Video-Based AI Misconduct Monitoring: Unintended Consequences on Operational Outcomes

Abstract

Employee misconduct poses significant challenges for organizations, yet traditional monitoring methods, such as closed-circuit television (CCTV), are predominantly reactive, offering little opportunity for timely intervention. Recent advances in computer vision and edge computing have enabled a new class of AI-enhanced video monitoring systems that detect employee misconduct in real time. This development marks a significant technological shift by providing immediate feedback upon detection, a capability that was not feasible until very recently. Despite its growing adoption, the operational implications of this novel AI-based monitoring approach remain largely unexplored. Although the deployment of AI technologies is often associated with the promise of improved efficiency and enhanced operational performance, we find a counterintuitive outcome in our empirical setting. Using a unique dataset from a restaurant chain that implemented video-based AI misconduct monitoring to deliver real-time feedback to employees, we uncover unintended consequences: applying this technology significantly reduces restaurants’ operational performance, as measured by revenue and customer rating scores. To understand this counterintuitive effect, we further investigate the underlying mechanisms driving these adverse outcomes. Interestingly, we find that the observed performance degradation is primarily driven by high-quality restaurants, whereas those with low service quality exhibit potential for operational improvement following the deployment. We further show that the type of feedback generated by the AI system plays a critical role in shaping its impact: feedback emphasizing individual operational misconduct, such as uniform violations, intensifies the negative effects, whereas feedback on process operational misconduct, such as leaving cashier stations unattended during customer check-out, mitigates these adverse effects. Taken together, these findings highlight potential unintended consequences of applying advanced AI surveillance tools and offer new managerial insights by illustrating how organizational context and the nature of feedback types jointly shape the effectiveness of real-time AI monitoring.