AI and financial advisers: what's still worth paying for?

AI will answer your pension questions in seconds, for nothing, and the answer will usually be pretty good. That is a real problem for anyone charging for routine financial advice. But the question of AI and financial advisers may turn out to be less about who gives the better answer than about who is answerable for it.

Say you're in your mid-50s, like the couple I have in mind here. Between you there are two pension pots, a mortgage coming off its fixed rate in March, £40,000 sitting in a savings account earning very little, and a message from your daughter about a flat she's found in Leeds and can't quite afford the deposit on. One of you would like to stop working at 60. The other thinks the investments are already riskier than they should be. You type the whole lot into ChatGPT and ask what you should do.

It takes about 20 seconds. Back comes a summary of your position, a set of calculations, a suggested order of priorities, a warning that you're light on emergency savings and a flag about the tax consequences of one of the options. It offers to build you a full retirement plan next. The answer is clear, it's specific, and most of it is right. It also costs nothing.

So what exactly would a financial adviser add? That isn't a question about the future. Research commissioned by the FCA found that 16 per cent of consumers had already used AI for at least one personal finance task; of those, 61 per cent had asked for suggestions based on figures they'd supplied and 24 per cent had uploaded personal documents. AI is already working as explainer, calculator and second opinion, and it's improving fast.

Whether it produces useful answers isn't really in dispute any more. It does. The question is where useful stops.

AI and financial advisers: the work that has already gone

Start with the concession, because it's a big one. A great deal of what people used to pay a professional for is now free, instant and perfectly decent.

AI will explain what the annual allowance is and why yours might be tapered. It will read a 40-page fund factsheet and tell you what you actually hold. It will compare product features, work out what happens if you pay in another £500 a month for eight years, and show you the difference between drawing four per cent a year and drawing five. It will even prepare you for a meeting with an adviser by identifying what information you're missing and what you ought to ask. None of that is trivial. Clear explanations matter, accurate sums matter, and until very recently you had to pay somebody for both.

The trouble starts when a useful answer gets mistaken for a complete financial plan.

Plenty of people won't need more than the answer, and it's worth an adviser saying so out loud. If your circumstances are straightforward, some combination of AI, regulated guidance, targeted support and the occasional piece of paid advice may give you everything you need. An ongoing relationship with an adviser is not something everybody has to have, and anyone in the profession who suggests otherwise is part of the reason people distrust it.

The advisers with most to lose are the ones whose service could be reproduced by a competent prompt and a decent template. Fund selection, a bit of market commentary, some routine sums and an annual review that could have been written for anyone. Which leaves a more awkward question. What can a good answer hide?

A correct answer and a suitable one are not the same thing

An AI system can explain a pension rule correctly, calculate a sustainable withdrawal rate accurately and put together a portfolio that looks sensibly diversified, and the overall recommendation can still be wrong for you. It may have missed a fact you didn't think to mention, misread which of your goals actually matters most, or treated a shaky assumption as settled. Correctness, suitability and completeness are three different tests, and passing the first says very little about the other two.

Research by Winder, Hildebrand and Hartmann gives a flavour of the risk. Portfolios generated by several large language models showed geographic and sector concentration, trend chasing, heavier use of actively managed funds and higher costs. Targeted prompts reduced some of those biases, though not consistently. The study used simulated investors and the models are improving quickly, so I wouldn't lean on it too heavily, but it does show how a recommendation can look reasonable while carrying weaknesses most users would never spot.

Presentation makes them harder still to spot. A study by Takayanagi and colleagues, small enough that I'd treat it as suggestive rather than settled, found that people trusted financial advice more, and were more satisfied with it, when it came from a more extroverted AI agent. The advice itself was worse. Confidence can be manufactured by tone as easily as by evidence, which is not a discovery unique to machines.

So the real danger isn't that a chatbot says something obviously mad. It's that it tells you what you were hoping to hear, glides past an inconvenient trade-off, or never asks the one question that would have changed your mind.

Which is where the usual framing of AI and financial advisers falls down, because it assumes the machine is biased and the human isn't. Research by Linnainmaa, Melzer and Previtero found that Canadian advisers recommended expensive, actively managed, return-chasing strategies to clients and then invested their own money in much the same way, some of them still doing it after they'd left the industry. Sales incentives don't explain that. They believed it. A 2012 audit study by Mullainathan, Noeth and Schoar found something adjacent: advisers often failed to correct the biases clients walked in with, and sometimes reinforced the ones that happened to be commercially convenient.

Neither study convicts every adviser, and I'm not suggesting it does. But a conflicted adviser and a sincerely mistaken one can cost you the same money, which suggests the useful question was never human against machine. It's whether anyone in the process is willing to challenge an assumption, and whether anyone is answerable when it turns out to be wrong.

The hard part is not the arithmetic

You don't lack calculations. You could clear the mortgage, top up the pensions or send the deposit to Leeds, and one of those probably produces the highest expected return, another gives you the most certainty, and the third might matter more to your family than either of the first two. AI will estimate the tax on each, project the retirement income and compare the withdrawal strategies, and it will do all of it faster than any human. What it cannot tell you is which of those three sacrifices you are actually willing to make.

