Mills Review: How AI Will Reshape Financial Services by 2030

The FCA's Mills Review maps how AI will reshape retail financial services by 2030, five human roles, four market shifts, and what we recommend firms must do now.

The FCA asked Sheldon Mills the key question. How will AI change retail financial services by 2030?  

He came back with 147 pages, 140 written submissions, and one answer. It will change everything.  

Firms will run on it. Consumers will delegate to it. Regulators will supervise with it. 

The question is no longer whether to allow AI. The question is who AI will serve. 

Mills does not deal in abstractions. He deals with gaps. The gaps are large and they are familiar. 

• Only 9% of consumers use traditional financial advice. 

• Just 30% hold life or income protection. 

• Around 900,000 people are unbanked. 

• £300bn sits in low-interest accounts. It sits there and does nothing. 

But the demand side is already moving.  

The Review's consumer research found 1 in 5 UK adults are open to AI making financial decisions for them, especially where the choices feel hard: debt, pensions, investments.  

And 26% already trust general-purpose tools like ChatGPT, Claude or Gemini for financial advice.  

The framework: five human roles 

The Review is built on one idea. As AI takes on more, the human role evolves.  

Mills maps levels of autonomy across five core roles: operator, collaborator, consultant, approver, observer where humans advance from performing task themselves towards setting boundaries, approving decisions and overseeing outcomes.  

The risks change with each role. Initially the risks are around accuracy. Later about consent, accountability and redress. Accountability gets harder to trace when models change, and if firms lean heavily on third-party providers.  

This autonomy spectrum is where the hard questions live, and few firms have answered them yet. 

Four shifts by 2030 

Firms transform. AI goes into every function. Underwriting, compliance, claims, product design, customer support. For leading firms it becomes the main way they process information, serve customers and evidence outcomes. People move from making decisions to monitoring them. That demands new skills and an honest review of what human oversight means. 

Consumer journeys go agent-led. Tools stop recommending and start acting. Agents manage money continuously, within limits the consumer sets. Done well, this fixes low switching and the advice gap. Done badly, hyper-personalisation becomes bias, opaque pricing and manipulation dressed as service. 

Market power moves. Whoever owns the AI interface owns the customer. Operating-system assistants, aggregators, consumers' own agents - they will decide which products are visible and how choices rank. The customer relationship drifts away from the financial firms that hold the licence. Upstream, a small number of model providers and hyperscalers control compute, data and frontier capability. Lock-in and sovereignty become board matters. 

Fraud accelerates. Deepfakes, synthetic identities, personalised social engineering. Attacks get faster, cheaper, more persuasive and harder to spot. Defenders need the same tools as attackers, and they need them first. 

Nobody asked for new rules. What firms want is clarity on applying the old regime to the new machines. 

What Mills tells the regulator 

The Review finds the existing regulatory framework sound.  

The Consumer Duty, the Senior Managers Regime, operational resilience - they were built to flex, and respondents did not ask to change them. They asked how to comply with them as autonomy grows.  

The Senior Managers Regime still applies when the model acts alone. Working out what that means in practice is now the industry's problem as much as the FCA's. 

The new risk is systemic. Shared reliance on the same models, datasets and infrastructure means correlated behaviour, herding and common points of failure. A change in one widely used model could move through firms and markets fast.  

Mills's answer is his flagship recommendation: an Agentic Supervisory Model - AI-enabled supervision that can see cross-firm patterns no individual firm can see.  

It is one of seven recommendations covering the regulatory perimeter, system-wide coordination, the transition to autonomous models, a scaled-up AI Lab, the foundations for agentic finance, and a public-interest financial capability service. 

What firms should do now 

The Review sets direction for the regulator. It sets homework for firms. Three pieces of it cannot wait for 2030. The team at Delta Capita recommend three priority actions: 

• Get governance ahead of autonomy. Extend model risk management to cover agentic systems and third-party model dependencies. Firms should decide now what human oversight means at each point on the autonomy spectrum, and evidence it. 

• Fix the data first. Agent-led journeys and outcome evidencing both run on data. If firms want to scale their AI past the pilot, the data must be trusted, governed and accessible. The Review makes the data problem a regulatory problem. 

• Map dependencies. Firms need to know where their AI supply chain concentrates, where lock-in sits, and where the regulatory perimeter could move. Boards will need to know. 

At Delta Capita, we help financial institutions build the data foundations, governance frameworks and AI operating models the agentic era demands. We benchmark your current state, and map a practical route to scaling safely and sustainably by putting AI at the centre with people providing the right oversight on outcomes. 

To discuss what the Mills Review means for your firm, get in touch. 

This article was co-authored by Karan Kapoor and Joe Chimento.