AI Has Potential to Improve Plan Design, Expand Retirement Plan Access

In the past few years, a retirement plan would be considered tech-savvy if it gave employees savings nudges on their birthday or hire date.

That’s changing—fast.

Tech-forward retirement plan providers, recordkeepers and third-party administrators are using artificial intelligence—including machine learning, large language models and even ChatGPT—to digitize and automate mundane plan administration tasks to reduce plan sponsors’ burdens and costs. Not only does it mean less work for plan sponsors, it is also making it easier for employees to save.

It’s necessary, too, as the SECURE 2.0 Act of 2022 mandates actions such as automatic enrollment and automatic escalation, which will require software updates.

Newer and smaller retirement plan providers are using technology to get a foothold in the industry and service more small- and midsize employers. In the future, technology may ease employee plan access, tailor investment offerings and perhaps offer advice, although sources interviewed for this story say technology is not there yet. The human touch is not going away, but it may just get more efficient.

Where We Are Now

Rakesh Mahajan, chief revenue officer at Human Interest Inc., a full-service 401(k) and 403(b) provider for small and midsize businesses, says digitizing and automating much of the retirement plan administration eliminates paperwork that was common even a few years ago. That efficiency makes it easier, faster and cheaper for plan sponsors to roll out benefits and simpler for employees to use. For new employees, accounts and investments are set up with a few mouse clicks. Eventually, he would like to make that even simpler.

“My aspiration is to have a magic button for everything,” Mahajan says.

Kevin Gaston, director of plan design consulting at Vestwell, says taking care of the administrative burden for plan sponsors also helps reduce their legal and compliance risks when offering the benefit, and technology can also flag other potential risks, such as alerting plan sponsors to late deposits or other issues.

On the Horizon

Chad Noorani, a retirement plan consultant at Benefit Financial Services Group, has experimented with using ChatGPT to gather compliance information and create a framework for blog posts that he would eventually rewrite to include his own resources and links. AI could help small retirement planners like Noorani become more efficient by, for example, writing code and personalizing materials to meet client needs. Efficiencies could allow him to scale his business and spend more time with clients.

“Business development may get easier; it may help you prospect better,” Noorani says, noting such automation could eliminate irrelevant discussions, such as pitching services that a plan sponsor may not need or want.

In the same vein, some retirement plans offer one-size-fits-most investment options, such as target-date funds, which may not be suitable for employees who have outside assets or competing savings goals.

Dani Fava, group head of product innovation at Envestnet, says machine learning and predictive analysis tools can make retirement plan design more customizable and help plan advisers create better savings strategies, even when employees have not entered all of their information. Envestnet’s Insights Engine uses machine learning to sift through its data and that of other connected data providers, plus an employee’s unique information, to find patterns. It can predict when a participant has outside assets and what type of assets with more than 75%accuracy. When Generative AI is included in the design of a retirement plan administration system, the toolcan learn from plan data and may be able to create more accurate or tailored offerings, she adds.

In coming years, plan sponsors are likely to have more participants who have more retirement savings, who will need greater support, Fava says. Generative AI could be used to create an asset allocation from the plan investment menu or identify the right level of savings without employees having to answer the usual risk tolerance questions.

Fava says this technology could be available in about a year, but getting regulators and plan sponsors comfortable using it is probably three to five years away.

Gaston says AI and fintech could expand access to robust savings tools, even at smaller businesses and for employees at all pay levels. “We look at it as a force multiplier, downwards and outwards.”

He says while AI’s effect on performance will be closely followed, he thinks AI could be used to offer investment guidance, such as spotting portfolio risk that is out of balance or finding new ways to increase savings. Gaston sees AI’s greatest benefit as improving financial wellness and finding that next dollar of savings, more than affecting asset management.

Industry sources reiterate that retaining a human touch will be important because of the emotions tied up with money. Humans will need to give AI boundaries, especially in a field of complex legal and rigorous compliance regulations. “You can tell Chat GPT to make me the best plan design. … It may not be fair or equitable, so it’s not good,” Gaston says.

The Far Future

Giuseppe Sette, president and co-founder of Toggle AI, says his firm is teaching large language models how to invest, educating the models  on financial concepts that will allow them to better understand and analyze investment problems. In the distant future, plan advisers might be able to assign some asset management decisions to AI, which he dubs an evolved version of “robo-advisory.” He also considers creating a drawdown strategy in retirement “another incredibly fertile opportunity” for AI.

However, Sette cautions that there is no room for error when using AI in finance or retirement planning because of the strict regulations and compliance. After all, AI still has a propensity to “hallucinate,” or make up answers, or it could identify investment opportunities using incorrect or inaccurate criteria.

To avoid those problems, Sette says AI developers would need to narrow the intent and type of questions a user asks an AI-enabled tool, so that the AI system can understand what the user wants. Critically important is establishing safeguards to control what the model says back to the user.

Relying on AI for financial advice remains some time off, but Sette is clearly bullish on the progress already made.

“It’s easy to get Pollyannish, because it’s very likely the most exciting technology we’ve seen [in] the last—I don’t know, ever,” he says.