US Regional Last-Mile Delivery Network
34% lower delivery costs with ML demand forecasting and route optimization.
The client is a regional last-mile delivery company in the mid-Atlantic US, serving about 140 e-commerce and retail clients. They run 85 vehicles out of three distribution hubs, handling an average of 9,400 deliveries per day across parts of Pennsylvania, New Jersey, Delaware, and Maryland.
The company had grown fast — 40% year-over-year volume increase over two years — on e-commerce acceleration. But the operational model that worked at 5,000 deliveries per day was failing at 9,400. The core problem was planning. Each morning, route planners at the three hubs manually built delivery routes using a legacy routing tool and personal knowledge of the territory. The process started at 4:30 AM and had to be done by 6:15 AM when drivers started loading. The planners were experienced and good, but 9,400 stops across 85 vehicles in 105 minutes left no room for optimization.
The consequences showed up in the numbers. Average vehicle utilization was 64% — trucks running with a third of their capacity empty. Failed first-attempt deliveries were at 14%, each costing about $8.70 in redelivery. Fuel costs had risen 22% over the prior year, outpacing the 8% increase in fuel prices — a sign of structural routing inefficiency. The company was spending $3.2M/year on contract drivers to handle overflow the existing fleet should have absorbed.
Demand prediction was guesswork. Monday volumes were typically 30-40% higher than Fridays, but specific daily volume by hub and zone wasn't forecasted. When actual volumes exceeded the morning plan — roughly twice a week — the response was reactive: contract drivers at premium rates, packages held for next-day, or routes extended past shift limits and into overtime.
The CEO had evaluated three commercial route optimization platforms. Each quoted $180K-$350K/year and required six to nine months of implementation. None addressed the demand forecasting problem, which the CEO correctly identified as the upstream root cause.
Two connected problems: knowing what tomorrow looks like, then planning the best response to it. The forecasting model came first because accurate demand prediction directly affects route planning, fleet allocation, hub staffing, and contract driver decisions.
Diagnose: data assessment and baseline (weeks 1-3)
Three weeks on extracting, cleaning, and structuring 26 months of delivery data from the client's TMS, fleet GPS records, and client order feeds. The data was messier than expected. GPS records had gaps for 12 vehicles, the TMS categorized deliveries inconsistently across hubs, and client order data arrived in four different formats. Getting to a clean, unified dataset took time, but nothing else worked without it. From that data we built a precise operational baseline: actual cost per delivery by hub, day of week, zone, and vehicle type. The baseline showed a 41% cost gap between the best and worst performing hub — something the company had never quantified.
Design and deploy: demand forecasting model (weeks 3-8)
The forecasting model predicted next-day delivery volume by hub and delivery zone, with a 72-hour forward view updated daily. Inputs: historical delivery patterns, day-of-week and seasonal trends, client-provided order forecasts (where available), and external signals like weather forecasts and regional event calendars. A gradient-boosted ensemble outperformed the LSTM and Prophet models we tested. Validated against eight weeks of held-out data, the model hit 96.8% accuracy at the hub level and 91.4% at the zone level — vs. the planners' implicit estimates, which post-hoc analysis showed were accurate to within 15-20%.
Deploy: route optimization engine (weeks 6-11)
The route optimizer took the demand forecast and produced optimized route plans for each hub, accounting for vehicle capacity, delivery time windows, driver shift limits, live traffic, and the physical constraints of each vehicle type. Proposed route plans were ready by 3:00 AM — three and a half hours before the old manual process started — giving planners time to review, adjust for local knowledge, and approve. Planners could override any route. The output was a starting point, not a mandate. Within four weeks, planners were accepting 88% of proposed routes without changes.
Scale: live dashboard and operations integration (weeks 11-14)
A live operations dashboard went up at each hub: live vehicle positions, delivery completion rates, predicted vs. actual volume, and flags for routes falling behind schedule. The dashboard also showed next-day and 72-hour forecasts alongside current fleet and staffing levels, so hub managers could make contract driver and overtime decisions a day ahead instead of reacting the morning of. Integration with the existing TMS meant route plans, delivery confirmations, and exception data flowed both ways without manual intervention.
34%
Per-package delivery cost reduction
Blended across all three hubs, first full quarter
96.8%
Forecast accuracy (hub level)
Next-day volume prediction, 12-week validation
82%
Vehicle utilization
Up from 64%
$1.9M
Annual contract driver savings
Contract driver spend reduced from $3.2M to $1.3M
6.1%
Failed first-attempt delivery rate
Down from 14% (~$8.70 saved per avoided redelivery)
3:00 AM
Route plans ready
3.25 hours earlier than the prior manual process
The three route planners, initially skeptical, became the system's strongest advocates after the second week. Overnight route generation gave them time for exception handling and driver communication instead of building routes under time pressure. The CEO said the contract driver cost reduction alone covered the full engagement cost within the first quarter. The company has since used the forecasting model to evaluate two potential new hub locations, modeling the economics from predicted volume distribution before committing to a lease.
“Our planners are the same people — they just have better information now. The routes are ready before they walk in, the forecast tells them what tomorrow looks like, and they spend their time on the problems that need a human. We should have done this two years ago.”
CEO, Regional Last-Mile Delivery Network
Fleet costs growing faster than volume?
If you are planning routes manually and reacting to demand instead of predicting it, we can show you what better looks like.