Logistics Network Design and Facility Location

Introduction

Logistics network design determines how products, information and associated services move between suppliers, plants, warehouses, distribution centres, retailers and customers. It is a strategic decision because facilities and transportation lanes usually involve substantial capital, long asset lives and contractual commitments. A poorly designed network can create excessive transport cost, slow delivery, duplicated inventory and weak resilience. A well-designed network balances cost, service, risk, flexibility and sustainability.

Network design differs from daily transport planning. Transport planning assigns shipments to existing routes and carriers; network design decides the number, location, role and capacity of facilities and the structure of flows between them. Typical decisions include:

• whether to ship directly or through an intermediate facility;
• how many distribution centres to operate;
• where facilities should be located;
• which customers and suppliers each facility should serve;
• whether to consolidate, cross-dock or hold inventory;
• which transport modes and lanes should be used; and
• how the network should respond to uncertainty and disruption.

The design problem is therefore a strategic form of systems analysis. Local improvements, such as selecting the cheapest carrier on an individual lane, may increase total cost if they create longer repositioning distances, additional handling or poor asset utilisation.

Learning objectives

After studying this chapter, the learner should be able to:

1. distinguish major logistics network configurations;
2. evaluate direct shipment, hub-and-spoke, consolidation and cross-docking choices;
3. apply centre-of-gravity and location-allocation methods;
4. formulate a total-cost and facility-location model;
5. interpret trade-offs among cost, service, resilience and sustainability; and
6. recommend a practical network design using quantitative and managerial evidence.

1. The structure of a logistics network

A logistics network consists of nodes and links. Nodes include suppliers, manufacturing sites, ports, terminals, warehouses, cross-docks, retail outlets and customers. Links are transportation connections over which products move. A network may also include reverse-logistics flows for returns, reusable packaging, repair and recycling.

A useful abstraction is a directed graph. Let [[M:G=(V,A)]] represent the network, where [[M:V]] is the set of nodes and [[M:A]] is the set of directed arcs. Each arc may have attributes such as distance, transit time, capacity, freight rate, emissions and reliability. Demand is associated with customer nodes or geographic zones, while supply and processing capacity are associated with origin nodes and facilities.

Network design decisions operate at several levels:

The levels are interdependent. A strategic decision to centralise inventory may reduce facility cost but increase delivery distance and exposure to a single disruption. Tactical and operational policies determine whether the expected benefits are actually achieved.

Service dimensions

Network design should specify service requirements rather than treating service as a vague objective. Relevant dimensions include order cycle time, delivery frequency, fill rate, on-time delivery, product availability, temperature integrity and return-processing time. A facility that minimises cost but cannot meet a required delivery window is not a feasible solution.

2. Major network configurations

Direct shipment

In a direct-shipment network, each origin ships directly to its assigned destination. The structure has few handling points and often provides short transit time. It is attractive when shipment volumes are large, demand is predictable, products are sensitive to handling or customers require dedicated service.

Its disadvantages are low load utilisation for small orders, many origin–destination lanes and limited opportunities for consolidation. If there are [[M:s]] suppliers and [[M:d]] customers, a fully connected direct pattern can require up to [[M:s×d]] potential lanes, although not all lanes will be used.

Hub-and-spoke

In a hub-and-spoke network, flows move from origins to one or more hubs and then to destinations. The hub consolidates freight and may provide sorting, transloading or service differentiation. The model can reduce the number of long-haul connections and improve vehicle utilisation.

However, the hub adds handling and may increase transit time. It also creates concentration risk: failure at a major hub can affect many customers. Hub design should therefore consider backup capacity, time windows, congestion and service-critical products.

Consolidation and break-bulk

Consolidation combines several small shipments into a larger movement. Break-bulk reverses the process by separating a large inbound load into destination-specific shipments. The economic logic is that transport cost does not always rise proportionally with weight or volume. Full-load or line-haul movements may have a lower unit cost than many small shipments.

Consolidation is most useful when shipments have compatible destinations, timing and handling requirements. It becomes less attractive when waiting time, incompatibility or additional handling exceeds freight savings.

