Logistics Network Design and Facility Location

Introduction

Logistics network design determines how products, information and associated services move from suppliers to customers through a system of plants, warehouses, cross-docks, hubs, gateways, transport links and final delivery points. Facility-location decisions determine where these nodes should be placed; network-structure decisions determine how they should be connected and operated.

These decisions are strategically important because they normally involve substantial capital, long asset lives and difficult-to-reverse commitments. A network with too few facilities may provide economies of scale but create long lead times and expensive transport. A network with too many facilities may improve responsiveness while increasing inventory, labour, property, management and coordination costs. The objective is therefore not simply to minimize transport distance. It is to design an economically viable, service-capable, resilient and compliant network.

At postgraduate level, network design should be treated as a multi-objective decision problem. Cost, service, capacity, risk, sustainability, regulatory exposure and strategic flexibility must be evaluated together.

Learning Objectives

After completing this chapter, the reader should be able to:

1. Distinguish logistics network structures, distribution channels, hubs and gateways.
2. Explain the strategic trade-offs between centralized, decentralized, hub-and-spoke and hybrid networks.
3. Apply centre-of-gravity and load-distance methods, while recognizing their assumptions and limitations.
4. Formulate basic facility-location and capacity-allocation models.
5. Use scenario analysis to evaluate uncertainty, resilience and service consequences.
6. Interpret a practical network-design case and recommend an implementable solution.

1. The Architecture of a Logistics Network

A logistics network consists of nodes and arcs. Nodes include suppliers, manufacturing plants, consolidation centres, regional distribution centres (RDCs), fulfilment centres, cross-docks, ports, airports, rail terminals, parcel hubs, retail outlets and customers. Arcs represent physical, information or financial connections. Each arc has attributes such as distance, travel time, transport mode, capacity, cost, reliability and emissions.

A useful abstraction is a directed graph G = (N,A), where N is the set of nodes and A is the set of directed links. Product flow on a link is constrained by its capacity, while customer demand must be fulfilled subject to service and policy requirements.

Network structures

Centralized network. One or a small number of distribution centres serve a wide market. This structure often pools inventory and lowers facility costs, but it may increase outbound distance and exposure to a single disruption.

Decentralized network. Multiple regional facilities serve geographically defined markets. It can reduce delivery time and last-mile distance but duplicates inventory and fixed operating resources.

Hub-and-spoke network. Flows are consolidated through hubs. It offers high vehicle utilization and broad connectivity with relatively few direct routes. However, hub congestion, handling, missed connections and hub failure can have system-wide effects.

Direct-ship network. Suppliers or plants ship directly to customers. It reduces intermediate handling and may suit large, predictable orders, but it can produce low shipment density and poor transport utilization.

Cross-docking network. Inbound shipments are sorted and rapidly transferred to outbound vehicles with little or no storage. Cross-docking is effective when demand, schedules, labelling and information systems are synchronized.

Hybrid network. Most contemporary networks combine structures. For example, fast-moving products may be stocked in RDCs, slow-moving products may be shipped from a central facility, and urgent orders may use an express hub.

Distribution channels

A distribution channel describes the institutional route through which goods reach the customer. Typical forms include manufacturer-to-customer, manufacturer-to-retailer-to-customer, manufacturer-to-distributor-to-retailer, and omnichannel combinations involving stores, e-commerce fulfilment centres and parcel networks.

Channel design is broader than warehouse placement. It includes ownership, order visibility, replenishment responsibility, service promises, returns, pricing authority and information sharing. A channel may be physically efficient but strategically unsuitable if it creates channel conflict or weak customer access.

Hubs and gateways

A hub consolidates, sorts or transfers flows among multiple origins and destinations. A parcel hub, airline cargo hub and rail intermodal terminal are examples. A gateway is a strategically important entry or exit point to a region or network, such as a seaport, border crossing, airport or inland port. Gateways are influenced by customs, trade policy, infrastructure, security and geopolitical conditions.

The distinction matters: a hub primarily performs network-consolidation functions, whereas a gateway connects a network to an external geography or transport system. A facility can perform both roles.

