top of page

Is AI any good at UK financial advice?

Nov 11
7 min read

AI can explain financial advice. But can it exercise judgement?



We gave five free AI tools the same five UK financial planning questions, from a simple ISA query to a complex estate-planning challenge.



They were often clear on the basics. But changing rules, personal circumstances and difficult assumptions exposed the limits.



Our verdict: AI can be useful for learning, but it won’t be surprising to you to learn that it isn't a good substitute for a human adviser.



Read more to see if you think AI is any good: https://lnkd.in/ejAGs6WG





The value of investments can go down as well as up, and you may get back less than you invested. This post is for information only and isn't personal investment advice.



Steve, I think this is a useful piece of research from 7IM, but its conclusion is considerably narrower than its headline suggests.

7IM has demonstrated some limitations of AI answering isolated financial questions. It has not demonstrated the limitations of AI-enabled financial planning.

That distinction matters enormously for the Academy.

1. My central criticism: they tested the wrong operating model

7IM tested five free AI tools using five questions, deliberately withholding important personal information and allowing no follow-up conversation.

That is a legitimate test of first-response quality. It is not a realistic test of a well-designed financial planning process.

Consider the difference:

7IM's test

An AI-enabled planning process

One question

Structured discovery

Missing personal information

Builds a personal information record

One response

Iterative conversation

No cash-flow model

Integrates lifetime cash-flow modelling

No verification process

Checks information against authoritative sources

No ongoing relationship

Maintains a plan as circumstances change

AI working alone

AI with episodic human expertise

The interesting question is not whether ChatGPT can produce a complete financial plan from an incomplete prompt.

It is whether a properly designed, AI-enabled planning system can help someone understand their circumstances, evaluate alternatives, make decisions and remain in control.

7IM hasn't tested that proposition.

They tested the intelligence of the answer, not the architecture of the planning system.

That could be a memorable distinction for an Academy response.

2. Their strongest criticism actually supports the Academy's model

Kelemen observes that the AI systems failed to challenge underlying assumptions.

For example:

  • Can someone actually afford to give money away to reduce inheritance tax?

  • Is saving £300 monthly sufficient to meet retirement objectives?

  • What happens if circumstances change?

  • Will the person's money last?

I agree these are important shortcomings.

But notice what they reveal.

The problem is not primarily investment selection. It is understanding the person's wider circumstances.

An inheritance tax question is actually a question about financial security, family relationships, mortality and personal priorities.

A £300 investment question is actually about future needs, earning capacity, retirement expectations and the adequacy of available resources.

This takes us straight to the Academy's central proposition:

Plan the person before the money.

Indeed, a person-centred process might discover that reducing inheritance tax or maximising retirement savings isn't the client's most important objective at all.

3. There is a revealing asymmetry in their comparison

7IM subjects AI to a five-scenario test and then compares its performance with what human advisers supposedly do.

But where is the equivalent evidence for human advisers?

Would five randomly selected advisers:

  • Identify every relevant tax change?

  • Ask every necessary discovery question?

  • Recognise every vulnerability?

  • Challenge the client's assumptions?

  • Produce accurate lifetime cash-flow projections?

  • Avoid product or asset-placement bias?

  • Recommend against investing when appropriate?

We don't know.

And that is a major methodological limitation.

The article says that nearly every human adviser offers cash-flow modelling. Even if that were true, offering a modelling service does not establish the quality of the assumptions, analysis or outcomes.

Nor does a human adviser's professional status guarantee impartiality.

This is especially important because 7IM is an investment management business operating within the existing adviser distribution ecosystem.

That does not invalidate its findings. However, its commercial position is relevant when interpreting its conclusion that consumers should continue relying on human advisers.

I would frame this as a question of symmetry:

If we demand evidence that AI produces good consumer outcomes, shouldn't we demand equivalent evidence from human advisers?

The appropriate comparator is not the ideal human adviser. It is the actual service consumers receive.

4. Their most interesting finding is almost buried

Only ChatGPT improved its performance as the scenarios became more complex.

That is intriguing.

It suggests that different AI systems may exhibit materially different capabilities when faced with complexity.

It also undermines treating AI as a single, undifferentiated technology.

But five questions, one response per question and one testing date are nowhere near enough to establish a reliable comparative ranking.

A stronger study would involve:

  1. Multiple realistic client cases.

  2. Repeated tests to assess consistency.

  3. Multi-turn discovery and clarification.

  4. Access to appropriate financial modelling tools.

  5. Verification against current UK legislation and tax rules.

  6. Independent assessment of consumer outcomes.

  7. A matched comparison with human-led advice.

Importantly, evaluators should distinguish factual errors, unsafe recommendations, failures to ask questions and failures to identify when professional intervention is necessary.

