Cross-Border Fulfillment Center Selection via Doubly Robust Estimation and Linear Programming
Organizations: Cross Border Tech & Science, Amazon · Agentic AI DevOps, Amazon
Abstract
We introduce the cross-border (XB) fulfillment center (FC) Selection Model: a multi-stage optimization framework combining forecasting, doubly robust estimation, and linear programming to select optimal source FCs. For each source-destination shipping arc, the objective is to choose a subset of source-country FCs that minimize delivery time and cost while maximizing profit and shipped units under a fixed domestic network. We evaluate the model on three production-scale international shipping corridors for a large e-commerce company using 12 months of shipment data, employing both aggregate offline replay and shipment-level counterfactual nearest-neighbor matching. Across all shipping corridors, the XB FC Selection Model achieves 8.5-51.1% reductions in delivery time and 4.4-48.3% reductions in logistics cost relative to human expert-defined FC selection policies, with up to 6.5 percentage-point improvements in 5/5 target compliance (5 days, $5 USD). Pareto analysis shows the model consistently dominates single-objective ranking strategies and unconstrained LP on the efficiency-reallocation frontier, with all results statistically significant.
Figures & tables
| US SG | US DE | US AU | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Pre-Deployment | Post-Deployment | Pre-Deployment | Post-Deployment | Pre-Deployment | Post-Deployment | |||||||||||||
| Strategy | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units |
| Profit Rank | 0.092 | -0.104 | -0.050 | 0.080 | -0.132 | -0.120 | 0.405 | -0.230 | -0.273 | 0.110 | -0.197 | -0.206 | 0.316 | -0.067 | -0.068 | 0.193 | -0.043 | -0.040 |
| Time Rank | -0.477 | -0.452 | -0.463 | -0.113 | -0.627 | -0.628 | -0.157 | -0.726 | -0.735 | -0.640 | -0.850 | -0.865 | 0.378 | -0.851 | -0.862 | 0.476 | -0.857 | -0.869 |
| ATE-Profit Rank | -0.015 | -0.111 | -0.048 | 0.013 | -0.131 | -0.126 | 0.274 | -0.460 | -0.471 | 0.010 | -0.436 | -0.439 | 0.626 | -0.375 | -0.374 | 0.419 | -0.316 | -0.292 |
| ATE-Time Rank | -0.412 | -0.162 | -0.142 | -0.274 | -0.414 | -0.421 | -0.379 | -0.691 | -0.667 | -0.627 | -0.706 | -0.690 | -0.187 | -0.582 | -0.589 | -0.094 | -0.596 | -0.584 |
| SG | DE | AU | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Evaluation | Metric | Actual | Expert | LP | XB FC | Actual | Expert | LP | XB FC | Actual | Expert | LP | XB FC |
| Same Product | Mean Time (Days) | 3.21 | 2.74 | 1.36 | 1.18 | 5.42 | 5.87 | 3.21 | 2.76 | 4.68 | 2.31 | 2.47 | 1.84 |
| Mean Cost (USD) | 8.47 | 8.31 | 7.14 | 6.52 | 13.76 | 13.21 | 11.84 | 10.27 | 14.63 | 12.46 | 14.17 | 11.92 | |
| % Meeting 5/5 | 54.72 | 47.35 | 51.86 | 59.41 | 11.64 | 6.27 | 9.18 | 12.83 | 6.02 | 6.47 | 4.16 | 8.21 | |
| Sample Size (N) | 1,842 | 1,842 | 1,842 | 1,842 | 1,176 | 1,176 | 1,176 | 1,176 | 963 | 963 | 963 | 963 | |
| Nearest Neighbor | Mean Time (Days) | 1.38 | 1.23 | 0.72 | 0.70 | 3.03 | 3.23 | 2.00 | 1.85 | 2.53 | 0.93 | 1.14 | 1.13 |
Appendix figures & tables12 assets
Supplementary material from the paper’s appendix.
