A few years ago I sat in a conference room with the operations leadership of a building products manufacturer that had never attempted any real logistics network optimization. Four plants, three warehouses, one very frustrated vice president. He opened the meeting by sliding a freight invoice summary across the table and saying, “Our carriers are robbing us.”

In reality, they weren’t. Their rates were actually decent. What we found after about six weeks of digging was far less dramatic and far more expensive: one plant was shipping finished goods to customers sitting less than an hour from a sister plant that made the exact same product. Nobody had decided to do that. Instead, a sourcing rule written during a capacity crunch years earlier had simply never been switched off, and the ERP kept obeying it.

That story is the whole reason logistics network optimization exists. Multiple site industrial networks are rarely designed. Rather, they accumulate. A plant comes with an acquisition, a warehouse gets leased because one big customer demanded faster service, and a third party logistics provider gets added during a busy season and never leaves. Each decision made sense on its own day. Put together, however, they form a network nobody would ever draw on purpose.

This article, therefore, is the playbook I use when a manufacturer, distributor, or industrial services company with several sites asks me to make sense of their footprint. It isn’t theory. Instead, it’s what actually moves the numbers.

Why Multiple Site Networks Need Logistics Network Optimization

Before talking about methods, it helps to understand why good companies end up with inefficient networks. In my experience, there are four usual culprits.

Acquisitions and Shifting Demand

First, there’s growth by acquisition. When you buy a competitor, you inherit their plants, their warehouses, their carrier contracts, and their customer promises. Integration teams typically focus on finance, IT, and sales. As a result, logistics integration tends to get postponed, sometimes indefinitely.

Second, demand migrates. Customers move. For instance, a region that was your strongest market ten years ago may be flat now, while a corridor you barely served has tripled. Your facilities, though, are still sitting where demand used to be.

Product Growth and Local Thinking

Third, products proliferate. A plant that once made forty SKUs now makes four hundred. Consequently, some of those SKUs can only be produced on one line at one site, which forces long shipments that nobody budgeted for.

Finally, and this is the one people least like to admit, local decisions get optimized while the network does not. Each plant manager runs a tight ship and hits their own cost targets. However, plant level efficiency and network level efficiency are not the same thing. In fact, a plant can look brilliant on its own scorecard while generating terrible freight economics for the company as a whole. That gap is why network design belongs alongside your broader manufacturing cost reduction strategies rather than in a separate silo.

What Logistics Network Optimization Actually Covers

People use the phrase loosely, so let me be precise. Samsung SDS describes logistics network optimization as working out the right number, location, and size of warehouses for a business while accounting for the many constraints inside its supply chain. That’s a good starting definition, and you can read their full explanation in their white paper on logistics network optimization.

For multiple site industrial operations, however, I widen the scope a bit. A proper study answers five linked questions:

  • Which facilities should exist, and where?
  • Which plant should make which product for which customer region?
  • How should goods move between suppliers, plants, warehouses, and customers, including mode and consolidation choices?
  • Where should inventory sit, and how much?
  • How does the network hold up when something breaks?

If you treat these as separate projects, you will get separate, conflicting answers. The value comes from solving them together, because the cheapest warehouse location depends on which plant sources it, and similarly, the best sourcing plan depends on where the warehouses are.

The Cost Picture Right Now

It’s also worth grounding this in current numbers. The 37th annual State of the Logistics Union report from CSCMP and Kearney put US business logistics costs at 7.8% of GDP, down from 8.8% in the previous edition. Supply Chain Digest has a useful breakdown in its State of the Logistics Union 2026 summary.

Here’s my honest read on that. Softer freight markets make everyone’s numbers look better, and that, in turn, creates complacency. I’ve watched companies congratulate themselves on falling transport spend when the decline came entirely from market rates, not from anything they did. When the market tightens again, and it always does, a poorly structured network gets exposed fast. So the time to fix your footprint is when you have breathing room, not when rates spike. If you want a framework for that moment, our guide to industrial procurement strategy during price shocks pairs well with this one.

