Human-machine interaction in order decisions: A multilevel analysis


Speaker


Abstract

Lab experiments have been widely used to obtain insights into human ordering behavior and the deviations from optimal order quantities. However, empirical work in real-world settings is still lacking. We fill this gap by examining approximately 180,000 order decisions of a material handling equipment manufacturer, where human planners place orders after receiving an order recommendation from a decision-support system (DSS). We focus specifically on the likelihood of deviations from a recommendation from the DSS and the direction of such deviations. We link such deviations to different item and demand characteristics, such as demand variability, lead time and supply uncertainty. Therefore, we develop two hierarchical generalized linear models (HGLMs) with three levels: the observation or order level, item level and planner level. We show that (i) planners adjust the recommendation when the item is important for production and  the supply uncertainty and lead time are high, and (ii) planners adjust the recommendation in the direction that aligns with their incentives, i.e., upward when uncertainties increase and downward to reduce inventory value. Surprisingly, planners significantly decrease the order quantity relative to the order recommendation when supply uncertainty is high.