More data means improved DC offerings

These days, the use cases for generative artificial intelligence are ever expanding — SEC Chairman Gary Gensler has called it the "most transformative technology of our time," on par with the internet and the mass production of automobiles — but while retirement industry stakeholders say the tool will have major implications for defined contribution plans one day, record keepers and service providers are taking a cautious approach.

"I think rushing too fast on this might come back to bite us from a brand and reputational damage perspective, and frankly also harm our participants themselves," said Swatee Singh, Iselin, N.J.-based executive vice president and chief data and AI officer at TIAA-CREF. "But where we think it gets applied? Every single place, whether it's … anything operational that you're looking to do, even basic things like public searches or internal searches, no question that it's going to be there very, very soon."

Among the use cases, generative AI, a tool focused on the automatic generation of content via massive neural network-based deep learning models, can potentially reduce plan costs by automating plan administration and back-end tasks, and produce personalized prompts and recommendations for participants to boost outcomes, sources said.

Although generative AI is an incredible tool, the application is risky for DC plans in its current state, said Emma O'Brien, Boston-based senior consultant for the NEPC LLC defined contribution team.

"The tool can become self-confident; there's the chance of hallucinations, meaning the model can generate false outcomes or information, so while it's powerful, the application needs to be regulated to protect participants," she said. "We're hearing from record keepers that are in the early stages of testing the application that they're going to be really cautious for now."

Like other record keepers, Fidelity Investments is using traditional AI or machine learning in its business, but isn't yet using generative AI, said Mike Gabree, Merrimack, N.H.-based head of technology for workplace investing. But, he added, the tool can eventually allow Fidelity to more rapidly onboard new clients by speeding up processing time and understanding incoming plan documents from previous record keepers.

Like other DC plan stakeholders, Fidelity, the nation's largest DC record keeper with $2.66 trillion in assets under administration as of Sept. 30, according to Pensions & Investments data, is taking a "cautious, iterative approach to implementing these solutions in our business, and are focusing on internal low risk use cases," Mr. Gabree said in an email.

Fidelity's first generative AI venture will be automating internal business processes to extract information from documents "in a faster, more reliable way, while still having a team member review to ensure the accuracy of results," Mr. Gabree said. "Plan sponsors and participants will experience a faster onboarding experience as a result."

At TIAA, which had $643 billion in DC AUA as of Dec. 31, Marc Jarmosevich, Charlotte-based senior managing director and chief analytics officer, said using generative AI to analyze TIAA's internal business data would one day "allow us to completely reimagine the business reporting and plan sponsor reporting that we do and really turn it into this more customized generative AI approach than sort of static off-the-shelf reports that are out there for folks. But that is very much something that is a twinkle in the eye right now, not something that is actively being worked on."

For now, TIAA is focused on having a generative AI policy that considers the pros and cons of the tool.

"What we're trying to do is put enough guardrails that we recognize at a governance level how to prevent harm," Ms. Singh said. Those guardrails include only permitting a generative AI program the ability to access public data and ensuring that there's a human overseeing its output. "We look to gen AI to give us the answer to problems … but we want a human with subject matter expertise there validating," she added.

One of the biggest problems with generative AI in its current state is inaccuracies, said Gregg Levinson, senior director of retirement at Willis Towers Watson PLC in Philadelphia.

He recently typed an inquiry on deferred compensation plans into ChatGPT, a large language model-based chatbot developed by OpenAI that launched last year, and said the response he got looked well-researched and informative, but the answer was totally wrong. Mr. Levinson, as a subject matter expert, knew the information was incorrect, but the general public would not.

"We can't validate with certainty everything that AI puts out; that's our problem right now," he said. "From a commercial standpoint, I think the risk is too great, but down the road the possibilities are definitely there."

Vaidyanathan Chandrashekar, co-founder and chief strategy officer of Cupertino, Calif.-based Congruent Solutions, a specialist technology solutions and outsourced plan administration service provider, said all retirement industry stakeholders should be considering how generative AI could impact them.

"Now is the time because AI is maturing like crazy," he said. "You need a strategy and a team to see what's out there and how is it going to be contextualized for your organization. That will be the determining, cutting-edge advantage over the market and the competition that you're going to have."

Fidelity since 2016 has used traditional AI or machine learning to route customers who reach its call center to the appropriate representative for parts of its business, Mr. Gabree said.

"AI takes away mundane, operational tasks, freeing up more time for our reps to engage more deeply with customers," he said.

Fidelity is also experimenting with combining its Fidelity data sets with generative chatbot technology to develop a new internally focused chatbot, which can help roles across the enterprise better service our clients and participant, he added. Specifically, with a chatbot interface, instead of manually searching for answers themselves, internal employees can ask the model to search for answers and bring back the results, according to Mr. Gabree.

AI and machine learning allow Vanguard Group Inc. to create "hyperpersonalized recommendations" that take into account not only what a participant's goals and demographics are, but how they've personally previously engaged with Vanguard's platform, according to Dave Stinnett, Malvern, Pa.-based head of strategic retirement consulting at Vanguard. He noted that Vanguard participants who have engaged with its emergency savings calculator are 74% more likely to take a positive action, like increasing their contribution rate.

