Network Design & Optimization
Supply chain network design is the strategic process of determining the optimal number, location, size, and role of facilities — factories, warehouses, distribution centers, cross-docks — and the transportation lanes that connect them. It is one of the highest-impact decisions a company makes because network structure locks in fixed costs, constrains service levels, and shapes transportation spend for years.
A well-designed network balances three competing objectives: minimizing total logistics cost, meeting customer service commitments (delivery speed and reliability), and maintaining enough flexibility to absorb demand shifts, disruptions, and growth.
Network design (also called network optimization or supply chain design) is the analytical process of selecting facility locations, assigning customer demand to facilities, and determining the flow of goods through the network to minimize total landed cost while satisfying service constraints.
Why Network Design Matters
Network decisions are among the longest-lived and most capital-intensive in logistics. Opening or closing a distribution center involves lease commitments, equipment investment, workforce hiring, and systems deployment — decisions that are difficult and expensive to reverse.
| Decision | Typical Commitment | Cost to Reverse |
|---|---|---|
| Open a new warehouse | 3–10 year lease | Early termination penalties, relocation costs |
| Close a facility | 6–18 month wind-down | Severance, inventory redeployment, customer disruption |
| Change facility role (e.g., DC → cross-dock) | 6–12 months | Racking removal, process redesign, system reconfiguration |
| Add/remove a transportation lane | 1–3 months | Carrier re-contracting, routing changes |
| Shift sourcing origin | 3–12 months | Supplier qualification, lead time adjustment |
Because of this permanence, network design relies on rigorous quantitative modeling rather than intuition. The analytical methods described below help companies evaluate thousands of possible configurations to find the best trade-offs.
The Network Design Process
A typical network design study follows a structured methodology, whether conducted internally or by a consulting firm:
Step 1: Define Scope and Objectives
Before modeling, the team must clearly define:
- Planning horizon — How far into the future should the network be optimized? (Typically 3–5 years)
- Objective function — Minimize total cost? Maximize service? A weighted combination?
- Constraints — Maximum delivery time, minimum service levels, budget caps, existing facility commitments
- Scope — Which products, customers, suppliers, and geographies are included?
- Decision variables — What can change? (facility open/close, capacity expansion, lane selection, sourcing shifts)
Step 2: Collect and Validate Data
Network modeling requires detailed data across multiple domains:
| Data Category | Examples | Source |
|---|---|---|
| Demand | Customer locations, order volumes, SKU mix, seasonality | ERP, OMS |
| Supply | Supplier locations, lead times, purchase costs, capacity | Procurement |
| Facilities | Current locations, capacity, fixed costs, variable costs, lease terms | Real estate, finance |
| Transportation | Lane rates, transit times, modal options, accessorial costs | TMS, carrier contracts |
| Inventory | Carrying costs, safety stock policies, service level targets | WMS, planning |
| Product | Weight, cube, temperature requirements, hazmat classification | Product master |
The most frequent cause of unreliable network models is poor data quality — especially transportation costs. If your baseline model cannot replicate actual spending within 3–5%, the scenario results will not be trustworthy. Invest time in data validation before running optimization.
Step 3: Build and Validate the Baseline
The baseline model replicates the current-state network. It includes all existing facilities, actual customer assignments, current transportation lanes, and real costs. The purpose is to create a validated starting point against which alternative scenarios are measured.
A good baseline should match actual total logistics cost within ±5% and replicate known customer service metrics.
Steps 4–6: Scenario Modeling, Optimization, and Analysis
With a validated baseline, the team designs alternative scenarios (add a facility, consolidate two warehouses, shift to regional distribution) and runs optimization algorithms to find the lowest-cost configuration for each scenario. Results are compared across scenarios and tested for sensitivity to key assumptions.
Types of Network Design Studies
Network design studies fall into three broad categories depending on whether existing facilities constrain the analysis:
Greenfield Analysis
Greenfield analysis (also called center-of-gravity analysis) ignores all existing facilities and asks: "If we could start from scratch, where would we put our facilities?" The algorithm places a specified number of facilities at locations that minimize the total weighted distance (or cost) to all demand points.
