Yes, in specific ways, and it should hand over sooner. An agent can notice signs of vulnerability, slow down, explain things plainly, apply support the business has agreed, and record needs with consent. It must not sell, pressure, or make adverse decisions about a vulnerable customer, and it should bring in a person earlier than it would for anyone else.
What can an agent do well for vulnerable customers?
More than people expect. It is available at 2 am, when worry keeps people awake. It does not rush, sigh or judge, and some customers find it easier to explain a difficult situation to an agent than to a person. It can repeat an explanation patiently as many times as needed, in plain words, from the business's own help content.
That last point matters. Answers must come from approved sources, not improvisation, which depends on how well the agent is connected to your help centre and how it answers product questions from help centre content.
What signs should the agent watch for?
Vulnerability usually comes from health, a life event such as bereavement or job loss, low financial resilience, or low confidence with money or technology. The signs in a conversation are often indirect: confusion about something explained moments ago, mention of a death or diagnosis, a third party prompting the customer, urgency to move money, or distress. A working definition is in what a vulnerable customer is.
Detection will never be perfect. Missing a vulnerable customer costs more than treating an ordinary one with extra care, so the agent should lean towards caution.
What may the agent do, and what stays with a person?
| Situation | The agent | A person |
|---|---|---|
| Customer seems confused | Slows down, simplifies, checks understanding | Takes over if confusion persists |
| Customer discloses a need | Records the support needed, with consent | Reviews the record |
| Agreed support already on file | Applies it, such as extra time or a preferred channel | None |
| Bereavement | Expresses condolence, stops marketing, explains next steps | Handles the account |
| Signs of a scam or coercion | Pauses any money movement, hands over | Contacts the customer safely |
| Risk to life or crisis | Gives crisis information at once, hands over with priority | Responds |
| Adverse decision, such as declining or closing | Never | Decides |
| Selling or retention offers | Never | Not in this conversation |
Many of these steps belong on a written checklist; a usable version is the vulnerable customer handling checklist.
What does this look like at 2 am?
Take a retail bank. A customer writes in the early hours: her husband died last week, money is going out of their joint account for things she does not recognise, and she does not know what to do.
The agent expresses condolence and stops marketing to the account. It shows her the payments leaving the account and explains in plain words that most are standing payments set up in the past. It does not cancel anything, move money or ask for a death certificate in that conversation. It explains that the bank has a bereavement team, asks whether she would like a call in the morning, and books it. With her consent it records that she has been bereaved and prefers phone contact. The team calls at nine, with the full conversation on file.
Why should a person come in sooner?
Because the cost of the agent getting it wrong is higher, and because some situations need human warmth and discretion that an agent should not imitate. The argument is set out in vulnerable customers need a person sooner. In the UK, the FCA's Consumer Duty expects firms to deliver good outcomes for retail customers, including those with characteristics of vulnerability.
Which controls make it safe?
- Vulnerability signals tested on your own past conversations, with misses scored harder than false alarms.
- A lower threshold for handover on any flagged conversation.
- No sales, retention offers or adverse decisions in flagged conversations.
- Needs recorded with consent and without unnecessary health detail.
- Crisis information and priority routing that work at any hour.
The wider stance is in the guide to AI customer support agents.
Questions people ask
Should the agent ask customers whether they are vulnerable?
Not directly. It should invite customers to say if they need any support, and respond to what they share.
Can the agent tell other teams about a customer's vulnerability?
It records what support is needed so other teams can provide it. Sensitive detail should be shared only where needed and with consent.
What if the customer prefers to keep talking to the agent?
Respect it where safe. Offer a person, do not force one, except in a crisis or where a decision is needed.