Supply Chain Network Design

Introduction and Context

Supply chain network design is the strategic determination of the facilities, suppliers, production sites, warehouses, transport links, inventory policies and customer-servicing arrangements through which products and information move. It answers questions such as: Where should plants and distribution centres be located? Which suppliers should serve which plants? Which customers should be assigned to which facilities? What capacities should be installed? How should the network respond to uncertainty, disruption and changing demand?

Network design decisions are long-lived, capital intensive and difficult to reverse. They therefore differ from operational routing or daily scheduling. A warehouse location, plant capacity or supplier qualification decision may shape cost and service performance for ten to twenty years. The central managerial challenge is to balance total cost, customer responsiveness, resilience, sustainability and strategic flexibility rather than minimising transportation cost alone.

A network can be represented as a directed graph containing nodes and arcs. Nodes include suppliers, plants, cross-docks, distribution centres, stores and customers. Arcs represent transportation, information or material flows. A useful design must satisfy material-balance, capacity, service, regulatory and financial constraints.

Learning Objectives

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

1. Explain the strategic role and structure of supply chain network design.
2. Compare facility-location, allocation, centre-of-gravity and optimisation models.
3. Calculate and interpret cost-to-serve and capacity decisions.
4. Construct scenarios for demand, supply, transport and disruption uncertainty.
5. Evaluate resilience, sustainability, governance and technology implications.
6. Recommend a network design using quantitative and managerial judgement.

1. Strategic Architecture of a Supply Chain Network

A network design begins with the customer proposition. A low-cost replenishment network may use centralised inventory and a small number of high-capacity facilities. A same-day e-commerce network may require forward fulfilment centres, urban microhubs and more inventory duplication. The correct design depends on product characteristics, demand density, lead-time tolerance and competitive strategy.

1.1 Network echelons

Common echelons are:

• Supply base: raw-material, component and packaging suppliers.
• Manufacturing system: plants, contract manufacturers and postponement facilities.
• International gateways: ports, airports, bonded warehouses and customs facilities.
• Distribution system: regional distribution centres, fulfilment centres, wholesalers and cross-docks.
• Customer interface: retailers, hospitals, industrial users, households and reverse-logistics points.

A two-echelon network has, for example, plants and distribution centres. A multi-echelon network includes several intermediate inventory and service stages. Adding an echelon may reduce transport distance and improve responsiveness, but it usually adds fixed cost, handling, inventory and coordination complexity.

1.2 Design trade-offs

The principal trade-offs include:

Network design is therefore a multi-objective problem. A financially attractive design may be unacceptable if it violates service, labour, environmental or continuity requirements.

2. Data and Decision Variables

Reliable design requires a common data model. Typical data include customer demand by product and period, supplier and plant capacities, facility fixed costs, variable processing costs, transport rates, lead times, duties, taxes, inventory parameters, labour availability, land constraints, emissions factors and disruption probabilities.

Important decision variables include:

• whether a candidate facility is opened;
• the capacity tier selected;
• the quantity shipped on each origin–destination lane;
• the supplier, plant or distribution centre assigned to each demand point;
• inventory positioning and postponement decisions;
• the quantity of strategic or emergency capacity retained.

Data should be time-stamped, reconciled to financial accounts and tested for unit consistency. Demand expressed in cases must not be combined with capacity expressed in tonnes without a validated conversion. Transport rates may be per shipment, pallet, kilometre, weight or volume; each produces a different cost structure.

3. Cost-to-Serve Analysis

Total network cost is broader than freight expenditure. A practical cost-to-serve model includes procurement, inbound transport, conversion, facility ownership or rent, labour, handling, inventory carrying, outbound transport, returns, taxes, duties, quality failures and stockout consequences.

For customer or segment j, a general annual cost-to-serve measure is:

where P is product or procurement cost, I inbound logistics, M manufacturing or conversion, F facility and fixed network cost allocated to the segment, O outbound logistics, R returns and reverse logistics, Q quality or compliance cost, and S shortage, expediting or service-recovery cost. All terms must be measured over the same period and in the same currency.

