Shortly after joining MassMutual in 2013 as its first data scientist, Sears Merritt created a small team to figure out how data, machine learning, and other forms of artificial intelligence could improve life insurance.
“A life insurance company, maybe to everyone’s surprise, is sitting on a wealth of information that no other business actually has access to,” says Merritt, who was promoted to chief information officer in 2022.
Merritt’s hiring signaled that 174-year-old MassMutual, ranked #102 on the Fortune 500, is focused on taking advantage of new AI technologies by revamping how it handles data. Merritt spent 36 months building an entirely new cloud-based data platform that would let MassMutual collect, store, organize, and ultimately leverage that data for AI.
“All the data that’s needed by things like generative AI, and any models that we created or bring in from outside providers, we integrate directly into that data platform,” adds Merritt.
To demonstrate the power of data science, Merritt and his team, within 18 months, created a mortality risk model. It factors in an individual’s medical history, wealth, and behaviors including physical activity, along with tobacco and alcohol consumption, plus other information, helping MassMutual manage underwriting risk and price policies appropriately. Because MassMutual is owned by its policyholders, and not shareholders, managing risk for each individual policy benefits the broader pool.
“The focus that we paid in laying the groundwork and building a solid kind of data foundation has really, really paid off,” says Merritt.
More recently, and in just nine months, MassMutual created and deployed AI virtual assistants in parts of the business, supporting customer service representatives and financial professionals. The company says it's seeing early gains in worker productivity from the virtual assistants, which still require a human in the loop for all decisions.
In customer service, employees can use the AI to field questions about policies, including some that were created decades earlier. The ChatGPT-like interface has trimmed the response time by 10%, according to Merritt. There has also been a 5% reduction in the number of calls that financial advisors place into a contact center to get their questions answered.
“What we are positioning MassMutual to do is evolve itself such that the virtual assistants that we are building today become autonomous agents in the relatively near future,” says Merritt.
He describes his approach to getting generative AI to MassMutual’s employees as more “measured” than “aggressive.” Generative AI capabilities have been broadly integrated into all of the company’s productivity and collaboration tools, including Microsoft 365 and Zoom. But for employees to gain access, MassMutual requires formal training to explain what generative AI can and can't do, while also teaching them about the technology's risks, including its tendency to make mistakes.
When it comes to choosing which AI to use, Merritt embraces both a buy and build strategy. Some products come from third-party vendors, including Microsoft's Copilot AI assistant and software code completion tool GitHub Copilot, while others are built in house. MassMutual also takes a multi-cloud approach to avoid vendor lock-in, with most data sitting with the primary cloud provider Amazon Web Services, but some workloads are also with Microsoft’s Azure.
And as of today, all of MassMutual’s generative AI products are used internally, meaning they're for employees, not customers. There are no plans to give policyholders access, at least in the near future.
“We still have a way to go,” says Merritt. “We're going to be very, very thoughtful and diligent before we put anything directly in front of a policyholder, because the cost of an error we believe right now is too high to just let these things loose.”
John Kell
Send thoughts or suggestions to CIO Intelligence here.
