AI and life insurance: Fast today, unpredictable tomorrow

Artificial intelligence is changing life insurance underwriting by making simple processes faster.

How far it will go from here is the big unknown.

"Carriers are still very much experimenting with how to bring AI into the underwriting process," said Michael Niemerg, principal and director of data science and analytics at Milliman.

Niemerg spoke during a Tuesday session on the AI impact on underwriting at the LIMRA Life Insurance and Annuity Conference.

Panelists agreed that AI is rapidly reshaping underwriting, improving efficiency and data analysis while raising new questions about accuracy, governance and the role of human decision-making.

Three main categories

Insurers are deploying three main categories of AI — generative AI, machine learning and rule-based systems — across underwriting workflows, Niemerg explained.

Generative AI, including large language models, has become the most visible form of the technology, widely used for tasks such as summarizing medical records and streamlining documentation.

“That’s how most people interact with AI in the real world,” Niemerg said. “It’s the most democratized version.”

In underwriting, generative AI is already being tested to condense complex documents such as attending physician statements, which often contain large amounts of unstructured text.

Machine learning, by contrast, has been used for years behind the scenes to predict outcomes such as mortality and morbidity risk. These models analyze structured data to generate risk scores that can guide underwriting decisions.

The third category, rule-based or “expert” systems, remains a foundational part of underwriting. These systems rely on predefined logic and decision trees to automate portions of the process.

All three forms of AI are increasingly converging within underwriting systems, Niemerg said, supported by a growing volume of data.

Data access grows

Over the past decade, insurers gained access to expanding data sources, including prescription histories, medical claims, electronic health records and traditional application data. Although these inputs can improve risk assessment, they also add complexity.

Electronic health records, for example, offer detailed clinical insights but are often difficult to process due to inconsistent formats and unstructured content.

“It's very semi-structured at an individual basis, and there's all sorts of versions of EHR," Niemerg said. "So it's very hard data to wrangle, put together and interpret."

AI is being applied across the underwriting lifecycle, from initial application intake to final decision-making. At the front end, new tools can analyze applicant behavior in real time — such as detecting anomalies in digital applications — to flag potential risks or misrepresentations.

AI is also helping insurers refine how they collect data, allowing companies to prioritize lower-cost sources and avoid unnecessary requirements.

Further into the process, predictive models and enhanced rules engines are enabling faster triage of applications, allowing simpler cases to be processed quickly while directing more complex cases to human underwriters.

However, experts cautioned that AI-driven summarization still requires human oversight.

“If underwriters ... have to go and check each fact, how much automation did you really create?” Niemerg noted.

Long ways away

Despite advances, panelists agreed that fully automated underwriting decisions driven by generative AI remain years away.

A central concern is reliability. Unlike traditional models, generative AI systems can produce different outputs from the same input — a characteristic known as stochastic behavior — making them harder to audit and trust.

“You need to know what level of accuracy you need," Niemerg said. "Sometimes, 90% accuracy is very useful, because you might have humans who are going to validate that information. Sometimes, you need 100% accuracy."

Testing has shown that even minor or irrelevant changes in input can alter AI-generated decisions, raising concerns about unintended bias or instability.

As a result, insurers are focusing on “human-in-the-loop” approaches, where AI provides recommendations but underwriters retain final authority.

In parallel, companies are building governance frameworks to manage AI use. These include clear definitions of what constitutes AI, documentation standards, monitoring systems and cross-functional oversight involving data science, actuarial, compliance and technology teams.

Looking ahead, industry leaders expect underwriting to become faster, more data-driven and increasingly automated — but not fully autonomous.

They also anticipate a shift in the role of underwriters, who may spend less time gathering data and more time making complex judgments and overseeing automated systems.

“Anytime something gets cheaper, easier to do, you're going to do more of it," Niemerg concluded.

© Entire contents copyright 2026 by InsuranceNewsNet.com Inc. All rights reserved. No part of this article may be reprinted without the expressed written consent from InsuranceNewsNet.com.

InsuranceNewsNet Senior Editor John Hilton has covered business and other beats in more than 20 years of daily journalism. John may be reached at [email protected]. Follow him on Twitter @INNJohnH.