Knowledge that Transforms

In this paper, we investigate the optimality of myopic inventory replenishment policies in a periodic-review single-echelon system, with nonstationary, correlated, stochastic demand and cost, and nonincreasing stochastic prices. Using the single-unit decomposition approach, we provide certain general conditions on the demand and cost processes under which a myopic policy is optimal. Under these conditions, the optimal policy is a myopic state-dependent base-stock policy, which can be expressed in closed form as a base-probability policy. Specifically, the order associated with a given customer should be placed if and only if its arrival probability within the leadtime is higher than a threshold. Our results generalize earlier conditions for the optimality of myopic policies. Namely, we show that myopic policies can be optimal even when the demand is correlated or stochastically decreasing.

The literature on mass customization generally focuses on the tradeoff between higher revenues from better matching customer preferences with product specifications, and higher costs of offering a broader—possibly fully customized—product line. Less well understood is the tradeoff between the increased ability to precisely meet customer preferences and the increased leadtime from order placement to delivery often associated with customized products. In this paper, we use a locational customer choice model to formulate a firm's integrated product line design problem that involves variety, leadtime (or inventory), and pricing decisions. We propose a dynamic programming based solution procedure that amounts to solving a shortest path problem on an acyclic network, and derive some structural results on the optimal product line design. We find that unimodal preferences generally result in hybrid product lines, with standard products clustering around the mode and custom products covering the tails, in contrast with the all-custom or all-standard product lines that are optimal under uniform preferences. We also numerically examine how the firm should adjust its leadtime and variety in response to changes in parameters such as customer dispersion and operational scale. We find that the tradeoff between leadtime and variety is sometimes nonintuitive and complex.

We consider the operational scheduling or dispatching problem of assigning servicemen to service requests that arrive in real time. The objective is to optimize responsiveness, i.e., to minimize waiting in excess of a promised response time. We study how responsiveness is influenced by modeling decisions and solution methods that arise when solving the dynamic problem by repeatedly solving real-time problems. Most results are derived using a set-partitioning based solution approach, which is shown to perform best among considered alternatives. Our research is based on a large-scale real-life application regarding roadside service assistance.

Previous experimental work showed that newsvendors tend to order closer to mean demand than prescribed by the normative critical fractile solution. A recently proposed explanation for this mean ordering behavior assumes that the decision maker commits random choice errors, and predicts the mean ordering pattern because there is more room to err toward mean demand than away from it. Do newsvendors exhibit mean ordering simply because they make random errors? We subject this hypothesis to an empirical test that rests on the fact that the random error explanation is insensitive to context. Our results strongly support the existence of context-sensitive decision strategies that rely directly on (biased) order-to-demand mappings, such as mean demand anchoring, demand chasing, and inventory error minimization.

Batch (wave) release policies are prevalent in warehouses with an automated sorter, and take different forms depending on how batches released consecutively may overlap downstream in the sorter. Continuous (waveless) release constitutes an emerging alternative recently adopted by several firms. Although that new policy presents several advantages relative to waves, it requires more expensive technology and involves the possibility of congestion-induced collapse (gridlock) at the sorter. Using an extensive data set of detailed warehouse flow information from a leading U.S. online retailer, we first develop a model with validated predictive accuracy for a warehouse operating under waveless release. We then use that model to compute operational guidelines for dynamically managing the main control lever of that policy with the goal of maximizing throughput while keeping the risk of gridlock under a specified threshold. Second, we leverage that model and data set to compare the performance of wave-based and waveless policies through simulation. The best waveless policy yields larger or equal throughput than the best wave-based policy in all scenarios considered, and thus appears to merit some consideration by practitioners.