When Dan, a 41-year-old software engineer from Florida, was wondering how to invest the excess cash in his Individual Retirement Account, he turned to ChatGPT for advice.
Initially looking for one investment to complement his existing portfolio, which included a large real estate mortgage Reit and a fund focused on dividend leaders, he ended up overhauling the whole $200,000 portfolio based on a compelling case made by the chatbot.
After telling ChatGPT how much he had in different holdings and how much cash he wanted to deploy, the AI told him that his risk exposure was too concentrated and he would be better off putting 80 per cent into a broad market equity index tracker and the remainder into a bond ETF — which could stay put for decades.
“I thought the AI gave really reasonable, sound, financially prudent advice,” says Dan, who, like the other case studies in this story, asked us to use only his first name. He adds that, while he could have come to a similar conclusion without it, speaking with the chatbot helped him to “commit to and actually execute” his plan.
Dan is one of millions of people on both sides of the Atlantic turning to generative AI for pension planning. A study of 5,000 Britons commissioned by Lloyds Banking Group late last year concluded that more than half of adults were using AI platforms for financial advice.
As many as one in three were using the tools at least once a week for information or advice on money matters, with OpenAI’s ChatGPT the most-used platform, followed by Google’s Gemini.
Escalating adoption has also caused turmoil in markets, with wealth managers and brokers seeing sharp falls in their share prices last month after a US-based fintech launched an AI-led tool to help financial advisers personalise clients’ investment strategies.
The worry for investors was over how the technology might undermine the traditional wealth management industry. Jason Wenk, chief executive of Altruist, which developed the tool, warned it made “average advice a lot harder to justify”.
Still, experts say that it’s critical for users to understand the limitations of AI and check its output before acting on its advice.
When Will, a 32-year-old strategy consultant, was moving back to London from the US, he was trying to understand distribution laws when it came to liquidating pensions in the US and how the UK tax authorities treat that.
He says ChatGPT told him that a new approach had been adopted last year that allowed you to do this and not incur tax. But when he went to check “it just absolutely wasn’t the case”.
“It’s just not very good at that level of detail because it’s just not able to actually use enough real reasoning to connect real concepts to real user situations,” says Will.
Another problem is that while chatbots are very good at sounding confident, they can slip up in mathematical calculations, although the most advanced models are improving all the time — in part by learning to use external tools such as code interpreters and calculators to handle computation more reliably.
They slip up because many of the models are not inherently performing actual calculations. Instead, they generate responses by predicting the most likely answer based on statistical patterns learned from their training data.
In a report published in December, the Financial Conduct Authority cautioned against AI “hallucinations” — where AI misinterprets information and generates things that sound right but are not true — while its confidence can make it seem authentic.
The watchdog suggested users ask AI to provide its sources, so they can verify the material against non-AI generated information.
Those using AI to help with retirement planning, be it working out how much to save or how much they can safely withdraw in retirement, need to know the right questions to ask.
“The quality of the output entirely depends on the inputs and it can occasionally default to generic or outdated assumptions if you are not specific,” says Paul, a 26-year-old who works in the finance department of a utilities company in Ireland.
He adds that he had found it particularly useful for translating broad advice into numbers: “If an adviser suggests a particular structure or strategy, I can quickly model what that implies over 25 or 30 years. That reduces the ‘black box’ element and makes it easier to see what I am actually paying for.”
When we asked FT Money readers if they were using AI to advise on their pension, dozens wrote in to tell us that they had been doing so for some time.
Some told us they thought AI tools may mean they never need to use a financial adviser, but those with more complex financial situations are using the tech to limit the amount of time they need to spend with their advisers — and check their workings — rather than replace them entirely.
John Bilton, global head of strategic research for the multi‑asset solutions team at JPMorgan Asset Management, says chatbots do not replace financial advice because if you don’t know very much about finance, chatbots will just present you with information structured in a more efficient way “that you still don’t understand . . . I think that’s the big risk”.
He says he has been working in markets for over 30 years, but when it comes to his pension he’ll always go to an adviser because: “I know what I don’t know — I think that’s the danger of AI is that people will assume they know what they don’t.”
He adds that if you use it as an investment tool rather than a data tool, it risks making underlying behavioural biases — such as the tendency to hold too much in cash or trade too often — stronger.
But it’s still early days for AI disruption and its impact. In January the FCA launched a review, led by Sheldon Mills, former FCA consumer chief, to explore how AI could evolve in the future, and what this means for markets, firms, regulators and retail investors.