2. Strategic Trade-offs in Network Design

The central trade-off is between scale efficiency and market responsiveness. Larger facilities can exploit purchasing power, automation and fixed-cost spreading. More facilities can reduce delivery distance and improve recovery after local disruptions. The relevant cost categories are:

• facility fixed cost and operating cost;
• inbound and outbound transport cost;
• inventory holding and pipeline inventory cost;
• order processing and handling cost;
• stockout, delay and expediting cost;
• taxes, duties, regulatory and compliance cost;
• carbon, congestion and environmental cost;
• risk and business-continuity cost.

A simplified total-cost expression is:

where TC is total relevant annual cost; F is facility cost; T is transport cost; H is inventory holding cost; P is processing and handling cost; S is service-failure cost; and R is risk, regulatory and other relevant cost. The expression is a managerial decomposition rather than a universal optimization formula. Each term must be defined consistently and double counting must be avoided.

Service dimensions

Facility location affects order-cycle time, delivery reliability, product availability, returns speed and customization capability. Distance is only a proxy for service. Actual performance depends on congestion, border processes, operating hours, carrier schedules, cut-off times and variability.

Inventory pooling

Centralization may reduce safety stock through risk pooling. If independent demand standard deviations are σ₁, σ₂, ..., σ_n, aggregate standard deviation under independence is:

If all locations have equal standard deviation σ, this becomes σ_total = σ√n. By contrast, separate stocking at n locations produces total standard deviation nσ under the same simplified assumptions. The pooling benefit exists only when demand correlation, replenishment lead times, service targets and allocation rules are properly considered. Positive correlation reduces the benefit.

3. Facility-Location Analysis

Facility-location analysis normally proceeds through five stages:

1. Define the decision scope, planning horizon, products, customers and service requirements.
2. Develop demand, transport, cost, capacity and risk data.
3. Generate feasible locations and network configurations.
4. Evaluate alternatives using screening, optimization and scenario analysis.
5. Select, implement and monitor the network.

A location should be assessed at the level at which the decision is actually made. A country-level decision may emphasize tax, labour, trade and political risk; a city-level decision may emphasize land, labour, road access and congestion; a site-level decision may emphasize zoning, utilities, drainage, security and truck circulation.

Centre-of-gravity method

The centre-of-gravity method provides a first approximation for a single facility serving geographically distributed demand. For customer or demand point i, let x_i and y_i be coordinates, and let w_i be annual demand or shipment weight. The weighted location is:

Coordinates must use a consistent system, and weights must be measured in a comparable unit. The method assumes that transport cost is approximately proportional to weighted distance, terrain and road topology do not materially distort the geometry, and there are no decisive constraints such as zoning or capacity. The result is a screening point, not an automatic site decision.

#### Worked example: centre of gravity

Suppose three markets have the following coordinates and annual demand.

Total demand is 450. Therefore:

The preliminary location is approximately (25.56, 24.44). A practical analyst would then test nearby industrial sites for road access, land availability, flood exposure, labour and legal feasibility.

Load-distance method

The load-distance method compares candidate sites by multiplying flow weight by distance. For candidate site j:

where d_ij is the distance from candidate j to demand point i. The preferred candidate has the smallest load-distance score if all other costs and constraints are equal. If transport cost per unit-distance differs by route, the more useful expression is:

where c_ij is cost per unit of weight per distance. This method is sensitive to the distance measure. Straight-line distance can be misleading; road distance or travel time is usually preferable for operational decisions.

Factor-rating method

Some location factors cannot be expressed reliably in monetary terms. A factor-rating method assigns a weight α_k to criterion k and a score s_jk to candidate j. The weighted score is:

Scores should be defined in advance, assessed by a cross-functional team and tested for sensitivity. The method provides transparency but does not eliminate managerial judgement or bias.

4. Optimization Models for Network Design

Optimization is appropriate when multiple facilities, flows, capacities and constraints must be considered simultaneously.

