Supply Chain Network Design

Introduction

Supply chain network design is the strategic determination of the facilities, capacities, locations, transportation links, inventory positions, and product flows required to serve markets. It answers questions such as:

• How many plants, warehouses, cross-docks, or fulfilment centres should be operated?
• Where should they be located?
• Which markets should each facility serve?
• What capacity should be installed, and when should it be expanded?
• Should production or inventory be centralized or geographically dispersed?
• Where should product differentiation occur?

These decisions are long-lived, capital-intensive, and difficult to reverse. They therefore require a total-cost perspective rather than a narrow focus on transport or facility operating cost. A network that minimizes freight expense may create excessive inventory, poor responsiveness, high tax exposure, or unacceptable risk. Conversely, a more expensive physical network may create strategic value through service, resilience, postponement, or market access.

Network design is consequently a multi-objective problem involving cost, service, risk, sustainability, flexibility, and regulatory compliance. The purpose of this chapter is to develop the concepts and analytical tools needed to design and evaluate such networks.

Learning Objectives

After studying this chapter, you should be able to:

1. Explain the strategic role and scope of supply chain network design.
2. Formulate total-cost models for facility location, capacity, allocation, and distribution.
3. Apply break-even, centre-of-gravity, and mixed-integer location logic appropriately.
4. Evaluate centralization, postponement, and alternative distribution structures.
5. Use scenario and sensitivity analysis to address uncertainty.
6. Interpret a network-design case and make a defensible managerial recommendation.
7. Recognize the limitations, governance requirements, and sustainability implications of network models.

1. The Network as a Strategic System

A supply chain network consists of nodes and arcs. Nodes include suppliers, plants, ports, warehouses, fulfilment centres, retailers, customers, and recovery facilities. Arcs represent transportation, information, financial, or material flows. In a basic forward network, suppliers feed plants, plants feed distribution facilities, and distribution facilities serve demand regions.

The design horizon is usually several years, whereas planning decisions may be monthly or weekly. This distinction matters. A network model should not merely reproduce today’s operating plan; it should test how the structure performs under expected growth, changing customer geography, product proliferation, disruptions, and regulatory change.

Figure: A simplified multi-echelon supply chain network.

A sound design process normally follows these stages:

1. Define objectives, scope, horizon, and service requirements.
2. Collect and validate demand, cost, capacity, lead-time, and risk data.
3. Generate feasible network alternatives.
4. Model flows and calculate total cost and service outcomes.
5. Test scenarios and sensitivities.
6. Select a robust design and develop an implementation roadmap.
7. Monitor assumptions and revisit the design as conditions change.

2. Total-Cost Logic

The central principle is that a network decision should be evaluated through its incremental total cost and strategic consequences. A general annual network-cost expression is:

where:

• TC = total annual cost;
• F = fixed facility, lease, maintenance, and overhead cost;
• P = production or procurement cost;
• T_in = inbound transportation cost;
• T_out = outbound transportation cost;
• H = facility handling and operating cost;
• I = inventory carrying cost;
• S = shortage, expediting, or service-failure cost;
• R = expected disruption and risk cost;
• E = relevant environmental, tax, duty, and compliance cost.

All terms must be measured over the same period and in the same monetary units. The expression is a decision framework rather than a universal accounting identity: terms should be included only when they are relevant and measured consistently. Sunk costs should normally be excluded unless the decision genuinely changes them.

Inventory carrying cost is often approximated as:

where h is the annual holding-rate fraction, V is average unit value, and Q̄ is average inventory in units. This assumes the holding rate captures capital, insurance, obsolescence, and storage effects and that average inventory is a reasonable approximation. If inventory varies by location or product, the calculation should be performed by item and node and then summed.

For a facility with annual fixed cost F_j, variable processing cost v_j per unit, and throughput q_j, operating cost is:

The model is linear only over the range in which fixed and variable costs remain stable. Overtime, congestion, stepwise labour, and capacity expansion can make the cost function nonlinear.

Cost-to-serve and service trade-offs

Total cost should be linked to customer segments rather than assessed only at aggregate level. Cost-to-serve may include order processing, picking, special packaging, delivery, returns, customer service, and failure costs. A profitable product or customer at gross-margin level may become unattractive after network-specific service costs are included.

Service is also multidimensional. Relevant measures include order-cycle time, on-time-in-full performance, delivery reliability, product availability, and response to demand peaks. A design should state the service constraint explicitly, for example, a required two-day delivery window or a minimum fill rate, rather than treating service as an unquantified aspiration.

