M. Megan Hunter UK & Europe · Remote
Work with me Testimonials Client results About Blog Book an intro call
AI and customers

AI and vulnerable customers.

I've spent a lot of time using AI in my own business, and I think there's a huge opportunity for financial services to do good things with it, especially around financial inclusion. It gets my brain whirring. But the customers who need the most care are also the ones a badly designed AI journey can let down fastest.

This page pulls together what the FCA expects when a firm uses AI with customers in vulnerable circumstances. It covers whether you can use AI to spot vulnerability, how to govern it, and how the CONC 7 rules apply if AI is part of arrears and collections. It isn't legal advice. Your compliance team owns the interpretation.

Are there special AI rules for vulnerable customers?

No. The FCA's approach to AI relies on the rules that already exist. In its October 2026 paper, AI in financial services: good practice and areas for improvement, the FCA says its examples don't create new requirements. The ones that matter most here are:

  • the Consumer Duty, including the consumer support outcome and outcomes monitoring
  • FG21/1, the FCA's guidance on the fair treatment of vulnerable customers
  • the Senior Managers and Certification Regime, so a named senior person is accountable
  • the rules on systems and controls, operational resilience and outsourcing
  • the Equality Act 2010, and data protection law overseen by the ICO

The FCA is clear that where firms use AI-supported tools, they stay responsible for the outcomes.

What the FCA says good looks like

The FCA says firms that use AI safely and responsibly:

  • define the customer need and target market before designing and deploying AI
  • test outcomes across different customer characteristics and scenarios
  • use proportionate controls to manage bias and accuracy, so decisions can be explained
  • tell customers how they've used AI, and its limits
  • monitor live outcomes so they can spot and deal with problems

For customers who need more help, the paper describes firms testing outcomes across vulnerability and digital or financial capability. It describes letting people choose between AI and non-AI support for straightforward tasks, and letting them switch channel without repeating themselves. It also describes firms using unexpected cases, like someone mentioning a food bank, to improve their prompts, controls and escalation.

On the other side, the FCA flagged firms that excluded people with vulnerability indicators from AI journeys by default without giving them another way to get support. It flagged firms that put too much faith in a large language model to spot vulnerability, and automated journeys that were hard to get out of.

Can you identify vulnerable customers using AI?

AI can help. The FCA describes firms using tools to spot possible signs of vulnerability, such as key words in what a customer writes, spending patterns, or signs someone is struggling with a digital journey.

One example in the paper stood out to me. A firm built a chatbot and worked with its frontline staff to decide which vulnerability indicators to look for. It kept a live list of terms the tool flags to staff, had a clear plan for supporting flagged customers through the tool or a person, and checked whether outcomes improved. In another, a firm's tool found spending patterns that suggested vulnerability, and the firm asked its ethics committee to decide how to use that fairly.

The warning is just as clear. The FCA saw firms place too much confidence in a model's ability to pick up vulnerability from text alone, especially temporary vulnerability. FG21/1 says vulnerability can be temporary, sporadic or permanent, so a model trained on one snapshot will miss people.

My view is simple. AI can help you notice. It shouldn't be the only way you notice, and it shouldn't decide on its own what happens next.

Should vulnerable customers be kept away from AI journeys?

Not by default. The FCA says excluding customers with vulnerability indicators may be right in some cases, but it can be unfair if the firm doesn't give them another way to get support that meets their needs.

I'd go a step further. A human alternative is fine if it gets people the same help. It's not fine if they get blocked or have a worse experience. And I don't think firms should decide for people whether they'd prefer AI. That choice should be the customer's.

Governing AI that prioritises or triages customers

The FCA asks firms to take particular care where AI affects people's access to services, how issues are prioritised, and the support customers get. If a tool decides who gets seen first, or who gets routed to a specialist team, this is the part to get right. From the paper, that means:

  • a named senior manager accountable for the use case from the design stage
  • human oversight designed for the job, with people who have the skill and authority to challenge, not a review that drifts into a tick-box
  • red, amber and green thresholds with agreed actions when performance slips
  • records of prompts, outputs, model versions and escalation decisions, so a complaint can be looked into
  • monitoring of real outcomes for different groups, not only technical metrics

The Consumer Duty already asks firms to monitor whether any group of customers is getting worse outcomes than others. That applies whether a decision was made by a person or a model.

Does AI debt collection comply with CONC 7?

CONC 7 is the part of the FCA Handbook on arrears, default and recovery. It doesn't mention AI. The rules apply however a firm contacts its customers, so an AI tool in collections has to meet them just like a person would. That includes:

  • clear, effective and appropriate policies for customers in or approaching arrears, and for the fair treatment of customers the firm understands or reasonably suspects to be vulnerable, with regard to FG21/1 (CONC 7.2)
  • treating customers in or approaching arrears with forbearance and due consideration, taking into account the individual circumstances the firm knows about or should know about (CONC 7.3)
  • telling customers, where appropriate, that free and impartial debt advice is available, and referring them to it
  • not pressurising customers to pay in very large amounts, too quickly, or by borrowing or selling their home
  • choosing the right way to communicate, and changing channel if that's what lets the customer engage
  • not contacting customers at unreasonable times, and explaining who is contacting them and why (CONC 7.9)
  • suspending recovery where the firm is told, or ought to be aware, that a customer may lack the mental capacity to make the relevant decisions (CONC 7.10)

CONC 7 also says firms should pay due regard to the Consumer Duty when dealing with customers in arrears, and that anyone acting on the firm's behalf has to comply with CONC too (CONC 1.2). The FCA's AI paper includes a credit example that worked well: the tool explained what it could and couldn't do, and when it couldn't help it pointed people to free debt advice or a member of staff.

