Supply Chain · Deliver

Network Design

Position warehouses between plants and customers: optimize service distance against transport, inventory and facility costs.

Time2 h
FormatSmall group
StageDeliver

Network Design: what it is and why it works

Network design decides how many warehouses you run, where they sit and which customers each one serves. The core logic is a total-cost trade-off: more locations shorten the distance to customers and cut outbound freight, but each new site adds fixed facility cost, handling labor and its own safety stock. Fewer locations do the opposite. The method turns origin-destination flows, candidate sites and cost rates into a model that adds transport, inventory, facility and a penalty for missing the service promise, then compares scenarios on that single total rather than on any one line item.

It works because most networks were never designed as a whole: a site was added for an acquisition, another for a big customer, and the map froze. A structured model exposes cross-subsidies that intuition misses, such as a cheap rural warehouse that forces expensive expedited deliveries. For a handful of sites a spreadsheet with a center-of-gravity check is enough; beyond that, a mixed-integer optimization tool earns its setup time. Once the footprint is set, Transport Optimization tunes the lanes, Cross-Docking decides which flows skip storage, and a Nearshoring Decision may shift where supply enters the network.

What you need

Twelve to twenty-four months of shipment history aggregated into origin-destination volumes by period
Customer demand by ZIP code or region, with service-time promises by segment
Candidate warehouse sites with fixed costs (lease, staffing baseline) and variable handling costs
Freight rate tables or rate models by lane and mode, inbound and outbound
Inventory carrying cost rate and current safety stock policy
Growth or mix assumptions for the planning horizon

What you get

A baseline model that reproduces current network cost within an agreed tolerance
Scenario comparison of total landed network cost and service level for each footprint option
Recommended number and location of warehouses with customer-to-site assignments
Sensitivity results showing how robust the choice is to freight, volume and rent changes
A phased transition plan with one-time costs such as moves, severance and dual running

When to use it

When the distribution network grew by history instead of by design.

How to do it, step by step

Gather flows: origin-destination volumes by period.
List candidate warehouse locations with fixed and variable costs.
Model the total: transport + inventory + facility + service penalty.
Test scenarios: fewer hubs, more hubs, nearshoring.
Choose the design with the best total, not the prettiest map.

Worked example: Consolidating five regional warehouses for an industrial spare-parts distributor

Illustrative scenario — figures are realistic but not from a real company.

A distributor of pump and valve spare parts runs five warehouses across the eastern half of the US, three of them inherited from acquisitions. Annual network cost is about $14.8 million: $7.1 million outbound freight, $2.3 million inbound, $3.4 million facilities and $2.0 million inventory carrying cost. The service promise is next-day delivery to 90% of customers by demand.

The analyst aggregated two years of shipments into 3-digit ZIP demand points and built a baseline model that landed within 3% of actual cost, which gave management confidence in the numbers.
Eight candidate sites were listed, including the five existing buildings and three greenfield options near interstate hubs, each with lease quotes and local labor rates.
The model priced four scenarios: status quo, three hubs, four hubs, and three hubs plus a small forward stocking point for the densest metro area.
The team added a penalty of $40 per order delivered outside the next-day window, so service could not be traded away silently.

Result. The three-hubs-plus-forward-point scenario cut total cost to about $13.1 million, an 11% reduction, while keeping next-day coverage at 91%. Outbound freight rose slightly, but facility and inventory savings more than offset it. The lesson: the two cheapest warehouses were closed, because their low rent did not compensate for long, expensive routes to customers.

Open it in the FieldKit Supply Chain app → Get the Supply Chain template pack — join the waitlist

Common pitfalls and how to avoid them

Building the model on average freight rates instead of lane-level rates.Use rate tables or fitted rate curves by distance and weight break; averages hide the lanes that drive the decision.
Skipping baseline validation and trusting the first scenario output.Reproduce last year's actual cost first; if the baseline is off by more than a few percent, fix the data before comparing options.
Ignoring the inventory side when adding sites.Model safety stock per location; splitting demand across more sites raises total safety stock even when sales stay flat.
Choosing the optimum without pricing the transition.Add one-time costs, dual running and customer disruption risk, and phase the change so service holds during moves.

Frequently asked questions

How many warehouses should a distribution network have?

There is no generic number. It depends on customer density, service-time promises, order profile and the relative cost of freight versus facilities and inventory. The right count is the one with the lowest total cost that still meets the service promise, which is why you test several footprints side by side instead of starting from a target number.

What is the center-of-gravity method in network design?

It finds the location that minimizes weighted distance to demand points, using volume as the weight. It is a quick sanity check for one site or a starting point for several, but it ignores real road networks, rate structures, labor markets and available buildings. Use it to narrow candidates, then evaluate those candidates in a full cost model.

What software is used for supply chain network design?

Dedicated network design tools solve mixed-integer optimization models with built-in rate and geography data. Smaller studies can be done in a spreadsheet with a solver add-in or an open-source optimization library. The tool matters less than clean flow data, a validated baseline and honest cost inputs for each candidate site.