United Parcel Service (UPS)

United Parcel Service operates one of the largest delivery networks in the world, with approximately 100,000 delivery vehicles and drivers completing roughly 19 million deliveries daily in the U.S. alone as of 2015. Route planning was historically a manual task: each driver followed a fixed route designed by supervisors using institutional knowledge and paper maps. The legacy approach was unable to account for real-time variables — traffic, customer preferences, new delivery stops, weather — leaving significant inefficiency in the system. Fuel represented one of UPS's largest operating costs, and driver labor accounted for roughly two-thirds of delivery cost per package. Even small reductions in miles driven per day compounded to enormous savings at UPS's scale: the company estimated that saving one mile per driver per day across the U.S. fleet would reduce costs by $50 million annually.

UPS began developing ORION (On-Road Integrated Optimization and Navigation) in 2003 and started rolling it out operationally in 2012. Full U.S. deployment was substantially complete by 2016. Key actions:

• Algorithm development: UPS built a proprietary routing algorithm using operations research techniques — including constraint-solving and machine learning — that generates optimized daily route sequences for each driver based on package volume, delivery time windows, fuel cost, and road network conditions.
• Data integration: ORION ingested data from UPS's package scanning systems, customer delivery instructions, GPS telemetry from vehicles, and historical traffic patterns to dynamically adjust routes throughout the day.
• Driver adoption program: UPS trained approximately 55,000 U.S. drivers on ORION over a four-year rollout. The company addressed driver resistance by demonstrating fuel savings and emphasizing that the system provided suggestions rather than mandates on genuinely novel conditions.
• Right-turn bias: A well-publicized optimization heuristic embedded in ORION minimizes left-hand turns (which require idling across oncoming traffic), reducing idle time and intersection wait times across the fleet.
• Continuous refinement: ORION's models are retrained continuously as new delivery data accumulates, improving accuracy over time.