10. Supply Chain Network Design and Optimization

Introduction and Context

Supply chain network design determines the physical and operational structure through which products, information and materials move from suppliers to customers. It addresses decisions such as how many plants, warehouses, cross-docks or fulfilment centres an organization should operate; where these facilities should be located; what capacity each should possess; which customers or markets each facility should serve; and how products should flow through the network.

These decisions are strategic because they normally involve substantial capital investment, long asset lives and significant switching costs. A poorly located distribution centre can increase transport expenditure every year, while excessive centralization may reduce inventory and facility costs but weaken delivery responsiveness. Conversely, a highly decentralized network may provide excellent service while creating duplicate inventories, underutilized facilities and managerial complexity.

Network design is therefore not simply a cost-minimization exercise. It is a structured attempt to balance total landed cost, customer service, resilience, sustainability, risk and strategic flexibility. The problem becomes more difficult when demand is uncertain, transport rates vary, facilities have capacity limits, products require special handling, or regulations constrain possible locations.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the strategic role and scope of supply chain network design.

2. Distinguish facility location, capacity, allocation and transportation decisions.

3. Calculate and interpret total landed cost and centre-of-gravity results.

4. Formulate basic facility-location and network-optimization models.

5. Evaluate trade-offs between cost, service, resilience and sustainability.

6. Apply scenario analysis, sensitivity analysis and simulation to network decisions.

7. Interpret technology-enabled approaches, including analytics and digital twins.

8. Recommend a network configuration and defend it from a managerial perspective.

The Network Design Problem

A supply chain network can be represented as a set of nodes and links. Nodes include suppliers, manufacturing plants, regional distribution centres, fulfilment centres, retailers and customers. Links represent transportation routes, information flows and, in some cases, reverse logistics. The design problem concerns the configuration of these nodes and links over a planning horizon.

Four decisions are especially important. The first is the facility-number decision: whether the network should contain one national warehouse, several regional warehouses, or a hybrid structure. The second is the location decision, involving geographic, infrastructural, regulatory and market considerations. The third is the capacity decision, which determines the throughput, storage and processing capability of each facility. The fourth is the allocation decision, which assigns customers, products or demand zones to facilities.

These decisions are interdependent. Adding a warehouse may reduce outbound distance but increase fixed operating cost and inventory duplication. Increasing capacity at an existing plant may be cheaper than opening a new plant, but it may increase lead times to distant markets. Assigning a customer to the closest facility may minimize transport distance but create an imbalance in workload or violate product compatibility constraints.

Strategic, Tactical and Operational Horizons

Strategic network design normally covers several years and includes facility opening, closure, major capacity and technology decisions. Tactical planning may revise sourcing regions, transportation modes, inventory policies or customer allocations over a shorter period. Operational decisions involve daily routing, dispatching, replenishment and order fulfilment. A robust design must support all three horizons: strategic choices establish the feasible structure, tactical choices exploit that structure, and operational decisions execute it.

Total Landed Cost and the Cost–Service Logic

The appropriate economic measure for network design is usually total landed cost rather than purchase price or transportation cost alone. A general expression is:

Total landed cost = procurement cost + inbound transportation + conversion cost + facility fixed cost + facility operating cost + inventory cost + outbound transportation + duties and taxes + handling cost + expected failure or penalty cost.

The exact components vary by industry. For international networks, customs duties, port charges, brokerage and currency effects may be material. For temperature-sensitive products, refrigeration, monitoring and spoilage costs must be included. For e-commerce, picking, packing, last-mile delivery and returns can dominate the economics.

A facility decision should also incorporate the cost of service. Relevant service measures include order cycle time, on-time delivery, fill rate, delivery-frequency capability and product availability. A design with the lowest accounting cost may be unacceptable if it produces lost sales, contractual penalties or reputational damage. In analytical models, service may be represented as a constraint—for example, 95 per cent of demand must be delivered within two days—or as a penalty cost for failing to meet the requirement.

Inventory is a major source of the cost–service trade-off. With independent demand and similar service targets, safety stock tends to increase as the number of stocking locations increases because inventory is duplicated. However, decentralization can reduce replenishment lead time and improve responsiveness. Risk-pooling theory explains why centralization can reduce aggregate safety stock: variability is pooled rather than held separately at every location. The benefit is strongest when demand patterns are not perfectly correlated and when transportation lead times remain acceptable.

