Evaluating Count Prioritization Procedures for Improving Inventory Accuracy in Retail Stores

Problem definition : We compare several approaches for generating a prioritized list of items to be counted in a retail store, with the objective of detecting inventory record inaccuracy and unknown out of stocks. Academic/practical relevance : We consider both “rule-based” approaches, which sort items based on heuristic indices, and “model-based” approaches, which maintain probability distributions for the true inventory levels updated based on sales and replenishment observations. Methodology: Our study evaluates these approaches on multiple metrics using data from inventory audits we conducted at European home and personal care retailer dm-drogerie markt. Results : Our results support arguments for both rule-based and model-based approaches. We find that model-based approaches provide versatile visibility into inventory states and are useful for a broad range of objectives but that rule-based approaches are also effective as long as they are matched to the retailer’s goal. We find that “high-activity” rule-based policies, which favor items with high sales volumes, inventory levels, and past errors, are more effective at detecting inventory discrepancies. The best policies uncover over twice the discrepancies detected by random selection. A “low-activity” rule-based policy based on low recorded inventory levels, on the other hand, is more effective at detecting unknown out of stocks. The best policy detects over eight times the unknown out of stocks found by random selection. Managerial implications: Our findings provide immediate guidance to our retail partner on appropriate methods for detecting inventory record inaccuracy and unknown out of stocks. Our approach can be replicated at other retailers interested in customized optimization of their counting programs.