An AI Approach to SME Inventory Management

Inventory management has been widely studied for large companies in countries with land borders. We focus on the application of Data Science to Inventory Management of Small and Medium Enterprises (SMEs) in Small Island Developing States (SIDs) which has the additional constraints that, the customer base is small and land-based delivery of products from other countries is not possible (i.e., only Sea and Air delivery is possible). In order to reduce costs, batch purchasing must be performed and this requires forecasting of products to be purchased as well as their quantities. We also need to take into account storage costs, shelf life of products and the reduction of the value of certain products (e.g., electronic items) over time. This requires a combination of accurate forecasting and opti-mized inventory management. We investigate both traditional and Machine Learning (ML) based approaches. Our results indicate that, although the ML models were generally good, the traditional models (such as seasonal naive and Holt-Winters) outperformed them in many cases due to the limited data available. We adopted a product-specific approach for the forecasting phase and then combined this with an optimized ordering strategy. With our approach, inventory was significantly reduced while maintaining the same level of customer satisfaction (i.e., availability of desired products).