The previous year’s report made the same point in sharper terms. Specifically, Kearney partner Korhan Acar argued that the industry needs to stop relying on short term fixes and instead build resilience into networks, technology, and decision making as a strategic requirement. Penske covered that in its 2025 State of Logistics Report highlights.

Logistics Network Optimization Starts With Data, Not Software

Every vendor demo starts with a beautiful map full of moving dots. Every real project, on the other hand, starts with a spreadsheet full of problems.

What Data a Logistics Network Optimization Study Needs

The data you need for a credible study includes:

  • Twelve to twenty four months of shipment level history, both inbound and outbound, with origin, destination, weight, cube, mode, carrier, and actual cost.
  • Customer demand by location and product family, ideally with service requirements attached.
  • Plant capacity by line or work center, including shift patterns and realistic utilization ceilings.
  • Facility cost structures, split into fixed and variable components.
  • Inventory levels and turns by site and product.
  • Supplier locations and inbound lead times.

The fixed versus variable split matters more than most people expect. For example, if you load a warehouse into a model with a single blended cost per pallet, the model can’t tell you what happens if you close it, because it doesn’t know which costs actually go away.

Expect Dirty Data

Naturally, the data will be messy. On one project, our shipment weights looked absurd for one plant. It turned out that site recorded weight in kilograms while every other site used pounds, and nobody had ever reconciled it because nobody had ever looked at all the plants in one dataset. That single unit of measure issue would have pushed the model toward closing a perfectly good facility. Therefore, budget real time for cleansing. On most of my projects it eats 40 percent or more of the schedule, and I’ve stopped apologizing for that.

Build a Baseline Leadership Will Believe

Before you model a single “what if,” you model “what is.” In other words, the baseline is a replica of your current network, run through the same model you’ll use for scenarios.

The test is simple. Does the modeled cost come within a few percentage points of what the finance team says you actually spent? If the baseline is off by fifteen percent, then every scenario is suspect, and the CFO will spot it in the first meeting.

I spend a disproportionate amount of time here because the baseline is where trust is built. When a plant manager sees their own lanes, their own volumes, and their own cost reflected accurately, they stop treating the project as an outside attack and start treating it as a tool. Above all, that shift is worth more than any algorithm.

In addition, the baseline tends to surface the first round of easy wins before you’ve run any optimization at all. Lanes that shouldn’t exist. Customers served from the wrong site. Partial truckloads going out daily when two shipments a week would do. In fact, my building products client from the opening story got a meaningful chunk of their savings from baseline findings alone.

Center of Gravity in Logistics Network Optimization: A Starting Point, Not an Answer

Center of gravity analysis is usually the first tool people reach for when asking where a facility should go. The center of gravity is the point that minimizes total weighted transportation distance or cost across all the nodes in a network. Moreover, the method became popular because it is simple, intuitive, and fast, so much so that teams ran it manually in spreadsheets long before modern network design software existed. The logistics team at 4flow has a clear explanation in their article on center of gravity analysis for new logistics sites.

Where the Method Falls Short

The trouble, though, is that a pure center of gravity result will happily put your new warehouse in the middle of a lake, or in a town with no labor pool, or forty miles from the nearest interstate. It knows nothing about lease costs, taxes, or road networks.

That’s why I treat it the way the better practitioners do. AIMMS, for example, recommends using center of gravity analysis as an early ideation step to identify candidate locations, which then feed into a broader and more complete network design study. Their write up on center of gravity best practices is worth reading if you’re new to the method. Treat that candidate list the same way you would any engineering feasibility study: a screen, not a verdict.

When Adding a Facility Saves Money

Used properly, the results can be very tangible. In one DSV project, a center of gravity analysis was used to compare a one DC network against a two DC setup for US demand. As a result, the two DC option delivered estimated savings of $289K a year, about 7% of baseline, while lifting two day customer coverage by 15% and three day coverage by 11%. You can see the details in DSV’s case study on adding an extra distribution centre after centre of gravity analysis.

