Quasi-experimental study reveals reduced mortgage approvals and employee autonomy during AI assistance in banking, indicating unintended rigidities in frontline nonroutine tasks.
Key Points
To examine how machine learning recommendation tools affect external customer outcomes and internal frontline employee job autonomy in nonroutine mortgage lending tasks.
Analyzed frontline banking employee lending decisions and mortgage outcomes using a quasi-experimental field dataset.
Conducted textual analysis and employee surveys to evaluate perceived rigidity and satisfaction with AI-derived recommendations.
Performed a preregistered experiment to determine the causal effect of AI response flexibility on employee response satisfaction and perceived job autonomy.
AI assistance led frontline employees to adopt more conservative behaviors, decreasing mortgage approval rates, loan default rates, and customer satisfaction with employees.
Textual analysis and survey data revealed that AI-generated recommendations appeared more rigid and less open to alternative solutions, reducing employee satisfaction.
Experimental findings showed that perceived inflexibility in AI responses directly reduced response satisfaction and indirectly constrained perceived job autonomy, regardless of perceived message helpfulness.