Build Your Policy

This challenge is highly accessible for participation—read the problem statement below, register here, download the prepared code, and build your policy!

* The winners and runners-up are awarded with, respectively, 500 and 200 Euro prize money.

** The winners will be invited to submit their solution as part of a (collaborative) paper to the International Journal of Production Economics.

Introduction

At ASML, efficient production and inventory planning is critical to ensuring competitive operational performance, yet it remains very challenging due to the complexity of our high-tech low-volume manufacturing environment. Demand is highly uncertain and non-stationary, often influenced by business cycles, while extensive production lead times necessitate planning based on long-term forecasts. A key characteristic of our engineering and manufacturing process is modular product design, which drives flexibility and efficiency. The resulting component commonality across multiple end items in the product portfolio enhances adaptability, but it significantly complicates the optimization of production and inventory levels due to the general-structure network. At the same time, tight production capacity constraints across multi-tier supplier networks add to the complexity of decision-making. Given the extreme cost of products and resources in this high-tech industry, even slight inefficiencies in inventory planning can have significant financial consequences, reinforcing the need for efficient solution methodologies. 

While the problem considered in this research challenge is not directly derived from ASML’s operational setting, it captures the fundamental challenges inherent to planning challenges in high-tech low-volume manufacturing. The developed methodologies and insights, therefore, are expected to enhance practical planning processes at ASML. 

Although the problem setting is complex, the core of the challenge is straightforward. You are asked to develop a policy that satisfies external customer demand as efficiently as possible, minimizing the total costs incurred in the supply chain. Here, your policy should prescribe the production-inventory decisions for all manufacturing steps given any state of the supply chain. 

Assessment Procedure

Assessment of the submissions will be conducted by the Committee using a blinded review process based on the following criteria:

• the quality, generalizability, scalability, and novelty of the solution approach;

• the significance of the findings and insights.

Timeline

• 📢 February 24, 2025: Launch of the ASML Stochastic Production-Inventory Planning Challenge

• 📢 July 11, 2025: Deadline submission

• 📢 August 1, 2025: Announcements finalists

The finalists will present their findings during the ISIR Summer School 2025 (exact date to be determined).

Registration Guidelines

Participants can register by filling out the form here. Upon registration, participants will be granted access to the repository with prepared code and data. The group size is limited to three members.

Submission Guidelines

Submissions can be done by sending an e-mail to challenge@isir2025.com, including the following:

• a title page with names, e-mail addresses, and affiliations of group members;

• a report (without author identifiers) exhibiting the solution approach, results and insights;

• the developed code (without author identifiers).

Committee

The ASML Stochastic Production-Inventory Planning Challenge Committee consists of:

• Tjum van Dijck (ASML; Tilburg University)

• Tijn Fleuren (ASML, Head of Planning & Delivery Strategy Research; Tilburg University)

• Maarten Hendriks (ASML, Head of Planning & Delivery Strategy; Tilburg University)

• Tarkan Temizöz (Eindhoven University of Technology).

• Willem van Jaarsveld (Eindhoven University of Technology)

Important Notes

• Although the case is motivated by the practice, data are anonymized for reasons of confidentiality.

• The goal is to develop an inventory planning policy, which can be directly implemented into an already fully coded simulation framework. The simulation is pre-built, requiring only the integration of your policy without extensive additional coding.

Problem Statement

The primary goal is to develop an effective inventory planning policy for managing the entire supply chain depicted below. This supply chain consists of nine nodes, each handling its own set of components. There are three end-products, i.e., C12, C13, and C14, each subject 2 to uncertain customer demand. These products are assembled at the final assembly stage (node 9) from three components, which are themselves produced from externally sourced materials. Each production and assembly process involves a deterministic lead time, indicated by the numbers on the arcs. These lead times are governed by either supply agreements or internal production capabilities. To initiate an assembly process, all required materials must be available, necessitating synchronization of incoming lead times. For example, assembling C12 takes two periods, meaning that the lead time from C9 and C11 to C12 is also two periods. Additionally, nodes have production capacity constraints, limiting the total output per period for that node—e.g., the aggregate production of components C6, C7, and C8 per period is restricted.

The supply chain incurs a per period unit inventory holding cost for maintaining positive on-hand inventory for each component. To ensure high service levels, any unmet demand results in a per period unit backlog penalty for each end-product. The uncertain external customer demand exhibits time correlation, for which we assume a stylized model. The aggregate demand for the three products, representing the overall market, follows an autoregressive (AR) process, and is stochastically allocated among the three end-products, whose demands are rounded to obtain integer quantities. Given the low-volume setting, the decision variables are also integer-valued. All input parameters can be found in the provided data. 

The objective is to establish an inventory planning policy that minimizes total expected costs over a five-year planning horizon. This policy prescribes (optimal) production-inventory planning decisions for all components across the supply chain in each time period. 

Repository Details

The framework (accessible upon registration here) has a pre-build simulation that runs the planning horizon for any given policy. The supply_chain_config file processes the input data and translates it into the necessary supply chain attributes. All time-period dynamics are handled in the dynamics file, which also includes a check of whether the prescribed actions from the policy are allowed. There are two main functions to evaluate our policy, both located in the evaluate file: one tests a single policy over a given number of trajectories and periods, while the other compares multiple policies. You are asked to implement your policy class in the policy file, which must contain a set_action function that takes the state as input (see state file) and returns the production quantity for each component in the supply chain.