Capacitated facility-location model

Let:

• J = candidate facilities;
• I = customer zones;
• f_j = fixed cost of opening facility j;
• c_ij = variable cost per unit shipped from j to i;
• d_i = demand of customer i;
• K_j = capacity of facility j;
• y_j = 1 if facility j opens, otherwise 0;
• x_ij = quantity shipped from j to i.

A basic mixed-integer model is:

subject to:

The first constraint requires demand satisfaction. The second prevents shipments from a closed facility and imposes capacity. The objective minimizes fixed plus variable cost. The model is suitable only when costs, demand and capacities are represented adequately. It does not automatically capture service-time distributions, congestion, strategic risk or nonlinear transport tariffs.

The model can be extended with constraints for maximum delivery time, minimum service coverage, multiple products, modes, carbon emissions, supplier qualification, facility-opening limits and flow conservation.

Transportation and transshipment logic

A transportation model assigns supply to demand at minimum shipping cost. A transshipment model allows flows through intermediate nodes. Flow conservation at an intermediate node can be written as:

unless the node is permitted to store, consume or generate material. In real networks, inventory balance, loss, returns and time windows may require a time-indexed model.

Multi-objective design

A single cost-minimizing solution may be unacceptable if it violates service, resilience or sustainability expectations. One approach is a constrained model:

Another approach uses a weighted objective, but weights should be justified because a numerical weight can conceal a value judgement. Pareto analysis is often preferable: a solution is efficient if no objective can be improved without worsening at least one other objective.

5. Capacity Allocation and Scenario Analysis

Opening a facility does not solve the network problem unless capacity is allocated appropriately. Capacity may refer to storage positions, pallet movements, dock doors, labour hours, vehicle departures, order lines or processing throughput. These are not interchangeable.

A basic capacity-allocation rule assigns customer demand to facilities while respecting:

If demand exceeds available capacity, the planner must use overflow facilities, subcontracting, postponement, demand prioritization or additional shifts. Allocating only average demand is unsafe when variability is material. Effective capacity should reflect peaks, maintenance, absenteeism and disruption buffers.

Scenario analysis

Scenario analysis tests a design under alternative futures rather than relying on one forecast. Scenarios may vary demand, fuel prices, exchange rates, port reliability, labour cost, carbon prices, tariffs, disruptions and customer service requirements.

If scenario s has probability p_s and cost C_s, expected cost is:

Expected value alone is insufficient when severe losses are unacceptable. Decision-makers should also examine worst-case cost, probability of capacity violation, service shortfall and recovery time. A robust design may have a slightly higher expected cost but substantially lower disruption exposure.

A useful process is:

1. establish a base case;
2. identify uncertain drivers;
3. construct internally consistent scenarios;
4. solve or evaluate the network under each scenario;
5. identify decisions that remain satisfactory across scenarios;
6. define trigger points for expansion, relocation or mode change.

6. Technology, Data and Artificial Intelligence

Modern network design uses geographic information systems, transport-management systems, warehouse-management systems, digital twins, telematics and optimization software. GIS supports road-distance, travel-time, zoning and hazard analysis. Digital twins allow planners to test flows, congestion and capacity before physical implementation.

Artificial intelligence can assist demand forecasting, anomaly detection, route-time prediction, site screening and scenario generation. However, AI output is only as reliable as the data and assumptions behind it. A model trained on historical demand may reproduce structural bias or fail during a regime change. Governance should include data provenance, explainability appropriate to the decision, human review, access controls, cybersecurity and monitoring for model drift.

Data quality problems are particularly damaging in facility studies. Common errors include duplicate customer locations, obsolete demand histories, inconsistent units, missing returns, incorrect postal geocoding and treating forecast demand as actual demand. Analysts should document definitions, time periods, currency, inflation treatment, distance metric and confidence ranges.

7. Sustainability, Resilience and Governance

Network design affects emissions through facility energy use, transport distance, modal choice, empty running, packaging and returns. A meaningful assessment should distinguish direct facility emissions from purchased energy and value-chain transport emissions. Carbon should not be reduced by shifting burdens to congestion, unsafe work or excessive inventory obsolescence.