Cross-docking

Cross-docking transfers inbound goods to outbound vehicles with little or no storage. A cross-dock is a flow-through facility rather than a conventional inventory-holding warehouse. It is effective when inbound and outbound schedules are coordinated, demand is relatively predictable and products are standardised.

Cross-docking can reduce storage and order-picking cost, but it requires accurate advance shipping information, disciplined appointment management, sufficient dock capacity and rapid exception handling. It is unsuitable for highly uncertain demand, long dwell requirements or products requiring substantial inspection or customisation.

Centralised and decentralised networks

A centralised network uses relatively few facilities, often one national distribution centre or a small number of regional centres. It can exploit inventory pooling, standardised processes and economies of scale. A decentralised network places facilities closer to demand, reducing last-mile distance and improving responsiveness but usually increasing inventory duplication and fixed cost.

The choice is not binary. Many organisations use a hybrid design: central inventory for slow-moving products, regional stock for fast-moving products and direct shipment for selected large customers.

Figure: Direct connections versus a hub-and-spoke structure.

3. Facility-location analysis

Facility location is the selection of sites for warehouses, plants, terminals or service centres. It combines geographic, economic, operational and institutional analysis.

Strategic location criteria

A professional location study normally evaluates:

• demand density and expected growth;
• supplier and production access;
• road, rail, port and airport connectivity;
• land, labour, utilities, taxes and construction cost;
• customer delivery windows;
• congestion, weather and disruption exposure;
• planning, environmental and safety regulations;
• energy availability and carbon implications; and
• community impacts and labour standards.

A low-rent site may be expensive in total if it is distant from customers or lacks reliable transport access. Conversely, a high-rent urban site may be justified for urgent, high-value or time-sensitive deliveries.

Centre-of-gravity method

The centre-of-gravity method provides an initial geographic estimate for a single facility. For demand points with coordinates [[M:(x_i,y_i)]], shipment weight or demand [[M:w_i]], the weighted location is:

Here, [[M:x^]] and [[M:y^]] are the estimated coordinates, [[M:x_i,y_i]] are the coordinates of demand point [[M:i]], and [[M:w_i]] is its weight, commonly annual demand, shipment volume or tonne-kilometres-related importance. Coordinates must use a consistent system, and weights must be non-negative and expressed in consistent units.

The method assumes that transport cost is approximately proportional to shipment weight and distance and that the geographic plane is a reasonable approximation. It does not account for roads, terrain, land availability, taxes, facility capacity or nonlinear freight rates. Consequently, it identifies a candidate area, not a final site.

#### Worked example: centre of gravity

Three markets have the following data:

Total demand is 600 tonnes. Therefore:

[[M:x^*=(100×10+200×40+300×20)/600=23.33]]

and

[[M:y^*=(100×20+200×30+300×60)/600=43.33]].

The initial candidate is approximately (23.33, 43.33). A subsequent feasibility study must test actual road distances, candidate parcels and regulatory constraints.

Factor-rating method

When qualitative and quantitative criteria both matter, the factor-rating method can compare candidate sites. Let [[M:r_{ij}]] be the score of site [[M:j]] on criterion [[M:i]], and [[M:w_i]] be the criterion weight, where weights sum to one. The site score is:

The method is useful for screening but depends on subjective scores, criterion selection and scaling. It should not conceal critical constraints: a site that violates a legal or safety requirement must be rejected regardless of its weighted score.

4. Location-allocation and total-cost models

Basic facility-location model

A location-allocation model jointly decides which facilities to open and which customers to serve. A simplified capacitated model is:

subject to

Here, [[M:F_j]] is the fixed cost of opening facility [[M:j]], [[M:y_j]] is one if it is opened and zero otherwise, [[M:c_{ij}]] is the variable cost per unit shipped from facility [[M:j]] to customer [[M:i]], [[M:x_{ij}]] is the quantity allocated, [[M:d_i]] is customer demand and [[M:K_j]] is facility capacity. The model assumes known demand, linear variable costs, divisible flows and no explicit service-level restriction. It can be extended with multiple products, transport modes, inventory costs, emissions, opening limits and delivery-time constraints.

