One more hub, fewer kilometres.
A real logistics client, a hub-and-spoke network, and one question: where does the next hub go?
What it is
Client project from the RSM Business Analytics Workshop. A logistics company wanted to expand its hub-and-spoke model with one additional hub, placed so that total kilometres driven are minimised. That turns into a travelling salesman problem and a vehicle routing problem over Dutch postal-code geography.
The pipeline: preprocess and geocode the client's route data with Haversine distances, generate candidate hub locations with K-Means clustering, then optimise. A Gurobi model came first; when it hit its limits, the final routing model was built in Google OR-Tools.
The work
- Cleaned and merged client route data with Dutch postal-code geodata, computing Haversine distances
- Generated candidate hub locations with K-Means clustering
- Built a kilometre-minimising optimisation model in Gurobi
- Delivered the final TSP/VRP routing model in Google OR-Tools when Gurobi hit scale limits
Concepts
What it taught me
Real optimisation projects are solver-agnostic: when the elegant model will not scale, ship the one that solves.