The Building Fit the Business.
The Business Moved.
One DC or three, insourced or outsourced — the network decision is expensive to get wrong and slow to reverse.
Distribution Network Strategy
Most network decisions arrive as a symptom rather than a question. A lease is coming up. Freight cost is climbing faster than volume. A channel has outgrown the node serving it. Service is slipping in one region and holding everywhere else. Or the building simply ran out of room, and the only options anyone has priced are a bigger building or another shift.
The underlying question is usually the same: how many buildings should this business have, where should they be, and what should each one serve. That question sits above the floor. Slotting and travel paths make a building faster, but no amount of optimization fixes a network pointed at the wrong map. GoVia Alliance Group answers that question with partners who ran the buildings before advising on them — network design from the floor, not from a model.
The versions of this that show up in mid-market operations are consistent. A company coming off a 3PL and standing up its own DC for the first time. An operation that has outgrown a single building and has to decide between expanding it, adding a second, or relocating entirely. Two networks bolted together after an acquisition, each still running its own logic. Or a footprint that made sense at a previous volume and quietly stopped making sense two years ago.
What the work actually is
The work runs from the question to the move. Model the network against real cost-to-serve rather than assumed rates. Evaluate sites against what the operation actually needs — labor market, inbound and outbound lanes, building specification, room to grow. Design the layout and the flow for the volume you are going to have, not the volume you have now. Then run the transition: sequence the move, protect inventory accuracy through it, and get the new building producing before the old one closes.
That last part is where most network projects come apart. The analysis is the easy half. Standing up a building and moving a live operation into it, without losing a season, is the half that requires someone who has done it.
Network modeling at this scale does not need a research department. It needs someone who can read the data, walk the candidate buildings, and tell you which model will survive contact with a real labor market on a real Monday.
What the Network Decision Actually Comes Down To
Five questions decide a network. Everything else is downstream of them, and getting any one wrong is expensive in a way that takes years to unwind.
1. Should You Operate It At All
The first question is whether the volume justifies running your own building. For a lot of companies it does not yet, and the honest answer is to stay with a third party until it does.
A 3PL converts fixed cost into variable cost. You pay for the space and labor you use, you absorb no lease, and you can leave. Your own building inverts that — lower unit cost at volume, and a fixed obligation that does not shrink when demand does. The crossover is not a single number, but the shape is consistent: below the volume where a building runs at real density, the third party is cheaper and far more flexible. Choosing the right partner and holding them to the agreement is its own discipline — 3PL advisory covers the selection, the contract, and the governance that keeps performance from drifting.
The trigger to insource is usually one of three things. Volume has grown to where you are paying 3PL margin on work you could run yourself. Handling requirements have become specific enough that a shared facility cannot serve them. Or service and control have become a competitive issue and you need the building to behave the way you decide it should.
Growing into your own DC is the normal path. The move in either direction is a project in itself — 3PL onboarding and transition is where inventory accuracy, system integration, and service continuity either hold or break. Standing up your own building before the volume is there is one of the more expensive mistakes in this category.
2. How Many Nodes
Node count is a service-versus-cost trade, and the curve is steeper than most operations expect. Every building carries a fixed floor that exists whether it ships anything or not — rent, a management team, a system instance, and a minimum staffing level just to open the doors. That floor does not scale down on a slow week.
One well-placed building in the eastern half of the country reaches most of the U.S. population within two days by ground. A second node buys another meaningful step. The third and fourth buy progressively less, because each is serving a smaller and more remote share of demand. The service curve flattens while the cost curve does not.
The cost nobody models is inventory. Safety stock does not split between buildings, it multiplies. Under the square root law, moving from one node to two raises the safety stock required to hold the same service level by roughly forty percent, and moving to four nearly doubles it. That carrying cost is permanent, it compounds with every SKU added, and it rarely appears in the network model that justified the second building.
Freight moves in two directions at once. Outbound cost falls as buildings get closer to customers. Inbound cost rises, because the same inbound volume now splits across more destinations and loses the density that made truckload economics work. Networks modeled on the outbound saving alone find the inbound bill waiting for them.
A second building does not halve the work of the first. It duplicates the fixed overhead and splits the volume, which means both buildings run at lower density than the one did.
3. Where — And It Depends Entirely on Who You Serve
Location logic is not universal. It changes completely with the channel, and this is where generic network models go wrong.
A retail DC replenishing stores locates near the largest cluster of brick-and-mortar locations. The outbound is a store delivery route, not a parcel shipment, and minimizing the miles on that route is the whole objective. An e-commerce facility locates against the parcel network. Outbound is zone-rated, so the site is chosen for its position relative to FedEx and UPS hubs and how many zones sit between the building and the customer base. This is why smaller e-commerce operations concentrate in the Midwest and Ohio Valley — Columbus, Indianapolis, Memphis, Louisville — where a single building keeps most of the country inside the lower zones. A B2B or wholesale operation importing containers locates near the ports instead, so inbound overseas freight merges with outbound wholesale without a second long haul. Los Angeles and Long Beach, Savannah, and the New York and New Jersey complex all exist as DC markets for that reason.
