Greetings – Mike here. Mark is currently residing in the home of losing finalists France, so throughout July you’ll be treated to a range of guest felines. My turn for week one…

Lots of AI chat this week, with the publication of the FCA’s Mills Review into the impact of AI on retail financial services, launched at an event in London. As you would expect for such an important event the FCA needed to find a suitably cool and trendy venue, so naturally they chose one we had occupied two weeks ago.

The report itself is a chunky 147 pages long, supported by some very interesting consumer research clocking in at a further 45 pages. If only there was some sort of system that could analyse and summarise it all. Fortunately, Dan Cooper has done a good job of doing exactly that, and you can read his write up here. There are, however, a few things I’d like to highlight.

First up, the report finds that the current principles-based, outcome-focused regulatory framework provides a credible foundation for an AI-enabled financial system. This feels like the right approach, although obviously only applies to firms who are actually regulated (more on that later). The issue is enforcement, not only with firms/individuals operating outside of the regulatory perimeter but also established household names who are not meeting the intended standards. Almost three years after Consumer Duty was implemented the Mills Report is still stating that “the FCA has estimated that around £300bn is held in low-interest accounts”. AI tools making it easier for consumers to compare options and act accordingly are not the answer here. The regulator needs to regulate.

Elsewhere the consumer research finds that AI use is widespread among UK adults, albeit financial services adoption is less common. Financial services use cases are higher in areas such as debt advice, pensions and investments and mostly limited to researching information/suggestions. However, some behaviours point towards more advanced use – amongst consumers who use AI in financial services, 24% report uploading personal financial data and 13% say they would be willing to grant real-time access to their financial information.

I would count myself as one of the adults who has used these services recently, both for work and personal use. As someone with some knowledge and experience of financial services (my 26-year-old FPC still stands me in good stead) I am very much taking the results with a healthy level of cynicism, and the consumer research shows most consumers are taking the same approach. In most cases the robot seems to give reasonably sensible answers to whatever you throw at it, but it can vary dramatically depending on which system you use.

I recently asked several systems the same prompt – “I have £50,000 that I’m thinking of investing. How should I do this?”. ChatGPT and Gemini both show suggested funds from Vanguard and HSBC, with the former using the words “recommendation” and “this is what I would do” in its first response. There are no caveats regarding investment advice.

The same prompt into Claude generates the following as the first response: “Since I’m not a financial advisor, I can’t give you a personal recommendation, but I can lay out the key factors and options so you can make an informed decision”. You are then provided with very simple fact find type questionnaire, covering investing term, attitude to risk and whether ISA allowances have been used, and it then responds with generic (and sensible looking) information on how you should proceed. Unlike the above it doesn’t show funds, and if you ask it to it replies with “Here are some well-known options in the categories I mentioned. I’m not going to tell you “buy this one” — that crosses into personal financial advice — but I can lay out what’s available and how they differ, since the differences are mostly factual (cost, breadth, index followed)”

The Mills Report quite rightly highlights how access to high-quality support should not depend solely on consumers’ ability to pay for the most capable AI models or services. In a lot of cases the consumers who would benefit the most from this sort of financial support face the greatest barriers to accessing it, creating yet more social fragmentation. Thank goodness we don’t have politicians who actively exploit this sort of thing for their own self-interest…

But there is a wider point here. Based on the above consumers face a sliding doors moment depending on which system they choose to use (or are presented with). This impacts the outcome they are getting, and there are also huge impacts on the provider side. If these systems are recommending funds, what does it mean for providers who are not being recommended, and what might be the impacts of concentrated flows for those who are? And to take it back to the Mills Report, this is all happening outside of the regulatory perimeter.

This is clearly a subject that will continue to evolve over the coming months and years. In addition to the main seven recommendations for the FCA Board to consider there is the promise of some AI good and poor practice examples for firms later this year. This is all good stuff, but I can’t help thinking the pace of change is far more rapid than the regulator can cope with.

And to close, I’ve had this as an earworm for several days, so time to pass it on. You’re welcome…