June 08, 2026
Finance Leaders Admit Their Strategies Risk Trapping Vulnerable Customers in AI Doom Loops
Most senior decision-makers across UK financial services now accept that their own approach to artificial intelligence could harm the customers who depend on support the most. New research from ArvatoConnect, which surveyed 1,000 leaders across banks, insurers, fintechs, building societies and credit providers, found that 77% believe their AI strategies risk harming vulnerable customers, and 28% rate that risk as high.
Almost all surveyed firms have increased their use of AI in customer-facing operations over the past year, and a little below 50% report a significant rise. Yet only 20% feel confident their approach carries little or no risk, and few say it carries none at all.
Adoption Is Racing Ahead of Safeguards
Ninety percent of leaders believe AI can worsen bias and digital exclusion for the people who need help most. The damage is already landing. More than a third (35%) of financially vulnerable customers reached a human only after significant effort, navigating automated menus and long waits, and a further 15% never reached one. Just 23% found a person easy to contact.
A similar share (52%) say automated systems rarely or never resolve their issue without escalation, while only 2% say automation consistently gives them the answer they need. Customers describe feeling frustrated, isolated and trapped in an “AI doom loop”, redirected through automated systems again and again with no resolution.
This is the friction many CX leaders have warned about. As one recent analysis of the contact centre job put it, the moment a customer feels stuck with no obvious route to a person, frustration spikes.
Naming the Risk Is the Easy Part
Leaders name algorithmic bias, fraud exposure and reduced human access as their top three risks. While 40% recognise digital exclusion as a major risk, only 24% assess it during AI rollout and just 29% build in escalation to a human. Only 31% test their systems for biased outcomes before launch, 27% test against vulnerable customer scenarios, and 26% run formal impact assessments.
“We’re seeing organisations lead with the technology, rather than the outcome,” said Debra Maxwell, ArvatoConnect CEO. “AI should be there to enable better experiences, not define them. That means going back to customer service 101.”
The majority believe AI can improve support and they have already deployed tools to help vulnerable customers. The strongest use cases include digital agents that assist with form-filling and signposting, intelligent triage and conversational analytics that spot vulnerability earlier. Used well, AI can free agents to handle the interactions that need empathy and human judgment.
A third of leaders (33%) do not know who should be accountable for AI-driven outcomes involving vulnerable customers, and similar numbers cannot say how to test, debias or measure these tools.
Firms Want the FCA to Spell Out the Rules
The Financial Conduct Authority has said its existing rules, including Consumer Duty, apply to AI, but it has not yet set out how. The absence of detail is enough to stall deployment. Some 85% of firms have delayed or are considering delaying AI projects, with consequences including slower product development and missed commercial opportunities.
In the meantime, many lean on a patchwork of standards such as ISO 42001, the world’s first international standard for artificial intelligence management systems and the US NIST framework (37%). The same 85% back the Treasury Select Committee’s call for the FCA to publish practical guidance by year-end.
