Global Network Design and Supply Chain Analytics

Introduction

Global network design determines where supply-chain activities occur, how products and information move, which suppliers are used, and how much capacity is installed. It is a strategic decision because facilities, contracts, transport lanes, technology platforms, and regulatory arrangements often remain in place for many years. A poor design can create structurally high landed cost, excessive working capital, fragile supply, or unacceptable service performance. A well-designed network aligns cost, responsiveness, resilience, sustainability, and strategic control.

Network design is not simply a transportation-cost exercise. It combines demand estimation, facility economics, sourcing, inventory, customs, taxation, capacity, lead time, risk, and uncertainty. Analytics provides disciplined methods for comparing alternatives, but managerial judgment remains essential because models simplify reality and rely on assumptions.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Explain the strategic role and major components of a global supply-chain network.
2. Calculate and interpret landed cost and total network cost.
3. Formulate basic facility-location, sourcing, flow, and capacity models.
4. Use optimization, simulation, sensitivity analysis, decision trees, and scenario modelling appropriately.
5. Translate analytical results into robust managerial decisions under uncertainty.
6. Identify governance, ethical, regulatory, technological, and implementation issues in global network design.

1. The Scope of Global Network Design

A global network normally includes suppliers, plants, contract manufacturers, distribution centres, ports, cross-docks, customer zones, reverse-logistics facilities, and digital information systems. The design problem has four linked dimensions:

• Location: where facilities and inventory points should be placed.
• Sourcing: which suppliers or production sites should serve each demand point.
• Flow: how products should move through the network.
• Capacity: how much production, storage, handling, and transport capability should be installed or contracted.

The design horizon is usually three to ten years, although strategic decisions should be reviewed as demand, trade policy, technology, and risk change. Tactical planning converts the design into sourcing allocations, production plans, shipment schedules, and inventory policies.

A useful distinction is between configuration and operation. Configuration concerns the structure of the network; operation concerns how that structure is used. A low-cost configuration may be operationally poor if it lacks flexibility, has long replenishment lead times, or depends on a single politically exposed corridor.

Core performance dimensions

A network should be evaluated against a balanced set of criteria:

Optimising one dimension in isolation can damage the others. For example, a single low-cost offshore plant may minimise unit manufacturing cost while increasing inventory, disruption exposure, and emissions.

2. Landed-Cost Analysis

2.1 Meaning and structure

Landed cost is the total cost of acquiring a product and making it available at a specified destination. Its boundary must be stated explicitly. A practical unit formulation is:

where:

• LC_u = landed cost per unit at the destination;
• P_u = supplier price per unit;
• C_src,u = origin handling and consolidation cost per unit;
• F_u = international freight per unit;
• I_u = cargo insurance per unit;
• D_u = destination handling, port, and documentation cost per unit;
• T_u = tariffs and non-recoverable import taxes per unit;
• B_u = brokerage and customs-compliance cost per unit;
• H_u = inland transport to the destination per unit;
• Q_u = quality, inspection, and compliance cost per unit;
• R_u = expected cost of loss, damage, obsolescence, or returns per unit.

The formula is additive when every term uses the same currency and unit basis. Recoverable value-added tax should generally not be included as an economic cost, whereas non-recoverable duties and taxes should be included. Incoterms determine which party arranges and pays for particular activities, but they do not by themselves determine the buyer’s economic cost; transfer prices and charges must be mapped to the actual decision boundary.

For a shipment, fixed costs should be allocated consistently. If a shipment costs F_s and contains Q_s saleable units, then the freight allocation is F_s ÷ Q_s. This allocation changes with shipment size, so landed cost is often nonlinear.

2.2 Worked landed-cost example

An importer evaluates a supplier quotation of $42 per unit. Origin handling is $1.20, international freight is $3.80, insurance is $0.25, destination charges are $1.10, brokerage is $0.45, and inland delivery is $2.20. The customs value is the supplier price plus origin handling and freight: $42 + $1.20 + $3.80 = $47.00. A 6% tariff therefore equals $2.82. Quality inspection is $0.40 and expected damage and returns are $0.70.

