2024 Top 50 DC Plans Report: How are plan sponsors using AI to affect pension member experience, savings outcomes?

For example, to provide members with a highly personalized experience, the CSS is looking into developing AI-powered nudges within the portal that will consider a member’s age and savings level to send them tailored suggestions.“Say this person’s about 40 years old — maybe it’s time to use a retirement planning tool or come to a webinar. Those behavioural nudges are something we’re looking at,” she says, noting the CSS is targeting a 2026 roll out for the tool.

Read: 2023 Top 50 DC Plans Report: Unleashing the power of multi-employer pension plans

One of Canada’s oldest and largest DC plans, the CSS is a multi-jurisdictional pension that serves more than 55,000 co-operative and credit union employees. Since the organization operates in eight provinces, Dove’s team is also looking at building an AI tool that can search an internal repository of pension statutes, rules and policies to answer member queries in plain English, with annotated references. “It’s good for compliance and the plan member experience as well.”

The CSS is also looking at adding an AI-powered chatbot to its website to help plan members find what they need — if they search ‘beneficiary,’ for example, the chatbot would point them towards web pages on pension beneficiaries and surface the plan’s beneficiary-related documents.

Improving back-office efficiency

Broadly speaking, AI is technology that allows computers and machines to simulate human learning, comprehension, problem-solving, decision-making and more.

Machine learning-powered systems learn from reams of historical data to make predictions or decisions using a set of rules or pre-defined strategies. Generative AI, which exploded into the public conscience in late 2022 when OpenAI released ChatGPT, is a type of deep learning model that can create complex original content — including images, video, audio and long-form text — in response to user prompts.

In the DC space, the use of AI is still relatively new. While chatbot and virtual assistant-style tools, powered by machine learning, have become somewhat commonplace among record-keepers, plan sponsors are still trying to understand what AI can do for them, says Bernadette Chik, DC advisory business leader at Mercer Canada.

Read:

“AI has captured our imagination and I think the reason it has is because the impact could be so far reaching in retirement savings programs. . . . It touches almost every aspect of the DC experience, potentially. . . . The key question that sponsors are asking is how to keep pace. There’s a sense that things are moving very fast [and] a sense that things are not moving fast at all, simultaneously. There’s anxiety of making sure they keep [up].”

AI has the potential to significantly improve multiple aspects of pension systems and management, according to a 2023 report by Mercer and the CFA Institute. On the back-end, it can help to automate repetitive and routine tasks, leading to improved accuracy and efficiency.

This is another opportunity the CSS is expecting to capitalize on, says Dove, noting the organization is planning to automate the processing of member documents — such as enrolment and beneficiary designation forms — as well as sending automatic messages to plan members to let them know their forms were received.

The Mercer report also highlighted the potential for AI to improve identity fraud detection and model expected member behaviour related to contributions, withdrawals and retirement rates. However, it noted plan sponsors need “accurate and complete data” on which to train their AI models in order to gain greater understanding of their operations.

Better communication

According to the report, one of the greatest opportunities for AI in the pension space is in improving communications to plan members and, in turn, boosting their engagement with their retirement savings.

“This development is needed as many pension arrangements are poorly understood or appreciated by individuals,” it said. “However, this is not solely the fault of members. It must also be accepted that the communication provided by . . . pension plans can be opaque, verbose or expressed in legal language.”

Read: AI presents opportunities, disruptions for DC pension plan investments

John Bradley, director of emerging technologies at Fidelity Investments, says this is an area where generative AI systems can be particularly impactful, based on where the technology currently is on the maturity curve. It could be used by DC plan sponsors to generate plain-language communications materials for members or with chatbots that can answer member queries.

Fidelity is currently working on a generative AI tool that would help members get information about their plans. “When we’re talking about implementing tools that benefit clients, a big thing we’re looking at is helping them find the information they need as efficiently as possible.”

Gen AI tools are backed by what are called large language models — neural networks that have been trained on billions of datapoints to be able to generate and process natural language. One challenge with these models is that, since they consume so much data, they can perceive connections or patterns that are nonexistent and create ‘hallucinations’ — outputs that are incorrect or nonsensical.

That’s why Fidelity has been taking a conservative approach, says Bradley, noting the company has been testing large language models from a range of providers and evaluating each for accuracy, transparency and cost. It’s planning to put guardrails on any future chatbot tool to limit the information it can consult to only relevant documents and policies, as well as requiring it to cite its sources when answering a plan member’s question. The company will be doing “rigorous testing” before rolling it out to ensure it’s as accurate as possible.

Read: AI can enhance DC pension communication, employee experience

“There’s a big trust component involved. We don’t want to release something too early and have it not meet the expectations of clients,” he says.

