AI and the Pension Industry

Artificial intelligence (AI) is rapidly transforming the business world, and the pension industry is no exception. With the rise of AI initiatives, the pension industry is poised to benefit from technological advancements. Always the curious actuary, I set out to investigate what sort of benefits those could be.

Being part of an organization at the forefront of innovation, I am fully immersed in exploring AI tools and capabilities, thinking through how they can be leveraged at an enterprise level as well as from a pension actuary’s perspective.

There are immediate, daily AI capabilities that we can all use—from summarizing information in emails and reports, to producing meeting notes and assigning tasks to meeting attendees. More sophisticated capabilities include creating charts, graphs or a presentation from a file (or numerous files). And still, we have barely scratched the surface.

We are in an era of maximizing efficiencies and scaling abilities, so generative AI couldn’t possibly have come at a better time, in my opinion. Aside from time-saving tasks, generative AI can create something quite different from previous stories, patterns or outcomes. That is, it connects many pieces of information to draw inferences, thereby innovating and providing new insights and relationships.¹

Insight From Andres Rojas at the Vector Institute

To better understand what initiatives exist for organizations looking to incorporate AI into their structure, I spoke with Andres Rojas at the Vector Institute, a nonprofit dedicated to AI research that works across sectors to advance AI application, adoption and commercialization in Canada.

Rojas, Vector’s director of applied AI projects, described how Vector partners with organizations and the industry at large, helping develop advanced AI tools, upskilling the workforce and facilitating talent acquisition in the AI space. “By exploring state-of-the-art techniques, organizations can identify areas worth investing in,” Rojas said.

Vector contributes to thought leadership and Canada’s role in AI, including economic and societal impacts. It brings together large enterprises, startups, AI researchers and policymakers to use AI to test, experiment and solve problems together—uncovering insights and kickstarting innovation. A recent Vector Computer Vision workshop discussed new work in this field. Generative modeling in Computer Vision analyzes and extracts information from images and videos. Machine learning capabilities in this field can help overcome human challenges ranging from facial, sound and action recognition to autonomous vehicles and medical imaging segmentation.

We discussed some of the enhanced efficiencies and productivity that AI systems can offer. Translating it into the retirement industry, imagine a world where the intake of all pension option forms, whether provided electronically or via a paper copy, is automated. Missing and incomplete information is immediately identified, and a missing documents letter to the member is automatically drafted.

More than automated procedures, AI systems can enhance the member experience and engagement. Many leading pension plans are developing chatbots to interact with members through either text or voice. Given the inherent characteristics of generative AI, such tools will continuously improve as they learn from previous questions, comments and answers.

The Finnish Centre for Pensions launched its chatbot, Tynne, in 2019, to answer frequently asked questions. Over 50% of all chats are now fully automated, and the Centre estimates Tynne saves 37 hours per month, enabling agents to focus on more complex questions that need human intervention. An extension of this can be using accessible data and analytics to personalize the communication to each member.²

Insight From Marshall Posner, Plan Actuary at the Ontario Municipal Employees Retirement System

Marshall Posner, plan actuary at the Ontario Municipal Employees Retirement System (OMERS), suggests personalization could be a key impact of AI on the pension industry. “Imagine an AI tool that prepares documents for individual plan participants, but the documents are customized for that individual to specifically enhance their interests. The customization could be based on previous interactions the plan administrator had with that participant, or using demographic information like their age, where they live, if they have a spouse. … The concept is not far-fetched and can go a long way to enhance pension terminology and familiarity with plan participants,” Posner said.

Improving engagement and communication is key to shaping the best outcomes for members, and the use of personalized data could propel this. To delve into this area, plans would need governance and data privacy policies. As this is a relatively nascent field, best practices and regulations surrounding it are limited, and many are currently in development. A joint initiative by the Office of the Superintendent of Financial Institutions (OSFI) and the Global Risk Institute (GRI) formed the Financial Industry Forum on Artificial Intelligence (FIFAI) to advance the conversation around appropriate safeguards and risk management in the use of AI. Preliminary AI considerations, including regulatory-related ones, can be found in the committee’s publication, A Canadian Perspective on Responsible AI. Vector also has published six AI Trust and Safety Principles that provide guidance for organizations developing their own code of conduct and AI policies.

I asked Posner, “Could AI potentially impact decision-making processes within pension plans?”

“An underutilized actuarial skill in the AI and machine learning space is analyzing past experiences of plan participants to set actuarial assumptions, something actuaries have been doing for decades but rooted in models that have potential for modernization,” he said. “Pension actuaries should keep an open mind on using machine learning models that are trained to seek hidden patterns or maximal correlations as part of their responsibility of setting best-estimate assumptions for a pension plan’s valuation. Assumptions with better predictive accuracy should result in lower experience deviations [‘gains and losses’] measured in future valuations. … The actuary needs large volumes of data to make this feasible.”

I also asked Posner, “With all these potential avenues for AI, how will the role of actuaries evolve within the pension industry?”

“AI won’t replace actuaries,” he said. (Phew, I breathed a sigh of relief). “AI takes the form of tools, whose continuous improvements will slowly transform the pension actuary’s role into focusing on overseeing and refining those tools, interpreting results and managing any ethical complexities that arise.”

The Future of AI and Actuaries

After talking to Rojas and Posner, I wondered, “What about strategy—can we incorporate AI on that front?”

Imagine having a program read through the last 10 years (or more) of a pension fund’s strategic plan, plus related meeting minutes leading up to those strategic plans, and combine that knowledge with information about emerging market risks and how legislation has developed over the last few years—and then put forward a proposed strategic plan for the next three to five years. Where should the pension fund’s focus be? What are the key risks? Where are the blind spots? Imagine having a digital strategy consultant with vast institutional knowledge (limited only by what has not been captured in an organization’s digital files) combined with up-to-date economic and demographic knowledge and superior analysis capabilities. I certainly would want to pick their digital brain.

I believe AI capabilities and expertise are continuously expanding and could be significantly leveraged in many aspects of the pension industry. There are immediate benefits we can all take advantage of, and there are advantages that likely require additional legwork to fully reap the benefits.

Consider the volume of documents being read as part of due diligence work for investment managers or research across provincial legislation. AI tools can read them and summarize the findings. And while such tools have been in use for years in one form or another, when starting to use machine learning for any type of research, it would be prudent to do this side-by-side with the current practice of individuals conducting the full review. A comparison can then be made, and feedback can be provided to AI systems to enhance the focus, key items to highlight and more.

Thankfully, there are organizations like the Vector Institute whose focus is enhancing AI capabilities and the ways in which organizations can leverage them. Such research could go a long way in maximizing efficiencies and productivity, which could have a significant positive economic impact on Canada.

How could AI affect actuaries? Actuaries often are described as critical thinkers, and I believe AI will help us further sharpen that skill. Leveraging AI tools means training them on what information to look for, how to connect dots and which dots are key to connect. What’s more, as machine learning and generative AI capabilities grow and improve, new dots and dots in our blind spots will be connected and inevitably challenge us to think even more critically.

Statements of fact and opinions expressed herein are those of the individual authors and are not necessarily those of the Society of Actuaries or the respective authors’ employers.

Copyright © 2024 by the Society of Actuaries, Chicago, Illinois.