Whatever the channel, the labor market decides whether the site is viable at all. A location that models beautifully and sits where every DC within twenty miles competes for the same hourly workforce will not staff at the rate the model assumed. Walk the market before signing.
4. What Each Building Serves — The Omni-Channel Question
Most mid-market operations are not single-channel anymore, and that is what makes this decision hard.
A network can split by geography, by channel, by velocity, or by customer. Geographic splits are simplest to manage and most common. Channel splits — retail replenishment in one building, direct-to-consumer in another — make sense when the pick profiles are genuinely different, and they usually are. A full-case store replenishment operation and an each-pick e-commerce operation want different layouts, different equipment, different slotting, and different labor.
The alternative is running both profiles under one roof, which is what omni-channel fulfillment actually means in practice: shared inventory, two picking disciplines, and a building designed to do both without either one degrading the other. It can work, and it is harder than it looks. The common failure is a building laid out for full-case retail that has an e-commerce operation bolted into a corner of it.
The split has to be decided before the buildings are designed, not after. Retrofitting a building laid out for one profile to serve another is the expensive version of this mistake.
5. The Transition
The move is where network projects fail, and it is the part least represented in the analysis. The model ends at steady state. The risk lives entirely in the months before steady state exists.
Sequence matters more than speed. Inventory accuracy has to survive the move, which means counting before, controlling during, and reconciling after — and a variance discovered three weeks later cannot be traced to which building created it. The new building has to be producing before the old one closes, which means running both for a period and paying the fixed cost of both.
Plan the overlap as a line item, not a contingency. Two buildings, two management teams, and two labor forces for sixty to ninety days is the normal shape of a network move, and it is the number most often missing from the business case that approved it.
The hidden failure is retention at the building you are closing. The people running it know the date, and the best of them leave first, because they are the ones with options. Service breaks at the old building while everyone is watching the new one. Retention agreements through the final ship date cost a fraction of what a missed quarter costs.
Nothing about a transition should be scheduled inside a peak window. If the calendar forces a choice between moving late and moving during peak, move late.
Straight Answers
Q: Should a smaller company run its own DC or use a 3PL?
Use a 3PL until the volume justifies the fixed cost, which for most companies is longer than they expect. A third party converts space and labor into variable cost with no lease and an exit. Your own building wins on unit cost only once it runs at real density. The trigger to insource is usually one of three things: you are paying 3PL margin on work you could run yourself, handling requirements have outgrown a shared facility, or service and control have become competitive issues.
Q: How many distribution centers does a mid-market company need?
Most start with one and add the second for service or freight, rarely both. A single well-placed DC in the eastern half of the country reaches a large majority of the U.S. population within two days by ground. The second building does not halve the work of the first — it duplicates fixed overhead and splits volume, so both run at lower density than the one did. Model the two-node case against the one-node case before assuming growth requires it.
Q: Where should I locate a distribution center?
It depends entirely on the channel, and this is where generic network models go wrong. A retail DC replenishing stores locates near the largest cluster of brick-and-mortar locations, because outbound is a delivery route and the objective is minimizing route miles. An e-commerce facility locates against the parcel network — near FedEx and UPS hubs, positioned to keep the customer base in low zones, which is why Columbus, Indianapolis, Memphis, and Louisville are dense with e-commerce DCs. A B2B importer locates near the ports so inbound containers merge with outbound wholesale without a second long haul.
Q: What is the difference between a retail DC and an e-commerce DC?
The pick profile, and it changes everything downstream. Retail replenishment moves full cases and pallets to a known set of stores on a schedule. E-commerce picks eaches to individual addresses with no schedule. They want different layouts, different equipment, different slotting, and different labor. Running both under one roof is what omni-channel fulfillment means in practice, and it works when the building is designed for it — not when an e-commerce operation is bolted into the corner of a full-case building.
Q: When should I relocate or add a building instead of expanding?
When the constraint is the map rather than the four walls. Expansion solves a space problem. It does not solve a location that has drifted from the customer base, a labor market that cannot staff the building, or a channel the site was never positioned to serve. If the model says the footprint is in the wrong place, adding square footage there makes the wrong answer bigger.
Q: What usually goes wrong in a DC relocation?
The move, not the analysis. Inventory accuracy has to survive the transition, which means counting before, controlling during, and reconciling after. The new building has to be producing before the old one closes, which means running and staffing both for a period. And nothing about a transition should be scheduled inside a peak window — if the calendar forces a choice between moving late and moving during peak, move late.