speeds up a yes carries little risk. AI that shapes a no must be explainable to the customer and to the regulator.

Two patterns stand out. The results that hold up keep a person on the decision that affects the customer. And the risk in this industry is not that AI fails quietly; it is that a denial made with AI has to be defended in public.

Scenario 1: The examiner asks for your AI program. State examiners start asking, during routine reviews, for a written list of AI systems, who owns each, and how each was tested. Assumes: the examiner tool now being piloted is adopted. Signal to watch: the national regulators' body formally adopts its AI evaluation tool, or your state includes AI questions in a market conduct exam. Last checked: September 2026. Twelve states began piloting the tool in March 2026.

Scenario 2: Simple claims settle with no person involved. Insurers let AI handle small, clear claims from first call to payment, and keep people for the complex ones. Assumes: customers accept it and regulators allow fully automated approvals. Signal to watch: a large insurer reports a meaningful share of claims settled end to end by AI. Last checked: September 2026. Aviva has said it is testing a voice-enabled AI claims agent for simple claims.

Scenario 3: The vendor's model becomes your lawsuit. A decision made with a vendor's model is challenged, and the insurer cannot explain how the model works. Assumes: vendors keep their models closed. Signal to watch: a court or regulator holds an insurer responsible for a vendor model's decision. Last checked: September 2026. Under the model rule, buying a tool does not transfer the insurer's liability.

All three reward the same preparation: a written inventory of AI decisions, a named owner for each, and documentation from every vendor that an examiner could read.

Large insurers gain first. They can afford data teams, model testing and compliance staff, and they shape what regulators come to expect.

Mid-sized insurers that start in claims gain. Routing, document reading and fast-track approvals pay without touching the riskiest decisions. A carrier that keeps a person on every denial can move quickly and defend what it did.

Agents and brokers change role. When customers and insurers both use AI to compare and quote, the agent's value moves from paperwork to advice on complex risks. Agencies that use AI for the paperwork keep more time for the advice.

Vendors hold more power, and more of the risk sits with the insurer. An insurer that cannot see inside a vendor's model still answers for it. Contracts that require documentation and testing records shift some of that balance back.

Policyholders gain speed, and some face new risk. Simple claims get paid faster. A customer whose claim is denied by a system nobody can explain faces a harder appeal, which is exactly what regulators and courts are now examining.

The cost ranges are illustrative and will be replaced with reviewed figures. The pattern holds regardless: start where AI speeds up a yes, and keep a person on every no.

One test before any contract. Ask the vendor to run its model on last year's closed claims or applications and show where it would have agreed or disagreed with what your people decided. Then ask for the documentation you would hand an examiner.

Keep the rules that do not change. Under the model rule adopted by 25 states, a decision made with AI must meet the same standards as one made by a person, and the insurer answers for a vendor's model.

Your customers: businesses and households. Commercial customers using AI to run their operations change their risks, sometimes lowering them, sometimes creating new ones such as AI errors and cyber exposure. Read the manufacturing, construction and real estate guides for how your commercial customers are changing.

Your distribution: agents and brokers. Agents using AI to compare quotes and fill applications will send more submissions, faster. Carriers that can answer quickly win the first look.

Your suppliers: data, models, repair and medical networks. Repair shops, medical reviewers and data providers are adding AI to their own work. Their outputs flow into your decisions, and under the model rule their errors can become yours. Read the legal guide for how AI is changing the advice you buy.