Bank staff using AI approved fewer mortgages and left customers less satisfied, study finds

The findings matter because AI is increasingly used to assist frontline staff in tasks that do not follow a routine. The authors argue the technology may not be inherently supportive in such work, and that it can affect both customers and the employees using it.

The study also looked at why. Textual analysis and a survey were consistent with AI-generated solutions appearing more rigid and less open to alternatives, and a preregistered experiment found such responses were seen as less flexible, which reduced satisfaction with the response and perceived job autonomy.

The research tracked external outcomes, meaning market offerings and customer satisfaction with frontline employees, and internal outcomes, meaning employees' job autonomy and satisfaction with the responses they received from the AI.

In the banking data, the AI-assisted staff's more conservative behavior showed up as fewer mortgage approvals and fewer default payments, alongside lower customer satisfaction with the employees.

To test the mechanism, the authors used textual analysis and a survey, then ran a preregistered experiment. It found AI-generated responses were perceived as less flexible, and that this reduced satisfaction with the response and, mainly through that route, perceived job autonomy regardless of how helpful the message seemed.

The authors conclude that AI should be aligned with the characteristics of nonroutine tasks so firms can use AI assistance while limiting unintended effects on employees and customers.