The new brain inside today's retirement plans

April 6, 2020

A/AText size:AAA

Progressive defined contribution plan providers are using AI to give individuals the nudges they need when they need them.

New Brain Inside

Plan sponsors want to help all of their employees prepare for retirement. Yet each individual is unique, with distinct income, debts, expenses, attitudes toward money, life goals, age, education … you name it.

Leading asset managers are helping plan sponsors overcome this hurdle by using artificial intelligence (AI) to personalize outreach efforts. They’re developing ways to tailor solutions to each employee based on that person’s behaviors and characteristics, much like targeted ads. The goal is to help every employee achieve a financially secure retirement.

BeFi meets AI

Insights from behavioral finance (BeFi) have been revelatory for defined contribution (DC) plans. They have led to hugely successful plan features that turned the human tendency toward inertia to employees’ benefit, including auto-enrollment and diversified default investments.

Shannon Nutter, head of participant strategy and development for Vanguard’s Institutional Investor Group, says the industry has been slow to move beyond that initial epiphany. But she says AI offers the promise to help the industry take the next step: giving each employee the right message for them at the right time. "For the last three or four years, we’ve invested heavily in using AI to personalize the participant experience," she says. "The more personalized the information is, the more relevant it will be—and the more likely the employee will take action."

Nutter says Vanguard had long been committed to developing a deep understanding of the implications that behavioral finance research has for its participants. Like many providers and plan sponsors, she says, the company also had huge volumes of information about employees, including their income, demographic details and behavior on the plan website and app. AI enabled the firm to speed up its research exponentially, allowing it to run thousands of simulations a day and test various hypotheses about which nudges worked best for particular types of individuals.

Of course, the better data there is (and the more of it), the more powerful AI can be, so the company is also experimenting with ways to gather key information. Vanguard recently identified participants for whom the firm lacked salary information and customized the website for those participants—with a pop-up quiz where they might typically show retirement income projections: "I earn $____ per year and I want to retire when I’m ____ years old."

In an era of real concerns about data privacy, it’s important for such queries to explain how providing the requested information will benefit the employee, Nutter says. "We make clear why we’re asking for specific information and why providing it can help us help them—because it will help us send only the messages that apply to that person," she says. "If the employee understands and values the outcome, they’ll be more comfortable sharing their data."

Pioneering personalization

If AI at Vanguard was first used as a research tool for developing a better retirement engine, it now has effectively become the engine, Nutter says—one that itself is in a perpetual state of further experimentation and refinement. Vanguard currently has a variety of ongoing, AI-enabled personalization initiatives in DC plans. The efforts use the full range of AI technologies: rules-based algorithms, machine learning and neural networks. Each plays a distinct role in moving participants along a critical path through the web of human emotions, impulses and actions (or lack thereof) mapped out by behavioral finance. Here are a few examples:

Determine the best next step for an individual employee. The hierarchy of actions can seem fairly straightforward for a generic plan participant: Save enough in a 401(k) to get the maximum company match, contribute to a health savings account (HSA) and so on. But this purely rational approach doesn’t account for human idiosyncrasies. "What if an employee thinks they don’t like HSAs because they had a bad experience with an old medical savings plan?" Nutter says. "We’re working on building algorithms that detect when an employee won’t act on a topic and problem-solves how to move them on to one that might better suit their personal situation and preferences."

Design messages to spur employee action. Vanguard is designing systems that automatically deliver each employee a single clear, personalized call to action across channels, based on research that shows one simple action is more effective than multiple options.

Say an employee isn’t contributing enough to get the full employer match. The personalization system might send a targeted email using the BeFi concept of set completion that says, "You missed a step. Click here so you don’t lose out on free money from your employer." The link takes the employee to a webpage that repeats the call to action, with a prominent button they can click to increase their contribution rate immediately. AI then measures the effectiveness of various approaches, rapidly increasing the understanding of what works.

So many aspects of a message can serve to make an employee pay attention or not. Vanguard is designing systems that, based on data, can customize not just the single call to action that an employee should take right now, but also the actual language the message should leverage, the visuals that will be most compelling and the decision-making approach that might prove most effective in getting them to make a change—all based on an understanding of how that specific individual has interacted with Vanguard in the past.

Reach out to a given employee at the right time for them. Vanguard is also experimenting with the most advanced, autonomous type of machine learning, called "deep learning," to deliver messages in the right frequency for each individual. "Your default might be to send a message every 30 days—but you may want your system to ignore that rule for someone who comes to your site weekly," Nutter says. "The idea is to see if our machines can come up with more optimal patterns of communicating."

The DC plan of the future

Nutter would be the first to say the new AI tools will be considered a failure if they don’t actually help employees achieve better retirement outcomes. Fortunately, for Vanguard and its clients, early results are promising.

According to the company, 81% of participants in Vanguard plans engaged with the company in the most recent 12-month period, and half of those employees took a positive action for retirement, such as registration for web access, boosting contributions to take advantage of employer matches, and more email click-throughs.

Those results are just the beginning. Researchers at Vanguard and elsewhere are using AI to continually test, measure and evolve efforts to personalize DC plans. Nutter expects these advances to give plan sponsors ever-greater flexibility to help everyone build their own successful retirement. "Everything we’re doing is for that goal," she says. "And we’re only on the cusp of what’s possible."

To learn more about Vanguard's 401(k) personalization programs, please click here.

Developed in collaboration with The Wall Street Journal.

Wall Street Journal logo


  • All investing is subject to risk, including the possible loss of the money you invest.
  • Diversification does not ensure a profit or protect against a loss.