This client operated a fleet of 280 trucks across the Midwest, but their routing was based on 15-year-old software that couldn't account for real-time traffic, weather, or dynamic customer windows. Fuel costs were 15-20% above industry average, and only 68% of deliveries hit their promised windows. The operations team was skeptical of any "AI" solution—they'd seen demos before that didn't work in practice.
We embedded with their dispatch team for 3 weeks before writing any code. The dispatchers knew things the data didn't capture: which customers were flexible, which drivers were good at which routes, union rules about breaks. We built a hybrid system where the ML optimizes routes but dispatchers can override any suggestion. The system learns from their overrides.
The platform uses reinforcement learning to optimize routes across the fleet, considering real-time traffic, weather forecasts, vehicle capacity, driver hours, and delivery priorities. We integrated with their existing TMS via API, though this took 6 weeks longer than planned because their TMS documentation was wrong in several places. Dispatchers have full override capability, and the system tracks which overrides improved outcomes.
Fuel costs decreased 18% ($1.4M annually). We projected 23%, but achieving that would require changes to their customer commitment windows that the sales team wasn't willing to make. On-time delivery improved from 68% to 89%. The system handles 900+ daily route optimizations. Dispatcher adoption was slower than hoped—took about 4 months before most dispatchers trusted the recommendations.
Unified patient data across 12 legacy systems. Reduced duplicate records by 73%—not the 89% we projected, but enough to materially reduce medication errors.
Migrated a legacy insurance platform to AWS. Zero customer-facing downtime, though we had three internal incidents during the parallel-running period.
Let's discuss how we can help you achieve similar results.