Network design: how many nodes, where, serving whom

The curve that drives the answer

As stocking points multiply, outbound distance and delivery time fall while inbound movement, facility overhead and total buffer stock rise. The total cost curve is therefore shallow near its minimum, which has a practical consequence: several network shapes usually cost about the same, and the tie is broken by service, risk and flexibility rather than by cost. Studies that present a single optimal answer to the nearest unit are overstating their own precision.

Serve areas before sites

It is more productive to decide which customer geographies must be reached within which service window, then find locations that satisfy those windows, than to start from candidate buildings. Demand geography, not property availability, should drive the shape. Once the serve areas are drawn, site selection becomes a constrained search over labour availability, land, access to the relevant corridors and cost, with the service promise already protected.

What a credible study contains

A defensible model uses real order-line history rather than aggregated shipments, values inbound and outbound movement separately, includes the stock consequence of each network shape, and tests the answer against scenarios rather than a single demand forecast. It should also state what it excludes. Models that omit inventory effects reliably recommend more facilities than the business can afford to fill, because the fragmentation cost lands in a budget the model never looked at.

Reversibility and the value of waiting

Lease length, automation depth and ownership all determine how expensive a wrong answer will be. Where demand geography is uncertain, a shorter lease or a shared facility buys the option to change your mind, and that option has real value even if it costs more per unit. The corresponding discipline is to name in advance what would trigger a network review, so the structure is revisited on evidence rather than on a change of leadership.

Cross-border structure changes the map

Where a network spans customs territories, the placement of stock relative to those borders affects duty timing, transit obligations and the paperwork burden of each flow. The design question is only where inventory sits and which market it serves from there; the procedural mechanics of clearing goods belong to trade operations, and national requirements differ enough that they belong with the administration publishing them rather than inside a network model.

Frequently asked questions

How often should a network be redesigned?

Only when something structural has moved: demand geography, channel mix, cost relativities between transport and property, or trade conditions. Redesigning on a fixed calendar tends to produce churn, while never revisiting the question leaves a network optimised for a business that no longer exists.

Does adding a warehouse always improve delivery speed?

It shortens the final leg for the area it serves, but it also splits stock, which raises the total buffer needed to hold the same availability and can create more out-of-stock events per site. The speed gain is real; it is simply not free, and the offsetting cost lands in inventory.

Should the model optimise for cost or for service?

Constrain service and optimise cost. Setting the delivery windows the business has committed to as hard constraints, then minimising cost within them, produces answers that can actually be implemented, whereas trading service against cost in a single objective function tends to yield networks nobody will approve.