2243. Optimizing online order fulfillment under capacity constraints: A comparative study of allocation strategies
Invited abstract in session TC-10: Fulfillment Operations II, stream Supply Chain Management and Production.
Thursday, 11:45-13:15Room: H16
Authors (first author is the speaker)
| 1. | Lais Wehbi
|
| Math Department, Vrije Universiteit Amsterdam | |
| 2. | Yasmin Roshandel
|
| Analytics & Optimisation, VU Amsterdam |
Abstract
In e-commerce, the task of fulfilling orders under capacity constraints across multiple distribution centers is both operationally critical and computationally challenging. This study presents a comparative analysis of real time order allocation strategies, examining trade-offs between computational efficiency and fulfillment performance. We evaluate five approaches: a baseline greedy algorithm, an estimated shadow price method, a repeated ILP with predictions and a capacity aware heuristic. We also assess the impact of allowing orders to be split among multiple centers.
Our empirical analysis, conducted using operational data from a large-scale retailer, shows that the capacity-aware heuristic reduces total costs by 15% and unfulfilled orders by 28% compared to the greedy baseline, while maintaining comparable computational speed. The ILP with predictions achieves near optimal performance with a 21% optimality gap, albeit with longer computational runtime. Additionally, we find that allowing order splitting further reduces total cost by 19% compared to the non-split greedy baseline, though is not always applicable in live environments. A novel contribution to our work is a dynamic capacity penalty factor incorporated to the heuristic, which works to prevent premature depletion of centrally located warehouses within reasonable computational speed, a common limitation that is typically observed in myopic strategies.
Our findings offer practical insights for retailers seeking to optimize fulfillment networks, particularly during peak demand periods, demonstrating that substantial efficiency improvements are achievable without complex computational infrastructure.
Keywords
- Linear Programming
- Supply Chain Management
- Mixed-Integer Programming
Status: accepted
Back to the list of papers