We, as the Zalando team BART, are excited to share our latest research paper, describing the optimization problem of order batching and picking in Zalando's warehouses. In this paper (preprint available on arxiv), we formally introduce our proposed order batching problem and provide benchmark instances, two baseline algorithms, and a solution validation tool, all made publicly available on GitHub. Our goal is to provide insights to the research community on planning and optimizing the warehouse order picking process in large-scale warehouses, such as Zalando's.

The Underlying Optimization Problem

Zalando Tech Logistics is responsible for creating the software that manages all Zalando warehouses and their processes. Team BART, part of Zalando's Logistics Algorithms department, provides the decision-making algorithms for order batching and picking. These decisions can be broken down into four parts:

1. Order Selection: Which customer orders are processed next?
2. Item Allocation: Which warehouse items are used to fulfill a selected order?
3. Batching: Which selected orders are picked together?
4. Picking: How are batches split up into pick tours?

Traditionally, these decision problems are considered individually and solved using simplified rules. For example, order selection could be done using a first-in-first-out approach. However, our experience and analysis of batching algorithms have shown that a purely sequential approach is far from optimal. While there has been some research on these problems in the literature, there is no closed formulation, to the best of our knowledge, that encapsulates all four problems into one. And this is exactly what we aim to achieve with our paper: We combine all of the four problems into one, named Joint Order Selection, Allocation, Batching and Picking.

Benchmark Instances

To ensure a clear understanding of the problem statement, we provide benchmark instances for the Joint Order Selection, Allocation, Batching, and Picking Problem. These instances allow anybody interested to immediately try out their ideas for solving this problem. Additionally, we share the implementation of two baseline algorithms described in the paper.

Outlook

We aim to stimulate academic discussion around the Joint Order Selection, Allocation, Batching, and Picking Problem. We believe there are practitioners and researchers interested in this type of optimization problem. By providing benchmark instances, we hope to establish a standard definition that can be easily adapted for further research.

Publishing this problem formulation also allows us to share insights on how we are solving this problem at Zalando. We look forward to sharing more in our next publication. In the meantime, we welcome any feedback and collaboration from the community: Feel free to share your feedback via GitHub.

We're hiring! Do you like working in an ever evolving organization such as Zalando? Consider joining our teams as a Machine Learning Engineer!