Resilience is the ability to prepare for, absorb, recover from and adapt to disruption. Relevant design measures include dual sourcing, geographic diversification, alternate gateways, emergency inventory, flexible contracts, interoperable systems and pre-qualified carriers. Redundancy has a cost, so it should be targeted at high-consequence vulnerabilities rather than applied indiscriminately.

Governance issues include land-use impacts, community consultation, labour standards, road safety, customs compliance, dangerous-goods regulations, privacy of location data and fair treatment of contracted workers. A low-cost location may be unacceptable if it relies on unlawful labour practices or creates unmanaged environmental harm.

8. An Integrated Design Procedure

A professional network study should integrate qualitative and quantitative methods rather than treating one technique as decisive.

Step 1: Define the service proposition. Specify delivery promise, order cut-off, product availability, returns and customer segments.

Step 2: Segment the flow. Separate products by value, velocity, temperature, hazard, demand variability and fulfilment urgency.

Step 3: Map the current network. Record nodes, lanes, modes, capacities, costs, lead times and failure points.

Step 4: Build a clean demand and cost baseline. Reconcile units, currencies, seasons, taxes, duties, inventory and transport rates.

Step 5: Screen locations. Apply regulatory, geographic, labour, infrastructure and risk filters before detailed optimization.

Step 6: Use centre-of-gravity and load-distance methods. These provide transparent initial guidance and help identify implausible candidates.

Step 7: Optimize feasible configurations. Include fixed cost, flow cost, capacity and service constraints.

Step 8: Test scenarios and sensitivity. Vary demand, costs, capacities, disruptions and service requirements.

Step 9: Evaluate implementation. Consider transition inventory, systems integration, workforce, contracts, permits and customer communication.

Step 10: Establish control metrics. Monitor cost per order, transport utilization, order-cycle time, on-time delivery, capacity utilization, inventory, emissions and disruption recovery.

Applied Case: NorthStar Medical Devices

The following is a fictional illustrative case.

Background

NorthStar Medical Devices supplies hospitals in three regions. It currently operates one national distribution centre near the manufacturing plant. The centre provides inventory pooling and low facility cost, but expedited shipments have increased because hospitals require next-day delivery for critical devices. A coastal port is also becoming unreliable during storm seasons.

Annual demand is 450,000 units: 100,000 in the west, 200,000 in the central region and 150,000 in the east. Two candidate RDCs are available: Site R, centrally located with annual capacity of 260,000 units, and Site E, near the eastern gateway with annual capacity of 220,000 units. The existing national facility can be retained with capacity of 450,000 units, or converted into a central replenishment centre with capacity of 180,000 units.

Problem

Management must decide whether to retain the centralized structure, open one RDC, or use a central replenishment centre plus an eastern RDC. The decision must account for cost, next-day service, gateway risk and medical-product continuity.

Analysis

The centre-of-gravity calculation using the three demand points provides a central screening location, but it does not represent the eastern gateway risk or the requirement for next-day delivery. A load-distance comparison favours a central site when all units are treated equally. However, service-weighted analysis gives higher importance to critical eastern demand and shows that an eastern facility reduces response time substantially.

A capacitated model reveals that Site E alone cannot serve all demand because its capacity is 220,000 units. Site R alone cannot serve all demand because its capacity is 260,000 units. A combined design can serve 480,000 units, creating 30,000 units of nominal capacity buffer. The buffer is not necessarily sufficient if peak demand or emergency stock is material; usable capacity must be assessed after reserving space for safety stock, quarantine and returns.

Scenario analysis produces three relevant cases:

Approach and solution

NorthStar selects a hybrid structure: the existing site becomes a central replenishment and quality-control centre, while Site E becomes an eastern RDC. Western and central routine demand continues to be served from the central facility, while eastern fast-moving and critical products are stocked at Site E. Qualified third-party logistics capacity is contracted as overflow rather than treated as permanent base capacity.

The design includes dual gateway capability, minimum emergency inventory for critical products, documented allocation priorities and a monthly capacity review. The model is recalculated quarterly using updated demand and disruption information.

