FCA Launches Live AI Testing in UK Finance: What It Means for Banks and Fintechs
The UK Financial Conduct Authority has launched an AI Live Testing initiative to trial AI systems in real market conditions. Announced December 3, the program pairs major banks and fintechs with FCA specialists and technical partner Advai to test, learn, and set guardrails before AI scales across the sector.
The FCA says this is the first program of its kind in UK finance. Participants include NatWest, Santander, Monzo, and subsidiaries of Lloyds Banking Group and Avantia Group.
What the first cohort will test
- How to build practical evaluation frameworks that go beyond accuracy to include calibration, drift, bias, and stability under stress.
- How to govern live monitoring: thresholds, alerting, human oversight, and rollback plans when models misbehave.
- How to manage risk as machine learning becomes embedded in core banking processes.
Early projects lean toward retail services: AI-supported debt resolution, financial guidance, and smarter customer engagement. Expect a focus on explainability, fair outcomes, and treatment of vulnerable customers.
Why this matters for finance leaders
AI is moving from pilots to production. The risk is no longer theoretical-it's operational, legal, and reputational. This initiative gives firms a clear path to deploy with control, and gives the regulator visibility into what "good" looks like at scale.
Practical steps to prepare (even if you're not in the cohort)
- Map AI use cases by risk: credit, fraud, collections, advice/guidance, servicing, marketing. Assign owners and risk tiers.
- Define pre-deployment tests: performance by segment, fairness checks, stability under data shifts, calibration, adversarial probes.
- Stand up live monitoring: drift, bias, false positives/negatives, latency, customer complaints signals. Set hard thresholds and auto-fallbacks.
- Tighten model governance: inventory, approvals, documentation, change control, audit trails, and clear sign-off from risk and compliance.
- Strengthen data controls: lineage, PII handling, retention, access, and synthetic data policies for safe experimentation.
- Build human-in-the-loop for decisions that affect customers' money, credit, or collections. Require second-line review for edge cases.
- Vendor management: performance SLAs, access to logs, retraining rights, model update notices, and incident reporting clauses.
- Run incident drills: model rollback, customer remediation, and communications playbooks.
Key risks to watch
- Drift and instability under real market conditions.
- Bias across protected characteristics or vulnerable segments.
- Over-reliance on opaque models without clear recourse for customers.
- Hallucinations or incorrect advice in generative systems used for guidance or support.
- Weak vendor controls and gaps in logging or reproducibility.
- Operational resilience: outages, cascading failures, or silent degradation.
What "good" looks like
- Clear link from business outcome to model metrics customers would consider fair and safe.
- Simple dashboards for live risk indicators and decision thresholds.
- Explainable outputs customers can contest, with fast human review.
- Routine red-teaming and challenge from risk and audit, not just data science.
- Evidence that controls hold under stress: backtests, shadow mode, and phased rollouts.
Retail use cases: special attention areas
- Debt resolution: affordability signals, treatment strategies, and impact on vulnerable customers.
- Financial guidance: clarity on scope, disclaimers, and clear handoffs to human advisors.
- Customer engagement: personalization limits, opt-outs, and consistent outcomes across segments.
How to engage
Firms outside the cohort can still align with emerging practices and track outputs from the regulator's innovation work. See the FCA's innovation hub for context and updates: FCA Innovation. The technical partner, Advai, focuses on AI assurance and adversarial testing, which can inform your evaluation plans.
Bottom line
AI is moving into live banking. The firms that win will pair speed with control-clear metrics, live monitoring, and human oversight baked in from day one. Start small, measure what matters, and keep customers' outcomes at the center.
Related resources
Your membership also unlocks:
