Reducing Inventory Counting Errors with Machine Learning: A Data-Driven Approach for E-Commerce

Inventory counting is one of the most critical processes in e-commerce, as it directly impacts stock control and, consequently, customer experience and profitability. In an internal analysis conducted at Hepsiburada, one of Turkey’s leading e-commerce companies, found that using traditional manual counting methods, the error rate can reach nearly 30%. In this study, to manage this error rate at lower levels and reduce operational costs, a machine learning based decision framework was developed. Data from the warehouse management system was analyzed, and feature engineering techniques were applied. Using PyCaret and Optuna for model selection and optimization, XGBoost was identified as the best performing model and deployed. The model achieved over 85% recall, demonstrating success in counting performance. This approach is expected to improve the error rate by at least 10-15%, enhancing efficiency and reducing costs in inventory management.