Thus:

The supplier price of $42 is therefore not comparable with a domestic offer unless all relevant cost elements are included. If the domestic alternative costs $55.40 delivered but has shorter lead time and lower disruption exposure, the apparent $0.48 import advantage may not justify the risk or working-capital requirement.

2.3 Beyond unit landed cost

Decision makers should calculate total cost to serve, not only unit landed cost. A network-level expression is:

The terms must be defined to avoid double counting. For example, facility cost may include fixed operating cost but not depreciation if the analysis separately uses cash capital expenditure. Inventory cost may include cycle stock, safety stock, pipeline inventory, and obsolescence. Risk and emissions should be monetised only when the organisation has a defensible valuation method; otherwise they should be treated as separate decision criteria.

3. Mathematical Models for Network Design

3.1 Facility-location model

A basic capacitated facility-location model selects facilities and assigns flows. Let:

• i index candidate facilities and j index demand zones;
• y_i = 1 if facility i is opened, otherwise 0;
• x_ij = units shipped from facility i to zone j;
• F_i = fixed cost of opening facility i;
• c_ij = variable cost per unit from i to j;
• d_j = demand in zone j;
• K_i = capacity of facility i.

The objective is:

subject to:

The first constraint requires demand satisfaction. The second prevents shipments from unopened facilities and limits capacity. This is a mixed-integer linear programme when costs, demand, and capacity are treated as known and linear. Additional constraints can represent minimum utilisation, supplier qualification, route restrictions, carbon budgets, service levels, or single-source limits.

The model is useful for identifying economic structure, but its assumptions are restrictive. Demand may be uncertain, transport rates may have quantity breaks, and facility performance may deteriorate during disruption. A model should therefore produce a portfolio of alternatives rather than a supposedly perfect answer.

3.2 Multi-echelon and capacity considerations

A multi-echelon model may include suppliers, plants, regional distribution centres, and customers. Flow conservation requires that, at each intermediate node, inflow plus opening inventory minus outflow equals the change in inventory. Production capacity, storage capacity, labour availability, and transport capacity can be represented as constraints.

Capacity should be evaluated in effective rather than nominal terms. If nominal annual capacity is 100,000 units, planned utilisation is 85%, and expected availability is 95%, effective capacity is:

This calculation assumes the factors are multiplicative and independently applicable. Maintenance, changeovers, product mix, and labour constraints may require a more detailed capacity model.

3.3 Service constraints

A cost-minimising model may allocate all demand to the cheapest facility. A service constraint can prevent this outcome. For example, if facility i can serve zone j within the required lead time, define a_ij=1; otherwise a_ij=0. Then:

where M is a sufficiently large valid upper bound, preferably the relevant demand or capacity rather than an unnecessarily large number. This constraint excludes unacceptable lanes. Alternatively, the model can minimise cost subject to a weighted average delivery-time limit, though average service can conceal poor performance for particular customers.

4. Optimisation, Simulation, and Scenario Modelling

Optimisation

Optimisation finds the best solution according to a stated objective and constraints. Linear programming is suitable for divisible flows and linear relationships; mixed-integer programming is needed for open/close, select/not-select, or fixed-charge decisions. Multi-objective problems can use a weighted objective:

where C, S, R, and E are normalised cost, service penalty, risk, and environmental measures, and weights w_c, w_s, w_r, and w_e sum to 1. Weights are managerial preferences, not objective facts; sensitivity testing is essential.

Simulation

Simulation evaluates how a proposed network behaves under variable demand, lead times, failures, customs delays, and replenishment rules. Monte Carlo simulation repeatedly samples uncertain inputs from specified probability distributions and records performance outcomes. If demand is normally distributed with mean μ and standard deviation σ, a simulated demand draw may be represented as D = μ + σZ, where Z is a standard normal random variable. The normal assumption is inappropriate for bounded, highly skewed, or intermittent demand unless transformed or otherwise justified.

Simulation does not automatically identify the best network. It compares policies or configurations under uncertainty. A sound study uses a warm-up period for non-terminating systems, adequate replications, common random numbers where alternatives are compared, and confidence intervals for performance estimates.

