Advisory Wizard
On 6 July 2026, the FCA published the Mills Review: the first regulator-led study of how AI will reshape retail financial services by 2030. It is the clearest signal yet of what good looks like when AI enters the advice process. We have read all 140 pages. Here is where we stand.
What the Review actually says
Beneath the headlines, three findings matter for every advice firm.
Accountability stays with you. Under the Senior Managers Regime, responsibility for client outcomes does not transfer to an AI provider, however sophisticated the technology. Firms must be able to demonstrate reasonable steps: what the AI contributed, what the adviser decided, and why.
Explainability is becoming the standard. The Review highlights systems that combine machine learning with rules-based, deterministic logic as the route to outputs a firm can actually evidence, showing how a result was reached and whether the relevant requirements were applied.
The human role defines the risk. The Review maps AI on a five-level autonomy spectrum, from AI as a tool the adviser operates, through to AI acting autonomously while a human merely observes. Its own analysis is direct: the Consumer Duty and the advice and guidance boundary operate effectively where the human decides, and come under pressure where the human is reduced to approving or observing.
Where Advisory Wizard stands on that spectrum
Deliberately, and permanently, on the side where the adviser decides.
Advisory Wizard's AI drafts, compares and challenges. It never advises a client, never executes a decision, and never acts autonomously. Every suitability report, every cashflow scenario and every strategy suggestion passes through a named adviser who reviews, amends and affirms it before it reaches a client. In the Review's language, we operate where the human is operator, collaborator and consultant: the zone the FCA's analysis says the existing regulatory framework handles well.
This is not a limitation we tolerate. It is the design.
Built to evidence, not just to assist
The Review's practical consequence for firms is an evidence question: when the regulator asks how an outcome was reached, what can you show?
Advisory Wizard is built so the answer is: everything.
- A complete audit trail. Every AI contribution is logged: what was suggested, when, by which version of the system, and what the adviser decided.
- Rules first, AI second. Deterministic suitability and risk logic sets the boundaries within which the AI works, the hybrid approach the Review identifies as the foundation of explainable systems.
- Human sign-off, structurally enforced. There is no path from AI output to client without an adviser's decision in between.
Your clients' data works for your clients. Full stop.
We do not train AI models on your client data. Not ours, not anyone's. Client information is minimised before any AI processing, sensitive fields are encrypted at rest, and our processing is governed by UK GDPR, with health-related data handled to the higher Article 9 standard. The Review found that trust and control decide whether consumers accept AI-supported services. We think trust starts with what you don't do with people's data.
Our commitments
- We maintain a published AI governance policy, structured on the principles of ISO/IEC 42001, the international standard for AI management systems.
- We will give every firm we work with the records they need to meet their Consumer Duty and SM&CR obligations, as standard, not as a premium tier.
- We will never move Advisory Wizard's AI into autonomous, client-facing or execution roles.
The Mills Review describes a future in which AI is everywhere in financial services, and in which the firms that thrive are those that can prove their technology serves the client. That is the product we are building. Built by an adviser, for advisers, after the Consumer Duty, in the Mills era, not retrofitted to it.
Read the Mills Review in full on the FCA website.
Advisory Wizard is technology for authorised financial advisers. It does not provide financial advice, and no output of the platform constitutes a personal recommendation.