Capacitated Production Planning and Inventory Control when Demand is Unpredictable for Most Items: The No B/C Strategy

In this paper, we examine a discrete-time, periodic-review production environment that assembles several hundred items and that possesses limited, perhaps random production capacity. The demand for a large subset of these items is highly erratic and extremely difficult, if not impossible, to predict accurately. Consequently, a coordinated production-inventory strategy, such as the No B/C Strategy presented in Carr, et al. (1993), is necessary. In such a strategy, inventory is carried only in high demand rate, predictable items and production priority is given to nonstocked items. Production is controlled for stocked items through a modified base stock policy. A key feature of our approach is that it does not rely on item-level forecasts for each item. Our objective is to develop and test a computationally efficient and accurate procedure for establishing base-stock levels that minimizes the expected holding and backorder costs per period over an infinite horizon. We consider two alternate operating scenarios and compare them. In the first, we consider an information poor environment and assume that production decisions must be made in advance of observing demand. In the second, we assume an information rich environment and observe demand prior to making production decisions. We quantify the value of this information in terms of operating performance. We demonstrate that as the capacity utilization increases, the value of obtaining advanced demand information, while still valuable, diminishes. In these environments, there are two types of safety stock needed to minimize expected costs: demand-driven safety stock and capacity-driven safety stock. We demonstrate that for capacity-driven safety stock, traditional equal-fractile allocation policies may lead to substantial inventory imbalances over time and result in lower than expected service levels. To resolve this, we present an alternative means for allocating capacity-driven safety stock that favors storing inventory in items for which the risk of not selling them is lower. Through a series of numerical experiments, the model is shown to be accurate and on average performs within 0.26% of a lower bound. The information rich case resulted, on average, in 35% lower safety stock and 57% lower expected costs per period than the information poor case. Using our proposed capacity allocation scheme resulted in 23% less units in imbalance than if the newsvendor allocation were used. We conclude with an experiment of the approach in an industrial environment and demonstrate its effectiveness over current management practices.