Navigating uncertainty in Amazon's middle-mile network

Summary

Amazon's middle-mile logistics network coordinates tens of millions of packages across hundreds of facilities while managing subtle demand variations and major operational disruptions. To solve this mixed-integer optimization challenge without combinatorial explosion, Amazon models routes at coarse time resolutions augmented by precomputed 15-minute timing bounds. The system uses Monte Carlo methods alongside a graph attention network featuring interconnected site graphs and origin-destination graphs to capture complex spatial correlations and flow interdependencies. By generating synthetic flow permutations, planners stress-test candidate network designs against hundreds of plausible scenarios to prioritize routing optionality. This risk-aware modeling approach preserves network flexibility, absorbs severe shocks, and yields potential cost savings while safeguarding delivery promises.

Context

Amazon's middle-mile network must move tens of millions of products across hundreds of facilities with limited capacities and strict operating schedules. Network planning requires solving mixed-integer optimization problems with millions of binary and continuous decisions, which explode combinatorially. Real-world planning faces both structural disruptions, such as severe weather or facility outages, and daily microvariations in demand and travel times. Standard methods that attempt to make solutions robust against every individual failure scenario fail to scale computationally.

Approach / What changed

Amazon developed a network design tool and risk-aware Monte Carlo modeling framework prioritizing operational optionality. The optimization tool operates at coarse time intervals while enforcing precomputed 15-minute timing bounds for feasible departures and arrivals. To capture complex flow dependencies, Amazon uses a graph attention network featuring two interconnected graphs: a site graph capturing facility proximity and shipping routes, and an origin-destination graph capturing interdependent package flows between facility pairs. Planners generate synthetic origin-destination flow permutations to stress-test candidate network configurations across hundreds of plausible operational scenarios.

Takeaways

• Even when isolating only demand variability, optimizing middle-mile planning for uncertainty delivers up to 0.5% in potential efficiency savings.
• Using precomputed 15-minute timing bounds within coarse-resolution optimization models prevents combinatorial explosion while keeping truck schedules operationally realistic.
• A dual-graph architecture combining a physical site graph with an origin-destination graph captures demand correlations and traffic competition that physical proximity alone cannot reveal.

Related reading

• Amazon · May 5, 2026

  How mechanism design theory helps optimize Amazon-vendor collaboration

  Amazon and its vendors face an asymmetric information problem when coordinating purchase orders across fulfillment centers because neither party shares proprietary cost, capacity, or demand data. To solve this coordination gap without exposing sensitive information, Amazon's Supply Chain Optimization Technologies organization designed Flo Pro by integrating the Vickrey-Clarke-Groves mechanism with the Consensus Planning Protocol. The distributed optimization framework uses the alternating-direction method of multipliers to coordinate iterative agent responses, calculating cost-benefit transfers across rolling six-week planning horizons. In lower-dimensional settings, Amazon also evaluated a transparent menu-of-contracts approach where vendors select optimal plans priced at Amazon's incurred costs. A nine-week pilot with a consumer-product manufacturer confirmed that Flo Pro successfully aligns incentives and achieves measurable cost savings.

  Dirk Bergemann

• Slack · May 28, 2026