Understanding variations in the order picking process: Data-driven and simulation-based approaches

Abstract

Warehousing plays a pivotal role in modern supply chains, especially in the face of growing e-commerce demands and consumer expectations for fast and reliable delivery. Among warehouse operations, order picking is the most labour-intensive and costly, and, despite technological advances, it is still largely performed manually. Many researchers have focused on improving order picking efficiency, and most of these optimisation models assume idealised, robotic compliance with system-generated instructions. This assumption overlooks one of the primary advantages of employing humans, their autonomy and ability to react to real-time conditions. Surprisingly, little research has focused on understanding the actual behaviour of order pickers, especially by using real-world data to reveal behavioural patterns. This disconnect between model assumptions and real-world behaviour motivates the research in this dissertation. More specifically, this thesis explores the nature and implications of deviations made by order pickers, instances where pickers diverge from prescribed instructions, using data from a real-world warehouse and simulation-based modelling.

The thesis makes two main contributions. First, it develops a structured approach to detect and analyse deviations in order picking, specifically, order deviations (when pickers change the prescribed pick sequence) and time deviations (unexpected variation in task duration). This methodology is applied to a rich dataset from a real-world warehouse, covering over two years of operations and over 500 workers. The analysis uncovers that order deviations occur in almost 5% of the picking tours and that order picking time is heavily right-skewed. The latter implies that while there are limited observations where order pickers perform their tasks faster than expected, there are many occurrences of significant delays. Through a combination of exploratory analysis and mixed-effects regression modelling, the study identifies variables associated with both deviation types and provides new empirical insights into the behavioural patterns of human order pickers.

Second, the thesis examines the operational impact of routing deviations through an agent-based simulation model that reflects modern warehouse environments. Researchers always assume that the optimal routing policy has superior performance over routing heuristics. However, this belief needs to be tested in a realistic environment where order pickers may not always strictly adhere to the routing instructions given to them. In our study, we simulate routing heuristics in storage environments in which they are most effective, a practice common in real-world settings, and compare them to the optimal policy in the same storage environment. The results show that intuitive heuristics, when aligned with their most efficient storage environment, offer comparable performance to optimal routing while reducing cognitive demands on workers. Our findings nuance the belief that optimal routing should always be preferred, and prove that human-centric modelling approaches are essential to recognise human behaviour and still achieve high operational performance.

Together, these contributions advance our understanding of human behaviour during the order picking process and offer practical insights for designing decision support systems that are both efficient and behaviourally realistic. Furthermore, the thesis provides valuable insights for managers by shedding light on how order pickers behave in practice, helping bridge the gap between system design and day-to-day operations.