The objective is not automatically “minimum freight cost.” It is the minimum of the specified total cost. If inventory, handling and facility costs are omitted, the model can recommend a network that appears inexpensive but is operationally unrealistic.

Total landed logistics cost

A practical total-cost model may include:

where [[M:C_F]] is facility cost, [[M:C_T]] transportation cost, [[M:C_H]] handling and processing cost, [[M:C_I]] inventory carrying cost, [[M:C_O]] order or administrative cost, [[M:C_R]] reverse-logistics cost, [[M:C_E]] environmental or compliance cost and [[M:C_Risk]] the monetised or scenario-based cost of disruption. Not every term must be monetised precisely; some can be treated as constraints or compared through scenarios.

Inventory carrying cost is often estimated as:

where [[M:I]] is average inventory in units, [[M:v]] is unit value and [[M:h]] is the annual holding-rate fraction. The units are currency per year. The holding rate may include capital cost, insurance, obsolescence and storage-related components, but double counting must be avoided.

Transport-cost comparison

For a shipment, a simplified cost can be represented as:

where [[M:C_0]] is a fixed shipment charge, [[M:r]] is the rate per unit-distance-unit, [[M:D]] is distance and [[M:q]] is shipment quantity. This is only appropriate when the carrier tariff is approximately linear over the relevant range. Real tariffs may include minimum charges, weight breaks, fuel surcharges, accessorial fees, dimensional-weight rules and capacity constraints.

5. Comparing network alternatives

A decision team should compare feasible alternatives rather than relying on a single optimisation output. A useful evaluation sequence is:

1. define demand, product and service assumptions;
2. identify candidate facilities and network configurations;
3. calculate transport, facility, inventory and handling costs;
4. test capacity, delivery time and regulatory feasibility;
5. evaluate risk, emissions and labour or community effects;
6. conduct sensitivity and scenario analysis; and
7. select a robust design and establish implementation milestones.

Sensitivity and scenario analysis

Sensitivity analysis changes one or more parameters, such as fuel price, demand, labour cost or facility capacity, to observe the effect on the recommendation. Scenario analysis changes a coherent set of assumptions, such as “rapid growth,” “port disruption” or “regionalisation.” A robust design may not be the lowest-cost design under the baseline scenario, but it may perform acceptably across a wider range of futures.

Important measures include:

• total annual logistics cost;
• cost per unit or order;
• average and maximum delivery time;
• on-time delivery probability;
• facility utilisation;
• transport load factor;
• inventory days and working capital;
• emissions per tonne-kilometre or order; and
• recovery time after a disruption.

6. Applied case: NorthStar Consumer Products

This is an illustrative case, constructed for learning purposes.

Background

NorthStar supplies packaged household products from two plants to 18 regional markets. It currently operates four warehouses. Demand has grown, online orders have increased and customers in major cities now expect two-day delivery. The existing network was designed when demand was more evenly distributed and store replenishment dominated.

Problem

Management observes high warehouse duplication, low truck utilisation on several lanes and rising emergency shipments. Closing facilities could reduce fixed cost, but a fully centralised design might increase last-mile distance and compromise the two-day service promise.

Analysis

The project team first segmented products by demand velocity and service requirement. Fast-moving products accounted for 70% of volume and were concentrated near three metropolitan regions. Slow-moving products had broad but sparse demand. The team then developed three alternatives:

A capacitated location-allocation model estimated annual cost, while operational managers checked delivery windows, dock capacity and labour availability. Scenario tests considered 20% demand growth, a fuel-price increase and a two-week closure of one facility.

Approach and solution

Alternative C was selected. One regional centre served the eastern demand cluster and another served the western cluster. Slow-moving products were pooled in the eastern centre and shipped to the western centre when economically justified. Selected high-volume supplier flows bypassed storage through scheduled cross-docking. A limited direct-shipment policy was retained for full-truckload customer orders.