The center of gravity method uses weighted averages of customer coordinates:
Optimal X = Σ(dᵢ × xᵢ) / Σ(dᵢ)
Optimal Y = Σ(dᵢ × yᵢ) / Σ(dᵢ)
Where dᵢ is the demand weight of customer i, and (xᵢ, yᵢ) is the customer's geographic coordinate.
Greenfield analysis assumes no existing infrastructure. It produces theoretical optimal locations — "demand-weighted centers" — that serve as starting points for more detailed analysis. The name comes from the idea of building on undeveloped ("green field") land.
Greenfield results are theoretical — the optimal point might fall in a lake or a residential area. But they reveal where demand concentration justifies facility placement and expose how far current facilities deviate from the ideal.
Brownfield Analysis
Brownfield analysis starts with the existing network and evaluates modifications — opening new facilities, closing underperforming ones, reallocating customers, or changing facility roles. It accounts for sunk costs, remaining lease obligations, and transition expenses.
Brownfield analysis is more practical than greenfield because it incorporates the real constraints companies face: existing leases, trained workforces, installed systems, and customer relationships.
Hybrid Approach
Most real-world network studies combine both methods:
- Run greenfield analysis to identify the theoretical ideal
- Compare greenfield results against the current network
- Design brownfield scenarios that move the current network toward the greenfield ideal within practical constraints
- Evaluate the cost and risk of each transition step
Mathematical Optimization Models
Behind network design tools are mathematical optimization models, primarily mixed-integer linear programming (MILP), that find the lowest-cost solution across millions of possible configurations.
The Facility Location Problem
The classic facility location problem (FLP) is the mathematical foundation of network design. It determines which facilities to open from a set of candidates and how to assign customers to those facilities.
Uncapacitated Facility Location Problem (UFLP)
The UFLP assumes facilities have unlimited capacity. The objective is to minimize the sum of fixed facility costs and variable transportation costs:
Minimize: Σⱼ fⱼyⱼ + Σᵢ Σⱼ cᵢⱼxᵢⱼ
Subject to:
Σⱼ xᵢⱼ = 1 for all customers i (each customer assigned to exactly one facility)
xᵢⱼ ≤ yⱼ for all i, j (can only assign to open facilities)
yⱼ ∈ {0, 1} (facility j is open or closed)
xᵢⱼ ≥ 0 (assignment fractions)
Where:
fⱼ= fixed cost of opening facilityjcᵢⱼ= cost of serving customerifrom facilityjyⱼ= binary decision: open facilityj(1) or not (0)xᵢⱼ= fraction of customeridemand served by facilityj
Capacitated Facility Location Problem (CFLP)
The CFLP adds capacity constraints: each facility can handle only a limited volume. This is more realistic for warehouse and plant location decisions where throughput is bounded by floor space, dock doors, labor availability, or equipment.
The key additional constraint is:
Σᵢ dᵢxᵢⱼ ≤ Qⱼyⱼ for all facilities j
Where dᵢ is the demand of customer i and Qⱼ is the capacity of facility j.
P-Median Problem
The p-median problem locates exactly p facilities to minimize the total demand-weighted distance. Unlike the UFLP, it fixes the number of facilities rather than letting the optimizer choose. This is useful when budget constraints dictate the maximum number of facilities a company can operate.
Multi-Period Models
Static models optimize for a single time snapshot. Multi-period models plan the network evolution over several years, deciding when to open, close, or expand facilities based on projected demand growth. This captures the timing of capital investments and avoids myopic decisions.
