Aplicación de técnicas de inteligencia artificial para reducir errores y tiempos en el proceso de picking de un centro de distribución automotriz.

Abstract

This research aimed to propose the application of data analytics and artificial intelligence techniques to optimize the picking process in an automotive distribution center. This was achieved through demand analysis, operational product segmentation, and the identification of relationships between SKUs. The study used a quantitative approach qualitative support, utilizing operational records from the first four months of 2026, direct observation, and technical advices from employees involved in selection and packing activities. The analysis comprised 66,153 data records or order items, corresponding to 15,066 different orders and 6,930 different products. An average order preparation time of 23 minutes and an estimated error rate of 20.0 % were determined, calculated from 3,308 errors recorded across 16,544 orders included in the incident analysis. These results highlighted opportunities for improvement related to product organization, control of critical SKUs, location updates, reduction of rework, and strengthening operational monitoring. As part of the proposal, association rules were applied to identify relationships between star products and slow-moving candidates within multi-item baskets. After applying common counting, confidence, and lift criteria, 202 rules were identified that could be used for logistic analysis. A ranking model was also integrated Random Forest to prioritize recommendations. An interactive prototype in Google Colab and a dashboard in Power BI to visualize process indicators. The proposal does not envision a final implementation, but rather a technical alternative to improve decision-making, reduce routes and transportation, increase the operational efficiency , and strengthening the operational efficiency of order picking.