What network design decides

Network design sets the physical skeleton of a supply chain: plants, warehouses and the lanes between them. These are long-term, expensive decisions. A warehouse lease or a new plant locks a company in for years, and it decides most of what transport and inventory will cost afterwards.

It comes down to four decisions:

1. Facility role: what each site is for. Is a plant a low-cost producer, the hub for a region, or the lead site for new products?
2. Location: where plants and warehouses should be.
3. Capacity: how big each one should be.
4. Market allocation: which plant supplies which warehouse, and which warehouse serves which customers.

Most of the trade-offs move with one dial: the number of facilities. Add warehouses and:

• delivery gets faster and last-mile freight falls, because you’re closer to customers;
• inventory rises, because each site needs its own safety stock (risk pooling works in reverse);
• facility costs rise, with more rent, staff and systems;
• inbound freight rises, because each site gets smaller, less efficient shipments.

So total logistics cost forms a U-shape: it falls at first, bottoms out, then climbs. A company competing on price sits near the bottom of the U with a few large warehouses. One competing on speed, such as same-day e-commerce or critical spare parts, deliberately builds past the cheapest point and pays for responsiveness.

Why each cost moves

• Inventory grows with the square root of the number of stocking points, the risk-pooling law in reverse.
• Facility cost grows roughly in line with the number of sites: rent, staff, systems.
• Outbound transport falls as sites get closer to customers. For customers spread over an area, the average distance to the nearest site shrinks roughly with $1/\sqrt{N}$.
• Inbound transport rises slowly, because each site receives smaller, less efficient shipments.

How a network study runs

1. Frame the question: what may change (sites, capacities, flows) and what is fixed.
2. Collect data: demand by location, freight rates by lane and mode, warehouse costs, inventory costs, service targets.
3. Build and validate a baseline: model today’s network and check it reproduces actual costs and flows closely. If it can’t, nothing it says about the future can be trusted.
4. Optimize and run scenarios: alternative designs, demand futures and cost shocks.
5. Recommend: a design, its cost and service, the risks, and a transition plan.

The network follows the strategy

A network can’t be cheapest and fastest at once. Fisher’s distinction between efficient and responsive supply chains applies directly: functional products with steady demand and thin margins want few, large, efficient sites; innovative or urgent products want more sites, closer to customers, and more stock. Many companies need both, so they segment the network: a lean backbone for the bulk, plus fast nodes for the products and customers that pay for speed.

How tax rules shaped networks in India

Before the Goods and Services Tax (GST) in 2017, inter-state sales attracted central sales tax, so companies often kept a warehouse in many states for tax reasons rather than logistics reasons. GST removed most of that incentive, and many firms consolidated into fewer, larger regional hubs placed for transport cost and service. It is a clear example of how a policy change can shift the bottom of the U-curve.

Building in flexibility

Network decisions last years, while demand and costs change faster. Ways to keep options open:

• Lease rather than buy, or use third-party logistics (3PL) warehouses for uncertain regions.
• Build in stages: modular capacity, with land or contract options for expansion.
• Design for scenarios, not one forecast, and prefer designs that stay good across them.
• Review regularly: every few years, or on triggers such as an acquisition, a new channel or a big shift in freight costs.

Classic modelling mistakes

• Designing for today’s demand instead of the demand the network will serve for its lifetime.
• Using averages: a network that works on average can fail at the festive peak. Check peak-week capacity.
• Straight-line distances: real roads are longer, and some lanes are much slower than distance suggests.
• Cost only: leaving out service, risk and carbon, then discovering the “optimal” network can’t meet delivery promises.
• Ignoring the transition: closing sites costs money (leases, severance, dual running), so the move matters as much as the destination.