Key Cost Components in Network Modeling
The total landed cost optimized in network design includes several categories:
| Cost Component | Description | Behavior |
|---|---|---|
| Facility fixed costs | Rent/lease, insurance, property tax, base utilities, base labor | Fixed per facility; increases stepwise with capacity |
| Facility variable costs | Labor per unit, handling, packaging, utilities per throughput | Increases proportionally with volume |
| Inbound transportation | Supplier → facility; often ocean/rail for long haul, truck for short | Depends on sourcing decisions and facility locations |
| Outbound transportation | Facility → customer; typically the largest variable cost | Heavily influenced by facility proximity to demand |
| Inventory carrying cost | Capital, storage, obsolescence, insurance on held inventory | Increases with number of stocking locations (safety stock multiplier) |
| Customs and duties | Import taxes, broker fees, compliance costs | Relevant for international networks; varies by origin/destination |
| Risk and penalty costs | Stockout cost, late delivery penalties, disruption impact | Captures service-level trade-offs |
The Inventory–Transportation Trade-off
One of the most important dynamics in network design is the trade-off between inventory cost and transportation cost:
- Fewer, larger facilities → lower inventory (risk pooling reduces safety stock) but higher outbound transportation (facilities are farther from customers)
- More, smaller facilities → higher inventory (safety stock duplicated at each location) but lower outbound transportation (closer to customers)
The optimal number of facilities balances these opposing forces:
The square root law of inventory provides a rough estimate: when consolidating from n locations to m locations, total safety stock changes by a factor of √(m/n). Consolidating from 4 warehouses to 1 reduces safety stock to approximately 50% (√(1/4) = 0.5). This relationship helps justify centralization — but only if the resulting transportation cost increase does not exceed the inventory savings.
Service-Level Constraints
Network design is not purely a cost-minimization exercise. Service-level constraints ensure that the optimized network can actually serve customers within acceptable timeframes.
Common service constraints include:
| Constraint Type | Example | Impact on Network |
|---|---|---|
| Maximum delivery time | 95% of orders delivered within 2 days | Forces facilities closer to demand; increases facility count |
| Same-day / next-day zones | Major metros require next-day ground delivery | Requires facilities within 150–250 miles of major population centers |
| Order cut-off times | Orders placed by 2 PM ship same day | Affects processing capacity and carrier pickup schedules |
| Fill rate targets | 98% line-item fill rate | Requires broader inventory assortment at more locations |
| Lead time from origin | International orders: 4–6 weeks; domestic: 1–3 days | Drives inventory positioning and safety stock levels |
Service-Distance Mapping
A practical tool in network design is mapping the percentage of demand reachable within various transit-time windows from each candidate facility. In the United States, for example:
- 1-day ground covers roughly a 250-mile radius from a DC
- 2-day ground covers approximately 500–750 miles
- 3-day ground reaches most of the contiguous U.S. from strategically placed locations
A common benchmark: 2 strategically placed DCs (e.g., East Coast + West Coast) can reach ~95% of the U.S. population within 3-day ground. Adding a 3rd central DC extends 2-day coverage to ~80%+ of the population.
Scenario Analysis and Sensitivity Testing
After the optimizer finds the minimum-cost network for each scenario, the results must be tested for robustness:
Scenario Comparison
| Scenario | # Facilities | Total Cost | Avg. Delivery Time | Risk |
|---|---|---|---|---|
| Baseline (current state) | 5 | $42M | 2.1 days | Medium |
| Consolidate to 3 | 3 | $38M | 2.6 days | Lower |
| Expand to 7 | 7 | $46M | 1.6 days | Higher |
| Relocate + add 1 | 5 | $39M | 1.9 days | Medium |
Sensitivity Analysis
Test how results change when key assumptions shift:
- Demand variation: What if demand grows 20% faster or slower than projected?
- Fuel cost: What if fuel surcharges increase by 30%?
- Labor cost: What if wage rates in a target market rise significantly?
- Service requirements: What if customers demand next-day instead of 2-day?
- Disruption: What happens if the largest facility goes offline for 30 days?
Selecting a network based solely on the lowest-cost scenario without testing sensitivity is dangerous. The "optimal" network may be brittle — a small change in demand or costs could make a different configuration better. Always evaluate the robust solution (one that performs well across multiple scenarios) rather than the optimal solution for a single scenario.
