In a tech-driven world advisers may soon have to evaluate the effectiveness and reliability of providers’ AI capability when assessing wider pension propositions and pricing.
TO DOWNLOAD A FULL PDF OF THE ROUND TABLE REPORT CLICK HERE
This was one of the striking observations to come out of a recent Corporate Adviser round table, looking at the emergence of default retirement solutions.
Advisers and providers attending the event agreed AI will have a transformative effect on the pensions market — particularly when it comes to helping people make better decisions at retirement, be it through default solutions, targeted support, or delivering personalised advice on a more cost-effective basis.
But this will create specific challenges for the intermediary sector. Muse Advisory associate Paul Armitage said: “We will have a new role as evaluators. Clients already pay us to assess products and propositions and whether they are delivering value for money. AI will rapidly become a key component of providers’ propositions, but will advisers have the skillset or capability to evaluate how providers are using this technology in-house, be it to offer guidance to members or create cohorts for default retirement solutions?”
Armitage wasn’t the only adviser to express these concerns. WTW director Mark French pointed out that governance will be key. AI won’t just be used to segment membership and guide them towards products or solutions, he said. It will also be behind chatbot interactions and used by call centre staff when dealing with queries.
“Governance will be critical, but this is true at present where you have humans operating call centres and having different conversations with lots of different people. The question is how you monitor, control and evaluate that.” He said advisers will want providers to be transparent about these processes.
Stephen Coates, head of proposition at Mercer Workplace Savings, said this AI revolution comes as the pensions industry is on the cusp of significant legislative change. “We’ve got targeted support and default retirement on the horizon. This brings into sharp focus the question of personalisation, and whether AI can power hyper-personalisation.”
Mercer head of engagement Tom Higham said the firm is looking at AI to help deliver this, creating more bespoke products where possible. “It is all about the data. Through our Destination Retirement product we’ve got more data about our members than some other providers.” AI has the potential to “super-charge” the use of data, he said, “moving very quickly to a more hyper-personalised approach”. Ultimately this should allow providers to segment and target cohorts far more accurately, he added, with the ultimate aim of delivering more individualised guidance and support to individuals.
This vision for the future clearly shows the huge potential for this rapidly-developing technology. But the discussion also highlighted potential limitations at present, and more immediate challenges for the workplace sector.
Data challenge
One of the biggest challenges providers face is managing this data. Under current proposals for targeted support, firms can use pre-defined data points to segment their customer base into key cohorts, targeting communications, calls to action or ready-made product solutions to these different groups.
However, there is the question of how many cohorts providers will identify, and how large or small these will be. Current FCA proposals suggest that segmentation shouldn’t be so broad as to render suggestions meaningless, but neither should they be so finely tailored that they become de facto personalised advice.
Coates said that while the thinking behind this policy initiative is to allow firms to offer more targeted support to consumers, particularly around retirement decisions, in practice this might not always happen.
“We may find ourselves in the almost paradoxical situation where you might get some organisations becoming more data-phobic, because if you build your targeted support cohorts you don’t want too much data.”
He pointed out that current FCA proposals state you can’t ignore certain pieces of data, which then might make it difficult to fit an individual into a pre-defined group. “The question is what do you do with individuals who don’t fit these cohorts? You can refer them for advice but what if they don’t want to pay for that?”
Advisers agreed it is clearly not going to be practical or cost-effective for providers to run hundreds of separate cohorts, with bespoke communications and different product choices.
Armitage also raised the question as to what might happen as providers gain more information and data on members. “This could see a member effectively moved into a different cohort, which might change the messaging they receive. This could contradict previous messaging or calls to action, which hardly sounds optimal.”
XPS head of DC investment Mark Searle added that different providers may use different data points for targeted support, again potentially leading to conflicting messages for consumers with more than one pension.
Searle said he would like greater clarity from the FCA when the final rules are published to ensure greater consistency across the industry.
Isio DC pensions consultant Thomas Chalkley said these various issues support a model where members get independent support, rather than five different solutions from five providers, all of whom have a vested interest in trying to keep hold of these funds.
Coates added: “This whole issue is potentially problematic. And perhaps you get to the point where maybe you’ve got so much data that it is difficult to match it to a targeted support model.
“I think there are two ways of looking at current regulatory developments. Are targeted support and default retirement options a step on the path to greater personalisation, or is this the antithesis of that?
