New Odyssey
Back to Experiments
Logistics Experiment· National Logistics Provider

Route Optimization Experiment

Hypothesis

ML-driven route planning and demand forecasting can reduce per-delivery costs by at least 10% and improve on-time delivery SLAs from 82% to 90%+.

15%

Cost Per Delivery

Down from $8.40 avg

97%

On-Time Delivery

Up from 82% baseline

$3.2M

Annual Savings

Projected at scale

3

Months to Full Deploy

Across 50+ centers

Disconnected Systems Causing Delivery Failures

A national logistics provider operating 50+ distribution centers was struggling with visibility across their supply chain. Their warehouse management, transportation management, and customer systems operated in silos.

Route planning was largely manual, with dispatchers spending 3-4 hours each morning optimizing loads based on incomplete information. On-time delivery sat at 82%.

Inventory forecasting relied on spreadsheets and gut instinct. Stockouts were common at high-demand locations while other facilities sat on excess inventory, tying up $18M+ in working capital.

Controlled Experiment Across Distribution Centers

We ran a controlled 90-day experiment across 3 distribution centers (representing 12% of total volume), comparing AI-optimized routing against existing manual dispatch logic.

Primary metrics included cost per delivery, on-time delivery rate, total miles driven, and dispatcher planning time. Secondary metrics tracked driver satisfaction and customer complaint volume.

We controlled for geographic and seasonal variations by selecting centers with similar historical performance and running the experiment across a full quarter.

Measured Variables

Cost per delivery (all-in)
On-time delivery percentage
Total miles driven per route
Dispatcher planning time
Driver utilization rate
Customer complaint volume

Unified Intelligence Platform

We deployed Phoenix AI Platform to create a unified intelligence layer connecting WMS, TMS, and customer systems for real-time decision-making.

The route optimization engine uses ML models trained on historical delivery data, traffic patterns, and customer preferences to generate optimal routes in minutes instead of hours.

Demand forecasting models analyze historical sales data, seasonal patterns, and external signals to predict inventory needs 2-4 weeks ahead with significantly improved accuracy.

Real-time route optimization engine
Predictive demand forecasting
Dynamic load planning
Automated carrier selection
Real-time visibility dashboard
Exception management workflows

Results Exceeded All Targets

Within the 90-day experiment, treatment centers achieved a 15% reduction in cost per delivery—exceeding our 10% target. On-time delivery improved to 97%, far surpassing the 90% goal.

Dispatcher planning time dropped from 3-4 hours to under 30 minutes per morning, freeing them to focus on exception management and customer communication.

Based on the experiment results, the company approved full deployment across all 50+ distribution centers, projecting $3.2M in annual savings at scale.

Cost per delivery reduced by 15% ($8.40 → $7.14)
On-time delivery improved from 82% to 97%
Dispatcher planning time reduced by 85%
Inventory carrying costs reduced by 22%
Customer complaints decreased by 40%
Full deployment approved within 30 days of experiment completion

The results speak for themselves. We went from spending hours on manual routing to having AI-optimized plans ready before the first truck leaves. Our drivers are happier, our customers are happier, and our margins have never been better.

David Chen

SVP of Operations, National Logistics Provider

Want similar results?

Start with an Integration Diagnostic to identify your highest-impact opportunities.