Geographic Screening and the Centre-of-Gravity Method

Before detailed optimization, organizations commonly screen candidate regions. Screening criteria may include customer density, supplier proximity, labour availability, land cost, utilities, road and rail connectivity, port access, taxes, environmental restrictions, security and disaster exposure. A location that appears attractive geographically may be unsuitable if it lacks skilled labour or reliable power.

The centre-of-gravity method provides a preliminary estimate of a facility’s geographic position. If demand points have coordinates \((x_i,y_i)\) and weights \(w_i\), the weighted location is:

X = Σ(wᵢxᵢ) / Σwᵢ

Y = Σ(wᵢyᵢ) / Σwᵢ

Weights may represent demand volume, shipment frequency or transport cost exposure. The method is useful for initial screening, especially when transportation rates are approximately proportional to distance. It does not, however, identify the final site. It ignores road networks, congestion, land availability, taxes, facility costs, service constraints and nonlinear transport rates. Its result should therefore be treated as a candidate area rather than a final recommendation.

A practical procedure is to calculate the centre of gravity, identify nearby feasible sites, estimate actual route distances and costs, and then evaluate those sites using a broader optimization model.

Models for Network Optimization

Transportation and Allocation Models

A transportation model determines how much product should move from each supply node to each demand node. Let \(x_{ij}\) denote the quantity shipped from facility \(i\) to customer or market \(j\), and let \(c_{ij}\) be the unit transportation cost. A basic objective is:

Minimize ΣᵢΣⱼ cᵢⱼxᵢⱼ

subject to supply capacity, demand satisfaction and non-negativity constraints. This model is useful when the facility structure is already known. It does not decide whether facilities should be opened.

Facility-Location Models

A mixed-integer facility-location model introduces a binary variable \(y_i\), where \(y_i=1\) if candidate facility \(i\) is opened and zero otherwise. A simplified objective is:

Minimize ΣᵢFᵢyᵢ + ΣᵢΣⱼcᵢⱼxᵢⱼ + ΣᵢΣⱼhᵢⱼxᵢⱼ

where \(F_i\) is the fixed cost of opening facility \(i\), \(c_{ij}\) is transport cost and \(h_{ij}\) represents other variable landed-cost elements. Typical constraints include:

• Demand satisfaction: each market receives its required volume.
• Capacity: shipments from a facility cannot exceed its capacity if opened.
• Logical linkage: a facility cannot ship unless it is open.
• Maximum number of facilities: useful when management limits complexity.
• Service constraints: assigned facilities must meet distance or time standards.
• Product compatibility: hazardous, frozen or regulated products may require specialized facilities.

Binary variables make the model computationally more difficult than a purely linear transportation model, but they accurately represent opening and closing decisions.

Multi-Echelon Models

In a multi-echelon network, products may move from suppliers to plants, plants to regional distribution centres, and distribution centres to customers. The model must prevent unrealistic flows, such as shipping more from a distribution centre than it receives, and must account for inventory, handling and transfer costs at each echelon. Direct shipping may be economical for high-volume customers, whereas smaller customers may be served through an intermediate warehouse.

Robust and Stochastic Models

A deterministic model assumes that demand, costs and capacity are known. This is rarely fully realistic. A stochastic model represents uncertain demand or disruption using probability distributions or scenarios. A robust model seeks a solution that performs acceptably across a defined range of conditions rather than optimizing one forecast.

Two-stage stochastic planning is a useful framework. Facilities are selected before uncertainty is resolved, while allocations, production and transportation are adjusted after actual demand becomes known. Robust optimization may impose limits on the worst-case increase in cost or service failure. These approaches are particularly important for pharmaceuticals, energy, food, semiconductors and other sectors exposed to disruption.

Scenario Analysis, Sensitivity and Simulation

A single optimal answer can be misleading because it depends on assumptions. Scenario analysis evaluates alternative futures such as high demand, low demand, fuel-price escalation, a port closure, a new competitor, a carbon tax or a change in import rules. The purpose is not to predict one future precisely, but to identify decisions that remain sound across plausible futures.

Sensitivity analysis changes one or more assumptions to determine which variables influence the recommendation. Important variables may include demand growth, facility rent, transport rates, labour cost, service requirements and inventory carrying cost. If a small change in fuel cost switches the preferred design from two warehouses to three, management should recognize that the decision is fragile.

