AI for Insurance Marketing


By Robert Eaton, FSA, MAA, Principal and Consulting Actuary, Milliman

Artificial intelligence (AI) chat bots are now in consumer hands. Companies like OpenAI (chatGPT), Google (Bard), and Meta / Facebook (LLaMa) are pioneering sophisticated, interactive, and intuitive capabilities that tens of millions of people routinely use today. This technology provides a new interface for insurers to reach their customers: machines that communicate easily and clearly with people, at low cost.

What changes are on the way in the distribution of insurance products, to individuals, groups, and affinity markets? PIMA members should expect sales and marketing to change dramatically on at least three fronts: product education, sales and participation, and insured persistency. This article is an example of the changes we might see in the group and affinity markets.

Product Education

Selling group insurance today (in 2026) means training a company-specific chat application to answer questions about specific insurance products. The company has dozens of data scientists and product directors talking in meetings with the AI to get it up to speed. Yesterday we had FAQs which (it’s in the name) were most helpful only to the median consumer. The difficult parts were the ‘IFAQ’, or the infrequently-asked questions.

Our company-branded AI Assistants interpret customer questions with high reliability, and the Assistants provide accurate responses with citations to their policy and its marketing material. We had this capability as early as 2023 with many AI services that began summarizing and interpreting documents. Insurers and enrollment companies were keen to adapt that technology to improve customer education on the front end. The service is so low-cost and prevalent, that today pretty much all consumers have come to expect it. Having a product-specific Q&A AI Assistant is now a table stake for any insurer.

Our chatty product Assistant also answers questions about how one product (say, our hospital indemnity product) interacts with other products (like major medical coverage). The Assistant answers questions about deductibles, coverage limits, and it even reasons out scenarios of family medical spending to illuminate the value of the policies to the customer.

Where yesterday’s marketing material provided some hand-picked examples of spending for the typical customer, and how policies provided coverage, today’s product Assistants field questions about any variety of events that may happen, and they do so promptly and courteously and with infinite patience (this last trait is still undervalued).

Sales and Participation

Since the improvement of our customer education (though frankly, so did all of our competitors’) we have seen better employee engagement and resulting participation rates. The affinity and group insurance sale is no longer “here are the three options”. Instead, we have an advanced AI Recommender system. Our AI recommenders provide distinct options for each individual: “I picked these options for you based on your other selections, your history of spending, and what I know about your family and personal situation.”

When we first implemented the AI Recommender, it caused the actuaries more than a bit of consternation because the AI was sharp enough to generally optimize the customers’ choices, including choices about risk (the actuaries feared “anti-selection on steroids”). But in the end, providing a higher quality service to the customer turned out to be better business: more educated customers provided a greater spread of risk within our group sales, even if certain customers used more benefits with the help of the AI.

What really impresses some of our customers, provided they can get comfortable relinquishing some privacy, is their ability to opt-in to sharing their personal medical histories and family situations with the company’s AI Recommender. The AI sometimes serves up the option to purchase a package of services that no person would usually suggest: a mix of accident and cancer insurance to people with a family history of certain disease – even when informing the customer that the disease has a much lower prevalence today. One executive without a spouse was recommended more DI insurance and less life insurance, while another executive with a family was offered great life insurance coverage and less DI coverage. It makes sense, though the kind of personalization was only really seen in independent financial advisors before the AI came along.

In addition, once customers allow their personal information to be used in decision-making, the AI Recommender sends them a notice on certain life events, describing what new coverages they may be eligible for across the suite of products within their group benefits.

All of these AI advancements have generally lowered acquisition costs and spread risk across the groups we sell, which has boosted profitability. The downside is that all of our competitors use their own AI tools, and even the employees and group members have a personalized assistant helping them evaluate options, so our offerings don’t always align with expectations. Who knew that an AI could determine whether customers would pass underwriting (or not) and suggest group insurance over individual options?

On balance I expect that worksite and affinity group insurance will continue to deliver value through higher participation and sales, as decision-makers are equipped with better tools.

Insured Persistency

Finally, the AI Recommender we have now engages with members and customers as they leave employment, and the Recommender takes care to remind them of options as they transition to new employment or retirement. We previously saw larger insured lapses when groups churned carriers, or when employees changed jobs or retired. Today with AI assistants everywhere, customers better perceive the value of the benefits they’ve been paying into for years (particularly those that are rated by age at issue) and provide a balanced guide for whether the customer should continue paying for coverage. Moreover, the brokers have a different incentive structure now that they understand that moving a group to a new insurer doesn’t produce as many new sales.

In the early 2020’s this technology didn’t exist – customers called insurers on the phone and waited for customer service reps to pull up their benefits and learn more about their situation. That’s all ‘baked in’ to the AI assistants today, who are armed with the policy knowledge and the personal information to suggest retaining their policy (or not!), or possibly reducing coverage or making other choices.

Most customers are grateful for the helping hand and continue to pay their premiums for a supplemental health or voluntary life product. We’ve probably lost some profits on those products that were lapse supported, but we’ve made back profits on the rest, and we’ve learned to up-sell those customers into other new products as they work towards retirement.


Overall, the AI resources we have now seem to be a benefit to customers, and for our company this means good business for us. It hasn’t been the case for everyone – after all, wallet share still adds up to 100% – but worksite insurers have really benefited from a stronger realization of their value proposition: economies of scale that reduce costs, spreads of risk that mitigate anti-selection and lower premiums, and ease of access to a large variety of benefits.

These days it’s still hard for an AI to provide peace of mind to people (in some cases, quite the opposite), so we’re fortunate in insurance to be in the peace of mind business.

Published in the Fall 2023 issue of Insights Magazine.

PIMA® (Professional Insurance Marketing Association®) is a member-driven trade association focused exclusively on the affinity market.