Backorder minimization in multiproduct assemble-to-order systems

Yingdong Lu, Jing-Sheng Song, David D. YaoIIE 2005 · DOI

TL;DR

This paper tackles the complex problem of optimizing component inventory in assemble-to-order (ATO) systems to minimize customer backorders. The authors develop novel analytical bounds and efficient algorithms to determine optimal base-stock levels for multiple components and product types under budget constraints, offering practical insights for system design.

What Problem Does This Paper Address?

In today's market, companies are increasingly adopting "assemble-to-order" (ATO) strategies to offer customized products efficiently. In an ATO system, components are stocked, but final products are assembled only after a customer places an order. This creates a challenge: a stockout of any single component can delay or prevent the fulfillment of an entire product order. Managing inventory for numerous components is costly, so companies need to find the optimal balance between inventory investment and customer service (measured by backorders or order fulfillment time). Traditional inventory models designed for mass-production systems are insufficient for the intricate interdependencies and multiple product demands in ATO systems. This paper specifically addresses how to optimally allocate a limited budget across component inventories to minimize a weighted average of backorders across different product types.

Methods

The core of this paper lies in developing analytical tools to approximate and optimize the complex backorder objective function.

  • Bounds and Approximations: The authors derive both lower and upper bounds for the expected number of backorders. These bounds are crucial because the exact calculation of backorders is computationally intractable for multi-product ATO systems.
  • Surrogate Optimization Problems: These bounds are then used to formulate "surrogate" optimization problems. The idea is to optimize a simpler, tractable function that approximates the original, complex objective.
  • "Stack Structure" and Greedy Algorithms: A key contribution is the identification of a "stack structure" in the relaxed (continuous) optimization problems derived from the lower bounds. This structure allows for the development of efficient greedy algorithms to find optimal solutions for these relaxed problems.
  • Marginal Allocation: The algorithms leverage marginal allocation principles, where resources (inventory budget) are allocated incrementally based on their marginal impact on the objective function.
  • Upper Bound Approach: For the upper bound, a common scalar parameter 'a' is introduced and optimized, leading to a separable convex programming problem that can also be solved greedily.

Datasets

The paper uses numerical examples to illustrate the effectiveness of their proposed methods. These examples are based on hypothetical scenarios for an ATO system with six components and six product types, with varying cost structures, demand rates, and budget levels. The paper also references a simplified version of a personal computer example previously studied in Cheng et al. (2002) for comparison.

Key Insights and Ideas

  • The "Equal Fractile Rule" is Suboptimal: The common practice of setting safety factors (k) to be equal across all components (the "equal fractile rule") is shown to be ineffective. The optimal safety factors vary significantly, suggesting that cheaper components or those with higher demand/leadtime characteristics might warrant higher safety stocks.
  • "Stack Structure" is Key to Efficiency: The discovery of a "stack-like structure" in the relaxed optimization problems is a significant analytical breakthrough, enabling the development of efficient greedy algorithms where more complex methods might otherwise be needed.
  • Backorder as a Proxy for Response Time: The paper highlights that minimizing weighted backorders is directly related to minimizing average order response times, making it a relevant and actionable performance metric for customer service.
  • Bounds Provide Practical Solutions: The derived bounds and approximations, when used as surrogates, yield solutions that are very close to optimal, demonstrating their practical utility for real-world inventory management.

Key Results

The paper demonstrates that their proposed lower-bound and upper-bound approaches, coupled with efficient algorithms, provide near-optimal solutions for the complex backorder minimization problem in multi-product ATO systems. Numerical results show that the solutions obtained from their surrogate problems are very close to the true optimal solutions, with relatively small "Relative Error (%)" as indicated in the tables. For instance, in Table 1, the approximation approach often achieves the same "Ave. BackOrder" as the optimal solution, or a very close value. The optimal safety factors derived also show significant variation across components, reinforcing the suboptimality of the equal fractile rule.

Limitations and Future Work

  • Assumptions: The model assumes Poisson demand streams for product types and i.i.d. exponential leadtimes for components. While these are common assumptions, real-world scenarios might involve more complex demand patterns or leadtime distributions.
  • Batch Demand Extension: The paper extends the model to handle batch demands, but the complexity of deriving exact analytical expressions for backorders in this case remains a challenge.
  • Future Work: The authors suggest that their models and algorithms can be extended to support system design decisions, such as evaluating product module alternatives and configurations. Further research could explore more complex demand and leadtime distributions, as well as incorporate other operational factors like setup costs or capacity constraints.

Summary updated

Abstract

We consider a multiproduct assemble-to-order system. Components are built to stock with inventory controlled by base-stock rules, but the final products are assembled to order. Customer orders of each product follow a batch Poisson process. The leadtimes for replenishing component inventory are stochastic. We study the optimal allocation of a given budget among component inventories so as to minimize a weighted average of backorders over product types. We derive easy-to-compute bounds and approximations for the expected number of backorders and use them to formulate surrogate optimization problems. Efficient algorithms are developed to solve these problems, and numerical examples illustrate the effectiveness of the bounds and approximations.
BibTeX
@inproceedings{lu2005backorder,
  author = {Yingdong Lu and Jing-Sheng Song and David D. Yao},
  title = {{Backorder minimization in multiproduct assemble-to-order systems}},
  booktitle = {IIE},
  year = {2005},
  doi = {10.1080/07408170590961139},
}
Figure