Simulation is valuable when the system contains queues, variable processing times, vehicle congestion, stochastic demand or complex service rules that are difficult to represent in a mathematical program. A discrete-event simulation can compare candidate networks by measuring throughput, waiting time, utilization, stockouts and delivery performance. Optimization identifies promising configurations; simulation tests how those configurations behave under operational variability. Used together, they provide stronger evidence than either method alone.

A Structured Network-Design Process

A disciplined project normally proceeds through the following stages, although iteration is expected.

First, define the decision scope, planning horizon, products, markets, service commitments and constraints. Second, collect and cleanse data on demand, origins, destinations, costs, capacities, lead times and facility characteristics. Third, segment products and customers because one network may not suit all demand. Fourth, identify feasible candidate sites through geographic and institutional screening. Fifth, construct a baseline representing the current network. Sixth, formulate and solve alternative network models. Seventh, test results through sensitivity analysis, scenarios and simulation. Eighth, evaluate implementation feasibility, including labour transition, systems integration, permits and stakeholder effects. Finally, select, phase and govern the implementation.

Data quality is often the weakest part of the process. Demand may be recorded in sales units while transport is priced by weight or cube. Facility capacities may represent theoretical maximum rather than reliable throughput. Transportation rates may omit accessorial charges. Analysts must reconcile units, validate assumptions with operations personnel and document data provenance.

Applied Example: Regional Distribution Network

Consider an illustrative consumer-goods company serving four demand regions. It currently operates one central distribution centre. Management is considering either retaining the existing facility, opening two regional warehouses, or operating one central and one regional warehouse.

The central facility has a low fixed cost because it already exists, but average delivery distance is high. Two regional facilities require additional rent and labour and would duplicate some inventory. Preliminary centre-of-gravity analysis places the largest demand concentration near the eastern region, suggesting that a regional facility there may produce the greatest transport benefit.

The company first calculates baseline total landed cost, including outbound transport, warehouse labour, inventory carrying cost, returns and service penalties. It then uses a facility-location model with candidate sites in the east, west and central regions. The model requires all demand to be served and imposes a two-day delivery constraint for priority customers. A high-demand scenario, a fuel-price scenario and a disruption scenario are then tested.

The results show that two regional facilities have the lowest transport cost but excessive inventory duplication under the base case. The central-plus-eastern design has slightly higher transport cost than the two-regional design, but lower total landed cost, better service to priority customers and greater resilience than the single-facility design. Simulation further shows that the eastern facility would become congested during seasonal peaks unless temporary labour and overflow capacity are planned.

The recommended strategy is therefore a phased central-plus-eastern network. The company retains the central facility for slow-moving products and western demand, while the eastern facility handles fast-moving products and priority customers. A later trigger—such as sustained demand above a defined volume—would justify a second regional facility. This illustrates the value of real-options thinking: delaying an irreversible investment can be rational when demand uncertainty is significant.

Managerial and Industry Applications

In manufacturing, network design links plant capacity, supplier geography, postponement and distribution. A firm may centralize final configuration while locating component inventories in several regions. In retail, store replenishment, distribution-centre ownership and omnichannel fulfilment must be considered together. In e-commerce, the economics of last-mile delivery and returns often favour a network of fulfilment centres, sortation hubs and local delivery stations.

In healthcare, location decisions must include emergency response time, cold-chain integrity, regulatory licensing and product availability. In humanitarian logistics, facility design emphasizes accessibility, pre-positioning and resilience rather than lowest cost. In reverse logistics, collection points, inspection centres, repair facilities and recycling destinations must be incorporated into the forward network.

Managers should also distinguish ownership from physical location. A facility may be owned, leased, operated by a third-party logistics provider or shared with other firms. Outsourcing can reduce capital requirements and increase flexibility, but it introduces contractual, information-sharing and service-governance risks.

Technology, Analytics and Digital Network Design

Geographic information systems combine demand, road, demographic and facility data to visualize candidate locations. Transportation-management and warehouse-management systems provide actual shipment, route and processing data. Cloud-based optimization platforms allow planners to compare many network configurations without building all analytical infrastructure internally.

Digital twins represent network structure and operating rules in a continuously updated analytical model. They can test the effect of demand changes, port disruption, labour shortages or facility expansion. Machine-learning models may improve demand forecasts, delivery-time estimation and disruption-risk prediction. However, predictive accuracy does not remove the need for managerial judgment. An algorithm trained on historical allocation patterns may reproduce past bias, ignore strategic change or optimize a narrow cost measure.