3. Facility Location and Capacity Decisions

3.1 Qualitative screening

Before quantitative optimization, candidate sites should be screened for land, labour, utilities, road and port access, supplier proximity, tax and customs conditions, environmental restrictions, community acceptance, cyber and physical security, and expansion potential. A mathematically attractive site is not feasible if permits, labour, or utilities cannot be secured.

A weighted-factor method can support preliminary screening. If site j receives score s_ij on criterion i, and criterion weight w_i satisfies ∑w_i = 1, then:

Scores are ordinal or judgement-based, so the result should not be presented as a precise economic optimum. It is most useful for eliminating infeasible candidates and exposing managerial assumptions.

3.2 Break-even location analysis

When two sites have different fixed and variable costs, their cost equality occurs at:

where F_A, F_B are fixed costs and v_A, v_B are variable costs per unit for sites A and B. The formula assumes one product or an equivalent volume measure, linear costs, identical relevant service conditions, and a positive denominator. If the denominator is zero or negative, there may be no meaningful positive break-even volume.

Example. Site A has annual fixed cost $1.2 million and variable cost $8 per unit. Site B has fixed cost $2.0 million and variable cost $6 per unit. Thus:

q* = (2.0 − 1.2) million ÷ (8 − 6) = 0.4 million units

Below 400,000 units, A is cheaper; above 400,000 units, B is cheaper, assuming other costs and service are equal.

3.3 Centre-of-gravity method

For a single facility serving geographically distributed demand, a preliminary location can be estimated using demand-weighted coordinates:

Here (x_i,y_i) are customer coordinates and w_i is demand, shipment weight, or another appropriate flow weight. The method assumes transportation cost is approximately proportional to volume and distance, ignores road topology, congestion, fixed facility cost, service constraints, and candidate-site feasibility. It is therefore a screening tool, not a final location solution.

3.4 Facility-location optimization

A common mixed-integer formulation uses a binary variable y_j equal to 1 if candidate facility j is opened, and a continuous flow variable x_jk} for quantity shipped from facility j to market k. A capacitated model is:

subject to:

where F_j is fixed opening cost, c_jk is variable cost per unit shipped, d_k is demand, and U_j is capacity. The first constraint requires demand satisfaction; equality may be used when over-supply is not permitted. The second constraint prevents flow through a closed facility and limits open-facility throughput.

The model can be extended with production cost, inbound transportation, minimum throughput, service-radius constraints, multiple products, supplier capacity, carbon emissions, taxes, duties, and multi-period capacity expansion. Because binary decisions make the problem computationally difficult in general, commercial mixed-integer solvers, decomposition, heuristics, or scenario reduction may be required for large networks.

4. Centralization, Decentralization, and Postponement

Centralization concentrates inventory and activities in fewer locations. It can reduce safety stock through risk pooling, exploit economies of scale, simplify control, and improve asset utilization. Its disadvantages include longer delivery distances, greater dependence on particular facilities, possible customs exposure, and a larger consequence if the central node fails.

Decentralization places inventory or processing closer to customers. It can improve response time, reduce last-mile distance, and protect service during regional disruptions. However, it duplicates stock and infrastructure, increases coordination complexity, and may reduce volume economies.

For independent demand, a simplified risk-pooling relationship illustrates why aggregation can reduce safety stock. If each of n locations has standard deviation σ and demands are independent, aggregate standard deviation is:

If each location separately holds safety stock proportional to zσ, total decentralized safety stock is nzσ, whereas centralized safety stock is zσ√n. Here z is the service-factor quantile. The result assumes identical demand variability, independence, a common replenishment lead time, and the same service interpretation. Positive correlation reduces the benefit; perfect positive correlation eliminates the square-root aggregation advantage.

A more general form with equal standard deviation and pairwise correlation ρ is:

This formula demonstrates that correlation is strategically important. It should not be used to justify centralization without examining lead times, product substitutability, minimum order quantities, and customer service requirements.

Postponement

Postponement delays final configuration, packaging, labelling, localization, or other differentiation until demand is better known. It combines upstream scale with downstream responsiveness. Typical designs include generic products finished near the market, late-stage customization, modular components, and delayed assignment of inventory to regions.

Postponement is attractive when product variety is high, demand for variants is uncertain, common platforms are feasible, and final processing is relatively quick. It is less attractive when customization requires specialized equipment, regulatory approval, long final-processing times, or large local skills investments. The decision should compare incremental conversion cost and lead time with avoided obsolescence and inventory duplication.

5. Distribution Structures and Flow Allocation

Important network structures include:

Allocation decisions should consider transport rate structures, shipment consolidation, mode choice, delivery windows, and capacity. A low unit freight rate does not necessarily produce low total cost if it requires high inventory or misses the service promise.

