Cutting fuel costs with an AI-assisted fleet and route platform for a regional logistics operator

Challenge

The client runs cross-border groupage and full-truckload traffic along the Adriatic corridor with a fleet of roughly 620 vehicles. Dispatchers planned routes in spreadsheets fed by three incompatible telematics providers, so fuel consumption varied widely between depots and delays surfaced only after customers called. Management wanted a single operational picture, measurably lower fuel spend, and delivery commitments the company could actually keep.

Solution

A Vendenis team of six — two data engineers, two backend engineers, a machine-learning engineer, and a product designer — delivered the platform in 16 weeks. We consolidated the three telematics feeds into a single ingestion pipeline, built a route-optimization service that accounts for border-crossing wait times, vehicle class, and driver hours, and trained a delay-prediction model on two years of historical trips. Dispatchers now work in a real-time console that proposes plans and flags at-risk deliveries hours before they slip.

Architecture

The platform runs as a set of containerized services behind a single API gateway, deployed inside the client's existing cloud tenancy and integrated with its transport management system.

• Telematics ingestion layer normalizing roughly 38 million GPS and CAN-bus events per day into a common schema
• Route-optimization service solving depot-level vehicle routing under border-wait and rest-time constraints
• Delay-prediction model retrained weekly, served with per-shipment confidence scores
• Dispatcher console with a live map, manual plan overrides, and a full audit trail
• Integration layer syncing orders and status updates with the existing transport management system

Outcome

Six months after rollout, fuel cost per kilometre is down 14 percent against the prior-year baseline, and the on-time delivery rate has risen from 81 to 94.6 percent. Dispatchers handle a third more shipments per shift, and the delay-prediction model now triggers proactive customer notifications instead of apology calls. The client is extending the platform to its warehouse yard operations in a follow-up engagement.