Outcome

The selected design does not minimize nominal transport cost in every month. It improves next-day coverage for critical eastern customers, reduces dependence on the vulnerable gateway and preserves centralized quality-control activities. Its success depends on disciplined inventory segmentation, validated medical-product handling, accurate inventory visibility and a tested contingency plan.

Learning points

1. A geometric optimum may be operationally infeasible.
2. Capacity must be measured in the resource that constrains flow.
3. Service-critical products should not be evaluated only on average demand.
4. Resilience can justify a higher expected cost.
5. A network decision includes implementation and governance, not merely a model output.

Challenges and Limitations

No single method captures all network realities. Centre-of-gravity analysis ignores fixed costs, road networks and service constraints. Load-distance analysis can reward a site with poor access or high labour cost. Factor ratings can disguise subjective preferences behind precise-looking scores. Optimization can produce false precision when demand and cost estimates are uncertain. Expected-cost analysis can undervalue low-probability catastrophic events. Simulation can represent variability more realistically but requires detailed data and careful validation.

Other limitations include demand endogeneity, competitor responses, changing regulations, construction lead times, land-price volatility and organizational resistance. A technically superior design may fail if the firm cannot recruit labour, integrate systems, obtain permits or manage the transition.

Best-Practice Checklist

• Define service requirements before selecting locations.
• Separate screening methods from final decision models.
• Use road distance or travel time when operationally relevant.
• Include fixed, variable, inventory, service and risk costs without double counting.
• Model capacity by the actual bottleneck resource.
• Test correlation and variability rather than assuming independent demand.
• Evaluate disruption, recovery and alternate-mode options.
• Document assumptions, data provenance and decision rights.
• Include sustainability, labour and community impacts.
• Use AI as decision support, with validation and human accountability.
• Revisit the network when demand, regulation, technology or risk changes.

Chapter Summary

Logistics network design determines the configuration and operation of facilities, links, channels, hubs and gateways. Centralized networks pool inventory and exploit scale; decentralized networks improve proximity and responsiveness; hub-and-spoke, direct-ship, cross-docking and hybrid structures provide different combinations of consolidation, speed and risk.

Centre-of-gravity and load-distance methods are useful transparent screening tools, but they rely on simplifying assumptions. Factor-rating methods incorporate qualitative criteria, while capacitated facility-location and transportation models represent fixed costs, flows and capacity more formally. Scenario analysis extends the evaluation beyond a single forecast by examining uncertainty, disruption and service consequences. Capacity allocation must consider peak demand and bottleneck resources, not merely average volume.

A robust design integrates cost, service, inventory, resilience, sustainability, governance and implementation feasibility. Technology and AI can improve analysis, but only when supported by reliable data, validation and accountable governance. The professional objective is therefore not the mathematically cheapest network in isolation, but a network that performs acceptably across plausible futures and can be implemented responsibly.

Key Terms

Network design: Strategic configuration of logistics nodes, links, flows and operating policies.

Facility location: Selection of a physical site for a logistics or production activity.

Distribution channel: Institutional and physical route by which products reach customers.

Hub: Node that consolidates, sorts or transfers flows.

Gateway: Entry or exit point connecting a network to an external geography or mode.

Centralized network: Network with relatively few facilities serving broad markets.

Decentralized network: Network with multiple facilities positioned near market regions.

Hub-and-spoke: Structure in which flows are consolidated through central hubs.

Cross-docking: Transfer of inbound goods to outbound vehicles with minimal storage.

Centre of gravity: Demand-weighted geographic estimate of a facility location.

Load-distance method: Evaluation based on shipment weight multiplied by distance.

Factor rating: Weighted scoring method for comparing qualitative and quantitative criteria.

Capacity allocation: Assignment of demand or flow to facilities subject to resource limits.

Facility-location model: Optimization model selecting facilities and assigning flows.

Demand pooling: Aggregation of demand that may reduce total safety-stock requirements.

Scenario analysis: Evaluation of decisions under alternative assumptions about the future.

Resilience: Ability to withstand, recover from and adapt to disruption.

Digital twin: Data-enabled representation used to analyze or simulate a real system.