The relevant unit of analysis may be a product-market pair rather than an entire facility. Different products can use different networks: fast-moving standard items may be stocked regionally, while slow-moving variants may be centrally held or postponed. Strategic segmentation is often more effective than imposing one network on all products.

6. Scenario, Sensitivity, and Robust Design

Demand, fuel prices, exchange rates, labour costs, tariffs, disruptions, and technology adoption are uncertain. A single forecast is therefore insufficient. Scenario analysis evaluates alternative states such as base growth, high growth, recession, port disruption, supplier failure, and regulatory change.

For scenario s with probability p_s and cost TC_s, expected cost is:

Expected cost is useful when probabilities are credible and risk neutrality is acceptable. It can be misleading where a low-probability disruption has severe consequences or where management is risk-averse. In such cases, evaluate worst-case cost, service failure, recovery time, or a risk-adjusted objective.

Sensitivity analysis changes one or more assumptions and observes the effect on the decision. Important tests include demand growth, capacity utilization, transport rates, inventory holding rates, lead time, correlation, carbon price, and facility opening cost. A design whose recommendation changes under small assumption changes is fragile and should not be treated as a definitive optimum.

Robust design may deliberately select a slightly higher expected-cost option that performs acceptably across many scenarios. Flexibility has economic value. Examples include leased rather than fully owned capacity, modular automation, dual sourcing, reserved carrier capacity, and facilities that can be expanded in stages.

7. Sustainability, Risk, and Governance

Network design influences emissions through facility energy use, transport distance, mode choice, packaging, refrigeration, and returns. A carbon-aware model may add emissions:

where e_a is emissions per unit of activity and f_a is the corresponding activity quantity for activity a. Units must be consistent, such as kilograms CO₂-equivalent per tonne-kilometre multiplied by tonne-kilometres. Emission factors should be documented and preferably based on recognized life-cycle or national reporting standards.

Risk analysis should cover natural hazards, geopolitical exposure, cyberattack, labour disruption, supplier concentration, single-port dependence, quality failure, and demand shocks. Redundancy is not automatically optimal: its cost must be compared with the probability and consequence of failure. However, models should not monetize all resilience benefits mechanically; strategic continuity and human consequences may not be adequately represented by expected cost.

Governance is essential. Decision makers should document data sources, assumptions, scenario definitions, solver settings, excluded costs, and model validation. A model should be independently reviewed, version controlled, and stress tested. AI-assisted forecasts or optimization tools can accelerate analysis, but they do not eliminate the need for explainability, bias checks, human approval, cybersecurity, and accountability. Historical data may encode past inequities or temporary disruptions and should not be accepted uncritically.

Regulatory considerations can include zoning, environmental permits, customs classification, labour law, tax substance requirements, product traceability, food or pharmaceutical controls, data localization, and competition law. A nominally optimal network is unacceptable if it violates law or creates unmanaged social and environmental harm.

8. Worked Network Example

An illustrative company serves two markets from two candidate distribution centres. Annual demand is 60,000 units in Market 1 and 40,000 units in Market 2. Centre A has fixed cost $600,000 and capacity 100,000 units. Centre B has fixed cost $450,000 and capacity 70,000 units. Variable delivery costs per unit are:

Assume all demand must be served, there are no fixed costs other than the stated facility costs, and variable costs include relevant handling and transport. Consider three feasible designs.

Design A only. Capacity is exactly sufficient. Cost equals fixed cost plus deliveries:

600,000 + (60,000×4) + (40,000×9) = $1,200,000.

Design B only. Capacity is only 70,000 units, less than total demand of 100,000, so this design is infeasible.

Both centres open. Allocate Market 1 to A and Market 2 to B. This uses 60,000 units at A and 40,000 at B, satisfying capacity. Cost is:

600,000 + 450,000 + (60,000×4) + (40,000×5) = $1,490,000.

Under the stated assumptions, A only is cheaper by $290,000. Yet this conclusion is not automatically strategic. If Market 2 requires a service level that A cannot meet, or if A-only creates unacceptable disruption exposure, the two-centre design may be justified. The example demonstrates why feasibility, total cost, service, and risk must be evaluated together.

9. Applied Case: Meridian Home Appliances

This is an illustrative case developed for learning purposes.

Background

Meridian manufactures small kitchen appliances for three regions: North, Central, and South. Demand is growing, product variants have increased, and customers expect two-day delivery for popular models. Meridian currently operates one national warehouse near its plant. The warehouse has low average cost but suffers from seasonal congestion and long delivery distances to South.

