How to Design a Distribution Network: Choosing How Many Sites to Run, Setting the Service Promise, and Trading Freight Cost Against Inventory

Where the Money in a Network Really Sits

Published 2026-09-01 · EuroQuest International

Most distribution network reviews start in the wrong place. Someone asks where the next warehouse should go, a property agent produces a shortlist, and the conversation is about rent per square meter before anyone has agreed what the network is supposed to do. The site question is the last one to answer, not the first.

A distribution network is a set of deliberate trades. Every site you add shortens the last leg to the customer and lengthens the bill for stock, staff and buildings. Every hour you shave off the service promise pushes inventory forward, closer to demand and further from the flexibility of a central pool. This guide works through those trades in the order a real design follows: what the network is deciding, how many nodes it needs, what promise it is built around, where the sites belong, what data the exercise requires, and how to stress the answer before signing a lease.

What Is a Distribution Network Design Actually Deciding?

A network design fixes four things at once: how many stocking points you operate, where they sit, which customers each one serves, and what each one holds. Change any of the four and the other three move. That coupling is why the exercise resists being broken into separate property, transport and inventory decisions, and why it belongs to one owner rather than three departments negotiating.

The output is not a map. It is a cost and service position that the business has chosen on purpose, expressed as a map. Teams that treat freight and distribution network optimization as a routing problem tend to arrive at a locally sensible answer that quietly fails the finance test, because the savings sit in transport while the new costs land in inventory and property.

The Four Cost Lines That Move Together

Outbound transport falls as sites multiply, because the average leg to the customer gets shorter. Inbound transport rises, because replenishment now goes to several destinations instead of one. Fixed facility cost rises in steps, one lease and one management team at a time. Inventory rises too, and this is the line most often underestimated.

Safety stock does not split neatly when you split a site. Held across independent locations it grows roughly with the square root of the number of stocking points, so going from one site to four does not quarter the buffer at each, it leaves you holding about twice the total. The Census Bureau monthly inventories release put the total business inventories to sales ratio at 1.30 at the end of June 2026, down from 1.39 a year earlier, and every extra node in a network pushes against that direction of travel.

How Many Sites Should the Network Have?

There is no universal number, but there is a reliable shape. Total cost falls steeply from one site to two, less steeply from two to three, and usually flattens somewhere between three and six for a national or regional operation. Past that point the curve turns upward as fixed and inventory cost overtake the shrinking freight saving.

The flat part of that curve is the useful finding. When several configurations sit within a few percent of each other on cost, the decision moves to criteria that a cost model does not capture: labor availability, resilience to a single site failure, room to expand, and how easily the network absorbs a change of channel mix. A design that is marginally more expensive but survives losing a building is often the right answer.

Why the Curve Flattens Where It Does

Freight cost per unit is dominated by distance on long legs and by drops per route on short ones. Once the average outbound leg is short enough that a vehicle can complete a full round of drops inside a shift, adding another site stops buying distance and starts buying only marginal density. Eurostat data shows the scale of the long legs still in play: most EU road freight performance in 2024 was carried over distances between 300 and 999 kilometers, at 41.3 percent of the total.

The vehicle side matters as much as the map. Across the EU in 2024, 83 percent of road freight in tonne kilometers was performed by heavy goods vehicles with a maximum permissible laden weight above 30 tonnes, on Eurostat vehicle statistics. A network built around full loads between nodes and smaller vehicles on the last leg behaves very differently from one that tries to run one vehicle type end to end, which is the practical content of logistics and distribution efficiency work.

What Service Promise Are You Designing To?

The promise is the single most powerful input, and it is usually the vaguest. Next day is not a specification. Next day to whom, ordered by when, measured how, and failing how often before it counts as a breach? Two businesses with identical demand and identical products will build different networks if one promises next day to ninety-five percent of revenue and the other promises it to ninety-five percent of accounts.

Write the promise as a coverage target against a cut off time, then test what it costs at several levels. The gap between ninety percent and ninety-eight percent coverage is frequently an entire additional site. Putting that number in front of the commercial team turns a service debate into a priced choice, which is the point of running distribution network optimization and last-mile delivery as a joint exercise rather than a logistics one.