For a network design, fixed and variable costs can be represented as:

Here TC is total cost; Fᵢ is the fixed cost of opening facility i; yᵢ is 1 if facility i is open and 0 otherwise; cᵢⱼ is unit variable cost from facility i to customer or downstream node j; xᵢⱼ is shipped quantity; hᵢₖ is holding cost per unit of inventory at facility i in period k; Iᵢₖ is inventory; pⱼ is the penalty per unit of unmet demand; and uⱼ is unmet demand. This is a simplified objective and may be extended for emissions, risk and working capital.

A cost-to-serve analysis often reveals that apparently attractive customers or channels are unprofitable because of small orders, high delivery frequency, returns or special handling. Managers can then redesign minimum order quantities, delivery calendars, packaging, service levels or channel assignments rather than simply raising prices.

4. Facility Location and Allocation Models

4.1 Qualitative screening

Before optimisation, candidate sites should be screened for land, utilities, labour, supplier proximity, customer access, transport infrastructure, security, legal restrictions, taxation, community impact and expansion potential. A weighted-factor method can rank locations:

where Scoreₗ is the score of location l, wₖ is the weight of criterion k with ∑wₖ = 1, and rₗₖ is the normalised rating of location l on criterion k. Ratings must be defined consistently; a high score should always mean better performance. This method supports judgement but does not replace cost modelling.

4.2 Facility-location model

A capacitated facility-location model chooses facilities and allocates demand. A simplified mixed-integer formulation is:

subject to:

Dⱼ is demand at customer j, Kᵢ is capacity of facility i, and all quantities use the same period and unit. The first constraint requires demand satisfaction; the second prevents shipments from a closed facility and limits open-facility capacity. If each customer must be served by exactly one facility, binary assignment variables are required. If split shipments are allowed, continuous xᵢⱼ may be appropriate.

The model assumes known demand and costs unless scenarios or stochastic terms are added. It may also omit lead-time, labour, inventory and service constraints. Consequently, the optimisation result is a decision aid, not an automatic decision.

4.3 Multi-echelon design

In a multi-echelon model, flows must be conserved. For an intermediate distribution centre j:

The left side is inbound flow; the right side is outbound flow plus the change in inventory. This balance prevents a model from serving customers with material that was never supplied. Product conversion yields, bill-of-materials relationships and substitution rules may require additional constraints.

5. Centre-of-Gravity and Geographic Methods

The centre-of-gravity method provides an initial location estimate when transport cost is approximately proportional to distance and volume. For demand points j with coordinates (Xⱼ, Yⱼ) and weights Wⱼ:

Weights may represent annual tonnes, pallets or shipment volume. The coordinates must use a meaningful projected distance system; longitude and latitude should not be averaged blindly over large geographic areas. The method ignores land price, roads, congestion, taxes, capacity, political risk and non-linear transport rates. It is therefore a screening tool, not a final site-selection model.

Worked example. Three markets have annual volumes and coordinates as follows:

Total volume is 4,000 tonnes. Thus, X = (10,000 + 80,000 + 20,000)/4,000 = 27.5 and Y = (20,000 + 60,000 + 50,000)/4,000 = 32.5. The preliminary location is (27.5, 32.5). A geographic and financial feasibility study must follow.

6. Capacity, Economies and Service

Capacity design should distinguish nominal, practical and surge capacity. Practical capacity reflects maintenance, breaks, changeovers, quality losses and planned inefficiency. If expected demand is D, the required practical capacity with a target utilisation u is:

For example, 80,000 units of expected demand at a target utilisation of 0.80 requires 100,000 units of practical capacity. This calculation assumes demand and capacity are expressed in the same units and that utilisation is defined consistently. It does not by itself account for variability, product mix or peak-period capacity.

A useful strategic measure is capacity flexibility: the ability to shift production among plants, suppliers or technologies. Flexible capacity may have a higher unit cost but lower expected disruption loss. Capacity decisions should evaluate both expected cost and the value of optionality.

Service constraints may include maximum delivery time, fill rate, order-cycle time, freshness and geographic coverage. A design that minimises cost while failing a contractual two-day requirement is infeasible, even if its objective value is low.

7. Optimisation under Uncertainty

Demand, supply, exchange rates, fuel costs, labour availability and disruption events are uncertain. Treating a single forecast as fact creates false precision. Scenario planning begins by defining internally consistent states such as base demand, high demand, recession, port disruption, supplier failure and carbon-price increase.