Notice what that example shows. Adding a facility reduced total cost. People instinctively assume fewer buildings means lower spend; however, outbound freight often dwarfs warehouse rent. Sometimes the right move is to add a node, not cut one.

The Plant Side: Sourcing and Capacity

This is where multiple site industrial operations differ from a pure distribution network, and where a lot of off the shelf studies fall short.

Why Manufacturing Networks Are Different

In a distribution network, any warehouse can usually ship any SKU. In a manufacturing network, by contrast, that’s rarely true. Certain products can only run on certain lines. Likewise, some plants have certifications others lack, and changeover times make it uneconomical to split small runs between sites. Your model, therefore, has to reflect all of this or its recommendations will be physically impossible.

A good example comes from an AIMMS engagement with a fast growing food producer. The company ran two manufacturing sites and four third party distribution centers moving over 100 SKUs. Meanwhile, management weighed options ranging from adding a fifth distribution center to removing underused ones, all while dealing with perishable products, a large SKU count, and the fact that some SKUs could only be made in one factory. The full story is in their case study, Network Analysis and Optimization Leads to More Efficient Operations.

Handling Single Source Products and Real Capacity

That last constraint, single source SKUs, shows up in nearly every industrial client I work with. The practical answer is usually some mix of three things. First, rebalance which plant serves which region for products that can be made in multiple places. Next, accept the long haul for truly single source items but consolidate them aggressively. Finally, flag single source products as a resilience risk for leadership to consider when making capital decisions.

I also push clients to model realistic capacity, not nameplate capacity. For instance, a line rated at 100 units an hour that actually averages 78 after downtime and changeovers needs to go into the model at 78. Otherwise, the optimizer will load that plant to the brim on paper, and you’ll be expediting freight in real life.

Transportation Flows: Where Logistics Network Optimization Finds the Money

Once facilities and sourcing are set, the next layer is how goods physically move. Typically, this is where I find the largest share of near term savings.

The levers include mode selection between parcel, LTL, full truckload, intermodal, and rail; consolidation of shipments across plants heading to the same customers; plant to plant transfers that should be scheduled rather than ad hoc; and backhaul opportunities between your own sites.

The Cross Site Blind Spot

Multiple site manufacturers have a particular weakness here. Each plant tends to book its own outbound freight, which means two plants may be sending half full trucks to the same customer on the same day. Yet nobody sees it, because nobody looks across sites.

RXO published a case that illustrates the point well. A multisite manufacturer with heavy outbound freight spend, serving both major retail chains and project site deliveries from several plants, set goals to better use its dispersed plant network and achieve at least 10% net savings against its freight baseline. Then the rollout covered all manufacturing locations over about four months, starting with the sites that had the highest savings potential, and the optimization was designed to cut outbound freight costs by 10% to 15% at full run rate. Their case study on solving network complexity for a multisite manufacturer is a good read.

What I like about that example is the sequencing. They didn’t try to fix everything at once. Instead, they started where the money was, and I recommend the same to every client.

Inventory Positioning Across Echelons

Inventory is the part of logistics network optimization that finance teams care about most and logistics teams sometimes forget.

When you add a warehouse, you don’t just add rent. You also add safety stock. Every new stocking location needs its own buffer against demand variability, and those buffers don’t scale down neatly. Consequently, three warehouses each holding two weeks of safety stock typically carry far more total inventory than one central warehouse covering the same demand.

The fix is to think in echelons. Which products should be stocked everywhere? And which should sit only at the plant or a central hub and ship on demand? For example, a fast moving maintenance part that customers need same day belongs close to them. On the other hand, a slow moving engineered component that customers order three weeks ahead can sit in one place.

