Will AI Replace Your Financial Adviser? What the Evidence Says

The statisticians changed their answer this year#

For years the safe reply to this question was that software could rebalance a portfolio but could not talk anybody out of selling at the bottom. That answer is doing less work than it used to.

The US Bureau of Labor Statistics currently counts 299,400 personal financial advisers, pays them a median $105,070, and projects employment growth of 1% over the decade to 2035. The handbook gives a reason: "artificial intelligence (AI) tools for financial advice may moderate demand" (BLS Occupational Outlook Handbook). A government statistical agency does not write that on a hunch.

In July 2026, four economists tested the question properly rather than surveying opinions about it. Their results are more interesting than either the hype or the reassurance.

What an adviser is actually being paid for#

Strip away the marketing and a financial adviser sells three separate things.

The first is technical: how much to save, which tax wrapper to use, how to split money between shares and bonds, and how much can be drawn from a pension without running out. The second is behavioural, the phone call in March 2020 that stops you selling everything. The third is accountability. A regulated UK adviser owes you a suitability duty, carries professional indemnity insurance, and can be taken to the Financial Ombudsman Service if the advice was wrong.

Software has been chipping away at the first item since robo-advisers arrived. A robo-adviser is a website that asks a short risk questionnaire and then runs a rules-based model portfolio for a fee. A large language model, the technology behind ChatGPT and Gemini, works differently. It produces text one word at a time from statistical patterns, so it can discuss your circumstances without holding any formal model of your finances underneath. That distinction matters for what follows.

The best test yet of how good the machine is#

Taha Choukhmane, Weidong Lin and Matthew Akuzawa of MIT Sloan, with Tim de Silva of Stanford, asked 1,000 US adults to write their own prompts asking for financial help. They fed those prompts to GPT-5.2, then simulated what would happen if households followed the answers across a lifetime of job losses, market falls and bereavements (Choukhmane, de Silva, Lin and Akuzawa, 29 July 2026).

The advice was mostly sound. It pushed people towards near-universal stock market participation, against a real world where about a third of households own no shares, and towards an equity share that falls after 45. Both sit close to what mainstream life-cycle economics prescribes.

The failures were specific rather than catastrophic. Saving rates bunched at round numbers, with 31% landing on a multiple of 10%. Retirement withdrawals stayed at or below 4% of assets in 98% of cases, whatever the circumstances. Portfolios drifted instead of being rebalanced. And when a simulated household lost its income, the model cut spending sharply even with cash in the bank to smooth it out.

The uncomfortable finding is about who asks. By 60, simulated savers with low financial literacy were about $46,000 behind those with high literacy, and people who had never used AI were roughly $100,000 behind experienced users. Women ended up about $60,000 behind men. Two thirds of that gap came from how prompts were written; the rest came from the model, which gave identical prompts labelled female equity recommendations 0.54 percentage points lower. As de Silva put it to Stanford Report, "The way you write the questions matters a ton."

Free advice that works well for confident users and less well for everyone else is a familiar pattern in financial services. It is the advice gap in a new outfit.

Advisers have already hired the machine, as an assistant#

Adoption inside the profession is running ahead of the public debate. The 2026 intelliflo UK Advice Efficiency Survey of 209 advice professionals found 74% using AI, up from 43% a year earlier. Of those, 87% use it for note-taking and 44% for writing reports, which matters when 30% of advisers spend three hours or more on a single suitability report.

In the US, research by Edward Jones and Morning Consult covering 201 advisers in May 2026 put adoption at 82%. What they use it for is telling: 59% want AI on scheduling and meeting preparation, 53% on routine client emails. Sixty-eight per cent said long-term client trust still needs a human, and 38% said clients now check their recommendations against AI answers.

Substitution so far is happening at the level of tasks, not jobs. The paperwork is going first.

Regulation, not capability, will set the pace#

Britain is running an experiment that will tell us more than any product launch.

The Financial Conduct Authority estimates that about 23 million consumers are underserved by the advice and guidance markets. Its own Financial Lives 2024 survey of 17,950 adults found that 8.6%, or 4.6 million people, had taken regulated advice in the previous year, while 29%, or 15.8 million, had not despite holding £10,000 or more in investible assets or a pension pot they planned to access soon. Cost was cited by 14% of that group and a lack of trust in advisers by 13%.

The regulator's answer is targeted support, a category between generic guidance and full personal advice, letting a firm suggest a course of action to a group of consumers with shared characteristics without knowing everything about any one of them. Final rules landed on 26 February 2026 and the regime went live on 6 April 2026. The FCA's stated aim is support "at a cost they can afford", and a service priced for millions of people is unlikely to be delivered by humans on the phone.

Supply is the other half of the story. The FCA's financial advice market survey of 23 April 2026 counts about 31,000 UK advisers, flat since 2021, at 5,500 firms, down 15% over the same period. They look after 4.1 million clients and £1 trillion, roughly 150 clients each. Cerulli reported on 4 August 2026 that 35% of US advisers plan to retire within ten years, taking 40% of industry assets with them.

Read those two datasets together and the five-year problem is not too many advisers. It is too few, serving too few people, at a price most cannot pay.

