Operations

Route optimization

Computing the best sequence and assignment of stops across a fleet against live constraints: time windows, vehicle capacity, cost, and traffic. Most of the complaint operators have about their optimizer is not solver quality, it is constraint blindness: a route scores well and still cannot be run because the vehicle, the certification, or the service time was never in the model. Solved once at plan time it decays by mid-morning, so Nash re-solves it continuously as the day changes.

Where this fits

The day-to-day vocabulary of running deliveries. Routes, waves, drops, manifests, ETAs, SLAs. These are the units operators reason in, and the inputs every downstream model needs.

Related terms

Delivery costs: The all-in cost to move one delivery from origin to destination, including labor, fuel, vehicle, packaging, exception-handling overhead, and any per-job carrier fee. The denominator behind every cost-per-stop and cost-per-order target. Compresses with route density, batching, and carrier mix.
Continuous re-optimization: Re-solving the plan as the day unfolds instead of freezing it at dispatch. A cancelled order, a late driver, or a new pickup reshapes the affected routes in place, so the plan tracks reality rather than the morning's assumptions.
Delivery success rate: The share of orders that complete on the first attempt, on time, with valid POD, and without exception. The cleanest single metric for whether dispatch is working. Above 99% is best-in-class; most operations sit lower.
Geofencing: A virtual boundary drawn around a place. Crossing it fires an event: arrival at a stop, entry to a zone, a status change, a customer notification. The trigger that turns raw location into an operational signal.
Manual Intervention: Any action a human takes in the dispatch loop that automation didn't make on its own: re-routing a stuck order, calling a driver, escalating an exception, manually reassigning capacity. Manual-intervention rate is a closely-watched KPI for autonomic systems; the goal is for it to drop toward zero as the system learns the operation.