Contents

Vulnerable customers are not a category. They are anyone, at any time, experiencing circumstances that affect their ability to make decisions or engage with financial services.

The FCA's guidance is clear: firms must understand the nature of customer vulnerability, equip staff to recognise and respond to it, and take practical action to ensure vulnerable customers achieve outcomes as good as other customers.

AI customer service must meet these same standards. When a vulnerable customer calls, the AI must recognise the signs, adapt its approach, and escalate appropriately. Getting this wrong means regulatory risk. Getting it right means genuinely helping people who need extra support.

Understanding Vulnerability

Vulnerability is not a permanent state or a customer segment. It is a condition that can affect anyone, temporarily or long-term, across four key drivers:

Health

Physical or mental health conditions that affect ability to engage:

Hearing or visual impairment
Cognitive conditions affecting memory or comprehension
Mental health conditions affecting decision-making
Chronic illness affecting energy or attention
Acute illness creating temporary impairment

Life Events

Major life events that create stress or distraction:

Bereavement
Divorce or relationship breakdown
Job loss or financial shock
Serious diagnosis
Caring responsibilities
Domestic abuse

Resilience

Low capacity to withstand financial or emotional shocks:

Low financial resilience (no savings, unstable income)
Low emotional resilience
Limited support network
Previous negative experiences with financial services

Capability

Challenges in engaging with financial services:

Low literacy or numeracy
Limited English proficiency
Lack of digital skills
Learning disabilities
Low financial knowledge

A customer may be vulnerable on one dimension and not others. Vulnerability may be visible or invisible, declared or undeclared. The FCA expects firms to identify and respond regardless.

Detecting Vulnerability in AI Conversations

AI can detect potential vulnerability through multiple signals.

Linguistic Signals

How customers express themselves reveals potential vulnerability:

Confusion indicators:

"I don't understand"
Repeated questions about the same topic
Answers that don't match questions asked
Requests for repetition or clarification

Distress indicators:

Emotional language (crying, shouting, expressions of despair)
Statements about being unable to cope
Mention of crisis situations
Language suggesting mental health difficulties

Capability indicators:

Very simple vocabulary
Difficulty with numbers or dates
Apparent confusion about basic insurance concepts
Taking unusually long to respond

Contextual Signals

What customers say about their situation:

Health mentions:

"I've been in hospital"
"I'm on medication that affects my memory"
"My eyesight isn't what it was"
"I suffer from anxiety"

Life event mentions:

"My husband just died"
"I lost my job last month"
"I'm going through a divorce"
"I've had a diagnosis"

Financial difficulty mentions:

"I can't afford..."
"I'm struggling to pay"
"I've been getting debt letters"
"I don't know how I'll manage"

Behavioural Signals

How customers interact with the AI:

Pacing issues:

Very slow responses (possible cognitive difficulty)
Rapid, pressured speech (possible anxiety or crisis)
Trailing off mid-sentence
Starting over repeatedly

Engagement patterns:

Asking the same question multiple times
Not seeming to process answers
Agreeing to everything without apparent understanding
Becoming silent or disengaged

AI Response to Vulnerability Signals

When vulnerability signals are detected, AI must adapt.

Adjust Communication Style

Slow down: Reduce speech rate. Allow longer pauses. Do not rush.

Simplify: Use shorter sentences. Avoid jargon. Explain terms.

Check understanding: "Would you like me to explain that differently?" "Shall I go through that again?"

Offer alternatives: "Would you prefer me to send this in writing?" "Would it help to have someone else on the call?"

Express Empathy

Acknowledge the customer's situation:

"I'm sorry to hear you're going through a difficult time."

"That sounds really challenging."

"I want to make sure we support you properly."

Empathy costs nothing and matters enormously. The customer should feel heard, not processed.

Offer Appropriate Support

Depending on the situation:

Time and patience: "There's no rush. Take all the time you need."

Alternative formats: "I can send this information by post if that's easier."

Third-party support: "Would you like to have a friend or family member join this conversation?"

Specialist help: "We have a team that specialises in helping customers in difficult situations. Would you like me to connect you?"

Escalate Appropriately

Some situations require human involvement:

Crisis situations: Customer expressing self-harm ideation, domestic abuse, immediate danger.

Complex vulnerability: Multiple vulnerability factors, unclear capacity to make decisions.

