ArvatoConnect has published research showing that most financial services leaders believe their AI strategies could harm vulnerable customers. The study also found that the use of customer-facing AI has risen across the sector.
Based on a survey of 1,000 senior decision-makers across banks, insurers, fintechs, building societies and credit providers, the research found that 77% believe their own AI strategies could harm vulnerable customers. More broadly, 90% said AI could worsen bias and digital exclusion for people who need the most support, yet only 23% were confident their approach posed little or no risk.
Adoption has continued despite those concerns. Almost nine in ten organisations said they had increased their use of AI in customer-facing operations over the past year, with 41% reporting a significant increase.
Safeguard gaps
The findings suggest a gap between risk awareness and the controls firms apply before systems go live. Only 31% said they sandbox-test AI systems for biased or unethical outcomes before deployment. Meanwhile, 27% test using vulnerable customer scenarios, and 26% carry out formal impact assessments on vulnerable customers.
The gap is also visible in service design. Although 40% of organisations recognise digital exclusion as a major risk for vulnerable people, only 24% assess that risk when implementing AI. Just 29% said they built in escalation to human support.
Across six core service design practices, including understanding customer needs, defining successful outcomes, testing against real customer scenarios and monitoring performance, only around four in ten organisations said any of those steps were fully embedded.
The survey also pointed to uncertainty over accountability inside firms. A third of leaders said they do not know who should be accountable for AI-driven outcomes involving vulnerable customers. Meanwhile, 32% do not know how to test AI tools for vulnerabilities in customer scenarios before deployment, and 31% are unclear about how to prevent bias against vulnerable groups.
Consumer frustration
Separate polling of financially vulnerable consumers suggests those weaknesses are already affecting customers. Nearly three-quarters (74%) said they had felt like giving up when trying to get help from their bank, insurer or provider, while 26% said they had abandoned their attempt altogether.
More than a third (35%) said they were only able to reach a human after significant effort, including navigating automated systems and long waits. A further 15% said they could not reach a human at all, while 23% said they could do so easily.
Automated systems often failed to resolve problems on their own. Some 52% said AI or automated systems rarely or never resolved their issue without escalation to a human, while only 2% said automated systems consistently provided the answers they needed.
Many consumers described the experience as frustrating. The polling found that 58% felt frustrated, 33% felt isolated and unable to get the support they needed, and 32% said they felt trapped in an "AI doom loop" in which they were repeatedly redirected through automated systems without resolution.
Concerns about what comes next were also evident. Almost half of respondents (48%) said they fear it will become harder to get in touch with a human as AI becomes more common in customer service, while 44% worry it will become more difficult to get help with complex financial issues.
Regulatory questions
The findings come as financial firms weigh the benefits of AI against uncertainty over how regulators expect those systems to be governed. The Financial Conduct Authority has said existing frameworks, including Consumer Duty, should apply to AI, but firms appear to want more detailed practical guidance.
According to the research, 85% of firms said they had delayed or were considering delaying AI implementation because of a lack of regulatory clarity. Among those who had delayed, 33% cited slower product development and innovation, 31% pointed to a reduced ability to improve customer outcomes, and 30% said they had missed commercial opportunities.
Many firms now rely on a mix of external and internal frameworks rather than a single UK standard. Respondents cited internal governance models, ISO 42001, the US NIST AI Risk Management Framework and the EU AI Act among the approaches in use.
At the same time, executives still see strong potential in AI. Almost nine in ten leaders said the technology could have a positive impact, and 82% said their firms had already implemented AI-enabled tools intended to support vulnerable customers. The most commonly cited uses included digital agents that help with form-filling and signposting, personalised support journeys, intelligent triage, conversational analytics to identify vulnerability earlier, and tools that adjust the reading age of communications.
"AI has enormous potential to improve customer experiences, particularly for identifying vulnerable customers, who often struggle to access timely, personalised support. But we're seeing organisations lead with the technology, rather than the outcome. AI should be there to enable better experiences, not define them. That means going back to customer service 101: understanding where customers struggle, what good looks like for them, and then designing services, with or without AI, around those needs," said Debra Maxwell, Chief Executive Officer of ArvatoConnect.
"For a sector that widely recognises the risks, the gap between knowledge and mitigation is considerable," the CEO added.