Pareto-efficient solution: Solution for which improving one objective requires worsening another.

Review Questions

Short Answer Questions

1. Distinguish a logistics hub from a gateway.
2. State two advantages and two disadvantages of a centralized distribution network.
3. What assumptions underlie the centre-of-gravity method?
4. Why can capacity utilization based on average demand be misleading?
5. Identify three governance issues in facility-location decisions.

Long Answer Questions

1. Compare centralized, decentralized, hub-and-spoke and hybrid network structures.
2. Explain how centre-of-gravity, load-distance and factor-rating methods can be combined in a location study.
3. Formulate and interpret the basic capacitated facility-location model.
4. Discuss how inventory pooling influences the choice between centralized and decentralized facilities.
5. Explain why resilience and sustainability should be integrated into network design rather than added after cost minimization.

Analytical / Application Questions

1. A firm serves four markets with unequal demand from two candidate RDCs. Design a data table and explain how you would compare the candidates using load-distance, fixed cost, capacity and service constraints.
2. Construct three plausible scenarios for a network exposed to port disruption. Specify the variables that change, the performance measures to calculate and the trigger points for managerial action.
3. A proposed AI model recommends closing a regional warehouse. Develop a governance and validation plan that tests data quality, service impact, bias, resilience and implementation feasibility.

Multiple Choice Questions with Answers

Question 1

Which network structure primarily consolidates flows through an intermediate node?

A. Direct-ship network

B. Hub-and-spoke network

C. Fully decentralized network

D. Point-of-use network

Correct Answer: B. Hub-and-spoke network

Question 2

In the centre-of-gravity method, the weighting variable normally represents what?

A. Annual demand or shipment volume

B. Number of employees at the facility

C. Building height

D. Product selling price only

Correct Answer: A. Annual demand or shipment volume

Question 3

Using demands 100, 200 and 150 at x-coordinates 10, 40 and 20, respectively, the weighted x-coordinate is approximately:

A. 20.00

B. 22.50

C. 25.56

D. 30.00

Correct Answer: C. 25.56

Question 4

The load-distance score for a candidate site is calculated as:

A. Fixed cost divided by capacity

B. Demand multiplied by selling price

C. Shipment weight multiplied by distance and summed

D. Capacity divided by lead time

Correct Answer: C. Shipment weight multiplied by distance and summed

Question 5

Which constraint in the basic capacitated facility-location model prevents shipments from exceeding an opened facility’s capacity?

A. ∑_j x_ij = d_i

B. ∑_i x_ij ≤ K_j y_j

C. y_j = 0 for every facility

D. ∑_s p_s = 1

Correct Answer: B. ∑_i x_ij ≤ K_j y_j

Question 6

What is a principal benefit of inventory pooling?

A. It always eliminates transport cost

B. It can reduce aggregate safety-stock requirements

C. It guarantees zero stockouts

D. It removes the need for forecasts

Correct Answer: B. It can reduce aggregate safety-stock requirements

Question 7

Why can expected cost alone be inadequate in scenario analysis?

A. It cannot be calculated when probabilities exist

B. It ignores all fixed costs by definition

C. It may undervalue low-probability, high-consequence disruptions

D. It always produces a decentralized network

Correct Answer: C. It may undervalue low-probability, high-consequence disruptions

Question 8

Which facility is most closely associated with rapid transfer from inbound to outbound vehicles?

A. Cross-dock

B. Long-term storage depot

C. Manufacturing plant

D. Retail showroom

Correct Answer: A. Cross-dock

Question 9

A gateway is best described as:

A. A local picking location only

B. An entry or exit point connecting a network to an external geography or mode

C. A demand forecast

D. A facility with no transport links

Correct Answer: B. An entry or exit point connecting a network to an external geography or mode

Question 10

Which practice is most appropriate when using AI for network-design decisions?

A. Accepting the model output without validation

B. Ignoring data provenance because the model is automated

C. Combining validation, human review and monitoring for model drift

D. Using historical data without testing regime changes

Correct Answer: C. Combining validation, human review and monitoring for model drift