Reluctant to decrease: Human-algorithm collaboration in tactical inventory decisions

Abstract

Problem definition: Setting base stock levels in spare parts inventory management is a high-stakes decision supported by advanced algorithms. Humans frequently overrule the algorithm recommendations. We partner with an OEM that has deployed best-in-class algorithms with humans still making changes to the 24% of the base stock decisions, and aim to understand what drives such overrides.

Methodology/results: We deploy a structural estimation model estimating the humans’ psychological cost for overage and underage and find that in particular, the psychological cost for underage are substantially higher than parameterized in the algorithms. Interestingly, we can show that such psychological cost for underage are particularly high in case the algorithm proposes to reduce the inventory from 1 to 0, and provide evidence of the existence of the endowment effect in inventory control.

Managerial implications: Managers and decision-makers can benefit from our findings by understanding the behavioral biases influencing decision-making in spare parts inventory management. Leveraging the parameter estimates from our structural estimation, our counterfactual analysis shows that a Pareto-optimal alternative solution can be obtained that is more aligned with the humans’ perceptions of underage costs, thereby decreasing the need for planner adjustments and offering actionable insight to improve the decision-making process.

Keywords: Human-algorithm Collaboration, Behavioral Operations, Inventory Management, Spare Parts, Structural Estimation