Now Vanguard, which has $540 billion in DC AUA as of June 30, can individualize its recommendations, reminders and guidance for participants based on what they have shared about their full financial pictures, Mr. Stinnett said in an email. "Thanks to better data, we can motivate plan participants with digestible and actionable insights on their next-best actions, rather than inundating every participant with all available information," he said.

The firm can now tell a plan participant who is falling short of meeting their employer match exactly how much more to contribute each paycheck to get the full benefit, or educate a participant with an extreme asset allocation on the availability of target-date investments, Mr. Stinnett added. Vanguard communicates with participants through emails, printed materials, educational content on the participant website, digital nudges on the web and tools/calculators, he added.

Moreover, Vanguard is also seeing demand for financial wellness features that bring together debt paydown, emergency savings and other non-retirement goals. Participants who engage with guides or action plans within Vanguard's 'My Financial Wellness' experience are "close to 50% more likely to increase their payroll deferral," Mr. Stinnett said. "That type of information is resonating with our plan sponsors, so we expect to see even greater availability in the future."

Broadly, there is now more and better data available for record keepers and service providers, said Jamie McAllister, Chicago-based senior vice president and defined contribution consultant at Callan LLC.

"Now we just have such a wealth of information being provided and the record keepers or other service providers like managed account providers, they can take all that — gender, salary, the state they live in — and just use all of that data to come up with different recommendations," she said.

Callan collects participant data in a number of ways, including asking record keepers to fill out a lengthy questionnaire each year that includes information on transfer activity or contribution dollars coming into plans.

With data more readily available, plan sponsors are now requesting more individualized solutions, sources said.

"I think we're beyond the days of participation rates and savings rates," said Will Hansen, Arlington, Va.-based executive director of the Plan Sponsor Council of America and chief government affairs officer at the American Retirement Association.

"While that data is still useful, I think some plan sponsors are definitely looking to get into the weeds more on the financial backgrounds and circumstances of their employees at a more macro level to determine if there are other communication methods and/or other benefits that they could provide to ensure that their employees have a healthy financial footing."

Record keepers are now "differentiating themselves by saying they have the tools or data to help with personalization and customization because that's what they're hearing from plan sponsors," Ms. McAllister said.

However, plan sponsors "definitely want the personalization, but they are a little cautious about the data that's being provided," she added. "It applies to the security aspect of the data, but also to the accuracy/completeness of the data."

Tim Rouse, Simsbury, Conn.-based executive director at the SPARK Institute, which represents retirement industry players such as record keepers, investment advisers, mutual fund companies and benefit consulting firms, said the more data a record keeper gets from a plan, the more data it can use to build a workable, tailored financial wellness program.

"Participants want help with 'where should I put my next dollar? 401(k), HSA, credit card bill, student loans?'" Mr. Rouse said. "In order to do that you obviously need a lot of information about them."

One issue is that data sharing in the DC plan space is currently "haphazard," according to Mr. Rouse.

Plan sponsors and payroll providers are understandably resistant to offer access to much of their data, though plan sponsors are more likely to "lend" data, Mr. Rouse said. "If there's a service provider hired to be the financial wellness provider, that provider would be granted access into the benefit systems but when and if that relationship terminates, access to the data terminates," he explained.

SPARK and the Defined Contribution Institutional Investment Association last year convened a meeting of retirement industry stakeholders in an effort to create more effective holistic financial wealth solutions that safeguard plan and participant data. A second such workshop was held in June.

"We need to be able to share data for the benefit of the participant, but we also need to make sure that those cybersecurity and privacy concerns are addressed," Mr. Rouse said.

Willis Towers Watson's Mr. Levinson and NEPC's Ms. O'Brien each said their firms do not ask for participants' personally identifiable information, but do seek data like age and the investments to which participants are allocating.

"Record keepers have the ability to provide very useful plan data which NEPC can leverage to provide customized advice for each plan based on the participant base," Ms. O'Brien said.

NEPC analyzes plan data to estimate each participant's expected risk and return based on their allocation as well as the expected risk and return for each option in the plan, she added. "Through that study we've been able to improve participant outcomes because they can help us quantify an overall risk profile that might be higher or lower than desired, and it can identify holes in the lineup or find unintended redundancies between investments," Ms. O'Brien said.

Willis Towers Watson is using client specific retirement and health data to address participants' financial and general well-being, according to Mr. Levinson.

For one client Mr. Levinson declined to name, Willis Towers Watson overlayed health insurance claims with potential signs of financial stress, like 401(k) withdrawals or reductions in plan contributions, to address the participants' overall well-being.

"We have measurable results that show when you put these programs in place and you include engagement so that employees trust it and actually use it, you start to see results."

Added, Mr. Hansen: "Data shows that if an individual has a strong financial footing, they are more productive at work, less distracted at work, so anything that an employer can do to ensure that more of their employees have good financial footing just helps the bottom line of that company."