So can AI in collections comply? It can. Whether yours does depends on how it's designed, tested and overseen, and on what happens when it spots someone in difficulty.

Outsourced providers and AI vendors

If a contact centre provider or an AI vendor runs part of your vulnerable customer process, the responsibility stays with you. The FCA says firms can outsource AI services but remain responsible for them, and it flagged firms that took vendor assurances on trust without testing them. The outsourcing rules in SYSC 8.1 still apply.

I'm not going to name providers here. What I'd ask any provider for is the same evidence I'd want in-house:

  • outcomes data split by customer group, including customers with characteristics of vulnerability
  • how their approach maps to FG21/1, in practice, not only in a policy
  • call or chat samples you can review yourself
  • how and when a customer reaches a person, and whether they have to repeat themselves
  • records of what the AI did, when, and who stepped in
  • what happens if their model changes or is switched off

What I'd check in an AI customer journey

  1. The customer need the tool is there to meet. If you can't name it, start there.
  2. The route for someone who needs more help, and whether it's as good as the AI route.
  3. Whether the customer can choose, and can switch without starting again.
  4. Whether human review is still review, or has slowly turned into a tick-box.
  5. Testing with people who need extra help, not only the average customer.
  6. The records you'd need if a customer challenged an outcome.
  7. A plan B if the supplier changes or retires the model.

How I help

I help regulated firms decide where AI belongs in the customer journey, what stays with a person, and how to show customers in vulnerable circumstances get good outcomes. I don't sell an AI product, and I'm not a law firm or a compliance monitor. More on that is on my AI and customer service page. The FG21/1 definition, the four drivers and the TEXAS and IDEA models are on vulnerable customers and the Consumer Duty. My notes on the FCA paper itself are in what the FCA found on AI in financial services.

Book an intro call

Common questions

What are a firm's obligations when it uses AI with vulnerable customers?

There are no separate AI rules. Firms using AI still have to meet the Consumer Duty, including the consumer support outcome and outcomes monitoring, and should follow FG21/1 on the fair treatment of vulnerable customers. SM&CR, the systems and controls and outsourcing rules, the Equality Act 2010 and data protection law also apply. The FCA says firms stay responsible for outcomes delivered through AI-supported tools.

Can we identify vulnerable customers using AI?

AI can help spot possible signs of vulnerability, such as key words, spending patterns or someone struggling with a digital journey. The FCA has warned against assuming a large language model can reliably identify vulnerability from text alone, especially temporary vulnerability, without enough testing and safeguards. It works best alongside people, with a clear plan for what happens when a customer is flagged.

Should vulnerable customers be excluded from AI journeys?

Not by default. The FCA says excluding customers with vulnerability indicators may be appropriate in some cases, but it can unfairly exclude them if the firm doesn't provide another way to get support that meets their needs. Good practice includes letting customers choose between AI and non-AI support for straightforward tasks and switch channel without repeating themselves.

Does AI debt collection comply with FCA CONC 7?

CONC 7 doesn't mention AI, so the same rules apply whether a person or a tool is in contact with the customer. An AI collections process has to support forbearance and due consideration, fair treatment of customers who may be vulnerable, signposting to free debt advice, no undue pressure, appropriate contact times and channels, and suspending recovery where there are mental capacity concerns. Whether a specific tool complies depends on its design, testing and oversight. This is not a legal opinion.

Who is responsible if an outsourced provider or AI vendor gets it wrong?

The regulated firm. The FCA says firms can outsource AI services but remain responsible for them, and it has flagged firms that relied on vendor assurances without testing them. CONC also requires firms to make sure people acting on their behalf comply with CONC.

Did the FCA create new AI rules in October 2026?

No. The FCA's AI in financial services good practice paper, published on 19 October 2026, says its examples do not create new requirements. They show how existing rules, including the Consumer Duty and FG21/1, apply to AI.

Related reading

  • AI and customer service
  • Vulnerable customers and the Consumer Duty
  • What the FCA found on AI in financial services
  • What the FCA's October 2026 AI findings mean for customer outcomes

Sources

  1. FCA, AI in financial services: good practice and areas for improvement. Published 19 October 2026. Updated 20 October 2026.
  2. FCA, FCA approach to AI.
  3. FCA, FG21/1: Guidance for firms on the fair treatment of vulnerable customers.
  4. FCA Handbook, CONC 7 Arrears, default and recovery (including repossessions). Viewed 16 November 2026.
  5. FCA Handbook, CONC 1.2.
  6. FCA Handbook, SYSC 8.1 (outsourcing).
  7. FCA, Firms' treatment of customers in vulnerable circumstances: review. First published 7 March 2025.
Get in touch

Let's work together.

Book an intro call hello@megan-hunter.com
M. Megan Hunter

Customer experience, strategy and operations leader

Explore
Services Fractional and interim Client results Testimonials About Blog FAQs Links
Topics
Consumer Duty Vulnerable customers Financial inclusion Digital inclusion Social sustainability AI and customer service Fair value assessments FCA targeted support AI and vulnerable customers
Elsewhere
LinkedIn The Making It Matter Podcast Founder, Nadi Megan's Substack
Contact
hello@megan-hunter.com Book a call
© 2026 Megan Hunter. All rights reserved.
Cookie Policy Privacy Policy