Take the retirement income question on its own. An annuity buys certainty and takes away access to your capital. Staying invested keeps the flexibility and hands you the market risk. Spending more in your sixties, while you're well enough to enjoy it, means less protection against what your eighties might cost. Every one of those is defensible.

Which one is right depends on things no model can read off a spreadsheet: how much each of you values security, whether a 30 per cent fall would genuinely cost you sleep even with the plan intact, how strongly you feel about helping your daughter now rather than in 20 years, and whether the most tax-efficient answer is still the right one if it leaves you feeling permanently skint. Behind all of it sits the hardest question in financial planning, which is working out what 'enough' actually means.

Planning is rarely a matter of finding the perfect answer. It's usually a negotiation between several worthwhile ones, and the compromise you reach in one place changes what's available to you somewhere else. Drawing pension income early reduces your flexibility later. Clearing the mortgage cuts the monthly outgoings and may leave you short of accessible cash. Helping adult children can be the most rewarding thing you ever do with your money and still thin out your resilience at 80.

Taken one at a time, each of those decisions looks sensible. Taken together, they can produce a life you didn't particularly want. Connecting the moving parts is most of the job, and it's the part that looks least impressive from the outside.

Then there's the bit nobody enjoys. You may need to be told that retiring at 60 isn't realistic on these numbers. That helping with the deposit now would put your own security at risk. That the portfolio you were hoping to overhaul is fine as it is and the sensible course is to leave it alone. Sometimes the honest answer is 'I don't yet have enough information to recommend anything', which is a good deal harder to say than something confident. You can prompt an AI to challenge you, and it will. But it has no stake in whether you take it badly, and it never has to sit opposite you and watch your face while you hear it.

A plan you can't stick to isn't a plan

Say you keep the money invested rather than clearing the mortgage. Six months later the market drops 25 per cent. The mortgage is still sitting there, the pension statements are worse than they were, and every headline you read seems designed to ask whether you knew what you were doing. Nothing about the plan has changed. You have. And that's the part the calculations never covered, because people who understand perfectly well that markets fall still want out when it happens, and the same people, 18 months into a bull run, will look at a sensibly cautious portfolio and feel like the only person at the party not making money.

The problem isn't ignorance. Our brains are wired to make poor financial decisions once fear or excitement or the sight of a neighbour's new car gets involved, and what shifts under pressure isn't your knowledge but your confidence, your priorities and your appetite for loss. Morningstar found that US fund investors earned less than the funds they held over the ten years to 31 December 2024, purely because of when they bought and sold. I wouldn't call that a fixed tax on being human, and it certainly isn't proof that an adviser would have stopped it, but it does show behaviour mattering at least as much as fund selection.

Good advice can't guarantee good behaviour. Neither can good information, which is the more uncomfortable finding. A field experiment involving around 8,000 customers of a large German brokerage found that only about five per cent took up an offer of free, unbiased financial advice, and those who did followed relatively little of it. A correct answer can sit there untouched for years.

Helping someone stay the course involves rather more than saying 'stay the course'. It means working out in advance which situations are most likely to trigger an expensive decision, agreeing what will happen when one of them arrives, and being able to tell the difference between a real change in circumstances and a bad week.

Occasionally the most valuable thing an adviser does is give you permission to do nothing at all, acting as a steady hand when greed or fear takes over.

I've never seen a credible universal figure for what this is worth, and I'd be sceptical of anyone who offers you one. The contribution is more mundane than a number. It lowers the odds that fear or impatience turns a workable plan into an expensive mistake, and for a lot of people that's the whole of the fee justified right there.

Who is answerable when the answer is wrong?

The debate about AI and financial advisers is usually framed as a contest over who gives the better answer. I think that's the wrong contest. The question that will matter more, and sooner, is who is on the hook when the answer turns out to be wrong.

A chatbot can produce something that feels intensely personal without creating any of the protections that come with regulated advice. You may have handed over your pension details, your mortgage, your spending and your family circumstances, and none of that gives you the complaints and redress rights you would have with an authorised firm. FCA-commissioned research suggests most people don't realise it: only 40 per cent correctly identified that acting on investment suggestions from a general-purpose AI tool does not, by itself, create a formal right to complain to the Financial Ombudsman Service. Whether you can use the Ombudsman at all, or claim from the Financial Services Compensation Scheme, depends on who provided the service, whether the activity was regulated and whether it falls within the scope of those schemes. An authorised firm that uses AI inside a regulated service stays responsible for meeting its obligations, and it can't blame the software when its advice falls short.

That accountability works on three levels. Regulatory, in that regulated firms must comply with the rules for the service they provide and clients have somewhere to go when they don't. Professional, in that an adviser should be able to tell you why a recommendation was suitable, what it assumed and how they arrived at it. And behavioural, in that good advisers don't hand over a plan and disappear. They revisit it, test whether it still fits and talk you out of abandoning it for the wrong reasons.

The UK is also building something in between. Targeted support is intended to let authorised firms with the right permissions offer ready-made suggestions to groups of consumers with similar characteristics, which could be genuinely useful for people who neither need nor want to pay for full financial planning. It isn't the same as a personal recommendation based on a proper look at your circumstances, and nobody should sell it as one.

