AI skills for UK payments careers: what the FCA examples teach

What the FCA published — and what it did not

Fact: the FCA published examples of good practice and areas for improvement on 19 October 2026. The material draws on supervisory work, industry and consumer engagement, AI Live Testing, the Supercharged Sandbox, an event for smaller firms and the AI Input Zone.

Scope: it is relevant to regulated firms that develop, buy or use AI in regulated activities, consumer journeys, operations or governance. The FCA says the examples are illustrative, non-exhaustive and likely to evolve. They do not create new requirements or prescribe one model.

Evidence limit: the publication does not state a numerical sample size for the firms or people behind the examples. Treat it as regulatory practice material, not a survey from which to estimate how common a behaviour is.

1. Start with the consumer need

The first skill is problem definition. Before proposing an AI assistant, scoring tool or automated journey, explain whose problem it solves, the target user, the decision it supports and the foreseeable harm if it is wrong.

In a payments role, that could mean separating a request for faster merchant support from a decision about onboarding, affordability, fraud or vulnerability. The risk, evidence and approval path may be different for each.

• State the user and the outcome in one sentence.
• List what the system may do, must not do and when it should abstain.
• Choose a measurable check: accuracy alone may not show whether the customer outcome is fair or understandable.

2. Explain the role and limits of AI

The FCA highlights clear explanations at the point of use. People should be able to understand that AI is involved, what it can and cannot do, when to use another route and how to challenge an outcome.

For a sales or customer-support professional, this means avoiding claims such as “the system is always right” or “AI approved you”. Use the firm’s authorised wording, distinguish assistance from a formal decision, and make the human support route clear.

Practise clear discovery and explanation →

3. Keep evidence that another person can follow

Useful records can include the input, output, model version, approved sources, applied filters, escalation decision and the person responsible. The purpose is not paperwork for its own sake: records support monitoring, complaints, investigation and correction.

For a candidate, the transferable skill is disciplined documentation. A clear CRM note, call summary or handover should separate what the customer said, what the system suggested, what you checked and what happened next.

• Fact: information directly observed or provided.
• Inference: a conclusion that still needs checking.
• Decision: the action taken, by whom and under which authority.

4. Design human oversight that can actually intervene

Human oversight is more than placing an approval button at the end. A person needs enough information, authority and time to question the output, stop the process and offer another route.

The FCA examples also warn against assuming a language model can reliably identify vulnerability from text alone. In a customer conversation, use approved indicators and escalation routes; do not diagnose a person or rely on a model’s confidence as proof.

5. Plan for failure and supplier dependence

AI systems can change, become unavailable or perform differently under stress. Payments teams should understand critical dependencies, fallback processes, monitoring triggers and how work continues if a model or supplier is unavailable.

This is a useful interview topic for sales, operations, product, compliance and customer-success roles. A strong answer links innovation to continuity: what would you monitor, when would you pause the tool and how would customers still receive support?

Prepare honest interview examples →

Explore selected UK payment roles →

Turn the guidance into career evidence

Choose one real task where you used automation or AI. Explain the need, the boundary you set, the source you checked, the way you reviewed outputs and the fallback when the tool was uncertain. Use only examples you can support.

If you have not used AI at work, create a fictional payments exercise instead: draft a merchant-support assistant for a failed payment, then specify what it may explain, which data it must not expose and when it hands over to a person. Label the exercise as practice, not employment experience.

Payment Sales is Awais Amir’s independent education and careers project. This lesson does not provide legal advice or promise a job, promotion or earnings.

Practise the distinction