Logistics Network Optimization: Lavazza Coffee

A linear-programming exercise redesigning the distribution network of a global coffee producer. The goal: decide whether to open a second satellite warehouse, where to put it, and which customers each warehouse should serve — all while minimizing annual transport and handling cost across 17,600 tonnes of demand.

Built in Excel Solver as part of DAT-3652 Business Modeling & Optimization (MS Business Analytics, Hult International Business School, Spring 2026).

---

The problem

A central warehouse in the Turin area serves six large distribution-center customers through LTL (less-than-truckload) shipments via a third-party logistics provider. Volumes have grown, delivery points have consolidated, and LTL-only service is now the cost bottleneck.

Opening a second satellite warehouse lets the company use the FTL (full-truckload) rate — roughly 40% cheaper than LTL — on the inter-warehouse leg, and shorten the LTL hop on the last mile to some customers. The optimization question is:

1. Which of six candidate locations should the satellite sit in?
2. Which customers should each warehouse serve?
3. How much does it save vs. keeping the central-only design?

The model

A linear program minimizing total annual network cost:

min  Σ (LTL cost × flow)     ← central → customer + satellite → customer
   + Σ (FTL cost × flow)     ← central → satellite backbone
   + Σ (handling cost × flow per warehouse)

subject to:

• every customer's annual demand is fully met
• only one satellite is opened (binary selector)
• flow conservation between the central warehouse and the satellite

Data inputs:

• Annual demand per customer (tonnes): 1,500 · 2,300 · 3,000 · 5,500 · 3,000 · 2,300
• LTL cost matrix (€/tonne) for all 7×7 origin-destination pairs
• FTL cost = 0.6 × LTL cost (the ~40% discount)
• Handling cost per warehouse location (€15–€22/tonne)

Decision variables: flow from each warehouse to each customer (Xij), plus a binary selector for which satellite location is active.

Seven scenarios were solved: one for each of the six candidate satellite locations, plus the central-only baseline. Sensitivity reports were generated for every scenario.

Results

Optimal configuration: open Location 4

• Total cost: €8,548,200 / year
• Savings vs. baseline: €1,551,000 / year (~15.4% reduction)
• Routing decision:
  • Customer 4 (5,500 t) → served by the new satellite at L4 via an FTL backbone from the central warehouse
  • Customers 1, 2, 3, 5, 6 → remain on the central warehouse via LTL

Location 4 wins because Customer 4 is both the largest single demand point (31% of total tonnage) and sits far enough from the central warehouse that the FTL-backbone + short-haul-LTL pattern beats a direct long-haul LTL shipment.

Why not L2?

L2 was the close runner-up (€8.60M, €1.50M savings) and actually had the lowest LTL cost of any scenario — but it split volume more aggressively between the two warehouses, driving handling costs up by ~40% and FTL costs up 29% vs. L4. L4 hits a better total by concentrating only the largest customer on the satellite.

Sensitivity takeaways

• The ranking of L4 vs. L2 is stable across reasonable variations in the LTL/FTL cost ratio.
• Handling-cost assumptions were kept flat per location in line with the case; in practice, the start-up fixed cost of a new warehouse would need to clear the ~€1.55M/year hurdle to keep the investment justifiable.
• The solution is sensitive to demand concentration: if Customer 4's volume dropped materially, the optimal satellite location would shift toward clustering multiple mid-sized customers (likely L2).

Deliverable

See lavazza_optimization_deliverable.pdf for the full write-up — model formulation, Solver screenshots, scenario comparisons, and sensitivity reports.

Stack

• Excel Solver (Simplex LP)
• Scenario Manager / manual scenario sweep for the binary warehouse selector
• Sensitivity reports for shadow prices and reduced costs

Case source

Based on the SDA Bocconi School of Management case "Redesigning the Lavazza Logistics Network" (Veronesi, 2018), used as the modeling basis for the course exercise.