For this reason, I usually segment products by velocity and service expectation before running any network scenarios. It takes a couple of days, and it consistently prevents the model from recommending a footprint that looks cheap on freight but quietly doubles working capital. Our piece on inventory management for manufacturers covers that segmentation in more depth.

Designing Logistics Network Optimization for Disruption, Not Just Cost

Five years ago, most network studies I ran had one objective: minimize total landed cost while meeting service targets. Today that’s no longer enough, and frankly it was never really enough.

Now I build resilience into the model from the start. That means testing how the network performs if a plant goes down for a month, if a major port slows, if a key supplier fails, or if tariffs reshape sourcing economics overnight. A structured supply chain risk assessment gives you the list of disruptions worth modeling, and a solid supplier qualification process reduces how often you’ll need those backup scenarios.

The financial case for this is strong. Writing in Supply Chain Management Review, one consultancy reported that across hundreds of network design projects it consistently finds between 5% and 15% in annual cost avoidance, with scenario testing letting companies balance cost, service, and flexibility using data. The full piece is Designing Supply Chain Networks That Matter.

Still, the key word there is balance. A resilient network usually costs a little more to run in a normal month. Therefore, leadership needs to see that trade off clearly, in dollars, so they can choose deliberately. My job is not to pick the answer for them. Rather, it’s to make the cost of each option visible.

Logistics Network Optimization Scenarios I Always Run

Every network is different, but there are a handful of scenarios I run on almost every multiple site industrial project:

  • Baseline optimized: Keep every existing facility but let the model reassign sourcing and customer allocation freely. This shows how much savings are available without touching real estate.
  • Footprint reduction: Test closing each facility one at a time, then in combinations.
  • Footprint expansion: Add candidate sites from the center of gravity work and see if any earn their keep.
  • Single plant outage: Take your largest plant offline for 30 days and see what breaks.
  • Demand shift: Grow your fastest growing region by 30 percent and shrink your weakest by 20 percent.
  • Cost shock: Raise fuel and freight rates by 25 percent to see how sensitive the network is.
  • Service upgrade: Tighten delivery commitments for top customers and measure what it costs.

Of these, the first scenario is the one I’d beg any company to run even if they do nothing else. It’s consistently the highest return, lowest risk finding, because it requires no capital and no layoffs. Just better rules.

Logistics Network Optimization Tools and Methods Worth Knowing

You don’t need to be a mathematician to lead a network project. Even so, you should know what’s under the hood.

Optimization and Simulation

Most serious network design tools rely on mixed integer linear programming. For example, a team of researchers built an optimization tool for forward logistics networks that combines center of gravity methods with mixed integer linear programming and business heuristics to find optimal solutions with low runtime, then recommends an alternative facility network that lowers overall costs compared with the existing one. Their paper, Advanced Quantitative Techniques to Solve Center of Gravity Problem in Supply Chain, is technical but readable.

Simulation is the other major method. While optimization tells you the best static design, simulation tells you how that design behaves day by day with real variability. Accordingly, I use optimization to pick the shape of the network and simulation to stress test it. Platforms like anyLogistix pair greenfield analysis with features such as road network integration, which goes some way toward fixing the “warehouse in a lake” problem. Their article, The Center of Gravity: A Smarter Way to Design Supply Chains, shows how that works in practice.

A Word on AI and Digital Twins

AI and digital twins get a lot of airtime now. SupplyChain247 notes that more businesses are pursuing network design approaches that weigh resilience and proximity as heavily as cost and efficiency, with AI adding predictive analytics, prescriptions, automation, and optimization. See Mastering Supply Chain Network Design. My honest take is that those tools are genuinely useful once your data foundation is solid. However, bolted onto messy data, they produce confident nonsense faster.

Implementation: Where Good Models Go to Die

I’ve seen excellent network studies end up as a slide deck nobody acts on. The model was right; the organization, however, wasn’t ready.

A few things separate the projects that deliver from the ones that stall.

