AI already runs through claims.

AI is already common in insurance. In surveys by state regulators, 88% of auto insurers and 70% of home insurers said they use AI or plan to. Claims is where most of it runs. Among home insurers, 54% used AI in claims, more than in any other part of the business.

The money is real. Aviva, one of the largest UK insurers, says more than 80 AI models in its motor claims saved more than £60 million in 2024. It says complex cases now take 23 days less to assess. It also uses AI to summarize doctors' reports for life cover, while its underwriters still make the call.

But insurance has something most industries do not: a regulator who can ask to see how every decision was made.

The rules have caught up. Twenty-five states and Washington, D.C. have adopted a model rule from the national body of insurance regulators. It says the law has not changed. A decision made with AI must follow the same rules as one made by a person. And an insurer that buys a vendor's model still owns every decision it makes.

The courts are watching too. A lawsuit says UnitedHealth used an AI model to cut off care for elderly patients. It claims nine in ten appealed denials were reversed. A federal judge let part of the case go forward in 2025. The claims are not proven. But the lesson for any insurer is plain: an AI that denies is far riskier than an AI that speeds up a yes.

Here is what that looks like in practice. A regional carrier handles 50,000 claims a year. If AI can route and settle the simple ones two weeks faster, customers are happier and adjusters spend their time on the hard cases. That is a clear gain, and every week saved is a customer not waiting on a check. If the same AI starts turning down claims, each denial needs a reason a person can explain to a customer, a regulator and possibly a judge.

Fraud is the quiet win. About four in ten home insurers use AI to flag claims that look wrong. A flag is not a denial. It sends the file to an investigator, who decides. That keeps the savings and the judgment in the right hands.

Underwriting is next. Reading applications, medical records and inspection reports takes an underwriter hours. AI can summarize them in minutes, while the underwriter still makes the call. The risk is the same as in claims: a summary can be wrong, so someone must still check the source before a customer is turned down.

There is one more catch. Hours saved are not dollars saved. An adjuster who finishes files faster lowers cost only if the company needs fewer adjusters, less overtime or fewer outside firms. Otherwise the gain is speed and service, which is worth something, but it is not the same thing.

What should a mid-sized insurer do first? Three plain things. Write down every place AI makes or shapes a decision about a customer. Name the person who answers for each one. Ask each vendor to show, in writing, how its model works and how it was tested. With those three in hand, you can answer a regulator, and you can judge any AI pitch. None of this needs a large budget. It needs someone who owns the answer, and the time to write it down.

The full guide shows where AI touches an insurer's money, what has actually happened, who gains and who loses, and what to ask before you spend.