تاثیر خطای تخمین در سیاست پذیرش سفارش پویا در محیط های MTO B2B

When swarming demands cause stringent capacity situations, order promising becomes a challenging job. However, a dynamic order admission policy by utilizing the concept of revenue management may find a good way to solve the problem. Unfortunately, the expected profit under different dynamic order admission policies is sensitive to the estimation error of order forecasts. In this paper, the impact of estimation error is investigated under various order structures. The post analysis is performed and shows significant statistical difference among the optimal unbiased DSKP policy, biased DSKP policy, and FCFS policy. The results reveal the robustness and superiority of DSKP policy in most scenarios.

In the make-to-order, business-to-business (MTO B2B) environments, swarming demand caused by seasonal factor or new product launching frequently interfere with daily operations of sales or product managers. These managers sometimes may fail to fulfill all order requests because the planning capacity is insufficient to satisfy all demands in the high season. Tool machine, fashion apparel and shoe making industries frequently face this stringent capacity problem (Franco et al., 1995 and Sridharan, 1998). Evidences show this problem not only bothers the planners of traditional manufacturing industries, but also annoys the leading companies in the semiconductor industries, e.g., TSMC, UMC, and Chartered, etc. (David & Andy, 2007). To handle this problem, Harris, de, and Pinder (1995) suggested applying the concept of revenue management to manufacturing industries. Because their work focused on the manufacturers with continuous production process, the order admission policy proposed by Harris and Pinder is a kind of inventory rationing policy for make-to-stock (MTS) environments. This inventory rationing policy is very similar to the booking limit control in airline, hotel, or car rental industries. Balakrishnan et al., 1996 and Balakrishnan et al., 1999 proposed another heuristic capacity rationing policy for a MTO company which segments orders into two classes by their margins. In each class, the number of orders follows Poisson distributions; therefore, the aggregate demand is an exponential random variable. Their capacity rationing policy is made according to the expected value of the approximated class demand. Later, Barut and Sridharan, 2004 and Barut and Sridharan, 2005 extended their researches for multi-class demands and proposed a revised heuristic. However, some properties of MTO B2B environments had not been captured in previous researches. For instance, MTO B2B companies usually have limited business clients for contact and face finite planning horizon. For each order, its margin and order size are possibly distinct. Furthermore, it is difficult to identify any specific arriving pattern of orders in MTO B2B environments. In order to handle the above characteristics in MTO B2B environments, David and Andy (2007) reformulated the problem as a discrete Markov Decision Problem (MDP), or more specifically, a Dynamic and Stochastic Knapsack Problem (DSKP). It turned out that the optimal policy follows a Markov deterministic policy with revenue-threshold decision rules. Unfortunately, all these researches assume that parameters of order size distributions are known with fixed quantities in their models. These situations rarely happen in the real world. Production and operations managers repeatedly express the view that forecasting is a critical activity since the accuracy of the forecast significantly impacts the quality of operation plans. However, if the forecast has considerable error, even well-conceived plans and excellent operating performance against the plan may result in very disappointing productivity ( Lee & Adam, 1986). Apparently, the more estimation errors in the forecast, the less willingness planners are likely to adopt a dynamic order admission policy in practice. In the researches of Becker et al., 1994 and Balakrishnan et al., 1999, they also concluded that if a model is not sensitive to estimation error, it can make the model more risk-averse and more attractive for the planners. Based on these statements, we intend to investigate the impact of different types of estimation error under various order structure. Three types of estimation error are discussed including error of spikedness, error of mean, and error of deviation. Also, various characteristics of order structure are explored such as margin attractiveness, capacity tightness, number of orders, demand lumpiness, and order size variation. Finally, a post analysis is performed to systematically test the model robustness and effectiveness of DSKP policy against estimation errors in order size parameters. We believe that understanding the impact of estimation error with respect to various order structures can help planners to choose the best policy to be applied. The remainder of this paper is organized as follows: Section 2 will briefly describes an optimal dynamic order admission procedure for MTO B2B corporations based on David and Andy’s work. Next, the structure of our experimental design is presented. Also, the results from our simulation experiments are analyzed in Section 4. At last, conclusions and further researches are followed.

In this paper, we propose the optimal order admission procedure and investigate the impact of estimate errors on the procedure. Three types of estimate errors and five order structure factors constitute a variety of scenarios in practice. The z-tests show that DSKP policy is robust in 64.5% of studied scenarios and is superior to base policy in 90.9% of cases. For estimation error factors, the impacts of error of spikedness and error of mean are significant to DSKP policy. However, DSKP policy is possibly independent on the error of deviation in our study. The phenomenon may exist because the effect of deviation error is averaged out in the risk-neutral model. On the other hand, the order structure effects, high-level attractiveness, tightness, and order size variation may damage the robustness of DSKP but will positively benefit the effectiveness of DSKP compared with base policy. Due to the impact of estimation errors, there is a trade-off between robustness and effectiveness when DSKP policy is applied in a margin-attractive and tight-capacity environment. The increasing number of potential orders will cause less desired effect to DSKP for both its robustness and effectiveness. Demand spikedness may reinforce the negative impact of estimate error on DSKP performance. In conclusion, in most cases of this study, DSKP policy is a good choice in order admission control for MTO B2B environments. In future research, more validation research for DSKP policy is needed if the order structure in the real world is more complicated. Furthermore, the study needs to be reinvestigated if manufacturers are risk-averse. Also, multiple criteria or performance measures, such as capacity utilization and specific client satisfaction, can be considered as the planning objectives to assess the performance of DSKP policy.