AI can make everyday money management more accessible, timely and personal. At Monzo, we are exploring how AI could reduce the effort involved in managing money by helping customers understand their options, make a plan and adjust it over time, while remaining in control.
In the products we’re building for customers, we combine AI’s ability to explain and personalise with clear safeguards and predictable, auditable systems for any actions that follow. We’re testing this in stages and using what we learn to improve both the customer experience and the systems around it.
Financial services is a high-trust environment, and generative AI behaves differently from traditional software. Its responses can vary, rely on unstated assumptions or sound more certain than the underlying information supports.
So the question is not only whether an AI assistant can give a helpful answer, but how to deliver an assistant that consistently delivers good customer outcomes. We joined the FCA’s AI Live Testing programme to collaborate on answering this question.
Why we worked with the FCA
Monzo was one of the first firms the FCA selected for AI Live Testing. The programme supports firms with mature AI systems that are ready for real-world testing, combining input from the FCA’s regulatory specialists with technical support from Advai.
The programme gave us a structured forum for challenge, bringing regulatory and technical perspectives together with expertise across product, engineering, design, data, risk, compliance and customer outcomes.
The value came from moving beyond broad ambitions and working through practical questions: What customer problem are we trying to solve? What would a useful and understandable experience look like? Where could it confuse customers or fall short? What evidence would tell us that? How should we respond and improve when its behaviour differs from our expectations?
What we tested
We’ve been testing an early savings experience designed to help customers set goals and make progress towards them. It can help someone turn a goal, such as a holiday or rainy-day fund, into a manageable amount to save each month. It can also understand how they’re progressing and revisit the plan when their circumstances change.
The scope was intentionally narrow. We wanted to learn where the experience is genuinely helpful, where it needs improvement and how to make its support clearer and more consistent. It’s there to help customers think through their plans and make their own decisions.
The FCA's AI Live Testing programme allowed us to examine our assumptions in controlled real-world conditions and discuss emerging risks early. It helped us strengthen our approach, contribute to developing good practice and share practical experience that can inform future policy on AI in financial services.
What we learned through working with the FCA
Our discussions with the FCA centred on five practical areas:
Start with the customer problem. A clear intended benefit and narrow scope make it easier to define what good looks like, where the experience could fall short and what evidence matters.
Evidence and testing. We explored how existing obligations translate into evidence of accuracy, relevance, appropriate handling of uncertainty and customer understanding.
Keep customers in control. We considered risks beyond factual errors, including hidden assumptions, overconfidence and over-reliance. The experience should be clear about what it can and cannot do, avoid creating misplaced confidence and support customers to make their own decisions.
Whole-system assurance. We worked through who owns the service, how its performance is monitored, how changes are reviewed and tested before release, and when issues should be escalated. AI assurance needs to continue as models, data and customer behaviour evolve.
Our experience within the FCA AI Live Testing reinforced the value of an outcomes-based, principles-led approach. Detailed rules tied to a specific model can date quickly, while high-level principles alone may leave uncertainty about what proportionate assurance looks like. The most useful approach combines clear outcomes with practical evidence and controls.
What happens next?
We’ll continue both sharing practical lessons with the FCA as Monzo delivers products that work for customers, and contributing to the wider discussion on responsible AI in financial services.
We’re going to keep testing, learning and building better services for our customers. This is only the start of the journey.