Explainability, access control, data security and model governance are essential. Decision-makers should be able to identify the data, assumptions and constraints that produced a recommendation. Sensitive information about supplier capacity, customer demand and facility vulnerabilities must be protected.

Challenges, Risks and Common Errors

The most common error is optimizing transportation cost while omitting inventory, facility, handling, service and risk costs. Another is treating the forecast as a fact. Forecast error can make a seemingly optimal facility underutilized or overloaded. Analysts also frequently use straight-line distances when road travel times, congestion and border delays determine actual performance.

Other risks include excessive model complexity, poor data definitions, hidden capacity constraints, double-counting costs and failure to involve operations managers. A mathematically optimal design may be impossible to implement because land cannot be acquired, permits are delayed, labour is unavailable or customers reject a service change. Network decisions can also create stranded assets and workforce displacement.

A further danger is false precision. Results reported to several decimal places may conceal uncertain assumptions. Scenario ranges, confidence intervals and qualitative judgement should accompany model outputs.

Governance, Ethics and Sustainability

Network decisions affect employees, communities, suppliers and the environment. Closing a facility may improve financial performance but cause substantial local employment loss. New facilities may increase traffic, noise, land consumption and emissions. Ethical design requires transparent criteria, stakeholder engagement and compliance with labour, land-use, safety and environmental regulations.

Sustainability should be embedded in the objective rather than treated as an afterthought. Models can include carbon emissions from transport and facilities, renewable-energy availability, packaging waste, water consumption and environmental penalties. Consolidation may reduce inventory and facility emissions but increase delivery distance. Conversely, a larger number of facilities may shorten delivery routes while increasing building energy use. Life-cycle analysis is therefore preferable to evaluating only one emission source.

Emerging Directions and Best Practices

Recent network-design practice is moving toward resilient, digitally connected and low-carbon configurations. Firms increasingly use multi-sourcing, postponement, nearshoring, supplier diversification and flexible third-party capacity to reduce dependence on a single node. Urban fulfilment, micro-fulfilment and alternative delivery models are developing in response to e-commerce density and customer expectations.

The following practices improve decision quality:

• Use total landed cost and explicit service metrics rather than isolated cost measures.
• Segment customers, products and risk profiles before modelling.
• Validate data and assumptions with finance, operations, procurement and commercial teams.
• Compare deterministic, scenario-based and simulation results.
• Test disruption, demand growth and capacity loss, not only the most likely forecast.
• Use phased investments and trigger points when uncertainty is high.
• Include carbon, compliance, workforce and community effects.
• Document model logic and maintain governance after implementation.
• Review the network periodically because demand, technology and risk conditions change.

Chapter Summary

Supply chain network design establishes the number, location, capacity and allocation of facilities and flows. It is a strategic decision that must balance total landed cost with service, resilience, sustainability and implementation feasibility. The centre-of-gravity method is useful for preliminary geographic screening, while transportation, facility-location, mixed-integer, multi-echelon, stochastic and robust models support more rigorous decisions. Scenario analysis and sensitivity testing reveal the effects of uncertainty, and simulation evaluates operational behaviour involving variability and congestion. Digital tools, GIS, optimization platforms and digital twins extend analytical capability, but data quality, transparency and managerial judgement remain essential. Effective network design is therefore not the selection of a theoretically perfect map; it is the creation of a feasible and adaptable supply system that performs acceptably across changing conditions.

Key Terms

Allocation: Assignment of customers, products or demand volumes to particular facilities or supply sources.

Capacity: Maximum practical volume a facility, route or process can handle over a defined period.

Centre of gravity: A weighted geographic point used to estimate a potentially efficient facility location.

Digital twin: A digital representation of a physical network used for monitoring, analysis and experimentation.

Facility-location model: An optimization model that selects sites and assigns demand while considering fixed and variable costs.

Fixed facility cost: Cost incurred from establishing or operating a facility, largely independent of short-term throughput.

Landed cost: Complete cost of making a product available at its destination, including purchase, transport, duties, handling and related costs.

Mixed-integer programming: Optimization involving continuous variables and discrete decisions such as opening or closing facilities.

Multi-echelon network: A network containing multiple sequential levels, such as plants, distribution centres and customers.

Network resilience: The ability of a supply network to withstand, adapt to and recover from disruption.

Opportunity cost: The value of the best alternative sacrificed by selecting a particular network decision.

Postponement: Delaying final product configuration or customization until demand information is clearer.