For scenarios s with probability pₛ, a basic expected-cost objective is:

where ∑pₛ = 1 and TCₛ is total cost under scenario s. This approach is appropriate when probabilities are credible and risk neutrality is acceptable. It can undervalue catastrophic but low-probability events.

A risk-averse formulation may add a penalty for downside risk. One common measure is conditional value at risk (CVaR), which focuses on the average cost in the worst α proportion of outcomes. Its interpretation depends on the chosen confidence level and scenario distribution; it should not be treated as an objective truth about risk.

Robust optimisation protects against a defined uncertainty set rather than relying on probabilities. Stochastic programming allows recourse decisions after uncertainty is observed. For example, a plant location may be a first-stage decision, while emergency shipments and temporary capacity are second-stage decisions. The choice among deterministic, scenario-based, robust and stochastic methods should reflect data quality, decision irreversibility and consequence of error.

Stress testing

A robust design should be tested for:

• a demand increase in a major region;
• loss of the largest supplier or facility;
• transport-lane closure;
• energy or fuel-price escalation;
• border, tariff or regulatory change;
• cyberattack or prolonged information-system outage;
• extreme weather and climate-related disruption.

Useful outputs include unmet demand, recovery time, cost increase, capacity bottlenecks, inventory depletion and alternative sourcing feasibility.

8. Supplier and Production Network Design

Supplier network design is not simply a price-ranking exercise. Evaluation should include total landed cost, quality, capacity, technology, financial health, labour practices, geopolitical exposure, intellectual property protection, environmental performance and recovery capability.

Single sourcing can support scale and supplier development but creates concentration risk. Dual or multi-sourcing may increase purchase and qualification cost while reducing expected disruption impact. A portfolio approach assigns strategic, leverage, bottleneck and routine categories different governance policies.

Production network design includes make-or-buy, plant specialisation, postponement and product allocation. A specialised plant may achieve lower conversion cost, whereas a flexible plant may absorb demand variation. Postponement delays final differentiation until demand information improves; it can reduce finished-goods variety while requiring modular product and process architecture.

9. Technology, Analytics and Digital Twins

Geographic information systems support spatial analysis of customers, roads, labour pools, utilities and hazards. Optimisation software solves large mixed-integer models, while simulation evaluates congestion, lead-time distributions and operating variability that static models may overlook. Digital twins connect transactional, sensor and external data to test network policies over time.

Artificial intelligence can improve demand forecasts, identify anomalous freight costs, estimate travel times and discover nonlinear relationships. However, an AI recommendation is only as reliable as its data, objective function and governance. Explainability, bias testing, access control, cybersecurity, model drift monitoring and human approval are essential. An algorithm trained on historical allocations may reproduce past inequities or conceal service failures in disadvantaged regions.

10. Sustainability, Ethics and Governance

Network design affects emissions through facility energy use, transport distance, packaging, refrigeration and returns. A carbon-aware model can add emissions cost:

where τ is the internal carbon price, eₐ is emissions per unit on activity or arc a, and qₐ is activity volume. Units must be consistent: if e is kilograms CO₂e per unit, τ must be currency per kilogram CO₂e.

Environmental assessment should include scope 1, 2 and relevant scope 3 emissions, but boundaries and data quality must be disclosed. Social considerations include worker safety, wages, forced-labour risk, community displacement and equitable access to essential products. Governance requires documented assumptions, approval authority, audit trails, conflict-of-interest controls and periodic model validation.

Regulatory constraints may involve zoning, environmental permits, customs, product traceability, data protection, competition law and sanctions. A low-cost location is not viable if it cannot legally operate or exposes the organisation to unacceptable compliance risk.

11. Illustrative Case: Meridian Medical Devices

Background

Meridian Medical Devices, a fictional manufacturer, supplies sterile diagnostic kits to hospitals in three regions. Demand is growing, delivery within 48 hours is required for emergency orders, and the current single distribution centre is near capacity. Management is considering a second distribution centre and a second qualified component supplier.

Problem

The existing centre provides low fixed cost and inventory pooling, but average delivery time to the western region is excessive. A western facility would improve service but add rent, labour and duplicated safety stock. A second supplier costs 3% more per component but could protect production against a regional flood.

