Human-Machine Interaction in Digital Control Rooms: The Dual Impact on Workload and Operational Performance

Abstract

In this article, we show a dual impact of humans intervening in automated decision-making systems, referred to as human-machine interaction (HMI). We find that HMI increases workload, with negative consequences on operational performance, but it simultaneously mitigates the negative effect of workload on operational performance. With accelerating digitization and increasing concerns about employees' workload, studying the link between HMI, workload, and operational performance is timely and practically relevant. For this research, we constructed a granular dataset on digital control rooms from the Belgian railway infrastructure company. This setting covers more than 4.5 million event-level observations, involving 664 controllers in 10 control rooms. It includes rich information on controllers' workload and HMI, train delays, and a broad range of contextual factors. We are the first to show that, in addition to the railway infrastructural and contextual factors, employees' workload and HMI impact train delays. In particular, utilizing an instrumental variable (IV) two-stage least squares (2SLS) approach, we find that changing workload from minus one to plus one standard deviation from the mean leads to a significant increase of 0.303 standard deviations (12.496 seconds, while the average is 0.836 seconds) of train delay between two consecutive train signals. Changing a controller's HMI in a similar way mitigates this negative effect by over 15%, but increases workload by around 0.079 standard deviations. Our findings have important managerial implications, such as how to factor in employees' HMI heterogeneity in task allocation to improve their operational performance. Our study opens new research avenues.

Keywords: employee workload, human-machine interaction, people-centric operations, digital control rooms, causal inference