These are different dimensions of quality.

5. The false choice: AI or human adviser?

This is where I think 7IM's argument is weakest.

The closing proposition is essentially that AI can educate consumers, but humans must provide judgement and sustained support.

Yet this assumes that judgement and support must remain bundled into the traditional advisory business model.

Why?

The Academy's model separates these functions.

A different financial planning architecture

Human agency

The person retains ownership of their objectives, choices and decisions

AI operating system

Discovery, records, learning, modelling, scenario testing and plan maintenance

Episodic human expertise

Complex judgement, emotional support, specialist questions and interventions

Independent financial planning

Self-directed by default. Supported when it matters.

The person doesn't need an adviser continuously making decisions on their behalf.

They need the capability to make informed decisions, with reliable access to appropriate expertise when necessary.

This is precisely why the Academy's separation between Total Wealth Plans (the AI operating system) and Total Wealth Planner (the human support layer) is so significant.

It changes the economic structure of financial planning.

6. The most important omission: financial agency

There is a deeper issue.

7IM measures AI against the standards of the existing financial advice industry.

Accuracy, suitability, regulatory awareness, safety and transparency are important measures.

But what about the consumer's capability?

Does the process leave someone:

  • Better able to understand their financial circumstances?

  • More capable of evaluating alternatives?

  • Less dependent on intermediaries?

  • More confident in their decisions?

  • Better equipped to identify misleading claims?

  • More able to maintain their own plan?

These outcomes barely feature in the comparison.

Yet they are central to whether AI genuinely improves people's financial lives.

I would argue that a successful planning intervention should be judged not merely by the quality of its immediate recommendation, but by the capability it leaves behind.

The missing metric is agency gained.

This is something the Academy could potentially measure using its Financial Activation Measure (FAM), alongside the 5Cs: Clarity, Control, Confidence, Care and Course.

7. There are genuine risks we shouldn't dismiss

I would not claim that sovereign AI is automatically superior to institutional AI.

Both can produce errors. Both can reinforce assumptions. Both can create overconfidence.

And an independent AI system may lack access to authoritative information, reliable models or qualified specialist oversight.

The FCA's October 2026 good-practice publication recognises the importance of safeguards, escalation to humans, testing across different consumer circumstances and monitoring actual outcomes.

FCA


The Academy should embrace those disciplines.

There is also an important regulatory distinction: general-purpose AI tools do not provide the protections associated with regulated financial advice. The FCA has explicitly warned about that limitation.

FCA


Therefore, the Academy's argument should be that a well-designed system with appropriate boundaries can improve financial capability—not that AI eliminates the need for regulated specialists.

That position is both more credible and more defensible.

8. My suggestion: challenge 7IM to a different experiment

Rather than simply criticising the research, I would propose a constructive test.

Proposed Academy experiment

The Financial Agency Challenge

Take the same realistic consumer cases and compare three operating models.

1

Conventional human-led financial advice

2

General-purpose AI working alone

3

AI-enabled independent financial planning with episodic expertise

Evaluate each model against:

Factual accuracy

Safety and suitability

Assumption testing

Cost and accessibility

Consumer understanding

Agency gained

I suspect the hybrid agency-led model would be competitive, but that remains a hypothesis to test—not a result to assume.

The experiment would also allow us to examine the degree to which product incentives influence recommendations.

That could produce genuinely useful research for the industry, rather than another AI-versus-human comparison.

My overall assessment

7IM's proposition

My assessment

AI explains financial concepts well


Agree


AI can miss important legal and tax changes


Agree


AI sometimes fails to question assumptions


Agree


One-shot AI responses are inadequate for complex planning


Agree


AI cannot reliably exercise judgement


Too broad


Human advisers are therefore the best alternative


Not demonstrated


Ongoing financial support is valuable


Agree


That support must follow the traditional adviser model


Not established


Steve, I think there is an excellent Academy blog here, but I would avoid making it a defensive argument about AI.

The strongest response would accept 7IM's observations and challenge the assumption underlying its conclusion.

My suggested headline is:

7IM Asked Whether AI Can Replace a Financial Adviser. What If That's the Wrong Question?

And the central argument:

The future of financial planning is not a competition between artificial intelligence and human advisers. It is a question of how we combine technology, human expertise and personal agency to produce better outcomes.The real test is not whether AI can replicate today's financial adviser. It is whether people can achieve better financial outcomes, greater independence and more control over their lives through a different model of planning.

The final question I would leave readers with is:

Are we designing AI to make financial advisers more productive, or to make consumers more capable?

That, to me, is the debate 7IM has unintentionally opened.


 
 
 

Recent Posts

See All

Comments


bottom of page