How AI agents can revolutionise the Defined Benefit and Defined Contribution pension markets 

The next generation of effective, personalised AI applications can transform the way people plan for their future – a crucial factor in the pension landscape.

Artificial intelligence (AI) is no longer on the periphery of operational efficiency and customer engagement. Generative AI has penetrated the mainstream through tools such as ChatGPT and Google Gemini with its technology powering everyday tasks such as autocompleting texts, theatre booking and Netflix recommendations. 

Against this backdrop, however, readily available AI models still have their limitations and even the most advanced models are still short of general human intelligence. For example, members searching online for technical pension questions might be shown inaccurate or non-compliant information, or simply material ill-suited to their circumstances. Trustees, in turn, need to be aware of these types of risks to members and will need to improve their own AI literacy.    

However, AI can also be harnessed proactively to better serve members. The emergence of AI conversational agents – designed to listen, probe and display accurate, relevant information – is a new approach to help members engage effectively with their retirement benefits and improve trustees’ knowledge, too. 

Two-way dialogues for a better AI experience 

In pensions, it’s crucial that technology must be in the service of what members do and need. 

AI’s hyper-personalisation capabilities enable us to engage with members of all types of pension schemes, whether defined contribution (DC), defined benefit (DB) or hybrid. 

This allows us to engage with people from across different financial literacy backgrounds and demographics, on an individualised basis, but at scale – especially helpful given the increasing prevalence of multi-scheme vehicles such as DC master trusts. 

The emergence of AI conversational agents means that we can now have two-way conversations with many thousands of people at once about things that matter to them on their terms. It will completely change the way in which people interact with financial products, and smash some of the barriers we face to making pensions and saving more accessible. 

Mercer’s Ask My Pension, partnering with Engage Smarter, is a conversational agent designed to provide personalised, compliant and accurate guidance and support to individual savers, with comprehensive information about UK pensions and the Mercer Master Trust.  

It’s not just about breaking down these accessibility barriers, but AI agents and supporting educational sessions also allow users the space to ask questions about pensions without fear of judgment. 

The human touch 

The ever-changing needs of trustees, such as applying The Pensions Regulator (TPR)’s DB funding code introduced in September 2024, prompted us to develop a human-centric AI solution to help trustees deliver the best member outcomes. 

Dear Sam is an engaging way for trustee boards to get hands-on training in a safe, simulated environment. This provides a welcome and more engaging alternative to 'traditional' methods such as PowerPoint presentations. 

By interacting with our fictional AI negotiator Sam in the role of sponsor, they can work together as a team towards getting the best outcome for members while facing a range of real-world challenges. As a result, trustees are more prepared and confident in their negotiations. 

Learn more about Dear Sam in our short video:

Outside of the DB funding code, we also support trustees through our AI-readiness workshops focusing on practical uses, potential risks and strategic planning for AI in pensions.

Advanced analytics 

Meanwhile, AI has also been improving analytics capabilities in pensions. Discover, our multi-award-winning AI-powered DB covenant portal, has two key AI components: “traditional” predictive AI and agentic news monitoring AI. It uses these to provide robust and proportionate real-time covenant analysis and monitoring, ideal for proportionate situations including sectionalised or multi-employer schemes, such as master trusts and local government pension schemes.

The predictive AI is trained on our proprietary database of covenant ratings to robustly but proportionately capture our core methodology, while the agentic AI monitors news flow about pension scheme sponsors and selectively processes and analyses it. The results of that monitoring are fed into the predictive model, allowing trustees to track their covenant over time and receive automatic alerts when there are noteworthy developments.  

These are only some examples of the myriad ways that generative AI is increasingly helping the pensions industry with advanced analytics, and agentic AI will be core to implementing these at scale while keeping tasks ‘bite-sized’ to work around present-day AI limitations.      

Learn more about Discover in our short video:

Potential issues and risks 

There are of course risks associated with providing inaccurate information via AI, and it is imperative to implement robust governance while leveraging AI for better outcomes. 

Our AI agent is heavily caveated to ensure users understand they are interacting with a piece of technology, not a financial adviser. At the same time, our user testing shows that people want their personal data reflected in AI agents, so we need to consider how we can create data-led interactions within the boundaries of regulation. 

More generally, as many of us are using readily available Large Language Models to ask our everyday questions, finances can come into that conversation. The danger is that we have no way of knowing where the answers are sourced from, opening the door for our pension scheme members to make poorly informed decisions based on inaccurate information. 

The introduction of a fully compliant, accurate expert agent, like ‘Ask My Pension’, is our counter to this risk. 

Opportunities for the future 

Future phases of our AI agents will include more advanced analytical or modelling capabilities, introduce a wider range of data sources (including personalised data), and expand the level of autonomy granted.  

For Ask My Pension, we are exploring new deployment via channels such as WhatsApp, so members and clients can use the tool where they feel most comfortable. 

As we already see with Discover’s monitoring function, hyper-personalised communications switch the dialogue from reactive to proactive, directing the conversation to a user’s specific circumstance. 

The implications for Ask My Pension are that it won't be reliant on you to talk to it. It will know everything about you and will come to you to tell you the things you need to do. For example, it might see that you need to nominate a beneficiary, tell you where you do it, and then perform that transaction in the back end. 

The direction of travel 

Whatever the direction of travel, our human-centric approach focusing on members will remain at the core of our AI development. Gradually, as functionality and trust builds, we may see benefit in moving the human from ‘in-the-loop' to ‘on-top-of-the-loop' for more use cases. For example, harnessing the power of AI conversational agents can bridge gaps in support, especially during non-working hours, to provide immediate assistance. 

Today is only the very beginning of the capabilities of agentic AI in pensions. For now, it is limited to low-risk analytical decisions, but we see potential for it to become more advanced, more autonomous and more action-oriented in the future. 

Authors
Tom Higham

- Head of Engagement, Mercer Master Trust

Felix Mantz

- Senior Director, Cardano

Related solutions

Related insights