Problem

Management is considering three alternatives:

1. retain one national warehouse;
2. add a regional warehouse in South;
3. retain national stock for standard products but postpone regional packaging and final configuration at a South fulfilment site.

The company initially focused on outbound freight savings. The supply-chain team argued that inventory duplication, labour availability, demand correlation, returns, disruption exposure, and packaging regulation also required analysis.

Analysis

The team segmented products into stable high-volume models, uncertain seasonal models, and highly localized variants. High-volume models were suitable for regional stock. Seasonal models had correlated peaks across regions, reducing the benefit of decentralization. Localized variants were good candidates for postponement because the core product was common and final labelling could be completed in less than one day.

The team built a mixed-integer model with facility fixed costs, transportation, handling, inventory, capacity, and service constraints. It tested base demand, 20% growth, a national-warehouse disruption, and a South transportation-rate increase. It also used sensitivity analysis on inventory correlation and the cost of temporary labour.

Approach and solution

The recommended design was a hybrid network. The national warehouse retained most seasonal inventory and supplied the regional site with generic units. The South site held selected fast movers and performed final configuration for localized variants. A contract logistics provider supplied flexible overflow capacity during peak months. The design included a contingency carrier and a documented transfer plan if either facility became unavailable.

Outcome

The hybrid design did not minimize transportation cost in every scenario. It reduced delivery time for South customers, limited variant obsolescence through postponement, and provided scalable peak capacity. The investment case was approved subject to quarterly review of demand, service, and utilization assumptions.

Learning points

The case illustrates five principles:

• network design should begin with customer and product segmentation;
• postponement can be more effective than simply adding finished-goods stock;
• scenario performance is more informative than one forecast result;
• flexible capacity can have strategic value even when its nominal unit cost is higher;
• implementation governance is part of the design, not an afterthought.

10. Implementation and Best Practices

Successful network design depends as much on decision discipline as on optimization technique. Recommended practices include:

1. Define the decision boundary. Separate strategic choices from operational routing and scheduling.
2. Use clean, reconciled data. Validate demand, units, addresses, rates, capacities, and cost definitions.
3. Model constraints honestly. Include labour, legal, service, storage, dock, and supplier limitations.
4. Segment the network. Products, customers, and regions need not share one structure.
5. Use scenarios rather than false precision. Present ranges and decision sensitivities.
6. Compare total cost and total value. Include inventory, service, risk, sustainability, and implementation cost.
7. Validate with operational experts. A mathematically feasible solution may be physically impossible.
8. Plan transition. Include inventory migration, customer communication, systems changes, workforce, and commissioning.
9. Monitor leading indicators. Track utilization, service, forecast error, inventory, emissions, and disruption exposure.
10. Review periodically. A strategic design should be revisited when demand geography, technology, regulation, or risk materially changes.

Chapter Summary

Supply chain network design determines the long-term structure through which products, information, and resources move. It integrates facility location, capacity, allocation, transportation, inventory, service, risk, sustainability, and regulatory requirements. Total-cost analysis prevents isolated decisions that reduce one cost while increasing another. Break-even analysis and the centre-of-gravity method provide useful preliminary insight, while capacitated mixed-integer models support integrated facility-opening and flow-allocation decisions.

Centralization can pool risk and reduce duplicated inventory; decentralization can improve responsiveness and regional resilience. Postponement delays differentiation until demand is clearer and can reduce variety-related inventory risk. Scenario and sensitivity analysis are essential because forecasts and operating conditions are uncertain. Robust designs may sacrifice some expected cost efficiency to gain flexibility and continuity.

Ultimately, network design is not merely a mathematical optimization exercise. It is a strategic governance process requiring reliable data, explicit assumptions, cross-functional judgement, ethical and regulatory awareness, and an implementation plan.

Key Terms

Supply chain network: The interconnected facilities, transportation links, flows, and relationships serving supply and demand.

Network design: Strategic determination of facility locations, capacities, flows, and structural relationships.

Total cost: The combined relevant cost of facilities, production, transport, handling, inventory, service, risk, and compliance.

Facility location: Selection of physical sites for plants, warehouses, or fulfilment facilities.

Capacity: Maximum feasible throughput or storage volume over a specified period.

Centre-of-gravity method: A demand-weighted coordinate technique for preliminary facility-location screening.

Mixed-integer model: An optimization model containing continuous flow variables and discrete decisions such as opening a facility.

Risk pooling: Reduction in aggregate variability achieved by combining demand or inventory positions.

Centralization: Concentration of inventory or activities in fewer locations.

Decentralization: Distribution of inventory or activities across multiple regional locations.