Coates added that Mercer may “dip a toe” into targeted support, but the proposals currently only extend to FCA-regulated firms, not master trust providers overseen by The Pensions Regulator.
He said the firm’s ambition was to move towards “hyper-personalisation” where members receive more bespoke advice. “That’s where we should go, and we should do this as cost-effectively as possible to democratise the process so all members get access to this help.”
Engagement challenge
Advisers discussed whether providers always have the relevant data needed to provide effective targeted support, as well as more bespoke solutions.
French agreed that this could be an issue. He said in some ways this might differentiate targeted support from default retirement solutions. “The more data you have, the more personalised and the more targeted your solution is going to be. The fewer data points you have, the more it looks like a default. So it will be a sort of sliding scale.”
LCP principal DC consultant Jessica Clayson said : “This goes back to one of the biggest and long-standing challenges in the industry: member engagement. We need to have good engagement with members to ensure schemes have meaningful data, enabling them to target appropriate solutions. Without this, then you could get five different solutions from five different providers.
“But the challenge isn’t just engagement, it’s also trust. Will members be comfortable sharing this data with every provider, particularly if this information is about their broader financial situation?”
Chalkley agreed this was an issue. “Across the industry members are largely disengaged, so providers don’t have access to the data they need. The challenge is gathering this information at a level where it can be effectively analysed and acted upon.”
Trust issues
Advisers at the event said this issue of trust was key. Members have to be comfortable sharing financial data with providers and other third parties. But they also need to be confident about the accuracy of the results of any outputs — whether this is answers to questions asked, recommended products, or solutions or a call to action. This could become more of an issue in a world of AI-driven outputs.
While providers are already developing their own specialist AI agents, many members are currently using more general large language models (LLMs), such as ChatGPT, to answer questions about pensions and retirement. As Coates pointed out the accuracy between different AI agents can vary considerably.
Coates said that broader LLMs may be adept when it comes to answering more general questions, but accuracy dips when it comes to more technical or specialist issues around pensions. He pointed to research that shows that specialist AI agents give wrong answers just 2 to 3 per cent of the time.
In contrast, human error means helpdesks gave inaccurate information around 18 per cent of the time, with ChatGPT4 giving wrong information between 38 and 40 per cent of the time.
Barnett Waddingham partner Andy Parker said this issue will be amplified with the launch of pensions dashboards. “People are going to be able to see these different accounts, but who is going to tell them what to do with this information and how to turn these different pots into an income in retirement?”
French said he expected that early iterations of the dashboard will come with integrated AI technology — designed at the very least to help people understand germane issues, such as charges or performance on different funds.
As these dashboards evolve and commercial versions start to come to market, he said he expected AI tools to be a more prominent feature, offering support with consolidation and guidance around retirement decisions.
Armitage added: “The more we’ve talked about this, the more it strikes me that individual providers are in a pretty poor place to give targeted support and it needs to come from somebody central, like a dashboard or adviser, that has an overview of an individual’s financial situation.”
But some advisers expressed concerns that this could open the door to providers with the biggest marketing budgets winning the ‘consolidation’ battle — by effectively promising to do this legwork for consumers, even if they do not represent the best value for money.
However, while French acknowledged the power of brand recognition and loyalty, he pointed out that AI-driven dashboards should show which schemes offer real value for money. “The differentiator is going to be the return, and the rates you’re getting.”
There was debate though as to how this might translate into decision-making at retirement, particularly in terms of consolidating pension pots.
Would people choose to consolidate into their biggest pot, the one with the best return to date, or would dashboards show more meaningful comparisons between default retirement options?
Those at the event said it remained unclear how this might all work in practice, as many default models, for example flex and fix solutions, would be comparing different withdrawal rates.
This is further complicated by the fact that the dashboard will initially focus on projected incomes from pension pots, rather than fund size.
Armitage said: “I would have thought pot values would have been a more engaging number than the relatively small level of projected income from these pots.”
But Higham disagreed. “I quite like the transparency of an income, because people can wrap their head around what this means. At the moment if a projection says you will have a pot worth £100,000 in 30 years’ time most people will have no idea what to do with that information.”
Across the workplace sector there is now a new focus on retirement support
as a means of driving better member outcomes.
It is clear that regulatory changes combined with new technology are opening the door to cost-effective advice options, targeted support and more bespoke retirement solutions. For advisers the challenge will be in having oversight of this complex landscape as it evolves, and being able to compare individual propositions, which might involve an AI-drive targeted support matrix or more personalised guidance and support options.
