This paper is inspired by a real case, which occurred in a fastener manufacturer that provides VMI (Vendor Managed Inventory) service to a client that produces IT products. Under the VMI collaboration model, the electronic purchase order (ePO) provided by the client are usually very diversification, and sometimes, a numbers of demand items will be cancelled and/or terminated forever suddenly. This situation will yield many inventory management problems. In this study, a decision support module for ERP system is designed. It is a module embedded with Data Mining (DM) technology. By analyzing ePOs given from the client, correlations among demand items of the client would be detected, and the demand structure of the client would be established. The discovered demand associations among items were further utilized to build a virtual demand BOM for supporting the sales to create marketing strategies, to reduce risk of uncertainty from rolling forecasts, and may get more niches in the supply chain system.