Blog
Ruby Wang
Financial services are moving ahead of the regulators on AI, and fintechs are moving ahead of the incumbents. Cambridge CCAF's 2026 research has 82% of fintechs using AI for customer support, against 67% of traditional institutions. That agility is a real competitive edge, and it comes with a responsibility, because whether a fintech is moving money, lending, insuring, investing or giving financial guidance, it is handling sensitive personal and financial data, and every customer-facing answer counts. The firms winning the AI race will be the ones that can move quickly and still stand behind every decision they put in front of a customer.
The same conversation is playing out in leadership meetings across UK fintech, again and again. UK fintech has done a lot to show what good looks like for customers, and the best teams treat this conversation as a strength.
CEO's want to move faster, with more customers, quicker decisions and more of the operation supported by AI. The CTO has the tools to do it, whether that's models, cloud platforms or a team ready to build. And the Chief Risk & Compliance Officer asks the question that holds all of this together, i.e. can we explain this decision, and could we defend it to regulators such as the FCA if ever asked? I'd argue that tension is one of the healthiest things a fast-growing firm can have, and the leadership teams that treat it that way tend to grow responsibly and at pace.
The reality is that risk teams are not against AI at all. Instead, they just need to sign off on decisions they can explain and audit.
Why customer-facing AI is a different test
Inside a firm, a wrong AI answer is an internal mistake that somebody catches. In front of a customer, it is the firm speaking, held to the same standard as everything else the firm says, which is fair, clear and not misleading. Consumer Duty raises that bar by requiring good outcomes for retail customers, including those in vulnerable circumstances, and responsibility stays with senior managers for every area where AI is used.
Vulnerability is the hardest part of this for any firm, fintech or incumbent. The FCA's own work found that only about 4 in 10 consumers with characteristics of vulnerability have told any of their financial providers. The signals are often buried in everyday language, which is why this is an industry-wide challenge and not one any single firm has solved.
The FCA has no AI-specific rulebook, and I don't think its question is whether a firm used AI. Under Consumer Duty and the Senior Managers regime, it is whether the firm can show how a customer outcome was reached (WilmerHale).
What finance leaders told us
UnlikelyAI commissioned Censuswide to survey 200 senior UK finance leaders for its AI Trust Report 2026, and the numbers explain why risk and compliance teams are careful.
Only 26.5% would trust AI with customer communications under minimal human oversight, which means three in four would not. 70% say compliance sign-off is a significant factor slowing AI adoption, against 60% in other regulated sectors. And before approving an AI solution, 32% say their compliance team requires accuracy thresholds and 27% require explainability evidence, yet only 18.5% require an audit trail.
That last number is the one I keep coming back to. Firms are testing whether AI is right, but few are keeping a record of how each answer was reached, and accuracy alone will not secure sign-off once AI is speaking to customers. An answer can be accurate and still be one a firm can't evidence.
The upside for getting this right is clear too. Asked what would change if an AI solution could guarantee accurate, consistent and explainable outputs for high-stakes tasks, 32% said they would increase AI budgets, 31.5% would trust AI for customer-facing work, and 29.5% would find compliance approval easier to gain.
What would get a risk officer to say yes?
I'd say it comes down to three things:
The same input gives the same answer, every time. A risk officer can't sign off on something that behaves differently day to day.
Each decision is backed by clear reasons. It shows which rule was applied, the policy behind it, and what the customer actually said.
The language of policy and regulation is theirs. The firm's own policies and regulators' guidance, applied as written and not as a model's best guess at them.
Supporting a risk officer with a capability that provides this changes the conversation drastically. Risk becomes the team that makes it safe to grow, which is the role most risk leaders already want to play.
Moving fast and staying safe
There are absolutely ways to move fast and stay safe, and many fintechs are already doing it. The challenge is that traditional generative AI works by producing a likely answer each time, so the same question can get two different answers, and asking a model to explain itself gives you another generated answer and not a record of how the decision was made. Settings and guardrails reduce the variation but don't remove it the underlying risk.
At UnlikelyAI we took a different approach, building AI with reliable decisioning that is consistent and auditable, so firms can use it where the cost of getting it wrong is highest. We're working with financial partners on these exact compliance and operational challenges, and fintechs are well placed to lead here because they meet the question earlier than anyone else.
Five questions to test any customer-facing AI
If you ask it the same question 100 times, is the answer the same every time, and can you prove it?
For any single answer, can you show how it was reached, as a record and not a generated explanation?
Who signed off this use case, on what evidence, and would they stand behind it in front of a regulator?
When a customer shows signs of vulnerability, does it spot them every time and hand over to a human, and how do you know?
If the model changes tomorrow, how would you find out before your customers do?
Any question you can't answer with evidence is where to start. If you're responsible for Risk & Compliance, Customer Servicing and Operations or CTO at a fintech, I'd like to hear how you're handling this.
Survey data: AI Trust Report 2026 (Censuswide), commissioned by UnlikelyAI. 1,000 UK senior decision-makers with budget or decision authority for AI at organisations with 250+ employees, surveyed 26 November to 8 December 2025. The finance sample (n=200) covers UK finance as a whole, not fintech only. Fintech vs incumbent AI use: Cambridge CCAF 2026 Global AI in Financial Services Report.
