Extensions to the Guaranteed Service Model for Industrial Applications of Multi-Echelon Inventory Optimization

Victoria G. Achkar, Braulio Brunaud, Héctor D. Pérez, Rami Musa, Carlos A. Méndez, Ignacio E. Grossmann

Abstract

Multi-echelon inventory optimization (MEIO) plays a key role in a supply chain seeking to achieve specified customer service levels with a minimum capital in inventory. In this work, we propose a generalized MEIO model based on the Guaranteed Service approach to allocate safety stock levels across the network at the lowest holding cost. This model integrates several existing and some novel features that are usually present in pharmaceutical multi-echelon supply chains into a single model: review periods, manufacturing facilities, hybrid nodes (nodes with both internal and external demand), minimum order quantities (MOQ), and different service level performance indicators (fill rate and cycle service levels). We include a polynomial regression to approximate fill rates as a possible target measure to set safety stocks. To improve efficiency, we propose a nonlinear programming model to support decision making, which can be reformulated as a Quadratically Constrained Program (QCP), which leads to order of magnitude reductions in computational time. The performance of the model is evaluated by solving illustrative and real-world cases, and is validated with simulation.

Introduction: The Challenge of Multi-Echelon Inventory

Managing inventory across a complex supply chain is a delicate balancing act. Organizations must maintain sufficient stock to ensure high service levels while minimizing the capital tied up in "dead" inventory. This challenge is particularly acute in the pharmaceutical sector, where supply chains are multi-echelon and subject to complex industrial realities.

Traditional academic models often assume demand occurs only at the final stage of the chain. However, in practice, large distribution hubs often function as "hybrid nodes," serving both external customers and downstream internal warehouses. Furthermore, standard models frequently overlook manufacturing lead times and Minimum Order Quantities (MOQs). This research addresses these gaps by extending the Guaranteed Service Model (GSM) to capture these "messy" industrial features without sacrificing solvability.

Methodology: Extending the Guaranteed Service Model

The researchers developed an extended version of the GSM, a classic framework for safety stock placement. The core of this approach involves managing the Net Lead Time (NLT) across various stages of the supply chain.

Net Lead Time (NLT) interpretation and its propagation through a multi-echelon supply chain.

Key innovations in this extended model include:

  • Hybrid Node Modeling: The model explicitly accounts for nodes facing both independent (external) and dependent (internal) demand. This is crucial for accurately calculating demand variance at upstream nodes.
  • Fill Rate Approximation: While many models rely on Cycle Service Level, industry practitioners often prefer Fill Rate (the percentage of total demand met). The authors utilized polynomial regression to create a surrogate model that translates fill rate targets into appropriate safety stock factors.
  • MOQ Integration: The research models how large batch sizes affect inventory. Interestingly, while large MOQs increase cycle stock, they can reduce required safety stock because the large shipments themselves act as a buffer.
  • QCP Reformulation: To address the computational difficulty of the original non-convex Nonlinear Program (NLP), the authors transformed the problem into a Quadratically Constrained Program (QCP). By defining new variables to replace square root terms, the model can leverage high-performance solvers like Gurobi or CPLEX.
Modeling hybrid demand at the plant level, accounting for both independent and dependent demand streams.

Key Insights: The MOQ Buffer and Risk Pooling

One of the most significant findings is the "MOQ Buffer" effect. In some scenarios, a large MOQ can entirely eliminate the need for safety stock. If a node is forced to order ten weeks of supply at once, that massive overshoot often covers demand variability more effectively than a dedicated safety stock pile would.

Additionally, the model confirms the benefits of risk pooling. Holding stock further upstream (at the plant) is often more cost-effective than downstream (at retailers) because the plant can aggregate the uncertainty of all its downstream partners, reducing the total safety stock required across the network.

Results and Validation

The model was evaluated using illustrative examples and real-world industrial cases, including a network with 20 finished goods and 120 raw materials. Results were validated using discrete-time simulation to ensure theoretical stock levels met service targets.

  • Computational Efficiency: In a medium-sized industrial case, a standard NLP solver failed to find a solution within 1000 seconds. The reformulated QCP model achieved optimality in just 3 seconds.
  • Cost Reductions: Compared to a leading commercial MEIO software, this model achieved a 17% reduction in holding costs, significantly outperforming the commercial tool's 10% reduction.
  • Accuracy: Simulation results showed that predicted safety stocks met target service levels with a maximum error of only 0.02 points, validating the reliability of the fill rate approximation.

Conclusion and Future Directions

While the model provides a robust framework for industrial MEIO, some limitations remain. The current iteration assumes demand follows a normal distribution, which may not hold for "lumpy" demand patterns common in rare medicines. Furthermore, the model assumes infinite warehouse capacity.

Future work will focus on extending the model to handle non-normal demand distributions and integrating it with tactical planning tools, such as production sequencing "rhythm wheels," to provide a truly end-to-end supply chain optimization solution.

Summary updated

Abstract

Multi-echelon inventory optimization (MEIO) plays a key role in a supply chain seeking to achieve specified customer service levels with a minimum capital in inventory. In this work, we propose a generalized MEIO model based on the Guaranteed Service approach to allocate safety stock levels across the network at the lowest holding cost. This model integrates several existing and some novel features that are usually present in pharmaceutical multi-echelon supply chains into a single model: review periods, manufacturing facilities, hybrid nodes (nodes with both internal and external demand), minimum order quantities (MOQ), and different service level performance indicators (fill rate and cycle service levels). We include a polynomial regression to approximate fill rates as a possible target measure to set safety stocks. To improve efficiency, we propose a nonlinear programming model to support decision making, which can be reformulated as a Quadratically Constrained Program (QCP), which leads to order of magnitude reductions in computational time. The performance of the model is evaluated by solving illustrative and real-world cases, and is validated with simulation.
BibTeX
@article{achkar2024extensions,
  author = {Victoria G. Achkar and Braulio Brunaud and Héctor D. Pérez and Rami Musa and Carlos A. Méndez and Ignacio E. Grossmann},
  title = {{Extensions to the Guaranteed Service Model for Industrial Applications of Multi-Echelon Inventory Optimization}},
  journal = {Eur. J. Oper. Res.},
  year = {2024},
  doi = {10.1016/j.ejor.2023.08.013},
}
Figure