KPMG in Canada (No. 28 in this year’s report) has a private generative AI-powered knowledge management platform called Kleo that its DC plan members can use to find basic information. The platform is a knowledge base that employees can turn to for a range of functions, including navigating internal policies and programs, summarizing the key points from lengthy documents and drafting emails and presentations. It also includes information on the company’s human resources policies and procedures, benefits plan and pension. Employees can ask about general eligibility rules to join the plans, as well as details like the enrolment process and contribution levels.

“It improves access to information and appeals to some younger demographics who may prefer to access information in that way,” says Emilie Inakazu, KPMG in Canada’s director of benefits and well-being.

Boosting member engagement

Lisa Kramer, a professor of finance at the University of Toronto, says AI tools that incorporate behavioural finance research can help members more optimally contribute to their DC plans and improve engagement.

“Left to their own devices, most people won’t contribute much without a little help. We can use AI, for example, to take account of people’s demographics and their income to tailor recommendations or maybe implement automatic contributions that lead to improved financial outcomes.”

Read: 2024 DC Plan Summit: How AI is impacting DC plan members’ financial decisions

Research has demonstrated that people rarely change their actions once set on a default course, she adds. “If we use AI to look at somebody’s personal circumstances and it makes a recommendation that’s more aggressive than what they’re currently doing, if you set them on that course they’ll tend to stick with it.”

Sun Life is trying to leverage technology — including AI — to improve plan member education and engagement and to provide retirement planning recommendations. For example, when members are making investment selections, the insurer uses AI to surface the products that are most likely to be the best fit based on the data they’ve filled in about themselves. It also embeds AI into its retirement modelling tools — based on the information members fill in, it will make suggestions like delaying receiving Canada Pension Plan benefits or drawing more or less money depending on the member’s tax burden.

And then there’s Ella, a chatbot backed by machine learning that delivers personalized suggestions to plan members through their Sun Life portal by taking into consideration their ages and life stages, earnings, how much they’re saving within the app, the specifics of their DC pension plans and their behaviours in the member portal. For example, Ella may recommend a plan member maximize their employer match, let them know if they’re ahead or behind on their retirement savings goal and suggest they invest their tax return into their pension.

Read: 2022 DCIF: Using behavioural finance concepts to improve DC plan participation, contribution rates

The phrasing, timing and complexity of Ella’s nudges are informed by behavioural finance research, says Eric Monteiro, senior vice-president of group retirement services at Sun Life. For a plan member who isn’t maximizing their employer match, for example, Ella might send a nudge to say, ‘You’re missing out on $1,500 per year in employer-match contributions,’ rather than, ‘You could make more money if you maximized your employer-match contributions,’ because research has demonstrated people have an adverse reaction to negativity.

Those nudges work. According to a Sun Life report, digitally engaged plan members have account balances almost 2.5-times higher and contribute 60 per cent more to their plan than those who aren’t digitally engaged. In addition, 60 per cent of digitally engaged plan members maximize their employer match compared to 30 per cent among those who aren’t digitally engaged.

There are also ethical considerations with nudge-type tools, notes both Monteiro and Dove. The nudges and suggestions have to be in the ultimate benefit of the member, not the organization.

“We have underlying co-op principles — we don’t try to sell to people and we’re not trying to convince them to make a financial decision that benefits them plus us,” says Dove. “We want to make sure that, when we ask questions, we’re always doing that to the benefit of the member.”

With great power comes great responsibility

There are other challenges to work through before AI tools are implemented; for generative AI, specifically, the most significant hurdle is ensuring accuracy.

DC plan sponsors that are considering their own internal generative AI tools need to be prepared to invest the time up front to make sure the available information will be properly interpreted by the tool, says Inakazu, adding plan sponsors need to articulate not only how a program can be used, but how it can’t. “When you’re reading it, you would understand the context. But when a bot sees a list of how you can use it, if you don’t say explicitly you can’t, it doesn’t assume you can’t. There is work up front that you need to do and testing before you roll that out.”

Read: AI holds uncapped potential for pension plan sponsors’ administration, communications: expert

KPMG’s pension has a contribution range of between three and nine per cent with employer match, with the maximum percentage employees can contribute depending on their seniority and tenure with the organization. “[We had to] ensure it had the right context to say employees can contribute between three to nine per cent based on their years of service,” says Inakazu.

The forthcoming federal Artificial Intelligence and Data Act, which is currently before the standing committee on industry and technology, will regulate the design, development and use of AI systems. It could also affect the tools record-keepers and plan sponsors build and implement, says Bradley, noting it will set new responsibilities for organizations using AI around the models themselves and the data that goes into them, including metrics that must be tracked; explainability, re-training and monitoring requirements; and liability for harm.

“That is going to have a big impact on areas we focus on and what we try to do and what we might avoid based on the guidelines and regulations put forward.”

Kelsey Rolfe is a Toronto-based freelance writer.

Download a PDF of the 2024 Top 50 DC Plans Report.