UPS

UPS operates one of the world's largest ground delivery networks, with drivers completing millions of package deliveries daily across the U.S. Managing 55,000 active routes using legacy static planning created compounding inefficiencies: routes were calculated days in advance and could not adapt to real-time conditions such as traffic, weather, or last-minute stop additions. In logistics and freight, even marginal per-route inefficiencies multiply dramatically at fleet scale. Fuel is among the largest operational cost lines for any major carrier, and empty or suboptimal miles account for roughly 35% of total U.S. truck miles driven — meaning the status quo imposed substantial financial waste and unnecessary carbon output with every dispatch cycle.

UPS developed ORION (On-Road Integrated Optimization and Navigation), a proprietary machine learning platform engineered specifically for fleet-level route optimization. Unlike static routing tools, ORION dynamically recalculates all 55,000 daily delivery routes using real-time inputs: traffic conditions, package volume fluctuations, time-window constraints, and stop-sequence requirements. The system processes a combinatorial problem of enormous complexity — one that conventional commercial routing software cannot handle at this scale. ORION was deployed progressively across UPS's U.S. driver network, requiring deep integration with existing dispatch, telematics, and logistics management systems. Because the platform is purpose-built rather than adapted from off-the-shelf software, it can execute full fleet-wide re-optimization on a daily cadence without degrading delivery performance.

ORION generates over $400 million in annual cost savings, primarily through reductions in miles driven and fuel consumption across the entire fleet. Key outcomes include:

• 55,000 routes dynamically optimized every day, replacing static multi-day planning cycles
• Measurable reduction in empty and inefficient miles, which represent approximately 35% of total U.S. truck mileage industry-wide
• On-time delivery performance maintained despite the shift to fully dynamic routing
• Simultaneous improvement in UPS's carbon footprint, as fewer miles driven directly reduces fleet emissions

The dual financial and environmental impact has made ORION one of the most widely cited examples of production-scale ML deployment in logistics.

• Daily route recalculation captures far more value than static weekly planning — optimization frequency is as important as model quality
• At fleet scale, cost reduction and ESG improvement are not in conflict: eliminating wasted miles achieves both simultaneously
• Problems of this combinatorial complexity require purpose-built ML platforms; off-the-shelf routing tools cannot handle tens of thousands of interdependent real-time variables
• Integration with existing dispatch and telematics infrastructure must be scoped from the outset — progressive rollout only works if the data pipelines are production-ready
• Proprietary platform development is justified when the optimization problem is core to competitive advantage and no commercial solution matches the required scale