The Finding
New research from ArvatoConnect, surveying 1,000 senior decision-makers across U.K. banks, insurers, fintechs, building societies, and credit providers, finds that 77% of financial services leaders believe their own organization’s AI strategy could negatively impact vulnerable customers, with 28% rating that risk as high. This comes as 88% of firms have increased AI use in customer-facing operations over the past year. On the customer side, 74% of vulnerable customers have felt like giving up while trying to get support from a financial provider, and 26% have abandoned the attempt altogether.
Why This Matters for Collections and BFSI
For collections operations specifically, this data lands directly on the customers who are hardest to get wrong. Financially vulnerable customers are the population collections teams interact with daily, and the research shows automated systems are currently failing them at scale: 52% of customers said AI or automated tools rarely or never resolved their issue without escalating to a human, and 32% described becoming trapped in what the report calls an AI doom loop, repeatedly redirected through automated systems without ever reaching resolution. In a hardship or arrears conversation, that failure mode is not just poor CX; it is the exact scenario regulators scrutinize most closely.
What the Numbers Do Not Say Out Loud
The most telling finding is not the risk percentage; it is the gap between awareness and action. Ninety percent of leaders believe AI could worsen bias and digital exclusion, yet only 24% actually assess that risk during AI deployment projects, and just 29% build escalation routes to human support directly into AI-enabled journeys. Governance testing lags even further behind: only 31% sandbox-test AI systems for biased outcomes before deployment, and just 27% test using vulnerable-customer scenarios specifically. Meanwhile, a third of leaders admit they do not know who should be accountable when an AI system’s outcome harms a vulnerable customer. Leaders are aware of the risk and still not building the safeguards to test for it, which is a more concerning gap than ignorance.
The Practical Read
As ArvatoConnect CEO Debra Maxwell put it, organizations are leading with the technology rather than the outcome, when AI should be there to enable better experiences, not define them. For collections operations, this is a direct design instruction. A voicebot handling a payment conversation needs a built-in, tested escalation path to a human the moment vulnerability signals appear, not as a fallback bolted on after a complaint, but as a core part of the journey design. The finding that 92% of leaders believe vulnerability receives sufficient board attention, while only a quarter actually test for it, signals that the gap sits in implementation, not intention.