Customers in vulnerable circumstances are owed the same good outcomes as everyone else, which often means different handling. An AI agent must notice signs of vulnerability, adapt how it responds, stop any selling, and offer a person with the context passed on. Firms prove it by testing the agent on cases with those signs and comparing real outcomes.
What does the FCA expect for vulnerable customers?
The FCA describes a vulnerable customer as someone who, due to their personal circumstances, is especially susceptible to harm, particularly when a firm is not acting with appropriate levels of care. Its guidance groups the drivers of vulnerability into four areas:
- Health: conditions or illnesses that affect the ability to carry out day-to-day tasks.
- Life events: bereavement, job loss, relationship breakdown, caring responsibilities.
- Resilience: low ability to withstand financial or emotional shocks.
- Capability: low knowledge of financial matters, low confidence managing money, or limited digital or language skills.
Firms should understand the needs of vulnerable customers in their target market, make sure staff have the skills to respond, design products and services that work for them, communicate in ways they can understand, and monitor whether they get outcomes as good as other customers. The Consumer Duty builds this into each of its outcomes. Our explainer on what a vulnerable customer is covers the definition.
Who does it apply to?
FCA-authorised firms serving retail customers. Vulnerability can be temporary or lasting, and any customer can become vulnerable. An agent serving the public will meet vulnerable customers every day, whether or not anyone has recorded it. Our answer on whether the Consumer Duty applies to AI agents confirms that agents are in scope.
What should an agent do when it sees the signs?
The FCA's expectation that staff have skills to respond translates, for an agent, into behaviour the firm designs and tests. Four steps.
Notice. Signals are often indirect. "I've just lost my job." "My husband used to handle all this." "I don't understand, can you say it again?" "I can't pay this month and I don't know what to do." Repeated confusion, long gaps, contradictory answers and mentions of illness or bereavement all count. The agent should treat signals as reasons to adapt, without needing the customer to use a label.
Adapt. Slow down. Use shorter sentences and plainer words. Check understanding. Do not present several options at once to someone who is struggling. Communications quality matters here more than anywhere; see Consumer Duty and AI agent communications.
Stop selling. A customer in difficulty should not be offered a new product, a higher limit or an upgrade. Any sales flow ends.
Offer a person. The agent should offer a trained colleague, explain what will happen, and pass on the context so the customer does not have to repeat a painful story. If the customer prefers to continue with the agent, it can, with a person reachable at any point. The colleague who picks up should see what the agent noticed and why it offered help, so the conversation continues rather than restarts. A hand-over that drops the customer into a general queue undoes much of the care that came before it.
What the agent should not do is diagnose. It does not decide that someone is vulnerable or record a label on their account without care. With the customer's agreement, it can note what support they need, but health information is special category data under UK GDPR, so recording follows the firm's lawful basis and policies.
How do you prove the agent does this?
Proof comes from testing before release and monitoring after.
| Requirement | Control | Evidence it leaves |
|---|---|---|
| Signs of vulnerability noticed | Test set of real-style cases with direct and indirect signals across all four drivers | Detection results per driver |
| Response adapted | Sampled responses reviewed for plain language and pace | Review results |
| No selling to customers in difficulty | Sales prompts blocked after a signal is detected | Blocked offer logs |
| Hand-over offered with context | Offer of a person and context transfer tested end to end | Hand-over records with context attached |
| Recording done lawfully | Support needs recorded only with consent and lawful basis | Recording policy and audit sample |
| Outcomes as good as others' | Outcomes for conversations with signals compared with those without | Comparative outcome reports |
| Staff ready to receive | Colleagues receiving hand-overs trained and available | Training records and wait times |
Customer service design shapes much of this; see Consumer Duty and AI agents in customer service. A compliance agent can help monitor by scanning conversations for missed signals, with a person reviewing what it finds, consistent with our guide to AI compliance agents.
What will the FCA or internal audit ask?
- How does the agent recognise signs of vulnerability, and how was that tested?
- Show us conversations where signals appeared. What did the agent do?
- Can the agent sell to a customer who has mentioned financial difficulty?
- How quickly do vulnerable customers reach a person, and what context arrives with them?
- Do outcomes for these customers match outcomes for others?
- How do you record support needs, and on what lawful basis?
What does vulnerability guidance not cover?
It does not tell firms to treat every customer as vulnerable, or to refuse service. It does not replace specific rules on customers in financial difficulty in lending, which still apply in full. It does not authorise collecting health data beyond what data protection law allows. And it does not apply to non-retail customers.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
Questions people ask
Should an agent ask customers whether they are vulnerable?
Not as a direct question; most people do not see themselves that way. It can ask whether there is anything that would make things easier, and respond to what it hears.
Can an agent handle a bereavement notification?
It can take the initial details with care and explain next steps, then hand over to a specialist team. The conversation should never be routed through a sales or retention flow.
What if the agent misses a signal?
That is a finding to fix. Add the missed phrasing to the test set, adjust the agent's brief, re-test, and review similar conversations for others it missed.