Regulation doesn't guarantee good advice. Neither does AI. What regulation does is stop responsibility evaporating when something goes wrong, and that difference is worth more than it sounds. So the practical question isn't whether your adviser uses AI, because before long they all will. It's what they add once the routine work is done, and who carries the can for the result.

Five questions to ask before you pay for advice

A good adviser should be able to explain what you're getting without falling back on vague talk of reassurance, relationships and expertise. Ask them:

1. What important decisions have you helped me make? The answer should involve retirement, tax, family, risk or cash flow, not a list of products and funds.

   
   

2. How does my investment plan connect to everything else? Pensions, tax, protection, estate planning, the mortgage and what you want to do for your children should form one joined-up plan, not six separate ones.

   
   

3. When have you told me not to do something? A good adviser sometimes recommends no change to the portfolio, no new product and no action at all.

   
   

4. What evidence sits behind your approach, and how do you test your own assumptions? Be wary of vague appeals to market insight or to experience that can't be explained clearly.

   
   

5. What will you do for me over the next 12 months, what will it cost in pounds rather than percentages, and who checks the work? That last part now includes who reviews anything AI has produced before it reaches you.

Warning signs are frequent fund switching, complexity for its own sake, market forecasts, vague promises of outperformance and an annual review that could have been written for anybody. What you want instead is clear priorities, transparent costs, evidence-based investing and somebody prepared to argue with you rather than simply agree. Professional help earns its keep when decisions are interconnected, hard to reverse or emotionally charged. When yours aren't, you probably don't need it.

What you're actually paying for

You'll almost certainly ask AI before you ever ask a human, and you'll turn up better informed than clients did five years ago, with sharper questions and a rough idea of the answer. That's progress, and advisers who resent it are missing the point. AI will keep making information cheaper, calculations faster and conventional portfolios easier to build, and it will expose every adviser whose service amounted to fund selection, a market view and a review once a year.

But information was never the job. The advisers who go on earning their fee won't be the ones with the best answers, because the best answers are about to be free. They'll be the ones who help you work out which answer you can actually live with, keep you in it when your circumstances or your nerve change, and remain answerable for what they told you. 20 seconds got you a decent answer. Deciding whether it's the right one for your life, and sticking with it when everything in you wants to do something else, is the part that still takes a person.

Resources

Ahmed, S., Almond, R., Belton, C., Bogiatzis-Gibbons, D., Patel, K., Phadnis, M., Sholl, P., & Spang, J. (2025). Money talks: Lessons from two LLM pilots on consumer guidance. Financial Conduct Authority.

Bhattacharya, U., Hackethal, A., Kaesler, S., Loos, B., & Meyer, S. (2012). Is unbiased financial advice to retail investors sufficient? Answers from a large field study. The Review of Financial Studies, 25(4), 975–1032.

Financial Conduct Authority. (2024). Artificial intelligence update: Further to the Government’s response to the AI White Paper.

Financial Conduct Authority. (2026). Supporting consumers’ pensions and investment decisions: Rules for targeted support (Policy Statement PS25/22).

Financial Ombudsman Service. (n.d.). Complaints we can help with.

Financial Services Compensation Scheme. (n.d.). What we cover.

Linnainmaa, J. T., Melzer, B. T., & Previtero, A. (2021). The misguided beliefs of financial advisors. The Journal of Finance, 76(2), 587–621.

MoneyHelper. (n.d.). Money purchase annual allowance (MPAA).

Mullainathan, S., Noeth, M., & Schoar, A. (2012). The market for financial advice: An audit study (NBER Working Paper No. 17929). National Bureau of Economic Research.

Ptak, J. (2025). Mind the gap 2025: The more investors traded, the less they made. Morningstar Research Services.

Takayanagi, T., Izumi, K., Sanz-Cruzado, J., McCreadie, R., & Ounis, I. (2025). Are generative AI agents effective personalized financial advisors? In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 286–295). Association for Computing Machinery.

Winder, P., Hildebrand, C., & Hartmann, J. (2025). Biased echoes: Large language models reinforce investment biases and increase portfolio risks of private investors. PLOS ONE, 20(6), e0325459.

Yonder Consulting. (2026). AI consumer research. Financial Conduct Authority.

Finding an adviser

If you'd like to put those five questions to somebody, we can help. Second Life Financial Planning introduces people to advisers who work the way this article describes: evidence-based, clear about costs and prepared to tell you to do nothing when doing nothing is right. Just click on 'Find a planner', Tell us a bit about your situation and we'll suggest advisers worth talking to. And if it turns out you don't need ongoing advice, we'd rather say so than sell you some.

Or start with the book

You might choose to manage without an adviser, but you'll still need to know what you're doing. The book How to Fund the Life You Want by Robin Powell and Jonathan Hollow was written with you in mind and is out now in a second edition from Bloomsbury. Tax rules, pension regulations, annuity rates and care costs have all shifted since we first wrote it, so the numbers are current, and every chapter points to an online page where we'll flag anything that changes again. AI can answer your questions. This is about knowing which ones to ask.