From Human-Led to AI-Enabled:
What the Mills Review Means for AI in Retail Financial Services
The Financial Conduct Authority has just published “The Mills Review: AI and the Future of Retail Financial Services”, the report was authored by senior FCA official Sheldon Mills at the request of the FCA Board and informed by 140 written submissions.
Its conclusion leaves little room for ambiguity. By 2030, AI will have reshaped UK retail financial services, shifting the industry away from human-led decision-making and towards financial activity that is AI-enabled, continuous and increasingly delegated.
Mills puts it plainly in his foreword: the question is no longer whether to allow AI, but “who is AI going to serve?” His answer is citizens, and he treats empowering them as one of the UK’s great opportunities for economic growth.
The report forecasts four shifts
1. Firms will be transformed from the inside out. AI is expected to become a core operating capability rather than a bolt-on, with customer service, underwriting, claims handling, compliance, risk management, product design and software development all in scope. The human role changes accordingly, moving from performing tasks to supervising the systems that perform them. For leading firms, the Review suggests AI may become the main way they process information, serve customers and evidence outcomes.
2. Consumer journeys will change shape. Most people currently use AI to look things up, but the Review anticipates a move towards trusted AI agents that act continuously on a consumer’s behalf within agreed limits, managing savings, comparing products, switching providers, optimising investments and supporting debt management. The prize is substantial, since the advice gap (only 9% of consumers use traditional advice), stubbornly low switching rates, financial exclusion (around 900,000 people are unbanked), under-saving (some £300bn sits in low-interest accounts) and protection gaps (just 30% hold life or income protection) are all problems that better tools might genuinely dent. The Review is candid that the reverse is also possible: unequal access to high-quality tools could widen inclusion gaps rather than close them.
3. Competition will move to the interface. In one of the Review’s sharper findings, market power may increasingly sit with whoever controls the AI layer between a consumer and a financial product. That gatekeeper could be a bank, an operating system such as Apple or Google, an AI provider, an aggregator, or an independent consumer agent. Firms may find themselves competing through recommendations, rankings, visibility and agent integrations rather than through the shop window as we understand it now. Much will also depend on upstream technology markets, where concentration, cost and vendor lock-in could shape smaller firms’ access to frontier capability on fair terms.
4. Fraud becomes faster, cheaper and more persuasive. AI makes fraud more scalable as well, and the Review flags deepfakes, cloned voices, synthetic identities and personalised social engineering as the threats to expect. It also warns that attacks will be harder to spot and stop. The uncomfortable symmetry is that the same technologies will be needed to defend against them. Firms and regulators, the Review argues, will need access to many of the same AI capabilities as the attackers.
The spectrum of autonomy
To describe how delegation from human to AI tools happens in practice, the Review sets out a five-level framework for the human role:
● L1 – human as Operator: AI is a tool that helps complete a defined task
● L2 – human as Collaborator: human and AI plan and act together
● L3 – human as Consultant: AI compares options and recommends, the human decides
● L4 – human as Approver: AI prepares actions, human authorises
● L5 – human as Observer: AI acts continuously within agreed limits, human monitors outcomes
The report stops well short of predicting full automation, but it does expect many activities to drift towards the Approver and Observer stages over time, which is where the interesting regulatory questions begin. Its own framing is that early concerns about accuracy and over-reliance give way to harder questions of consent, accountability and redress as systems move along the spectrum.
What the regulator thinks it needs
The Review finds the overall regulatory framework remains sound. Its principles and outcomes-based approach, including the Consumer Duty, the Senior Managers Regime and operational resilience requirements, was designed to flex across changing business models, and respondents did not ask for it to be rewritten. What they wanted was clarity on how to interpret and govern AI within the existing regime. Greater autonomy nevertheless raises new challenges around accountability, consumer understanding, liability, systemic risk and where the regulatory perimeter should sit.
The Review makes seven priority recommendations to the FCA, with decisions on taking them forward resting with the FCA Board and Executive. They are to: secure and adapt the regulatory perimeter; strengthen system-wide coordination and oversight; monitor the transition to autonomous models and adapt regulatory frameworks; scale up the FCA’s AI Lab to support AI model and system innovation; enable the foundations for agentic finance; build and adopt an AI-enabled agentic supervisory model; and develop a trusted public-interest AI-enabled financial capability service.
Taken together, the recommendations describe a regulator preparing to work at the same speed as the firms it oversees. The perimeter, the supervisory model and the AI Lab all point in the same direction: the FCA is being asked to understand agentic finance from the inside, and to have the tools to intervene before harm accumulates rather than after it has been documented. The Review calls the resulting approach an Agentic Supervisory Model, one designed to support and enhance human supervision rather than replace it. It is notably warm about the FCA’s existing work, describing the AI Lab and Scale-up Unit as groundbreaking and urging the regulator to build on them.
A view from inside the FCA AI Sandbox
Some of this is already visible up close. Iga Sloan, Founder and Chief Consultant at Digital Regs is a mentor for the FCA AI Supercharged Sandbox, which is part of the AI Lab the Mills Review recommends scaling up. Digital Regs has been observing these changes first-hand as they happen, watching firms innovate and opening new avenue underserved consumers. Sloan sees huge potential in AI to make finance more inclusive and accessible, reaching the people who have never been able to afford advice, who have never switched a product because the process was too opaque, and who have been quietly priced out of good financial outcomes.
It is easy to read a report like this as a warning about risk. Sloan’s view from inside the Sandbox is a reminder that it is equally a report about who financial services currently fails, and who it could reach instead. The Review says much the same thing in its seventh recommendation, which asks the FCA to convene a free, inclusively designed AI service so that reliable financial guidance does not become another thing only the well-served can afford.
The bottom line
The Mills Review treats AI as a once-in-a-generation opportunity to improve financial outcomes, competition and productivity, while making clear that the opportunity is conditional on trusted governance, clear accountability and strong consumer protections. Realising it, the Review argues, would also strengthen the UK’s position as a leading, trusted global financial centre.
The direction of travel is towards a more agentic retail finance sector, where consumers, firms and regulators all delegate more decisions and actions to AI while retaining meaningful human oversight. Getting that oversight right is the substance of the work ahead.