Network Archetypes
Different business models and product characteristics lead to different network structures:
Direct-Ship Network
Goods move directly from manufacturer or supplier to the end customer, bypassing intermediate storage.
- Best for: Heavy/bulky products, low-volume / high-value items, made-to-order
- Advantages: No warehousing cost, no inventory holding cost
- Disadvantages: Longer lead times, higher per-shipment transport cost, limited carrier consolidation
Hub-and-Spoke Network
A central hub receives and consolidates shipments, then redistributes to regional spokes (or customers).
- Best for: LTL carriers, parcel networks, airlines
- Advantages: Consolidation economies, simplified lane management
- Disadvantages: Additional handling, hub becomes single point of failure
Regional Distribution Network
Multiple DCs positioned across geographies, each serving a region. Inventory is replicated at each location.
- Best for: Fast-moving consumer goods, e-commerce with next-day delivery promises
- Advantages: Fast delivery, carrier zone reduction
- Disadvantages: Inventory duplication, higher facility costs
Tiered Network
A multi-echelon structure with central DCs feeding regional DCs, which feed local delivery points.
- Best for: Large product catalogs where not all items can be stocked everywhere
- Advantages: Full assortment at central level + fast delivery for top sellers at regional level
- Disadvantages: Complexity, inter-facility transfers, more safety stock
Choosing the Right Archetype
| Factor | Favors Centralization | Favors Decentralization |
|---|---|---|
| Product value | High (reduce inventory investment) | Low (inventory cost manageable) |
| Demand variability | High (risk pooling benefits) | Low (predictable demand at each location) |
| Delivery speed requirement | Relaxed (2–5 day acceptable) | Aggressive (next-day, same-day) |
| Product variety (SKU count) | High (cannot replicate all SKUs) | Low (can stock everything everywhere) |
| Shipment weight | Heavy (transport cost dominates) | Light (transport cost per unit is low) |
| Order frequency | Low (few large orders) | High (many small orders) |
Network Design for International Supply Chains
Global networks add complexity beyond domestic distribution:
Additional Factors
- Customs and duties: Import duty rates vary by country of origin and HS code classification. Free trade agreements can eliminate duties if rules of origin are met.
- Foreign trade zones (FTZ) and bonded warehouses: Defer or reduce duty payments by storing goods in designated zones. Particularly valuable when goods are re-exported or duty rates are high.
- Transfer pricing: Goods moving between affiliated entities across borders must follow arm's-length pricing rules. Network design must align with the company's transfer pricing strategy.
- Currency risk: Operating costs in multiple currencies introduce exchange rate variability.
- Lead time variability: Ocean freight transit times (4–6 weeks) require larger safety stock buffers than domestic supply. See Ocean Freight.
- Regulatory environment: Local labor laws, environmental regulations, and import/export restrictions constrain where facilities can operate.
Near-Shoring and Reshoring Considerations
The decision to locate facilities closer to end markets (near-shoring) versus in low-cost countries (offshoring) involves a total cost analysis:
| Factor | Offshoring | Near-Shoring |
|---|---|---|
| Unit production cost | Lower | Higher |
| Transportation cost | Higher (ocean + drayage) | Lower (truck/rail) |
| Lead time | 4–8 weeks | 1–2 weeks |
| Inventory requirement | Higher (pipeline + safety stock) | Lower |
| Duty exposure | Varies (may benefit from FTAs) | Varies (may benefit from FTAs) |
| Disruption risk | Higher (distance, port congestion, geopolitics) | Lower |
| Responsiveness to demand | Slow (long replenishment cycles) | Fast (short cycles) |
| Minimum order quantities | Typically higher | Typically lower |
Network Design Tools and Software
Several commercial platforms support network design modeling:
| Capability | Description |
|---|---|
| Greenfield / center-of-gravity analysis | Identifies theoretical optimal facility locations |
| MILP optimization | Solves facility location, allocation, and flow problems |
| Scenario management | Stores and compares multiple network configurations |
| Multi-period modeling | Optimizes network evolution over time |
| Demand mapping and visualization | Geospatial display of demand, facilities, and flows |
| Sensitivity analysis | Tests robustness of solutions against assumption changes |
| What-if simulation | Models disruption, demand shifts, and cost changes |
Common platforms include Coupa (LLamasoft), AIMMS SC Navigator, anyLogistix, Llamasoft Supply Chain Guru, Blue Yonder, and Oracle Supply Chain Planning. Open-source tools like PuLP (Python) and Google OR-Tools can solve smaller facility location problems.