Robust optimization: An approach seeking solutions that perform acceptably across uncertain conditions.

Scenario analysis: Evaluation of a decision under alternative plausible future conditions.

Sensitivity analysis: Examination of how changes in assumptions affect a model’s result.

Service constraint: A requirement limiting delivery time, availability, fill rate or another customer-service measure.

Simulation: Computational representation of system behaviour under specified rules and variability.

Total landed cost: The aggregate economic cost of sourcing, producing, moving, storing and delivering a product.

Transportation model: A model determining shipment quantities between known origins and destinations.

Review Questions

Short Answer Questions

1. What four primary decisions are addressed in supply chain network design?

2. Why is total landed cost more useful than transportation cost alone?

3. State the centre-of-gravity formulas for the horizontal and vertical coordinates.

4. What is the purpose of sensitivity analysis in network design?

5. Distinguish strategic network design from operational transportation planning.

Long Answer Questions

1. Explain the cost–service trade-off involved in centralizing versus decentralizing inventory.

2. Describe the structure and major constraints of a mixed-integer facility-location model.

3. Explain how scenario analysis and simulation complement optimization.

4. Discuss the role of resilience, sustainability and governance in network design.

5. Describe a complete process for designing and implementing a distribution network.

Analytical / Application Questions

1. A firm can operate one national warehouse or two regional warehouses. Explain the data and cost elements you would require before recommending either alternative.

2. A centre-of-gravity calculation identifies a location in an environmentally restricted zone. How should the organization proceed, and why should it not accept the calculated point automatically?

3. A two-warehouse solution minimizes expected cost but performs poorly when one facility is disrupted. Recommend an analytical approach for comparing this design with a higher-cost resilient alternative.

22. Multiple Choice Questions with Answers

Question 1

Which decision is most directly represented by a binary variable in a facility-location model?

A. The quantity shipped on a route

B. Whether a candidate facility is opened

C. The demand forecast for a market

D. The average delivery time

Correct Answer: B. Whether a candidate facility is opened

Question 2

Why may a centralized inventory network require less aggregate safety stock?

A. It eliminates all transportation lead time

B. It guarantees perfectly accurate forecasts

C. It pools demand variability across markets

D. It removes the need for service targets

Correct Answer: C. It pools demand variability across markets

Question 3

What is the principal limitation of the centre-of-gravity method?

A. It cannot use demand weights

B. It ignores geographic coordinates

C. It always produces multiple locations

D. It may ignore roads, regulations and facility feasibility

Correct Answer: D. It may ignore roads, regulations and facility feasibility

Question 4

Which cost is most likely to be omitted if a firm evaluates only purchase price and freight?

A. Inventory carrying cost

B. Customer demand

C. Product specification

D. Facility coordinates

Correct Answer: A. Inventory carrying cost

Question 5

A model requires 95 percent of demand to arrive within two days. This is best represented as a:

A. Service constraint

B. Binary demand variable

C. Centre-of-gravity weight

D. Procurement discount

Correct Answer: A. Service constraint

Question 6

What is the main purpose of scenario analysis?

A. To prove that one forecast is correct

B. To examine performance under plausible alternative futures

C. To eliminate the need for data validation

D. To replace all optimization models

Correct Answer: B. To examine performance under plausible alternative futures

Question 7

Which situation most strongly supports using simulation in addition to optimization?

A. All demand and processing times are fixed

B. The network has no capacity restrictions

C. Queues, congestion and variable processing times affect performance

D. Only the purchase price is relevant

Correct Answer: C. Queues, congestion and variable processing times affect performance

Question 8

In the illustrative case, why was the central-plus-eastern design preferred?

A. It had the lowest fixed cost in every scenario

B. It balanced cost, priority service and resilience

C. It eliminated the need for inventory

D. It required two new regional facilities

Correct Answer: B. It balanced cost, priority service and resilience

Question 9

Which practice best addresses false precision in network analysis?

A. Reporting more decimal places

B. Ignoring uncertain assumptions

C. Using scenarios and documenting assumptions

D. Selecting the mathematically smallest objective value without review

Correct Answer: C. Using scenarios and documenting assumptions

Question 10

What is a major governance requirement when AI or digital twins support network decisions?

A. Preventing managers from reviewing model outputs

B. Using only historical data without validation

C. Ensuring transparency, security and explainability

D. Removing all qualitative constraints

Correct Answer: C. Ensuring transparency, security and explainability