AI Tapped to Mine Form 5500s, Leverage Chatbots for Retirement

Tech-focused insurance brokerage and retirement advisory firm Newfront held an event not usually associated with risk benefits and 401(k) plan management: a hackathon to incorporate artificial intelligence into client-facing products and services, earlier this year.  

The results contributed to an AI-driven workplace benefits chatbot that is currently in use by employers. While the bot is geared toward general benefit questions, retirement plan services and inquiries are part of the offering and are on track for further development, says Newfront retirement services practice leader Greg Kaplan.

“We take all the benefit guides, all the policy documents—any information that exists from employee benefits and retirement plans—and ingest it into our AI engine,” says Kaplan, a senior vice president for Newfront Retirement Services Inc. “You can throw any questions, such as what health plan does a company offer, or what is the company 401(k) match? It is trained to provide the correct answers.”   

It is hard to open a financial newsletter these days without coming across an article mentioning AI. But the practical implications of the technology can sometimes be hard to identify. According to Kaplan, at San Jose-based Newfront the AI revolution is starting to get very real, very fast.

In late June, a firm representative presented before the Congressional AI Caucus in Washington with other industry players. Newfront’s Lin Yuan, vice president of engineering, showed the caucus how the firm’s benefits bot answers questions about areas such as health savings account plans and flexible spending accounts, according to an announcement.

For Auditors and AI

Back in San Jose, Kaplan’s retirement team has been putting AI to work for plan sponsor clients. The team, he says, is using an AI technology known as “large language model” to mine and analyze Form 5500 retirement plan filings in order to identify trends and insights for client plan design.

“We generally benchmark to the extreme,” Kaplan says. “The more specific we can get to an individual client, the more relevant [the analysis] becomes….We like to go really granular and industry specific, to the plan size, to the number of participants, to the client’s peer group and who they are competing with for talent.”

In the past, the team would rely on industry reports such as The Vanguard Group’s How America Saves, along with its own manual inputting of 5500s. This October, when the new 5500 forms are filed by plan sponsors, the team will be using large language modelling to ingest the reports with the goal of providing deeper and richer insights in faster time and less human work hours.

“When a client comes to us and says these are our competitors, we can quickly go in and mine the data to look for insights,” Kaplan says.

HR Benefits Assistant

Kaplan has a technology background from having spent time at Microsoft Corp., and like many of his colleagues at Newfront, brings a tech-startup ethos to employee benefits. But much of his interest in retirement plans, he says, comes from having worked on the leadership team of a company and spending a lot of time explaining benefit offerings to employees who would “flag me in the hallways.”

In the process of creating its chatbot, the Newfront team found that some of the benefit guides the AI was using were not clear enough, according to Kaplan. The result was a “feedback loop” that ultimately helped create better materials and therefore better responses from the machine.

Newfront notes in its materials that use of its chatbot can shed as many as four weeks off the workload of human resource teams that would otherwise be answering emails or, as Kaplan experienced, getting flagged down in the hallways.

How AI technology will flow into the retirement space more directly is still “in early innings,” Kaplan says. But this fall’s round of 5500s, he believes, will take his team one step closer to practical use.