On picking operations in e-commerce warehouses: Insights from the complete-information counterpart

Catherine Lorenz,Alena Otto,Michel Gendreau

2024-10-18

Abstract:Major players in e-commerce process dynamically incoming orders in real-time and already use advanced anticipation techniques, like AI, to predict characteristics of future orders. However, at the warehousing level, there are still no unambiguous recommendations on integrating anticipation with intelligent online optimization algorithms, nor an unbiased benchmark to assess the improvement potential of advanced anticipation over myopic techniques, as optimal online solutions are usually unavailable. In this paper, we compute and analyze Complete-Information Optimal policy Solutions (CIOSs), of an exact perfect anticipation algorithm with full knowledge of future customer orders' arrival times and items, for picking operations in picker-to-parts warehouses. We provide analytical properties and leverage CIOSs to uncover decision patterns that enhance simpler algorithms, improving both makespan (costs) and order turnover (delivery speed). Using a metric similar to the gap to optimality, we quantify the gains from optimization elements and the improvements remaining for advanced anticipation mechanisms. To compute CIOSs, we design the first exact algorithm for the Order Batching, Sequencing, and Routing Problem with Release Times (OBSRP-R), based on dynamic programming. Our analysis advises on the largely overlooked intervention - the dynamic adjustment of started batches, and the strategic relocation of an idle picker towards future picking locations. The former affected over 60% of CIOS orders and triggered consistent improvements across warehousing policies. The latter occurred before 39%-62% CIOS orders and could decrease a myopic policy's gap to optimal turnover by 4.3% on average (up to 14%). Notably, we challenge the debated concept of strategic waiting, revealing why the latter resembles an "all-in" gamble and harms both makespan and order turnover when intervention is allowed.

Optimization and Control

What problem does this paper attempt to address?

This paper attempts to solve the problem of optimizing order - picking operations in e - commerce warehouses. Specifically, the paper focuses on how to balance the need for fast and low - cost delivery when processing dynamically arriving orders in real - time. Although advanced prediction techniques (such as AI) are currently used to predict certain characteristics of future orders, at the warehousing level, there is still a lack of clear guidance and unbiased benchmarks on how to integrate these predictions with intelligent online optimization algorithms and how to evaluate the improvement potential of advanced prediction techniques relative to myopic optimization techniques.

To fill this gap, the paper calculates and analyzes Complete - Information Optimal policy Solutions (CIOSs), that is, the optimal solutions for picking operations when the arrival times of all future customer orders and the ordered items are known. Through this method, the authors reveal the decision - making patterns under the perfect prediction strategy, which can be used to enhance simple algorithms while improving picking time (i.e., cost) and average order turnaround time (i.e., delivery speed).

The main contributions of the paper include:
1. **Designed a special dynamic programming algorithm** for solving the Order Batching, Sequencing and Routing Problem with Release Times (OBSRP - R), which is the first exact algorithm for OBSRP - R proposed in the literature.
2. **Provided an analysis of CIOSs** and extracted actionable suggestions from it, such as dynamically adjusting batches that have already started and strategically relocating idle pickers to future picking positions, which can significantly improve picking efficiency.
3. **Challenged a controversial concept in warehousing literature - strategic waiting**. Through the analysis of CIOSs, the paper explains why strategic waiting is not only as unreliable as an "all - in" gamble but also damages picking time and order turnover rate when intervention is allowed.

In conclusion, by calculating and analyzing the optimal strategy under complete information, this paper provides a new set of optimization methods and insights for order - picking operations in e - commerce warehouses, aiming to improve picking efficiency and reduce costs.