Segment the Promise Before You Segment the Network

Few businesses need one promise for everything. Fast moving lines that drive repeat purchase may justify forward stock in several locations; slow moving lines almost never do, and holding them centrally protects both availability and working capital. A network that stocks the top decile of lines regionally and everything else centrally often outperforms a network that treats every line the same, at lower total cost.

How Do You Choose Where the Sites Go?

Start from demand weight, not from the existing estate. Plot where volume actually goes, weighted by the cost of getting it there, and the map produces a small number of natural gravity points. Those points are a starting hypothesis, never the answer, because the real world adds constraints that a center of gravity calculation cannot see.

Road access is the first constraint. A site twenty minutes from a motorway junction loses that time on every vehicle movement, every day, for the life of the lease. Labor pool is the second, and it is now the binding one in many markets: a cheaper building in a thin labor market is not cheaper. Planning consent, power availability for automation, and neighboring uses that restrict night operations complete the list, which is why infrastructure planning for transportation and distribution sits alongside the network model rather than after it.

Comparing the Common Network Shapes

Most designs resolve into one of four archetypes. The table below sets out how each behaves on the cost lines that matter, so a shortlist can be argued on evidence rather than preference.

Note what the third row buys. A cross dock carries reach without carrying inventory, which is the cheapest way to shorten the last leg when the range is broad and the promise is not extreme. It is under-used because it requires tighter inbound discipline than a stocking site does, and that discipline is an operating capability rather than a design choice.

What Data Does the Exercise Need?

A network study needs less data than people fear and better data than they usually have. Twelve months of order lines at delivery-point level, with weight and volume, will support a defensible answer. What derails studies is not missing data but dirty data: addresses that do not geocode, weights recorded as ones, and returns held in a separate system that nobody merges in.

Build the file before building the model. Confirm that order volume reconciles to the ledger, that the geographic spread matches what the sales team recognizes, and that seasonality is visible in the weekly profile. Teams that run advanced data analytics in logistics and distribution as a standing capability get to this point in days rather than months, because the reconciliation is already routine.

The Cost Inputs People Get Wrong

Freight rates should reflect what the network would actually pay at the new volumes and lane balance, not today's rate card at today's flows. A design that concentrates volume on a few lanes earns better rates on them and worse rates on the thin ones, and a model that applies one blended rate hides that entirely. Getting this right is the substance of freight cost analysis, and it changes conclusions more often than any other single input.

Empty running deserves its own line rather than burial in a rate. In the year to March 2026, heavy goods vehicles registered in Great Britain covered 5,714 million kilometers empty, which the Department for Transport reports as 30 percent of the total loaded and empty distance for the period. A network whose flows are one-directional pays for that geometry whether or not the model names it.

How Do You Test a Design Before Committing to It?

Three tests separate a design that survives from one that reads well. The first is peak. Rerun the network at the busiest four weeks rather than the annual average, and check whether the recommended sites still hold the throughput. Many designs that are optimal on mean demand run out of dock doors in November.

The second is failure. Remove each site in turn and measure what the remaining network can still serve. A configuration that collapses when one building goes offline is a concentration risk dressed as efficiency, and the cost of the insurance is usually smaller than it looks. The third is drift. Test the design against a plausible shift in channel mix or geography over five years, because a lease outlives most demand assumptions.

Transition is the step most often skipped. Moving from the current estate to the recommended one has a cost and a service risk of its own, and a design whose benefits arrive in year three while the disruption lands in year one needs that shown honestly. Businesses expanding across borders face this most acutely, which is why optimizing distribution for global expansion treats sequencing as part of the design rather than an implementation detail.

Where Teams Build the Capability

Network design sits between operations, finance and commercial, and the people who do it well are usually those who have seen several networks rather than one. Practitioners take this work in London, Amsterdam, Singapore, Barcelona and Brussels, and the full range sits under logistics and distribution management.

Frequently Asked Questions