The design included a contingency carrier contract, predefined overflow capacity and visibility requirements for advance shipment notices. The team rejected a purely mathematical recommendation that would have assigned several customers to a facility unable to meet their delivery windows.

Outcome

The illustrative evaluation projected lower duplicated inventory, improved average vehicle utilisation and preservation of the two-day promise for priority markets. The network was not the lowest-cost option in every scenario, but it had better performance under growth and single-facility disruption. NorthStar also established quarterly reviews because demand concentration and transport capacity were expected to change.

Learning points

1. Facility count should be evaluated with transport, inventory and service cost together.
2. A hybrid network often outperforms an extreme centralised or decentralised design.
3. Feasibility constraints and managerial judgement are essential complements to optimisation.
4. Resilience has economic value and should be tested explicitly.

7. Technology, sustainability and governance

Modern network design uses geographic information systems, transport-management systems, warehouse data, digital twins and optimisation software. Machine-learning forecasts can improve demand estimates, but forecast accuracy does not eliminate structural uncertainty. AI-generated recommendations require traceable data, documented assumptions, validation against operational constraints and human approval for high-impact decisions.

Digital twins can simulate facility throughput, dock congestion and disruption scenarios before physical changes are made. Real-time visibility supports dynamic reallocation, but data quality, interoperability and cybersecurity remain significant limitations.

Sustainability should be integrated rather than treated as an afterthought. Design choices affect vehicle kilometres, energy use, packaging, land use, local air quality and labour conditions. A network that reduces transport emissions by moving activities to a poorly regulated facility may create unacceptable social or environmental impacts. Organisations should apply emissions accounting consistently, verify supplier claims and consider regulatory requirements governing hazardous materials, worker safety, zoning and data protection.

8. Limitations and best practices

No model captures reality completely. Common limitations include uncertain demand, inaccurate distances, omitted congestion, assumed linear costs, static capacity, behavioural responses and weak data on disruption probability. Model precision should not be confused with decision accuracy.

Best practices include:

• define service requirements before optimising cost;
• use actual road distances and carrier tariffs where available;
• separate hard constraints from preference scores;
• include inventory, handling, reverse logistics and implementation cost;
• test multiple demand and disruption scenarios;
• validate outputs with facility and transport operators;
• document data sources, assumptions and model version;
• use phased implementation and pilot testing; and
• establish governance for periodic network review.

Chapter Summary

Logistics network design determines the structure, location and flow of a supply chain. Direct shipment reduces handling but may produce many low-utilisation lanes. Hub-and-spoke systems and consolidation exploit scale but introduce handling and concentration risk. Cross-docking reduces storage when schedules and information are well coordinated. Centralised networks pool inventory, whereas decentralised networks improve proximity and responsiveness.

The centre-of-gravity method provides a preliminary geographic estimate, while factor rating incorporates weighted qualitative criteria. Location-allocation models combine facility-opening and customer-assignment decisions subject to capacity and service constraints. Total-cost analysis must include facility, transport, handling, inventory, reverse-logistics, environmental and risk considerations. Robust network design requires sensitivity analysis, scenario testing, operational validation and attention to governance, sustainability and resilience.

Key Terms

Network design: Strategic determination of facilities, flows, lanes and capacities.

Node: A facility or point in a logistics network.

Arc: A directed connection through which flow moves.

Direct shipment: Movement from origin to destination without an intermediate facility.

Hub-and-spoke: Network connecting origins and destinations through a consolidation hub.

Consolidation: Combining shipments to improve transport utilisation.

Break-bulk: Separating a consolidated shipment into destination-specific loads.

Cross-docking: Transferring inbound goods to outbound transport with little storage.

Centralisation: Concentrating inventory or operations in relatively few facilities.

Decentralisation: Distributing facilities closer to demand markets.

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

Factor rating: Weighted scoring method for comparing candidate sites.

Location-allocation: Joint selection of facilities and assignment of demand.

Capacity constraint: Limit on the quantity a facility or lane can process.

Total logistics cost: Combined facility, transport, handling, inventory and related cost.

Inventory pooling: Sharing stock across markets to reduce aggregate uncertainty.

