Last Mile Delivery ETA Predictor

Last-mile route time decomposes into four terms that behave quite differently. Driving time is distance divided by an effective speed, where the traffic factor acts as a multiplier on free-flow speed: urban van speeds of around 30 km/h in light conditions fall to roughly 20 km/h in moderate congestion and below 10 km/h in severe congestion. Service time is the per-stop dwell at the door, covering the walk from the vehicle, any wait, capture of a signature or photograph, and the return. Access and parking time is tracked separately because it is vehicle dependent, close to zero for a bicycle and several minutes for a rigid truck in a city centre. Failed first attempts add a cost charged twice, once for the wasted visit and again when the parcel is reattempted on a later route. The economics turn on stop density rather than total distance. A dense urban round with drops 100 to 200 metres apart spends most of its time in service and parking, so adding stops is comparatively cheap, whereas a rural route with several kilometres between drops is dominated by driving and scales close to linearly. That is why deliveries per hour vary so widely, from roughly 3 to 5 in rural territory to 12 to 15 on a well-sequenced urban round, and why the last mile is routinely cited as the largest single component of parcel delivery cost. This model is deterministic and treats the route as one aggregate. It does not solve the sequencing problem, assumes the distance entered is an already-optimised route length rather than straight-line distance, ignores vehicle capacity limits that force a return to depot, and applies a flat 8 percent failure rate where the real figure depends on whether the address is residential, whether a time window was communicated, and whether an unattended drop is authorised.

Formal study of route planning begins with the vehicle routing problem, introduced by George Dantzig and John Ramser in 1959 in a paper on the truck dispatching problem, which generalised the travelling salesman problem to a fleet of vehicles serving demand points from a central depot. Dantzig and Ramser framed it concretely as gasoline delivery to service stations. The problem is NP-hard, so practical systems have always relied on heuristics rather than exact solutions, and the savings algorithm published by Clarke and Wright in 1964 remained a working method for decades and still appears as a construction step inside modern solvers. Commercial application waited for cheap satellite positioning and vehicle telematics. UPS began deploying its ORION route optimisation system across United States routes in the early 2010s and completed the rollout to domestic drivers around 2016, reporting savings on the order of a mile per driver per day, which is substantial at fleet scale. Pressure on the last mile intensified with the growth of e-commerce parcel volume, and again during the 2020 pandemic surge, which pushed carriers toward narrow delivery windows, parcel lockers, and micro-fulfilment hubs feeding cargo bikes in dense city centres. ETA prediction shifted over the same period from static planned schedules to models trained on historical telematics and live traffic feeds, which is what now allows carriers to publish a tight arrival window to the recipient on the day of delivery.