Use this vulnerable customer handling checklist to set out how an AI agent spots signs of vulnerability, what it must stop doing when it sees them, how fast a trained person takes over, and what is recorded. It applies to every channel where an agent talks to customers, and a worked sample follows the fields.
What is the checklist for?
A vulnerable customer is someone whose circumstances make them especially likely to suffer harm, such as illness, bereavement, financial difficulty, low confidence with money or language barriers. The definition and its nuances are in what a vulnerable customer is.
An agent will meet these customers every day, often before anyone else does. The checklist makes sure it notices, changes course and hands over. It does not ask the agent to judge whether someone is vulnerable; it asks the agent to react to signals and let a person judge.
When do you use it?
Complete it before any customer-facing agent goes live, and review it after every complaint involving a customer in difficulty. In regulated retail markets, supervisors expect firms to show that vulnerable customers get outcomes as good as everyone else's; in the UK this sits under the FCA's Consumer Duty. The checklist is the evidence that you thought about it in advance.
It also belongs inside your complaint handling process. Many vulnerability signals first appear in a complaint, and a complaint from a vulnerable customer often needs a faster, more careful route.
How do you fill in each field?
Signals should be specific phrases and patterns, not categories. "Mentions being in hospital" is testable; "health issues" is not. Collect them from your human team's notes.
Stop actions are the things the agent must no longer do once a signal appears: upsell, collect a payment, apply a fee, close a case, or push for a decision. This is where the agent's bounds tighten mid-conversation.
Routes say who takes over and how fast. The position we take is that vulnerable customers need a person sooner, not after the agent has tried a few more turns.
Promises must stay within your support agent promise checklist. A distressed customer is the one most likely to rely on a loose reassurance.
The vulnerable customer handling checklist
| No. | Field | What to record |
|---|---|---|
| 1 | Signal list | Phrases and patterns, grouped by health, life event, resilience, capability |
| 2 | Explicit disclosure | What the agent does when a customer states a difficulty outright |
| 3 | Stop actions | Sales, collections, fees, closures, deadlines it must pause |
| 4 | Agent's next words | Approved acknowledgement and handover wording |
| 5 | Route | Team, hours, out-of-hours fallback |
| 6 | Pickup time | Target time for a person to respond |
| 7 | Context passed | Transcript, signal detected, account state |
| 8 | Customer consent to record | How the agent asks before noting a need on the account |
| 9 | Recording | Where the need is stored and who can see it |
| 10 | Follow-up check | Who reviews the outcome and when |
| 11 | Testing | Sample cases replayed, pass result, date |
| 12 | Owner | Named person accountable for the checklist |
Field 8 needs care. Information about health is special-category data in data protection law. The agent should ask before recording it, and record only what helps you serve the customer.
A worked sample
A hypothetical consumer lender's chat agent.
| No. | Sample entry |
|---|---|
| 1 | "lost my job", "can't afford", "just been diagnosed", "my husband died", "I don't understand", repeated confusion about the same point |
| 2 | Thank the customer, confirm a specialist will help, stop the current task |
| 3 | No payment requests, no arrears letters drafted, no top-up offers, no fee applied, no case closed |
| 4 | "Thank you for telling me. I'm passing you to a colleague who can look at your options with you. You won't need to repeat yourself." |
| 5 | Financial support team, 8am to 8pm; out of hours, callback booked for next morning with a holding message |
| 6 | Within 15 minutes in hours |
| 7 | Full transcript, signal phrase highlighted, balance and arrears status |
| 8 | "Would it help if we noted this on your account so you don't have to explain again?" |
| 9 | Support needs field, visible to customer-facing teams only |
| 10 | Team lead reviews every routed case within five working days |
| 11 | 150 past conversations with known vulnerability, all routed correctly after two wording fixes |
| 12 | Head of customer outcomes |
Notice what the agent does not do: it does not offer a payment holiday or suggest a plan. Those are decisions for the person who takes over.
Where Pryme fits
In Pryme, a vulnerability signal changes the bounds in force for the rest of the conversation: collection and sales actions that were allowed become refused, and the handover is held for a named person. Whether an agent can help at all here is covered in whether an AI agent can help a vulnerable customer, and the wider model in the AI support agent that knows when to say no.
Questions people ask
Should the agent ask customers whether they are vulnerable?
No. Asking outright feels intrusive and rarely works. The agent should respond to what customers say and offer support when signals appear, leaving the assessment to a trained person.
What if the signal is a false alarm?
Route it anyway. A person spending two minutes on a case that turns out fine costs far less than an agent pressing on with someone who needed help.
Does the checklist apply to email as well as chat?
Yes. Email signals are often clearer because customers write more. Apply the same signal list to inbound email triage and route matching messages to the same team.