We’ve long believed that closing the advice gap means making trusted advice accessible at scale using technology. Last week, the FCA’s own review of AI in financial services set out the framework for the industry to actively embrace AI and move rapidly in this direction.
The Mills Review from the FCA is the first of its kind commissioned by a financial regulator anywhere in the world. Led by FCA Executive Director Sheldon Mills and drawing on over 140 submissions and a survey of 5,000+ UK consumers, it sets out how AI is expected to reshape retail financial services by 2030.
The FCA is not introducing a new AI rulebook. Instead, the Review leans on existing frameworks: Consumer Duty, the Senior Managers Regime, model risk management, as well as raising the bar on what “good” looks like under each as AI becomes embedded across financial services.
Here are four things advice firm leaders should discuss at your board.
1. Accountability doesn’t move to the machine
The Review is unambiguous: the Senior Managers Regime continues to apply as AI takes on more of the work. A senior manager remains personally accountable for outcomes even where the model’s behaviour sits partly with a third-party provider.
Boards need to be confident that the burden of proof can still be met. How do we continue to demonstrate, recommendation by recommendation, what went in, what came out and why?
Board question: If a regulator asked us tomorrow to justify one AI-assisted recommendation from six months ago, could we?
2. The bar for “explainable” AI has been raised
The Review is candid about the problems it sees with generative AI in financial contexts, particularly hallucination, inconsistent answers to similar questions and model drift over time. A key response is to look toward hybrid model architectures that pair machine learning with rules-based logic, often called neuro-symbolic AI, on the basis that they offer more consistency and are easier to evidence than a general-purpose model alone.
It also makes a point that’s easy to miss: assurance “cannot be a one-off exercise.” A model card or a certificate from launch day isn’t enough. The Review expects ongoing monitoring for drift. Boards need to ask not just “how was this tested?” but “how is it still being tested after it went live?”
Board question: Are we still testing our AI six months after launch, or did our due diligence stop at go-live?
3. Your clients are already using AI you don’t control
Only around 9% of UK adults currently receive regulated financial advice. At the same time, the Review’s own research found that roughly a quarter of consumers already trust general-purpose tools like ChatGPT, Claude or Gemini for financial advice, and only around 40% correctly understand there’s no formal route to redress if that advice goes wrong. One in five UK adults say they’re open to AI making financial decisions on their behalf.
The FCA’s response is to launch a review of the advice/guidance boundary within three to six months, looking at how consumers are using general-purpose LLMs and whether the regulatory perimeter needs adjusting. Whatever that review concludes, the underlying dynamic is already here: a meaningful number of people are getting “advice-like” answers from somewhere today, outside any regulated relationship.
The question for boards is how do you engage clients and prospective clients in this context, and how do you ensure your proposition is as frictionless and value adding as possible?
Board question: If a prospective client is already asking ChatGPT about their pension, what are we offering that’s better, and do they know it exists?
4. Autonomy is arriving in stages: know your own transformation journey
The Review sets out a five-point autonomy spectrum for how AI and humans work together: Operator, Collaborator, Consultant, Approver and Observer, running from AI as a straightforward tool through to AI acting continuously with a human monitoring exceptions. It also takes the view that full autonomy “seems unlikely to sit well with the current regulatory framework or UK societal appetites” any time soon although it does acknowledge and indeed welcome the idea that some advice or advice-like services could soon be fully automated.
Practically, this is an invitation to map your own AI use, current and planned, against that spectrum, which we believe is a good one, function by function. Most regulated advice work will sit at Operator, Collaborator or Consultant for some time yet: AI drafting, comparing and recommending, with an adviser directing and signing off.
Boards should consider using this framework so they can point to a clear, deliberate view of where each AI-assisted process sits. This is good governance, and it is the language supervisors will increasingly use.
Board question: Could we tell a supervisor today exactly where each of our AI-assisted processes sits on this spectrum, and why we chose that level?
Your firm’s transformation journey
None of this requires a new AI strategy overnight. But it does mean transforming your business over time to address these four challenges in parallel:
- Increasing automation while evidencing decisions so accountability can continue to be demonstrated, and is never assumed
- Monitoring AI on an ongoing basis so that explainability holds up after launch
- Engaging with and adding value to clients with minimal friction, recognising that generic tools are competing for their attention
- Mapping your AI activity against the autonomy spectrum so you can show, function by function, exactly where control sits
Closing the advice gap and unlocking trusted advice at scale
Addressing these challenges is precisely what Dynamic Planner’s Advice-grade AI has been built for: the best of both worlds, deterministic evidence for every recommendation, combined with generative AI evaluated through a structured, staged process from synthetic testing through to continuous production monitoring. It’s used by more than 200 firms, and we’re the first UK fintech certified to ISO 42001, the international AI management standard, as well as setting out our own AI charter. Tram, our white-label client app, already provides frictionless client engagement and onboarding for thousands of clients, with APIs available for firms who want to build their own experience on top of it.
The Mills Review confirms what we’ve believed from the start: closing the advice gap and making trusted advice accessible at scale isn’t a trade-off firms have to make between speed and safety, it’s exactly what evidenced, accountable AI now makes possible for firms and consumers alike.
If you’d like to talk through what that means for your firm’s own AI roadmap, we’d welcome the conversation.