Analysis

The team collected twelve months of demand, order lines, transport rates, facility costs, supplier lead times, quality records and emergency freight data. It first used centre-of-gravity analysis to identify candidate western zones, then screened them for hospital access, validated utilities, regulatory requirements and labour availability. A capacitated location-allocation model compared the current design with alternatives. Scenarios represented base demand, 25% growth, closure of the eastern centre for four weeks and supplier disruption.

The base case favoured one centre on annual accounting cost. When emergency freight, contractual service penalties and disruption recovery were included, the two-centre design produced a slightly higher normal cost but substantially lower downside exposure. The team also found that a full second plant was unjustified. A qualified second component supplier, limited safety stock and a contract for temporary warehouse space provided better flexibility.

Solution and governance

Meridian opened a western distribution centre at a moderate capacity tier, retained centralised slow-moving inventory, and assigned fast-moving emergency kits to both facilities. It qualified a second supplier and established quarterly capacity and quality reviews. The design authority required any change in demand assumptions, service targets or disruption probabilities to be documented and approved.

Outcome and learning points

The design improved western service while avoiding excessive duplication. Its main lesson is that network alternatives must be evaluated across normal and stressed conditions. Fixed cost, transport cost, inventory, penalties, resilience and regulatory feasibility must be considered together. The case is illustrative; the numerical outcomes are managerial conclusions rather than claims about an actual company.

12. Implementation Best Practices and Limitations

A disciplined implementation sequence is:

1. Define service, scope, horizon, decision rights and success measures.
2. Build and cleanse a node–arc data model.
3. Generate feasible candidate sites and supplier options.
4. Establish the baseline network and reconcile it to finance.
5. Develop location, allocation and capacity alternatives.
6. Test scenarios, sensitivities and disruption stress cases.
7. Validate results with operations, finance, procurement, legal and sustainability teams.
8. Pilot reversible changes and create a transition plan.
9. Monitor actual cost, service, utilisation, emissions and resilience.
10. Re-optimise when structural conditions change.

Limitations include poor data, uncertain probabilities, aggregation bias, omitted behavioural responses, unrealistic transport assumptions and overreliance on a single objective. Models can also create an appearance of precision that exceeds the quality of their inputs. Sensitivity analysis should identify which assumptions actually drive the recommendation. Managerial judgement remains necessary for reputation, political risk, organisational capability and ethical concerns that are difficult to monetise.

Chapter Summary

Supply chain network design determines the long-term structure through which materials, products, information and returns flow. It integrates facility location, supplier selection, production allocation, distribution, capacity, inventory and service decisions. Cost-to-serve analysis prevents narrow focus on freight by incorporating fixed, variable, inventory, quality, returns and shortage costs. Centre-of-gravity analysis provides an initial geographic estimate, while capacitated location-allocation and mixed-integer models evaluate discrete facility and flow choices. Capacity must reflect practical utilisation, product mix and peak requirements. Scenario, robust and stochastic approaches address demand, supply, transport and disruption uncertainty. Effective designs balance cost with service, resilience, sustainability, compliance and flexibility. GIS, optimisation, simulation, digital twins and AI can improve analysis, but require data quality, cybersecurity, explainability and human governance. The strongest designs are validated across normal and stressed conditions and implemented through cross-functional control.

Key Terms

Network design: Strategic configuration of nodes, facilities, flows and policies in a supply chain.

Echelon: A stage or level in a supply chain, such as plant, distribution centre or retailer.

Cost-to-serve: Full cost incurred to supply a customer, segment, product or channel.

Facility location: Selection of sites for plants, warehouses or other network nodes.

Allocation: Assignment of demand or flows to facilities, suppliers or production sites.

Mixed-integer programming: Optimisation involving continuous quantities and discrete decisions.

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

Capacity: Maximum feasible throughput over a defined period and resource basis.

Practical capacity: Capacity after allowing for maintenance, breaks, changeovers and normal losses.

Utilisation: Proportion of available capacity used by actual output.

Postponement: Delaying product differentiation until demand information is more certain.

Total landed cost: Purchase price plus logistics, duties, taxes, handling and related costs.

Scenario analysis: Evaluation of a decision under alternative internally consistent futures.

Robust optimisation: Optimisation designed to perform acceptably across a defined uncertainty set.

Stochastic programming: Optimisation that represents uncertain outcomes probabilistically.

