AI and Your Money: Strong Results, but Limits Remain

Research by investment firm 7IM tested five AI systems across five UK financial planning scenarios, from comparing ISAs to complex estate planning. ChatGPT ranked highest overall, scoring 4.30 out of 5, followed by Claude and Microsoft Copilot.

All systems showed strengths and weaknesses. ChatGPT received the highest factual-accuracy rating, while Claude performed particularly well on regulatory responsibility. However, the research identified limitations in handling personal circumstances, judgement and complex trade-offs.

The article concludes that AI can help people understand financial topics and explore their options, while qualified professional advice remains valuable for significant or complicated decisions.

See article.

---

The findings are interesting, but the conclusion goes further than the evidence presented. The article asks whether AI can replace an adviser. Your model raises a different question: can AI help people become more capable, with expertise brought in when needed?

There are five points worth separating.

1. A perfect score is not proof of flawless accuracy. The reported 5/5 for ChatGPT means it received the highest factual-accuracy rating within this test. It does not establish reliability across financial decisions. Likewise, better scores on the harder scenarios do not prove that complexity improves its performance generally. Investing Insiders Money AI

2. The methodology matters more than the leaderboard. I couldn’t locate 7IM’s original report. The supplied article does not identify model versions, exact prompts, repeated trials, assessment dates, assessor agreement or a human-adviser comparison. Without those details, the rankings are provisional.

3. Incomplete context is a process problem as well as a technology problem. A brief chatbot exchange is different from an ongoing process that maintains records, asks follow-up questions, checks sources and records uncertainty. A Customer Operating System should be evaluated as a whole workflow. That does not guarantee safety; it changes what needs testing.

4. The alternative to AI is not automatically good human advice. A fair comparison would assess both approaches for omissions, suitability, conflicts, cost and outcomes. Human expertise can add substantial value, but its superiority needs demonstrating for the task concerned.

5. Understanding options and making decisions are connected. Explaining trade-offs, identifying missing information and testing assumptions can materially improve someone’s decision. The person retains judgement; support helps them exercise it.

There is also a commercial frame to recognise: the article ends with an adviser-matching invitation. That does not invalidate the research, but it makes scrutiny of the move from findings to referral particularly relevant.

The strongest Academy response is to propose a better test: compare a person working alone, a person using AI, a person using a structured Customer Operating System with access to expertise, and a conventional advised process.

Measure accuracy and safety—but also cost, understanding, retained control and the person’s ability to handle their next decision.

“Does the support produce a good answer—and a more capable person?”