Customer request: Customer asks to speak to a person.

AI limitation: AI cannot adequately support this customer's needs.

Escalation must be warm: full context transferred, no repetition required.

"I'm going to connect you with one of our specialist team. I'll pass on everything we've discussed so you won't need to repeat yourself. They're really good at helping in situations like this."

Building Vulnerability Handling into AI

Effective vulnerability handling requires explicit design.

Vulnerability Detection Model

Train the AI to recognise vulnerability signals:

Keyword and phrase detection (distress language, health mentions, life events)
Sentiment analysis (negative sentiment, emotional intensity)
Behavioural pattern recognition (confusion, repetition, slow response)
Contextual assessment (claim circumstances, account history)

The model should produce a vulnerability score or flags that guide AI behaviour.

Adaptive Response Logic

Configure the AI to adjust based on vulnerability signals:

Low signals: Standard interaction with awareness
Medium signals: Adjusted pace, additional check-ins, offer of support
High signals: Immediate adaptation, proactive support offers, escalation consideration

Escalation Rules

Define when escalation is mandatory:

Any mention of self-harm or immediate danger
Customer explicitly requests human assistance
Vulnerability score exceeds threshold
Customer appears unable to understand or engage

Define when escalation is offered but optional:

Moderate vulnerability signals
Complex situations
Customer frustration or distress

Support Resource Integration

Connect AI to support resources:

Internal specialist teams (vulnerable customer team, complaints, bereavement)
External resources (mental health helplines, debt advice, domestic abuse support)
Customer documentation (notes from previous interactions, declared needs)

The AI should know what support is available and how to access it.

Training and Review

Regularly review AI interactions involving vulnerability:

Were signals correctly identified?
Was the response appropriate?
Was escalation handled well?
What could be improved?

Use reviews to refine detection models and response logic.

Recording and Documentation

Vulnerability interactions require careful documentation.

Interaction Records

Record in full:

Vulnerability signals detected
AI adaptations made
Support offered
Customer responses
Escalation events
Outcomes

These records support:

Regulatory compliance (demonstrating appropriate treatment)
Future interactions (continuity of care)
Quality review (identifying improvements)

Customer Records

Where vulnerability is identified, consider flagging the customer record:

Type of vulnerability (with customer consent where appropriate)
Preferred communication approach
Support needs
Named contacts if authorised

Subsequent interactions can reference this information, providing continuity rather than requiring the customer to re-explain their situation.

Privacy Considerations

Vulnerability information is sensitive. Apply appropriate protections:

Access controls limiting who can view vulnerability flags
Clear purposes for data use
Retention appropriate to purpose
Customer rights to access and correct

Balance duty of care with privacy rights.

The Regulatory Imperative

The FCA has made vulnerability a priority. Expectations include:

Understanding: Firms should understand the nature and scale of vulnerability in their customer base.

Skills and capability: Staff (including AI systems) should have skills to recognise and respond to vulnerability.

Product and service design: Products and services should be designed to meet vulnerable customer needs.

Customer service: Vulnerable customers should receive outcomes as good as other customers.

Monitoring: Firms should monitor outcomes for vulnerable customers and take action where needed.

AI is not exempt from these expectations. If AI handles customer interactions, it must handle vulnerable customer interactions appropriately.

Supervisory focus on vulnerability is increasing. Firms that cannot demonstrate appropriate handling (including in AI interactions) face regulatory risk.

The Human Element

AI can detect signals. AI can adapt responses. AI can escalate appropriately.

But vulnerability ultimately requires human connection. The customer who has just been bereaved may need not just information but empathy that only humans can fully provide. The customer in crisis may need human judgement about how to help.

The role of AI in vulnerability handling is:

Detect - Recognise signals that humans might miss
Adapt - Adjust interaction style appropriately
Support - Provide immediate assistance within AI capability
Escalate - Connect to humans when human support is needed
Inform - Provide humans with full context for effective support

AI is not replacing human care for vulnerable customers. It is extending the reach of that care, ensuring every customer interaction is assessed for vulnerability and responded to appropriately.

Ready to build vulnerability-aware AI?

SwiftCase Switchboard includes vulnerability detection, adaptive responses, and seamless escalation to specialist teams. Supporting vulnerable customers, supported by technology.

Book a demo | Learn about Switchboard | See the insurance solution