Scenario modelling

A scenario is a coherent combination of external conditions, such as demand growth, tariff policy, fuel prices, exchange rates, or disruption frequency. Scenarios should not be confused with forecasts. A forecast expresses expected conditions; a scenario explores plausible conditions, including adverse ones.

A simple expected-cost calculation is:

where p_s is the probability assigned to scenario s and TC_s is the total cost under that scenario. Expected value is useful for repeated decisions with reliable probabilities, but it can be misleading for one-time irreversible investments or when extreme downside matters more than average cost.

5. Sensitivity Analysis and Robust Decisions

Sensitivity analysis changes one or more assumptions and observes the effect on the recommendation. Important variables include demand, freight rates, tariffs, exchange rates, facility cost, capacity, service penalties, disruption probability, and carbon price.

A one-way sensitivity analysis changes one input while holding others constant. A two-way analysis varies two inputs, such as demand and tariff rate, to identify the switching point between alternatives. A tornado chart ranks variables by their effect on the outcome, although the chart does not prove causality.

For a decision alternative a, regret in scenario s can be defined as:

The robust alternative may minimise maximum regret:

This minimax-regret approach is appropriate when scenario probabilities are unreliable and management wants protection against a very poor relative outcome. It may select a more expensive expected-cost option because it preserves flexibility.

Robust design commonly includes dual sourcing, postponement, modular production, transferable tooling, qualified alternate lanes, reserved logistics capacity, and staged investment. These options have value even when they increase nominal cost, because they preserve the ability to respond as information improves.

6. Decision Trees and Real Options

Decision trees represent sequential decisions and uncertain events. A decision node is controlled by management; a chance node contains possible outcomes with probabilities. For a decision made before uncertainty resolves, expected monetary value is:

where V_a,s is the value or cost outcome of alternative a in scenario s. Costs can be represented as negative values, provided the convention is used consistently.

Worked decision-tree example

A company may establish a regional plant for a fixed investment of $12 million or continue importing. If demand is high, with probability 0.6, the plant produces a benefit of $20 million net of operating costs; if demand is low, with probability 0.4, the benefit is $4 million. The plant’s expected net value is:

If importing has an expected net value of $1.1 million, the plant is preferred on expected value. However, the low-demand outcome is $4 million − $12 million = −$8 million relative to the stated benefit convention. Management should test whether probabilities are credible and whether a staged lease, contract manufacturing, or expansion option can reduce downside.

Real-options thinking treats flexibility as an asset. A company may deliberately choose a modular facility or short-term contract even if its initial cost is higher, because it retains the option to expand, defer, abandon, or switch suppliers. Option value is difficult to estimate precisely, but it should be discussed explicitly rather than ignored.

7. From Model Output to Managerial Decision

A model output is not a decision. Managers should use a disciplined translation process:

1. Validate the data boundary. Confirm demand ownership, currency, tax treatment, capacity definitions, lead-time measurement, and cost allocation.
2. Check feasibility. Ensure that supplier qualification, labour, permits, customs, transport availability, and implementation timing are realistic.
3. Challenge structural assumptions. Ask whether demand is divisible, whether costs are linear, whether disruption events are independent, and whether service is adequately represented.
4. Compare alternatives, not only the optimum. Examine near-optimal solutions that offer greater resilience or simpler implementation.
5. Test uncertainty. Use scenarios, sensitivity analysis, and simulation rather than a single deterministic forecast.
6. Identify trigger conditions. Define thresholds for adding capacity, changing suppliers, or activating contingency routes.
7. Stage commitments. Use pilots, leases, flexible contracts, or modular investment when uncertainty is high.
8. Assign accountability. Specify owners, review dates, data sources, and governance controls.

A practical decision memorandum should report the recommended configuration, assumptions, total-cost range, service outcomes, major risks, sensitivity results, implementation dependencies, and conditions that would reverse the recommendation.

8. Illustrative Case: Northstar Medical Devices

Background

Northstar Medical Devices, a fictional company, sells diagnostic equipment in North America and Europe. Its single plant in East Asia supplies both markets. Unit manufacturing cost is low, but the company experiences long replenishment times, port congestion, and exposure to tariff changes. European customers require 95% on-time delivery, while North American demand is growing faster than forecast.

