How a 5,000-Vehicle Fleet Saved $18M a Year on Fuel
$18M
4–6 months
$500K+
A national carrier ran a $120M annual fuel budget with 20% wasted on suboptimal routing. One in five deliveries missed its window, routes were planned once a day and never adapted to traffic, weather or closures, and fleet utilization sat at 64%.
A dynamic routing platform: (1) a streaming layer fuses live traffic, weather, GPS telemetry and delivery schedules into one operational picture; (2) a constraint-based optimization engine using genetic algorithms and reinforcement learning; (3) continuous re-routing that detects disruptions — accidents, weather, closures — and recalculates affected routes with driver notifications; (4) a fleet command center with real-time tracking, fuel analytics and predictive ETAs.
Bolt treated every routing problem as a data problem. A streaming platform fused live traffic, weather, GPS telemetry and delivery schedules into a single operational picture, replacing routes that were planned once a day and never updated. A constraint-based optimization engine — combining genetic algorithms with reinforcement learning — produced routes that balanced fuel, time and delivery windows. Continuous monitoring detected disruptions like accidents, weather and closures and recalculated only the affected routes, pushing changes straight to drivers. A fleet command center gave operations real-time tracking, performance metrics, fuel analytics and predictive ETAs across all 5,000 vehicles in 200 cities. The payback landed in the first fuel cycle: $18M saved a year, on-time delivery up 22%, and utilization lifted from 64% to 82%. Delivered in 22 weeks, principal-led and production-first.
Large fleet operators with high fuel spend and static daily route planning, where real-time re-routing against traffic and weather drives both cost and on-time performance.
$18M Fuel Saved
82% Fleet Utilization
22% Faster Delivery
Real-time data fusion (traffic, weather, GPS), constraint optimization with genetic algorithms and RL, continuous re-routing, and a fleet command center across 5,000 vehicles in 200 cities. 22-week build, principal-led.
Dean Sloves
A large transportation and logistics operator replaced static, once-a-day route planning with a streaming platform that fused live traffic, weather, GPS telemetry, and delivery schedules into one operational picture. A constraint-based optimization engine combining genetic algorithms and reinforcement learning produced routes balancing fuel, time, and delivery windows, with continuous re-routing recalculating only the affected routes when disruptions hit. The payback landed in the first fuel cycle, with roughly $18M saved a year.
The build used Azure OpenAI as the model, with a stack of Apache Kafka, Python, PyTorch, PostgreSQL, Redis, Kubernetes, Mapbox, and AWS. The routing engine combined predictive analytics with genetic algorithms and reinforcement learning, plus process automation for driver re-routing.
The operator saved an estimated $18M a year on fuel (roughly 15% of fuel spend), lifted on-time delivery by 22%, and raised fleet utilization from a 64% baseline to 82%, across 5,000 vehicles in 200 cities.
The platform was delivered in 22 weeks, within an overall 4–6 month engagement range.
Large fleet operators with high fuel spend and static daily route planning, where real-time re-routing against traffic and weather drives both cost and on-time performance.
Have a similar challenge?
