Vulnerable customer safeguards

What it does

It gives an automated channel some of the care a trained person would show. The system recognises signs of vulnerability, whether stated by the customer, flagged on their record or evident from behaviour, and responds: it slows or pauses the journey, withholds actions that could harm, offers support and hands over to a person. The business owner of the journey is accountable; the vulnerability lead sets the standard. It operates on every interaction.

How to implement it

• Define the indicators: disclosures of health, bereavement, debt or abuse, confusion, distress, third-party involvement.
• Pass existing vulnerability flags to the AI channel.
• Detect disclosures in free text and speech, and test detection on realistic conversations.
• Specify the response for each indicator, including actions the system must not take.
• Hand over to trained staff with the conversation attached.
• Record the disclosure, with the customer's agreement, so they need not repeat it.

Evidence it operates

Design: indicator definitions and response rules; detection test results. Operation: referral volumes and reasons; sampled conversations showing the response; outcomes for flagged customers reported separately; training records for receiving staff.

Where it fits

Within the live journey. It is the primary control for failure to recognise vulnerable customers, and supports emotional reliance on conversational AI and age-inappropriate AI interaction. It depends on escalation to a human for somewhere to send the customer.

Limits

Detection is imperfect: many vulnerable customers say nothing, and some who are flagged are not. Inferring vulnerability is itself sensitive processing and needs a lawful basis and care. A referral helps only if a trained person is available promptly. Over-cautious rules can deny capable customers a channel they prefer.