Insurance underwriting & claims make the shift to AI tools

Listen to this article

Insurance underwriting – the process of evaluating an individual or organization’s risks – is changing rapidly due to the advance of generative artificial intelligence (AI) tools and use and so too are the regulations that guide it.

“ AI in insurance underwriting and claims administration has been a hot topic over the last two years or so,” said, Lauren R. Mendolera, an attorney who is a partner at Harter Secrest & Emery, LLP, who explains that while insurance companies have used machine learning for many years, the implementation and regulatory overlay of generative AI is different.

The biggest drivers for insurance companies to use AI in their underwriting process, Mendolera says, are a reduction in overhead and a decrease in the time it takes to underwrite.

One technical analysis published in the International Journal of Research in Computer Applications and Information Technology in January 2025 found AI has reduced the average underwriting decision time of three to five days to 12.4 minutes for standard policies while maintaining a 99.3% accuracy rate in risk assessment.

“Anybody who walked into an insurance company’s building in the 80s or the 90s could see how the underwriting was done,” said Mendolera, recalling a time in underwriting that was highly personalized and involved insurance agents physically pulling different endorsements from tangible folders. “But a part of it incorporated an element of human error, which I think had implications in terms of coverage and what happens after that.”

Mendolera explains that many states have a legal doctrine called contra proferentem (Latin for ‘against the offeror’) which, in the insurance realm, says that if an insurance policy is ambiguous, it is interpreted against the insurance company and for the insured.

“[For example] if an insurance agent grabs two endorsements and they’re both contradictory, the court will say, ‘Whatever’s in favor of coverage, we’re going to interpret it that way,’ she said. “With AI, you’ve got a situation where you have a non-person who presumably can do that same process with less risk of human error. You’ve also got an efficiency in terms of turnaround time.”

While AI in underwriting can reduce mistakes that could cost insurance companies money and get quotes out quicker, there are legal and compliance challenges insurers must seriously consider.

“One of the biggest ones that seems to be the focus of a lot of state regulatory actions and proposed bills at the legislative level is protecting consumers from discrimination,” Mendolera said.

She explains the information an AI model provides is only as accurate as what it’s taught.

“So, if you’ve got a model that’s relying on metrics that under law insurance companies aren’t allowed to rely on, then you’re going to create an AI model that is inherently discriminatory, or in some way violates laws,” Mendolera said.

New York addressed this topic last year when the Department of Financial Services, which is the entity in New York that regulates insurance companies, finalized a circular to provide guidance regarding the use of AI in underwriting and pricing decisions.

“It explained that AI used by insurance companies shouldn’t be used unless the insurance company can be confident that it can be used in a way that is compliant with state and federal laws, especially as to not be discriminating against protected class members,” Mendolera said. “Like discriminating against people based on race and giving different price quotes because of race or gender or sex or things of that nature.”

Joseph Wilson, special counsel at Barclay Damon LLP, says 27 states have some form of regulation or guidance on the use of AI by insurance companies.

He explains that for underwriting one key role of AI is to analyze large amounts of data to better predict risks. For example, software that recognizes patterns, analyzes large datasets, learns from those datasets, and then makes accurate predictions by analyzing new datasets.

“For property and casualty insurers, AI plays an important role in fraud detection — identifying anomalous patterns in a large set of data that suggests or identifies potential fraud,” he said. “For life insurance, AI can pull data from records — such as digital medical reports — and automate review and assessment of medicals and health risks. Essentially, AI can apply a large series of underwriting rules — thousands of rules — to assess medical risks, etc.”

However, it is critical that the data sources used by insurers are accurate and reliable and have some method of testing and verifying that AI’s decisions and outcomes are wholly accurate.

“When utilizing AI systems for functions that were traditionally performed by humans — reviewing records or investigating a policyholder’s history for example — an insurer needs to confirm that the output is accurate and complies with existing law and isn’t arbitrary or discriminatory,” Wilson said.

For example, insurers can’t unfairly discriminate among consumers.

“Even though race and ethnicity shouldn’t govern an underwriting decision, race or ethnicity may be derived from an individual’s name or locations (past and present) and could subtly impact the AI decision-making process without actually being included in the data process or dataset,” Wilson said.

Overall, haphazard or disordered implementation and use of AI have the potential to increase inaccurate, arbitrary, or discriminatory outcomes, which is something business owners should be aware of when it comes to insurance.

“While businesses may not know the extent of an insurer’s use of AI, they should be vigilant for adverse or unusual outcomes, such as insurance denials and unexpected pricing, which should be investigated to confirm the data that informs the decision is accurate,” Wilson said.

Caurie Putnam is a Rochester-area freelance writer.