Recourse: A later decision made after uncertainty has been observed.

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

Digital twin: Data-connected representation used to analyse or simulate a real network.

Cost-to-serve: Full cost incurred to supply a customer, segment, product or channel.

Carbon-aware design: Network design that explicitly accounts for greenhouse-gas emissions or carbon prices.

Review Questions

Short Answer Questions

1. What distinguishes strategic network design from operational transport planning?
2. Why can centralisation reduce cost while increasing risk?
3. State the main assumptions of the centre-of-gravity method.
4. What is the difference between nominal and practical capacity?
5. Why should a location model include material-balance constraints?

Long Answer Questions

1. Explain the major trade-offs involved in designing a multi-echelon distribution network.
2. Describe how cost-to-serve analysis can change customer, channel or service decisions.
3. Compare deterministic, scenario-based, robust and stochastic network-design approaches.
4. Explain how supplier portfolio design can improve resilience without eliminating efficiency.
5. Discuss the governance, ethical and sustainability requirements for AI-supported network design.

Analytical / Application Questions

1. Three markets have volumes of 500, 1,500 and 1,000 tonnes and coordinates (10, 10), (30, 20) and (20, 40). Calculate the centre-of-gravity coordinates and explain why the result requires further feasibility analysis.
2. A company faces annual demand of 120,000 units and targets 75% practical utilisation. Calculate required practical capacity. Then identify two reasons why the result may understate peak capacity needs.
3. Design a scenario-analysis framework for a regional distribution network facing 20% demand growth, a four-week port closure and a major supplier failure. Specify decisions, performance measures and governance controls.

Multiple Choice Questions with Answers

Question 1

Which decision is most characteristic of strategic supply chain network design?

A. Sequencing trucks for tomorrow’s deliveries

B. Selecting the location and capacity of a regional distribution centre

C. Printing a daily picking list

D. Authorising an individual purchase order

Correct Answer: B. Selecting the location and capacity of a regional distribution centre

Question 2

In the simplified facility-location model, what does the binary variable yᵢ represent?

A. Quantity shipped from facility i

B. Demand at customer i

C. Whether facility i is open

D. Inventory held at facility i

Correct Answer: C. Whether facility i is open

Question 3

Using the centre-of-gravity method, what is the correct weighting basis for market coordinates?

A. Annual demand or shipment volume

B. Number of employees at the market

C. The alphabetical order of customers

D. Facility rent alone

Correct Answer: A. Annual demand or shipment volume

Question 4

Expected demand is 80,000 units and target utilisation is 0.80. What practical capacity is required?

A. 64,000 units

B. 80,000 units

C. 96,000 units

D. 100,000 units

Correct Answer: D. 100,000 units

Question 5

Which cost is most likely to be omitted in a narrow freight-only analysis?

A. Line-haul transportation

B. Fuel surcharge

C. Inventory carrying cost

D. Distance travelled

Correct Answer: C. Inventory carrying cost

Question 6

What is the principal purpose of a scenario analysis?

A. To guarantee the forecast is correct

B. To evaluate decisions under alternative plausible conditions

C. To remove all managerial judgement

D. To replace capacity constraints

Correct Answer: B. To evaluate decisions under alternative plausible conditions

Question 7

A model requires that inbound flow to a distribution centre equals outbound flow plus inventory increase. This is an example of what type of constraint?

A. Material-balance constraint

B. Binary-opening constraint

C. Carbon-price constraint

D. Weighted-rating constraint

Correct Answer: A. Material-balance constraint

Question 8

What is a principal advantage of dual sourcing?

A. It always reduces purchase price

B. It eliminates quality management

C. It reduces dependence on one supplier

D. It guarantees zero inventory

Correct Answer: C. It reduces dependence on one supplier

Question 9

In the carbon-cost expression, the product of emissions per unit and activity volume represents:

A. Total emissions from the activity

B. Facility fixed cost

C. Customer demand probability

D. Practical capacity

Correct Answer: A. Total emissions from the activity

Question 10

Which statement about AI in network design is most appropriate?

A. AI outputs require no validation if the model is complex

B. Historical data automatically eliminate bias

C. AI should be governed through data quality, explainability and human oversight

D. AI can replace regulatory feasibility analysis

Correct Answer: C. AI should be governed through data quality, explainability and human oversight