Key Performance Indicators
Once a network is implemented, ongoing monitoring ensures it continues to perform:
| KPI | Definition | Target |
|---|---|---|
| Total logistics cost as % of revenue | All network costs (transport, warehouse, inventory) / revenue | Industry-dependent; typically 4–10% |
| Order-to-delivery time | Time from order placement to customer receipt | Per service-level commitment |
| % demand within X-day ground | Fraction of customers reachable within a transit-time window | 90%+ within 2-day (common U.S. benchmark) |
| Facility utilization | Actual throughput / capacity | 70–85% (allows for peak absorption) |
| Inventory turns | Annual COGS / average inventory value | Higher is better; varies by industry |
| Transportation cost per unit shipped | Total outbound transport cost / units | Trending downward after optimization |
| Perfect order rate | Orders delivered complete, on time, undamaged, with correct documentation | 95%+ |
Best Practices
-
Rerun network design every 2–3 years — or sooner if demand patterns, sourcing, or service requirements shift materially. Networks optimized for yesterday's demand may not serve tomorrow's customers.
-
Use total landed cost, not just transportation cost — A facility that reduces transport cost but increases inventory carrying cost may not improve total economics.
-
Model realistic constraints — Include existing lease obligations, workforce availability, building suitability, and time-to-open. An unconstrained model gives theoretically perfect but practically useless results.
-
Test for disruption resilience — Remove the largest facility from each scenario and measure the impact. A robust network degrades gracefully; a fragile one collapses.
-
Involve cross-functional stakeholders — Network design affects procurement (inbound lanes), sales (delivery promises), finance (capital allocation), and HR (workforce planning). Decisions made in isolation often fail in implementation.
-
Start with greenfield, refine with brownfield — Use greenfield analysis to understand the gap between current and ideal, then design a practical brownfield migration path.
-
Document assumptions explicitly — Every model contains assumptions about demand growth, cost inflation, service requirements, and exchange rates. Document them so future analysts can understand why decisions were made and when re-evaluation is needed.
-
Consider phased implementation — Moving from a 5-DC network to a 3-DC network is disruptive. A phased approach — close one facility, observe, adjust, then close the next — reduces risk.
Resources
| Resource | Description | Link |
|---|---|---|
| CSCMP Supply Chain Management Process Standards | Industry-standard definitions and process frameworks for network design | cscmp.org |
| MIT Center for Transportation & Logistics | Research on supply chain network optimization, facility location, and inventory theory | ctl.mit.edu |
| Google OR-Tools | Open-source optimization library supporting facility location and vehicle routing problems | developers.google.com/optimization |
| INFORMS (Institute for Operations Research) | Academic and practitioner resources on operations research and optimization methods | informs.org |
| ASCM (Association for Supply Chain Management) | Certifications (CSCP, CPIM) and body of knowledge covering network design principles | ascm.org |
Related Topics
- Supply Chain Strategy — Introduction — strategic framework that governs network design decisions
- Reverse Logistics — designing networks for reverse product flow
- Warehouse Management — the operational execution within network nodes
- 3PL & Contract Logistics — outsourcing facilities within the network
- FTL vs LTL — transportation mode decisions shaped by network structure
- Intermodal Transport — multi-modal lanes in network planning
- Temperature-Controlled Logistics — cold chain network design constraints