Supporting different forms of vulnerability
Two-thirds of firms monitor outcomes for customers with characteristics of vulnerability and 70% consider accessibility and vulnerability during design. Yet specific provision is much less common when vulnerability arises through changing circumstances.
Provision varies considerably across different forms of customer vulnerability
Vulnerability is not a single or static condition, and customers’ needs can change considerably over time. For firms deploying AI, this means designing services that can respond to a wide range of circumstances, from physical or cognitive needs to changes caused by bereavement, illness or financial difficulty.
The research shows that many firms are already building inclusion into the way they design and monitor AI-enabled services. Most report considering accessibility and vulnerability during the design process, while human support is widely available when an automated journey does not work as intended.
However, provision is not consistent across all forms of vulnerability. Firms report stronger support for some customer groups than others, particularly where needs are easier to identify or anticipate. The findings in this section examine where provision is strongest, where gaps remain and how firms are responding when customers need additional support.
Provision varies significantly by type of vulnerability. Firms report the greatest levels of specific support for older customers and those with physical disabilities, while provision is less common for customers whose circumstances have changed through events such as bereavement, redundancy or illness. This suggests that firms are better equipped to support needs that can be anticipated than those that emerge as circumstances change.
Most firms routinely monitor outcomes for vulnerable customers, but a significant minority do not assess these customers separately. Monitoring only overall performance risks masking differences between customer groups, making it harder to identify where vulnerable customers encounter greater friction or poorer outcomes.
Inclusive design is already embedded relatively early in the development process for many firms, with most considering accessibility and vulnerability before or during implementation. However, designing for vulnerability at the outset does not guarantee that provision will cover the full range of circumstances customers may encounter.
Human support is widely available when an AI-enabled journey does not work as intended. The challenge is ensuring that customers can find and access that support easily at the point of need, particularly when they are already experiencing difficulty with an automated service.
Key statistic
Provision is lowest for recent adverse life events
Coverage reaches 67% for customers aged 65 and over, but only 34% for those facing bereavement, redundancy or illness. The best-covered groups are defined by a permanent characteristic, the least-covered by circumstance. Vulnerability that arrives suddenly is the hardest for firms to have planned for.
Have specific provisions for customers experiencing recent adverse life events
Industry perspectives
Industry leaders reflect on the report’s key findings.
Kellie Youles head of risk & deputy MLRO UK Segpay
"The widening gap between perception and reality reveals a dangerous AI blind spot for firms. This creates severe regulatory and ethical risks. Unmonitored AI models can quietly entrench bias, misprice risk, or unfairly exclude the very consumers regulators require firms to protect. True inclusion readiness simply cannot rely on self-assessment. It requires concrete behavioral data, rigorous algorithmic auditing, and proactive safeguards to ensure automated systems do not leave vulnerable populations behind."
Noyan Nihat co-chief executive officer Cardaq UK
"Payments is infrastructure, and infrastructure carries obligations that a product does not. When a retailer's checkout or a bank's authentication step runs on AI, the people who fail it do not go elsewhere. They go without. Offering a human is an activity, not an outcome. Our industry earned its licence by being the thing that always works. That standard applies to the models now, too."