Skip to content
Back to case studies
·12 weeks

AI routing platform for logistics · 17% fewer km, 22% higher on-time rate

Built an AI-assisted routing platform for a mid-size EU logistics operator · OR-Tools + GraphHopper + LLM-based dispatcher explainer. 17% fewer route km, 22% higher on-time delivery rate.

THE PROBLEM

[1/3]

  • 01Dispatchers routed manually on a legacy tool · optimisation left on the table.
  • 02Driver acceptance of computed routes was historically poor (trust issue).
  • 03Last-mile constraints (time windows, vehicle types, driver rest) were informal.
  • 04Management couldn't see why a route was chosen · no explainability.

THE SOLUTION

[2/3]

  • OR-Tools + GraphHopper for the optimiser · PostGIS-indexed geometries.
  • Constraint DSL where dispatchers write time windows + vehicle rules in plain text.
  • LLM explainer generates a 2-sentence justification per route, reviewed by dispatcher.
  • A/B shadow-mode against manual dispatch for 4 weeks before full rollout.

Technologies

Next.jsPythonortoolsGraphHopperPostgreSQLPostGIS

THE OUTCOME

[3/3]

  • 0117% fewer total route km per month.
  • 0222% higher on-time delivery rate.
  • 03Driver acceptance of computed routes: 42% → 87%.
  • 04Management satisfaction with explainability: 4.6/5.

Let's get started.

Send an email or book a 30-minute call.