Appendix
| US SG | US DE | US AU | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Pre-Deployment | Post-Deployment | Pre-Deployment | Post-Deployment | Pre-Deployment | Post-Deployment | |||||||||||||
| Strategy | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units | Time | Cost | Units |
| Expert | -0.959 | -0.452 | -0.331 | -0.983 | -0.462 | -0.328 | -0.955 | -0.641 | -0.582 | -0.957 | -0.628 | -0.609 | -0.991 | -0.519 | -0.441 | -0.992 | -0.456 | -0.378 |
| Profit Rank | -0.957 | -0.507 | -0.350 | -0.981 | -0.531 | -0.408 | -0.947 | -0.713 | -0.670 | -0.953 | -0.703 | -0.693 | -0.988 | -0.553 | -0.480 | -0.991 | -0.479 | -0.402 |
| Time Rank | -0.980 | -0.731 | -0.672 | -0.986 | -0.765 | -0.702 | -0.980 | -0.924 | -0.912 | -0.986 | -0.941 | -0.945 | -0.988 | -0.925 | -0.919 | -0.989 | -0.919 | -0.912 |
| ATE-Profit Rank | -0.962 | -0.514 | -0.352 | -0.982 | -0.532 | -0.411 | -0.949 | -0.788 | -0.755 | -0.958 | -0.794 | -0.786 | -0.985 | -0.695 | -0.645 | -0.989 | -0.627 | -0.560 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | Expert | ATE-Profit Rank | 0.717 | 0.569 | 0.888 | 0.042 |
| Leg Cost (1000 USD) | Expert | ATE-Profit Rank | -66.354 | 0.001 | 0.000 | -1.775 |
| Shipped Units (1000 Units) | Expert | ATE-Profit Rank | -9.257 | 0.052 | 0.053 | -0.626 |
| Time Units (Days) | Expert | ATE-Time Rank | -26.363 | 0.001 | 0.005 | -1.021 |
| Leg Cost (1000 USD) | Expert | ATE-Time Rank | -162.802 | 0.001 | 0.000 | -1.426 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | Expert | ATE-Profit Rank | 14.461 | 0.109 | 0.092 | 0.690 |
| Leg Cost (1000 USD) | Expert | ATE-Profit Rank | -213.611 | 0.008 | 0.000 | -2.286 |
| Shipped Units (1000 Units) | Expert | ATE-Profit Rank | -24.708 | 0.008 | 0.000 | -2.219 |
| Time Units (Days) | Expert | ATE-Time Rank | -113.836 | 0.008 | 0.000 | -2.245 |
| Leg Cost (1000 USD) | Expert | ATE-Time Rank | -347.407 | 0.008 | 0.000 | -2.303 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | Expert | ATE-Profit Rank | 0.000 | 0.002 | 0.000 | 2.359 |
| Leg Cost (1000 USD) | Expert | ATE-Profit Rank | -0.753 | 0.002 | 0.000 | -4.143 |
| Shipped Units (1000 Units) | Expert | ATE-Profit Rank | -0.080 | 0.002 | 0.000 | -4.061 |
| Time Units (Days) | Expert | ATE-Time Rank | -0.000 | 0.006 | 0.005 | -1.170 |
| Leg Cost (1000 USD) | Expert | ATE-Time Rank | -1.284 | 0.002 | 0.000 | -3.453 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | XB FC Model | ATE-Profit Rank | 26.475 | 0.000 | 0.001 | 1.216 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Profit Rank | -28.615 | 0.001 | 0.000 | -1.705 |
| Shipped Units (1000 Units) | XB FC Model | ATE-Profit Rank | -9.578 | 0.000 | 0.000 | -2.104 |
| Time Units (Days) | XB FC Model | ATE-Time Rank | -0.606 | 0.910 | 0.710 | -0.110 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Time Rank | -125.063 | 0.000 | 0.004 | -1.033 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | XB FC Model | ATE-Profit Rank | 108.042 | 0.008 | 0.000 | 2.182 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Profit Rank | 38.362 | 0.109 | 0.035 | 0.923 |
| Shipped Units (1000 Units) | XB FC Model | ATE-Profit Rank | 4.750 | 0.055 | 0.034 | 0.933 |
| Time Units (Days) | XB FC Model | ATE-Time Rank | -20.254 | 0.008 | 0.000 | -2.397 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Time Rank | -95.434 | 0.008 | 0.001 | -2.029 |
| Baseline | Strategy | Mean Difference | Wilcoxon p-value | T-test p-value | Cohen’s d | |
|---|---|---|---|---|---|---|
| Metric | ||||||
| Time Units (Days) | XB FC Model | ATE-Profit Rank | 0.000 | 0.002 | 0.000 | 2.975 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Profit Rank | -0.003 | 0.375 | 0.873 | -0.052 |
| Shipped Units (1000 Units) | XB FC Model | ATE-Profit Rank | -0.001 | 0.846 | 0.678 | -0.136 |
| Time Units (Days) | XB FC Model | ATE-Time Rank | -0.000 | 0.432 | 0.328 | -0.327 |
| Leg Cost (1000 USD) | XB FC Model | ATE-Time Rank | -0.534 | 0.002 | 0.000 | -2.197 |
| Metric | Comparison | t-stat | p-value |
| Time (Days) | |||
| ATE vs Actual | -72.74 | ||
| LP vs Actual | -70.26 | ||
| Expert vs Actual | -72.74 | ||
| ATE vs Expert | -78.79 | ||
| LP vs Expert | -75.21 | ||
| Metric | Comparison | t-stat | p-value |
| Time (Days) | |||
| ATE vs Actual | -62.50 | ||
| LP vs Actual | -53.28 | ||
| Expert vs Actual | -62.50 | ||
| ATE vs Expert | -83.94 | ||
| LP vs Expert | -72.97 | ||
| Metric | Comparison | t-stat | p-value |
| Time | |||
| ATE vs Actual | -514.73 | ||
| LP vs Actual | -507.15 | ||
| Expert vs Actual | -514.73 | ||
| ATE vs Expert | 174.52 | ||
| LP vs Expert | 177.90 | ||