Abstract Inventory management is a critical task for distribution companies, as forecasting errors can lead to excess inventory, higher storage costs, or stockouts. This practice and policy paper presents a single-case study describing the development and deployment of a machine learning-based predictive model designed to optimize inventory management for a medium-sized Brazilian construction supplies distributor. Using the Amazon Forecast and Amazon SageMaker Canvas platforms, the study demonstrated how advanced demand forecasting techniques can potentially reduce stockouts and enhance operational efficiency. The research compared various predictive algorithms and assessed their performance using metrics such as RMSE, MAPE, MASE, and WAPE. The model achieved a weighted absolute percentage error (WAPE) of approximately 0.69% during the evaluation period, with performance varying across SKUs and demand patterns. The evaluation employed an 80/20 train-test split with a 60-day forecast horizon. Rather than claiming algorithmic novelty, this study’s primary contribution is architectural and socio-technical since it documents an end-to-end implementation blueprint for deploying cloud-based AutoML in an SME that previously relied solely on descriptive analytics. The reliance on a single aggregate metric, the absence of baseline comparisons, limited transparency into proprietary AutoML, and the lack of quantitative measurement of business impact are discussed in the limitations of the study. The study details the technology architecture, data governance workflow, and human–AI interaction protocols, providing technical and managerial recommendations for SMEs seeking to transition from retrospective reporting to simplified predictive analysis, even with limited IT resources and without specialized Data Science teams.
Goulart et al. (Tue,) studied this question.