In November, consumer research group Which? put six of the most popular generative AI platforms to the test, by asking 40 personal finance-related questions and scoring each on the quality of the answer.
Perplexity received the highest overall score, followed by Google’s Gemini AI overview. Meta’s responses proved weakest, with the lowest level of detail and specificity provided in its answers.
Andy Laughlin, a principal researcher at Which?, says that when he tested them, the scores and outputs across some of the platforms were “quite similar”, but Meta felt more “rudimentary” and “generic” than the other tools. Gemini and Perplexity stood out for accuracy, while the popular ChatGPT was only average.
He adds that “the seeming goal of AI chatbots to please us can cloud their accuracy and constructive usefulness” — they try too hard to simplify complex issues into easily readable summaries, often losing the nuance, detail and context in the process.
For example, his testing found that the tools often didn’t understand how users in Scotland or Northern Ireland might be impacted by specific financial rules and regulations.
Meta declined to comment.
Gary Smith, partner at Evelyn Partners, a wealth manager, says he uses AI from an efficiency and productivity standpoint, but not for actual advice, owing to the mistakes it can make.
“Following the introduction of the lump sum allowance and the lump sum death benefit allowance [in 2024] alongside the proposed changes to pensions and inheritance tax (IHT) from April 2027, pension planning is becoming far more complex and our advice has to be matched to individual client circumstances,” he says, noting cases linked to recent pension legislation where AI tools had given the wrong advice.
“The issue with pensions is that there are so many layers of legislation and product vagaries to consider and if these are not fully understood by the person inputting to AI, it can be nigh-on impossible for AI to generate the correct answer,” he adds.
Still, experts agree that AI tools are a great way to organise data, make suggestions, provide a second opinion on assumptions and compare existing financial products. They can also be used to prepare a list of questions before meeting a financial planner.
FT Money readers have also found AI tools helpful for more holistic retirement planning — including how to integrate their finances and hobbies.
Greg, a 51-year-old cloud consultant from Northamptonshire, says he has been using ChatGPT to come with a plan on how to spend his retirement, including learning a language, running marathons in different countries and places he wants to visit.
He uses AI to suggest and map out what he might want to do every year, and how much he would need to take out of his pension each year to support this lifestyle. He also found it good at checking his plans, which he has built by mapping out income streams across different pensions in Excel.
“I don’t use AI to create the plan but I find it really useful to validate it,” he says, adding that it had pointed out that he had failed to incorporate the tax-free allowance into his plan.
Given the boom in AI, pension providers are now scrambling to keep up, with many launching their own generative AI tools, which they say can avoid the pitfalls of more mainstream platforms by guiding users on what questions to ask and ensuring responses refer to the latest set of regulations.
Scottish Widows, part of Lloyds Banking Group, is launching “InvestAI” next month — which will be more effective than using more generic chatbots because it will have more information about the customer, according to investments director Manuel Pardavila-Gonzalez, who is overseeing the project. “It’s very easy [for users] to miss critical information,” he says, adding that Lloyds’s tools will suggest follow-up questions based on a client’s own circumstances.
But others are further behind. The UK’s largest defined contribution workplace pension provider, National Employment Savings Trust, says it is not currently looking to launch its own AI tool but recognises that “in the near future AI may help us offer widespread customisation to give members a better experience, and ultimately better outcomes”.
Another advantage of using one of these tools, according to the people building them, is that the firms are authorised and regulated by the Financial Conduct Authority. This means that they have a duty of care to their customers, unlike those provided by big, largely US-headquartered, technology companies.
But investment providers’ use of computers to cut costs and streamline processes has historically been mixed. A flurry of so-called “robo advisers” were launched in the UK in the early 2010s, but struggled to succeed.
Nutmeg, one of the earliest and biggest with a quarter of a million customers reported last year — never turned a profit despite managing assets of £7.4bn.
JPMorgan, which bought the robo-adviser in 2021, dropped the Nutmeg brand last autumn, citing the benefits of “leveraging JPMorgan expertise and heritage to provide consumers with exceptional investment products and services”.
Lloyds’ Pardavila Gonzalez says one of the problems with robo-advisers had been that they pushed people to invest too quickly, whereas the bank’s new tool would help with information and decision-making rather than quickly pushing people to transact.
Dan from Florida says he has found AI most useful for “financial education and willpower” adding that “the experience is great because you can chat with it at any time, I don’t have to schedule a meeting with someone, I don’t feel rushed to get off the phone and I don’t feel like I’m being hustled into buying something”.
Leave a comment