The Value of Auto-Replenishment--Evidence from a Field Experiment
Abstract:
Many firms leverage data-driven algorithms to aid operational decision-making and meanwhile allow managers’ discretion to override the algorithmic recommendations. However, this practice may introduce managers’ non-compliance with algorithmic recommendations due to their behavioral biases. In this study, we study the impact of the auto-replenishment system on store managers’ ordering behavior and the resulting inventory performance. We find the managers may not comply with the algorithmic order recommendations regarding both when and how much to do. Managers constantly postpone the ordering when recommended to order or rush to order when no order is recommended. They also tend to inflate the recommended order quantity when following the recommendation to place an order. In particular, when the algorithm precision is better, while compliance is most beneficial, managers’ non-compliance is worst; When the algorithm precision is worst, while managers’ discretion is most valuable, managers are more prone to comply. We refer to this phenomenon as compliance trap. This is likely because managers will comply when the decision task is more challenging (i.e., demand is non-stationery and forecasting is difficult) while managers will resort to themselves when the decision task is less challenging.