Load factor: Utilised transport capacity divided by available capacity.

Resilience: Ability to withstand and recover from disruption.

Digital twin: Digital representation used to simulate a physical network.

Review Questions

Short Answer Questions

1. What is the difference between strategic network design and operational routing?
2. State two advantages and two limitations of a hub-and-spoke network.
3. What assumptions underlie the centre-of-gravity method?
4. Why can cross-docking reduce logistics cost?
5. Identify four components of total logistics cost.

Long Answer Questions

1. Compare direct shipment, consolidation and hub-and-spoke configurations for a national retailer.
2. Explain how centralisation affects transport cost, inventory and customer service.
3. Describe the variables and constraints in a capacitated location-allocation model.
4. Discuss how sustainability and resilience should influence facility-location decisions.
5. Explain why optimisation results must be validated by operational managers.

Analytical / Application Questions

1. Using the three-market data in the centre-of-gravity example, explain what additional analysis is required before selecting a site.
2. A company is considering one national distribution centre versus two regional centres. Design a scenario analysis covering demand growth, fuel prices and facility disruption.
3. Propose a network configuration for a business with predictable fast-moving products, uncertain slow-moving products and strict urban delivery windows. Justify the roles of direct shipment, inventory pooling and cross-docking.

Multiple Choice Questions with Answers

Question 1

Which network configuration sends freight through an intermediate consolidation point before reaching customers?

A. Direct shipment

B. Hub-and-spoke

C. Point-of-use manufacturing

D. Reverse logistics only

Correct Answer: B. Hub-and-spoke

Question 2

In the centre-of-gravity method, the weight wᵢ commonly represents which quantity?

A. A site’s legal status

B. Annual demand or shipment volume

C. The number of candidate facilities

D. The maximum road speed

Correct Answer: B. Annual demand or shipment volume

Question 3

Using the example data in the chapter, what is the approximate centre-of-gravity x-coordinate?

A. 10.00

B. 20.00

C. 23.33

D. 40.00

Correct Answer: C. 23.33

Question 4

What is the defining operational characteristic of cross-docking?

A. Long-term storage of all inbound goods

B. Transfer from inbound to outbound transport with little storage

C. Exclusive use of air transport

D. Shipment only from suppliers to factories

Correct Answer: B. Transfer from inbound to outbound transport with little storage

Question 5

In the capacitated location-allocation model, yⱼ = 1 indicates that facility j is:

A. At full capacity

B. Closed permanently

C. Opened for operation

D. Assigned to every customer

Correct Answer: C. Opened for operation

Question 6

Which cost is most directly associated with inventory pooling?

A. Potential reduction in aggregate safety stock

B. Mandatory increase in every delivery distance

C. Elimination of all facility costs

D. Removal of customer demand uncertainty

Correct Answer: A. Potential reduction in aggregate safety stock

Question 7

A facility has annual inventory of 2,000 units, unit value of $50 and holding-rate fraction 0.20. What is the annual carrying cost using [[M:C_I=I×v×h]]?

A. $2,000

B. $10,000

C. $20,000

D. $50,000

Correct Answer: C. $20,000

Question 8

Which statement best describes a limitation of a factor-rating method?

A. It cannot compare qualitative criteria

B. It makes all regulatory constraints automatically feasible

C. Its results depend partly on subjective scores and weights

D. It requires every site to have identical capacity

Correct Answer: C. Its results depend partly on subjective scores and weights

Question 9

Why should a logistics network be tested under disruption scenarios?

A. To guarantee that disruption will never occur

B. To evaluate robustness beyond the baseline assumptions

C. To eliminate the need for capacity data

D. To replace all quantitative analysis

Correct Answer: B. To evaluate robustness beyond the baseline assumptions

Question 10

Which situation most strongly supports direct shipment?

A. Small, irregular orders with incompatible destinations

B. Large predictable volumes requiring dedicated movement

C. Highly uncertain demand requiring inventory pooling

D. Products that must be sorted through a common hub

Correct Answer: B. Large predictable volumes requiring dedicated movement