Order Picking Performance in Manual and Semi-Automated Systems: Empirical Evidence from Grocery Retail Warehouses

This cumulative dissertation examines order picking performance in manual and semi-automated picker-to-parts order picking systems, guided by the theory of swift and even flow (TSEF). Six empirical articles—organized into Part A (manual order picking system) and Part B (semi-automated order picking system)—investigate how physical task design, workload visibility, and the integration of automated guided vehicles (AGVs) affect performance outcomes. Linear mixed-effects regression models constitute the primary methodological approach applied throughout the dissertation, complemented by controlled experiments such as eye-tracking studies and large-scale field experiments with difference-in-differences (DiD) designs. Part A focuses on manual order picking and identifies key drivers of performance. Drawing on task switching theory, evidence from an eye-tracking experiment on switching and search behavior in combination with empirical data from a grocery retail warehouse indicates that switching between pick location heights increases task performance time. Ground-to-chest transitions are found to result in longer task performance times than chest-to-ground transitions, with product weight moderating these effects. Drawing on goal-setting and signaling theories, a large-scale field experiment provides empirical data, further demonstrating that workload visibility only affects work speed when the frequency of updates matches the decision tempo on the shop floor. Lower-performing workers are found to speed up in response to signals, while high-performing workers might slow down if signals are too infrequent. Part B introduces semi-automated order picking into an existing manually operated environment and follows its staged operational rollout. The results show that AGVs improve order picking performance, especially under favorable conditions such as short travel distances, light product weight, and low pick density. However, at the system level, performance follows a U-shaped pattern as moderate robot shares improve flow, while excessive AGV density creates congestion and stop–go patterns. The findings further reveal that cumulative experience improves task performance time in both manual and semi-automated settings, albeit with flatter learning curves and greater variability in AGV-assisted order picking. Finally, AGV-assisted order picking is associated with higher physiological workload, highlighting a trade-off between performance and worker well-being. This cumulative dissertation refines and extends the TSEF for picker-to-parts order picking by specifying mechanisms that affect order picking performance in manual and semi-automated systems. It advances task switching theory by showing that performance losses stem not only from absolute pick heights but also from switching between heights, with asymmetric switching costs moderated by product weight adding a sequencing perspective to ergonomic design. It further extends signaling theory by identifying signal frequency as a critical boundary condition for workload visibility, showing that pacing only stabilizes when updates match the decision tempo. For semi-automated systems, it demonstrates that performance gains depend on the context and follow a U-shaped relationship at the system level, while learning curves are flatter and execution more variable than in manual settings. Importantly, the results show that these performance improvements are accompanied by higher physiological workload, highlighting a trade-off between performance and worker well-being. Conceptually, the dissertation expands the TSEF by incorporating worker well-being as a complementary outcome dimension, emphasizing that sustainable swift and even flow requires aligning speed, stability, and ergonomic relief. From a managerial perspective, the results highlight actionable levers for warehouse operations. Batching and routing should account for height-switching costs while balancing ergonomic strain, supported by clear shelf labeling to reduce disorientation. It is necessary to align workload visibility with the decision tempo, as lower-performing workers benefit from signals whereas high-performing workers might slow down when updates are too sparse. Semi-automated systems should be deployed selectively under favorable conditions, with system-level robot shares calibrated to aisle length, travel distance, and picker density to avoid congestion. AGVs should be scaled gradually, and standardized guidelines are necessary to reduce variability in semi-automated settings. Finally, because AGV-assisted order picking elevates physiological workload, well-being indicators should be integrated into workforce management to trigger recovery breaks and prevent long-term productivity losses.