Where the line currently sits#

Task	Where the evidence points today	Likely position by 2031
Portfolio construction and rebalancing	Already automated; AI allocations sat close to theory but drifted	Automated, human oversight on exceptions
Cash flow and savings-rate planning	Reasonable, but rounded and slow to adapt after income shocks	Automated with better data feeds
Tax wrappers and pension rules	Rule-heavy and country-specific, so plausible errors are costly	Automated where rules are codified, checked by a person
Talking a client out of a panic sale	Vanguard measured a 56-point gain in peace of mind from human advice against 12 for digital	Still human for most clients
Divorce, business sale, estates	Untested in the research, and dependent on facts a model cannot see	Human-led
Liability for the outcome	Sits with the authorised firm, not the software	Unchanged, and the reason firms move slowly

Sources: BLS, Choukhmane et al., Edward Jones, Vanguard, FCA. The 2031 column is interpretation, not fact.

Key takeaways#

The official employment forecast has already shifted. US adviser employment is projected to grow 1% to 2035, and the BLS names AI as one reason.
Chatbot advice tested well on the basics and badly on the details: round-number saving rates, rigid 4% withdrawals, no rebalancing.
The quality of the answer depends heavily on the quality of the question, which is precisely the opposite of what an advice gap needs.
Advisers are adopting AI faster than clients are, and using it on administration rather than on judgement.
Nobody has moved regulatory liability onto a model. Until that changes, a human firm answers for the outcome, which caps how far replacement can go.

Frequently asked questions#

Will AI replace financial advisers within five years?

The evidence does not support that. Labour statistics point to flat employment rather than collapse, a third of US advisers are retiring, and liability still sits with authorised humans. Expect fewer routine tasks per adviser and more clients each.

Is it safe to ask a chatbot about my pension?

For understanding how something works, it is a reasonable starting point. For a decision with tax consequences or an irreversible transfer, the 2026 study found systematic errors, including withdrawal rates applied regardless of circumstance. Check specifics against the official rules or a regulated adviser.

What is targeted support, and will a robot deliver it?

It is a UK regime, live since 6 April 2026, letting firms suggest a course of action to groups of similar consumers. The FCA has not mandated a delivery method, but the economics of serving millions cheaply favour digital channels.

Are human advisers worth their fees?

Research by the International Longevity Centre and Royal London, using the government's Wealth and Assets Survey, found people who took advice in the early 2000s were £47,706 better off a decade later. That compares advice with no advice, not advice with AI. Nobody has run the newer comparison.

Should I cancel my adviser and use AI instead?

That is a decision about your own circumstances and this article is not advice. The evidence suggests AI is strongest where you already know what to ask and weakest where you do not.

What should I watch over the next five years?

Whether targeted support reaches large numbers of people, whether any regulator lets a firm rely on model output without human sign-off, and whether adviser headcount falls rather than merely flattening.

Glossary#

Robo-adviser A website or app that allocates your money to a model portfolio after a short questionnaire, with no human conversation.

Large language model (LLM) Software trained to predict text, which lets it answer questions in ordinary language. It has no built-in model of your finances.

Life-cycle theory The mainstream economic framework for how much a household should save and invest at each age.

Equity share The percentage of a portfolio held in shares rather than bonds or cash, usually reduced as retirement approaches.

Withdrawal rate The percentage of a pension pot taken as income each year. The often-quoted 4% figure is a rule of thumb, not a rule.

Advice gap The distance between the number of people who would benefit from advice and the smaller number who get it.

Targeted support A UK regulatory category, live since April 2026, allowing firms to suggest actions to groups of consumers with shared characteristics.

Suitability The regulatory duty on an advice firm to ensure a recommendation fits the individual client's circumstances.

References#

US Bureau of Labor Statistics, Personal Financial Advisors, Occupational Outlook Handbook, employment 2025, projections 2025 to 2035.
Taha Choukhmane, Tim de Silva, Weidong Lin and Matthew Akuzawa, AI Financial Advice: Supply, Demand, and Life Cycle Implications, 29 July 2026.
Stanford Report, The quality of AI's financial advice depends on how you ask, July 2026.
Financial Conduct Authority, Financial Lives 2024 survey: financial advice and support, published May 2025, fieldwork February to June 2024.
Financial Conduct Authority, PS25/22: rules for targeted support, final rules 26 February 2026.
Financial Conduct Authority, FCA opens authorisation gateway for targeted support, 2 March 2026.
Financial Conduct Authority, Understanding the advice market: financial advice firms survey, published 23 April 2026.
Cerulli Associates, Advisor Retirements Underscore Need for Stronger Rookie Development, 4 August 2026.
intelliflo, 2026 UK Advice Efficiency Survey, reported by IFA Magazine, 23 July 2026.
Edward Jones and Morning Consult, AI and the Future of Financial Advisors, fieldwork 15 to 27 May 2026.
Paulo Costa and Jane E. Henshaw, Vanguard, Quantifying the investor's view on the value of human and robo-advice, February 2022.
International Longevity Centre UK and Royal London, What it's worth: revisiting the value of financial advice, 28 November 2019.