The FCA Consumer Duty requires firms to deliver good outcomes for retail customers, and it makes no exception for work done by AI agents. Staff who supervise agents must be trained to spot foreseeable harm in what the agent drafts or holds, to recognise vulnerability, and to remove the friction an agent can create, then evidence the outcomes.
What does the Consumer Duty ask of people who supervise agents?
The Duty sets a higher standard of conduct for firms serving retail customers in the UK. In substance, firms must act in good faith, avoid causing foreseeable harm, and enable and support customers to pursue their financial objectives. It then sets outcomes on products and services, price and value, consumer understanding and consumer support, and requires firms to monitor and evidence those outcomes, including for customers with characteristics of vulnerability. Our explainer on what the FCA Consumer Duty is covers the structure in full.
The question of whether the Consumer Duty applies to AI agents has a simple answer: the Duty applies to the firm and its outcomes, however the work is done. What changes with agents is where the judgement sits. A supervisor who releases a held collections letter, approves a complaint response or reviews an agent's batch of service replies is the person who decides whether the outcome is good.
Which firms and staff does it cover?
The Duty applies to FCA-regulated firms whose products and services reach retail customers, including firms in the distribution chain that do not deal with the customer directly. Inside those firms, the staff who supervise agents typically include:
- Customer service team leads who review agent replies and release held actions.
- Complaints handlers who approve agent-drafted final responses.
- Collections and arrears staff who release payment plans, forbearance offers and default notices.
- Lending operations who review agent-prepared affordability and eligibility decisions.
- Product and pricing owners who read agent-produced outcome and value data.
- Quality assurance and risk staff who sample agent work for outcomes.
Under the Senior Managers and Certification Regime, a named senior manager is usually accountable for the Duty. That person depends on the competence of everyone on this list.
What competence does the Duty imply when the system is an agent?
The Duty is about outcomes, so supervisor training should be about recognising bad ones before they leave the building. Five skills matter.
Seeing foreseeable harm in a draft. An agent-drafted letter may be accurate and still harmful: a demand that ignores a payment plan already agreed, or a response that technically answers a complaint while missing its point. Supervisors must read for the customer's position, not only for errors.
Recognising vulnerability. Agents can flag signals such as missed payments, bereavement or mentions of illness. Supervisors must know what the agent can and cannot detect, and act on cues that appear only in tone or context.
Removing sludge. The consumer support outcome expects firms not to put unreasonable barriers in customers' way. An agent bounded too tightly can refuse a cancellation, loop a customer through questions or fail to hand over to a person. Supervisors should treat repeated refusals as a signal to review the agent's bounds, not as a measure of success.
Testing understanding. Agents produce clear-looking text quickly. Supervisors need to judge whether a particular customer will understand it, especially in letters about arrears, charges or product changes.
Reading outcome data. Every verdict is filed, so supervisors can see how many actions were held, refused or allowed for different customer groups. They must be able to read that data and notice when one group fares worse.
Depth gives these skills a frame. An agent answering balance questions at depth 1 needs light oversight. One drafting arrears letters at depth 2 needs a reviewer who reads every letter. One setting up payment plans at depth 3 within the customer's limits needs supervisors who sample its decisions for fairness. Sector detail is in Consumer Duty and AI agents in lending and in customer service.
How do you meet the Duty and evidence it?
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Avoid foreseeable harm | Agent actions with customer detriment risk are held for trained supervisors | Hold and release log with supervisor and training status |
| Support vulnerable customers | Vulnerability module for all supervisors; agent flags routed to trained staff | Training records; flagged-case outcomes |
| Consumer understanding | Supervisors test agent communications against the intended audience | Communication test results; changes made |
| Consumer support, no sludge | Review of refusal and handoff rates by journey | Refusal and handoff reports; actions taken |
| Monitor outcomes by group | Supervisors trained to read verdict data by customer segment | Outcome MI with commentary and owner |
| Competent supervisors | Supervisors pass an assessment on the agents they oversee before release rights | Assessment results; access grants |
| Board oversight | Outcome reporting includes agent-supported journeys | Board report sections and minutes |
What will the FCA or an auditor ask?
- Which customer journeys involve agents, and where does a person decide?
- How do supervisors know an agent-drafted letter is fair to this customer?
- Show us outcomes for vulnerable customers in agent-supported journeys.
- What do you do when the agent refuses a customer request repeatedly?
- Who approved this final response, and what training had they completed?
- How does the board see agent-related outcomes?
What does the Duty not cover?
It does not apply to wholesale or non-retail business in the same way. It does not set a training curriculum or require certificates. It does not replace data protection duties under UK GDPR, including those on solely automated decisions, which the ICO supervises. And it does not replace specific product rules; the Duty sits on top of them. Support teams can read the narrower case in Consumer Duty and AI support agents. The guide to AI literacy and certification for people who work with agents sets out the general model.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Where the Academy fits
Pryme Intelligence Academy's Operator track is built for the supervisors this article describes: team leads, reviewers and operations managers who run a live workspace. Its modules include human review, reading the workspace and when something goes wrong, with tests set in financial services and the learner's own role. Certified Pryme Operator needs 32 of 40 exam questions right and a practicum signed off by a Pryme assessor, giving firms checkable evidence of competence.
Questions people ask
Does a supervisor need to read every agent reply?
Not always. Replies at depth 1 can be sampled. Anything with possible detriment, such as arrears letters or complaint responses, should be held and read before it goes.
How do we show good outcomes when an agent did the work?
Use the record of verdicts. Report outcomes for agent-supported journeys by customer group, with commentary from a trained supervisor on what changed.
Is a high refusal rate a good sign?
Not on its own. Refusals protect against harm, but repeated refusals of reasonable requests can be sludge. Supervisors should review the bounds behind them.