Involve plant leadership early. If a model recommends shifting volume away from a plant, that plant’s manager will find a hundred reasons it won’t work. Admittedly, some of those reasons will be real. So get them into the room during baseline validation, not after the recommendation is final.

Phase the changes. Sourcing rule changes can happen in weeks. Carrier consolidation, meanwhile, takes a quarter, and facility moves take a year or more. Sequence them so early wins fund and justify the harder moves. Good engineering project management best practices apply directly here.

Set governance. Decide who owns network decisions going forward. Without clear ownership, the same drift that created the problem will recreate it within a few years. A formal engineering change management process is a useful model for controlling sourcing rule changes.

Measure against the model. Track actual cost and service against what the model predicted. When they diverge, find out why. In some cases the model missed something; in others, the operation didn’t follow the plan.

Keep the Model Alive

The biggest mistake I see is treating logistics network optimization as a one time project. You run a study, you implement it, and then you put the model on a shelf. Three years later, demand has shifted, a new plant has come online, and you’re back where you started.

Fortunately, the industry is moving away from that approach. 4flow observes that supply chain professionals increasingly see strategic network design as a continuous improvement process rather than a single project, and that modern software makes it practical to revisit the analysis regularly as the supply chain changes.

My recommendation for most midsize industrial companies is a light refresh every quarter, mainly updating demand and cost data and rerunning the baseline optimized scenario. In addition, plan a full strategic review every two to three years or whenever there’s a major event like an acquisition.

A Quick Logistics Network Optimization Readiness Checklist

If you’re considering a network study, ask yourself these questions honestly:

  • Can you pull twelve months of shipment level data from every site in a consistent format?
  • Do you know your true fixed and variable costs by facility?
  • Do you know the realistic, not nameplate, capacity of each plant?
  • Is there an executive sponsor willing to make decisions that cross plant boundaries?
  • Do you have a clear definition of service by customer segment?

If you answered no to more than two, start there. After all, fixing those gaps will deliver value on its own and will make any future study faster and more accurate.

Final Thoughts

Ultimately, logistics network optimization isn’t about software or maps or algorithms, though all three help. It’s about stepping back far enough to see your operation as one system instead of a collection of sites that happen to share a logo.

Every multiple site industrial network I’ve worked on had money hiding in it. In one case it was a forgotten sourcing rule. Another time, two plants were sending half empty trucks to the same customer. Elsewhere, a warehouse should have been closed years ago, or one should have been opened. Either way, the only way to find it is to look at the whole thing at once, with clean data and an honest baseline.

So if your network grew the way most do, one reasonable decision at a time, it’s probably worth a fresh look.

Frequently Asked Questions

What is logistics network optimization?

It’s the process of deciding how many facilities a company should run, where they should be, which sites should serve which customers, and how goods should flow between them to balance cost, service, and risk. The Samsung SDS white paper is a solid primer.

How much can a multiple site manufacturer save through logistics network optimization?

Results vary, but published figures often land between 5 and 15 percent of the relevant cost base. For instance, Supply Chain Management Review reports that range in annual cost avoidance across many network design projects, and RXO’s case study targeted 10 to 15 percent in outbound freight.

Is center of gravity analysis enough to choose a warehouse location?

No. It’s a useful first step for generating candidate sites; however, it ignores labor, real estate, taxes, and road access. That’s why AIMMS and 4flow both recommend feeding its results into a full network model.

How often should a network study be updated?

A light quarterly refresh plus a full strategic review every two to three years works for most industrial companies. Additionally, run a review after any acquisition or major demand shift.

Does adding a distribution center always increase costs?

Not necessarily. In fact, lower outbound freight and better service can outweigh added facility costs, as DSV’s two DC case study showed with roughly 7 percent savings.

What role does resilience play in network design?

A growing one. Modern studies, therefore, test plant outages, supplier failures, and cost shocks alongside pure cost scenarios. Penske’s summary of the State of Logistics Report highlights resilience as a strategic priority.

References