ArvatoConnect’s research found 89.7% of BFSI leaders believe badly implemented AI could worsen bias and digital exclusion, while financially vulnerable consumers are already reporting the cost of poor automation: 32% say they have felt trapped in an “AI doom loop” and 52.0% say AI or automated systems rarely or never resolve their issue without escalation to a human.
We believe avoiding those issues comes down to five practical commitments:
The practical assessment should not be presented as a verbatim FCA checklist. Instead, it should show whether the partner can evidence the areas the regulator repeatedly focuses on: clear accountability, capable frontline teams, responsible use of vulnerability data, quality customer support and consistent outcome monitoring.
Testing is where the market is weakest. ArvatoConnect’s Care by Design report found 41% of leaders name algorithmic bias a major risk, yet only 31.3% sandbox-test against biased outcomes and just 26.8% test with vulnerable-customer scenarios. A partner that can evidence this is already operating above the market norm, and it is the standard ArvatoConnect designs and tests to.
If a partner influences how vulnerable customers are supported, firms should be able to demonstrate that staff capability, service design, escalation routes and monitoring are robust and fit for purpose. ArvatoConnect’s research found that 76.9% of leaders believe their own AI strategy risks excluding vulnerable customers, so due diligence should assess whether a partner can actively reduce that risk in practice, not simply describe good intentions.
Before signing, compliance teams should get answers to four questions:
If a provider cannot evidence these, the risk still sits with the firm. ArvatoConnect’s glass-box, deterministic approach is built to answer these questions with evidence, keeping decision pathways auditable from the start.
The key point is choice. More than half (52.1%) of vulnerable consumers surveyed by ArvatoConnect said they would rather speak to a real person than AI when discussing complex or sensitive issues. But the answer is not to force every customer down the same route. Some customers may prefer AI, self-service or a less exposing channel in certain situations, so good design should make it easy to choose and move between support options.
ArvatoConnect designs real-time detection to support agent judgement and trigger the right next step, not simply to create another flag.
The most immediate exposure is financial and regulatory: complaints, redress, remediation, enforcement scrutiny and the cost of fixing failing journeys after customers have already been harmed. The FCA can impose financial penalties, public censure and other regulatory action where firms fail to meet its standards, so firms need to show how vulnerable-customer outcomes are tested and improved, not simply asserted. 11 firms are currently under investigation by the FCA for not meeting Consumer Duty – suggesting the FCA has made the move to enforcement
The operational risk is that poorly designed automation increases repeat contact, abandonment and escalation. ArvatoConnect’s research found 35.2% of financially vulnerable consumers could only reach a human after significant effort, while a further 14.6% could not reach one at all. That is not just a service issue; it creates avoidable demand and makes already complex cases harder to resolve.
The reputational risk is just as material. More than a quarter (28.2%) of vulnerable consumers said interacting with AI made them feel the organisation simply did not care. For regulated firms, that kind of sentiment can translate into complaints, attrition and loss of trust, especially when the same automated failure is repeated across thousands of interactions.
The stronger answer is not to avoid AI, but to design AI-enabled support around evidence, transparency, human accountability and customer choice.
ArvatoConnect is a UK-based customer experience and business optimisation provider that published Care by Design: Making AI Work for Vulnerable Customers in Regulated Environments, based on a survey of 1,000 senior UK BFSI decision-makers, alongside a separate survey of 1,000 financially vulnerable consumers. The report shows how to balance AI and human support to prevent digital exclusion and deliver fair outcomes for vulnerable customers under the FCA Consumer Duty. With XpertRule, ArvatoConnect builds glass-box, deterministic AI for regulated environments, where decisions must be explainable and a human stays accountable for the outcome.
