Greenfield studies

A greenfield study designs the network as if nothing existed yet: no plants or warehouses you must keep. A brownfield study starts from today’s network and asks what to keep, close, move or add. Greenfield answers are useful even for companies with networks, because they show the ideal shape and how far the current one is from it.

A typical study for an FMCG company expanding across India runs like this:

1. Map the demand. Where are customers, and how much does each region buy, say North 2.5 lakh units, West 3.5 lakh, South 2.5 lakh, East 1.5 lakh? Look at the future too: design for the demand in five years, not today’s.
2. Build scenarios. Low, base and high demand, because a network has to work across a range of futures.
3. List candidate sites. Mumbai, Pune, Ahmedabad, Delhi, Hyderabad, Bengaluru, Kolkata, Chennai, filtered by land, labour, roads and rail, power, taxes and incentives.
4. Collect costs. Fixed cost per site, handling cost, freight rates per lane, inventory carrying cost, and service targets such as delivery within 48 hours.
5. Optimize. A mixed-integer model chooses which candidates to open and how product flows, at minimum total cost within service limits.
6. Stress-test. Rerun with fuel prices up, a site closed or demand shifted. The best design stays good across scenarios, not just the base case.

The output isn’t just a map. It is a recommended network with the cost, service and risk of each alternative, which is what leadership needs to commit years of investment.

A toy model with real structure

The FMCG example from the gist: annual demand of 10 lakh units, split North 2.5 lakh, West 3.5 lakh, South 2.5 lakh and East 1.5 lakh. One plant near Pune. Candidate warehouses in Delhi, Mumbai, Bengaluru and Kolkata, with approximate road distances between them.

Assumptions:

• Outbound: ₹3 a unit for local delivery plus ₹0.01 per unit per km from the warehouse.
• Inbound from the plant: ₹0.004 per unit per km (full truckloads are cheaper).
• Fixed cost: ₹1.2 crore a year per warehouse.
• Inventory: ₹0.8 crore × $\sqrt{N}$ for $N$ warehouses.

That is the real output of a greenfield study: not one answer, but the price of each level of service, so leadership can choose knowingly.

What a full study adds

• Many more candidate sites, demand zones and products.
• Capacity limits and options per site.
• Freight rates by lane and shipment size, and transit times instead of distances.
• Several demand scenarios, with the design evaluated across all of them.

Robust designs, not just optimal ones

The design that wins the base case may lose badly if demand shifts or costs rise. Common ways to choose under uncertainty:

• Expected cost: weight each scenario by its probability and pick the lowest average.
• Minimax regret: for each design, find its worst shortfall against the best possible design in any scenario, and pick the design whose worst regret is smallest.
• Stochastic optimization: choose sites now (first stage), and let flows adjust in each scenario (second stage), minimizing expected total cost.

The scenario modelling stop works through a regret table.

What the model doesn’t see

Shortlisting candidate sites and checking the final choice needs on-the-ground factors:

• land availability, title and zoning; flood and earthquake risk,
• labour availability, skills and wage trends,
• road, rail and port access; reliable power,
• state incentives, permits and local regulations,
• proximity to key customers and to suppliers’ sites.

Many studies score these factors with weights and combine the scores with the model’s cost results.

From ideal to real

Most companies aren’t truly greenfield, so the ideal design becomes a target, and the real work is the path to it:

• Phasing: which sites to open, close or resize first, often timed with lease expiries.
• Transition costs: dual running, severance, moving stock and equipment, temporary service dips.
• Quick wins: sometimes changing flows (which warehouse serves which zone) captures much of the saving without any new site.

Data pitfalls

• Demand by bill-to address (where invoices go) instead of ship-to address (where goods go) puts demand in the wrong places.
• Freight rates from contracts can differ from what is actually paid; use invoice data where possible.
• Growth assumptions by region matter more than any detail of the base year.