AI’s Growing Role in the Life Insurance Sector
Artificial intelligence is beginning to reshape the life insurance industry, as both new digital players and established companies explore its potential.
AI has the capacity to significantly streamline the life insurance process, especially in refining application and review procedures. However, as AI moves closer to essential risk management roles like underwriting, it encounters more sensitive issues.
Some insurers are using AI to handle queries and certain client interactions, while others are testing the technology for underwriting.
Some firms are using generative AI to manage client inquiries and enhance client interactions. According to Laura Money, chief information and technology innovation officer at Sun Life Financial Inc. in Toronto, “We’re piloting Gen AI across Sun Life globally. For example, in the Philippines, we launched Advisor Buddy to help new advisors onboard quickly so they can serve clients sooner.”
Andrew Ostro, chief executive officer and co-founder of PolicyMe, an insurance provider based in Toronto, hopes to speed up AI adoption. PolicyMe has so far focused on using generative AI to answer advisor questions and analyze advisor phone call recordings to identify issues. But what about AI’s potential for underwriting?
“That’s exactly what AI should be used for,” Mr. Ostro says. However, he cautions that there are difficulties. “It’s going to take a long time to know if [AI-based underwriting] is right or wrong, and by then, it could be existential to a big insurer or reinsurance company.” He adds that the explainability of AI algorithms is also a challenge.
Philippe Cleary, iA Financial Group’s vice-president of underwriting, new business, and claims for individual insurance in Quebec City, says that the company already uses AI for some underwriting. The company made sure that an application underwritten automatically using its predictive model would have the same outcome as if an underwriter had manually underwritten it.
“We therefore also adopted a model risk management corporate policy in relation with the design, development and use of models,” says Mr. Cleary. “It monitors associated risks such as data quality, operational opacity, and confirmation of discrimination bias. The implementation of these policies is ongoing.”
The simplest applications for automated underwriting are instances where a policy should be accepted outright. However, Mr. Ostro views opportunities to streamline choices when approvals aren’t immediately granted. For instance, a distributor might need extra requirements, such as a nurse visit or a set of medical records. He believes that AI might refine those criteria. It might potentially offer an alternative product or reduce the number of tests required to gain approval.
Enter Agentic AI
As businesses grapple with the intricacies of the life insurance application and underwriting process, the next frontier could be agentic AI. Classical AI is trained to excel at one task, such as identifying fraud. Generative AI is skilled at many things, including data mining, writing e-mails, and transcription. However, it still acts like an intern without any initiative, requiring instructions for every micro-task.
Agentic AI is much better at dividing difficult issues into manageable steps on its own. “With agentic AI, it’s more a case of, ‘I need to solve this problem. You go ahead and figure out what to do,’” Mr. Ostro says.
This new type of autonomous AI agent could manage complex internal workflows for insurance firms. However, Byren Innes, managing director of Jennings Consulting, which consults on technology strategy for financial services companies, said that it is still largely untested in the industry.
“I’m not sure we’ve seen anything that’s been vaunted as a success story yet on this front,” he says.
Companies like Sun Life are cautiously considering the potential of agentic AI. “Eventually, we think some simple tasks will be executed using agentic AI, such as enabling our employees or advisors to order a new monitor or phone,” Ms. Money says.
Data Challenges
Data availability is a challenge in both of these use cases. AI algorithms depend on large amounts of data for training and analysis. According to Mr. Cleary, “Currently, what’s standing in the way of pushing further is that there are less data available in Canada.”
The complexities of data flows in the insurance industry are another challenge. According to Mr. Innes, agents work with several carriers, and distributors consolidate carrier data into their agent management systems. “The reality is the data are poor and inconsistent and, as a result, not trusted,” he says.
Legacy infrastructure is another problem. “One of our clients has 14 policy administration systems that they’re still running based on stacks of acquisitions they made over the decades, so they don’t have a single source of data,” Mr. Innes adds.
While traditional insurers often begin and end their data gathering with the policy application, PolicyMe is focused on offering more data for AI-powered analysis. PolicyMe was originally a broker that partnered with traditional insurers, but it subsequently expanded into underwriting, policy administration, and issuance. PolicyMe does not carry its own risk, but it manages all other elements of the process through a single, integrated system that it created.
According to Mr. Ostro, “We handle everything from how they landed on the site to how much time they spent on each page to what blog articles they read first to how they’re answering the application, all the way through to settlement data, and then into claims and beyond. That allows us a much richer set of analytics.”