Postponement: Delaying product differentiation until demand or destination is more certain.

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

Cost-to-serve: The total cost of serving a particular customer, product, or segment.

Scenario analysis: Evaluation of network performance under alternative future conditions.

Sensitivity analysis: Examination of how results change when assumptions vary.

Robust design: A design that performs acceptably across a range of uncertain conditions.

Risk pooling correlation: The degree to which demands move together and affect aggregation benefits.

Service constraint: A specified requirement for delivery time, availability, reliability, or fill rate.

Review Questions

Short Answer Questions

1. Why is total-cost analysis essential in supply chain network design?
2. State two advantages and two disadvantages of centralization.
3. What assumptions underlie the centre-of-gravity method?
4. Define postponement and identify one situation in which it is appropriate.
5. Why should scenario analysis accompany a single-forecast network model?

Long Answer Questions

1. Explain the stages of a strategic supply chain network-design process, from problem definition through implementation.
2. Compare direct shipment, centralized distribution, regional distribution, and cross-docking structures.
3. Describe the formulation and managerial interpretation of a capacitated facility-location model.
4. Discuss how inventory risk pooling depends on demand correlation, lead time, and service requirements.
5. Critically evaluate the role of sustainability and resilience in total-cost network design.

Analytical / Application Questions

1. Two sites have fixed costs of $900,000 and $1,500,000, with variable costs of $10 and $7 per unit, respectively. Calculate the break-even volume and recommend the lower-cost site at 300,000 units, assuming all other conditions are equal.
2. A firm must choose between one central warehouse and two regional warehouses. Develop a scenario table containing demand growth, disruption, and transport-rate scenarios, and specify the performance measures you would use to select a robust design.
3. A product has a common core but four region-specific labels. Propose a postponement network and explain the operational, inventory, quality, regulatory, and customer-service conditions that must be satisfied.

Multiple Choice Questions with Answers

Question 1

Which cost perspective is most appropriate for strategic supply chain network design?

A. Outbound transportation cost only

B. Facility rent only

C. Total relevant cost including service, inventory, risk, and compliance

D. Historical accounting cost without adjustment

Correct Answer: C. Total relevant cost including service, inventory, risk, and compliance

Question 2

In a capacitated facility-location model, what does a binary variable typically represent?

A. The quantity shipped to a customer

B. Whether a facility is opened

C. The distance between two locations

D. The average inventory level

Correct Answer: B. Whether a facility is opened

Question 3

Site A has fixed cost $1,200,000 and variable cost $8 per unit. Site B has fixed cost $2,000,000 and variable cost $6 per unit. At what volume are their costs equal?

A. 200,000 units

B. 300,000 units

C. 400,000 units

D. 600,000 units

Correct Answer: C. 400,000 units

Question 4

Which condition generally increases the inventory benefit of centralization?

A. Perfectly positively correlated regional demand

B. Independent regional demand

C. Longer final-assembly time after ordering

D. Higher duplication of safety stock

Correct Answer: B. Independent regional demand

Question 5

What is the principal purpose of postponement?

A. To move all inventory upstream permanently

B. To delay differentiation until demand or destination is clearer

C. To eliminate all final processing

D. To maximize finished-goods variety at every warehouse

Correct Answer: B. To delay differentiation until demand or destination is clearer

Question 6

Which statement about the centre-of-gravity method is correct?

A. It guarantees the globally optimal facility site

B. It accounts fully for zoning and road congestion

C. It is a preliminary demand-weighted location technique

D. It can be used only when all demand is zero

Correct Answer: C. It is a preliminary demand-weighted location technique

Question 7

Expected network cost across scenarios is calculated by:

A. Selecting the highest-cost scenario only

B. A probability-weighted sum of scenario costs

C. A simple average regardless of probabilities

D. Multiplying all scenario costs together

Correct Answer: B. A probability-weighted sum of scenario costs

Question 8

A design that performs acceptably across several uncertain futures is best described as:

A. A sunk-cost design

B. A robust design

C. A single-echelon design

D. A variable-cost design

Correct Answer: B. A robust design

Question 9

In the illustrative network example, why was the B-only design infeasible?

A. Its fixed cost was zero

B. Its delivery cost was identical to A’s

C. Its capacity was below total demand

D. It served no market

Correct Answer: C. Its capacity was below total demand

Question 10

Which is an important governance practice for optimization models?

A. Hide assumptions to avoid managerial disagreement

B. Ignore operational expert review

C. Document data sources, assumptions, and validation results

D. Use historical data without checking relevance

Correct Answer: C. Document data sources, assumptions, and validation results