Problem

Northstar must choose among three alternatives over a five-year horizon:

• retain the existing plant and add inventory;
• establish a European contract-manufacturing and postponement operation;
• build a second owned plant in Eastern Europe.

The analysis must include purchase and production cost, freight, duties, inventory, quality, capacity, service, disruption exposure, and investment.

Analysis approach

The team first defined the decision boundary and constructed a landed-cost model by market and product family. It then used a mixed-integer model to allocate demand subject to plant and contract capacity. Demand growth, tariff conditions, and port disruption were represented through scenarios. A discrete-event simulation tested order-cycle time and stockout performance under variable lead times. Sensitivity analysis varied European demand by ±25%, the tariff by ±5 percentage points, and the contract manufacturer’s yield.

The deterministic cost-minimising solution retained the Asian plant and used high inventory. Under disruption scenarios, however, European service fell below 88%. The contract-manufacturing option cost 2.4% more in the median scenario but maintained service above 95% in most tested scenarios. The owned European plant had the best long-run unit cost at high demand, but required substantial capital and had a two-year qualification lead time.

Solution

Northstar selected a staged design: qualify a European contract manufacturer for final postponement, retain the Asian plant for core components, and defer the owned plant decision for eighteen months. A trigger was established: if European demand exceeded 70% of the proposed owned-plant break-even volume for four consecutive quarters, management would reopen the capital-investment decision. The company also reserved alternate ocean and airfreight capacity and created a tariff-monitoring process.

Outcome and learning points

The selected design was not the lowest-cost solution in every scenario. It was preferred because it reduced time to responsiveness, preserved the option to expand, diversified final assembly, and limited irreversible capital exposure. The case demonstrates that robust network design often means selecting a flexible and implementable solution rather than maximising expected short-run cost efficiency.

9. Governance, Ethics, Regulation, and Technology

Global network decisions can shift environmental burdens, labour risks, and community impacts across borders. Governance should include supplier labour standards, modern-slavery controls, anti-corruption procedures, product traceability, cybersecurity, responsible minerals policies, and transparent escalation channels. A low landed cost is unacceptable if it depends on unlawful labour practices or falsified customs documentation.

Regulatory analysis should consider rules of origin, tariff classification, sanctions, export controls, product safety, data localisation, tax nexus, transfer pricing, and privacy. Legal advice is necessary where the network changes customs valuation, intellectual property ownership, or intercompany transactions.

Digital twins, cloud-based control towers, machine learning forecasts, and optimisation software can improve analysis. AI may identify demand patterns, estimate disruption probabilities, or generate candidate scenarios. Nevertheless, models can reproduce biased data, conceal uncertain assumptions, or create false precision. Human review, model documentation, data lineage, access controls, and independent validation are required. Generative AI should not be allowed to make autonomous decisions about sanctions compliance, supplier exclusion, or safety-critical allocation without accountable expert oversight.

10. Limitations and Best Practices

Important limitations include poor data quality, omitted costs, unstable probabilities, correlated disruptions, model overfitting, excessive aggregation, and implementation resistance. A model may recommend a mathematically feasible flow that is impossible because a supplier lacks certification or a port lacks the required handling capability.

Best practices are to:

• establish a cross-functional steering team;
• use a clearly documented decision boundary;
• reconcile model data with finance and operational records;
• separate fixed, variable, avoidable, sunk, and opportunity costs;
• model capacity and service at the level where constraints actually occur;
• include disruption correlation and recovery time where material;
• use scenario ranges rather than a single forecast;
• compare robust and near-optimal alternatives;
• validate results with operational experts and pilots;
• document assumptions, version changes, and approvals;
• review the network periodically using performance triggers.

Chapter Summary

Global network design integrates facility location, sourcing, material flow, capacity, service, risk, sustainability, and compliance over a strategic horizon. Landed-cost analysis establishes a comparable economic basis, but total cost to serve must also consider facilities, inventory, quality, risk, and environmental effects. Facility-location and mixed-integer models provide structured optimisation, while simulation tests performance under variable demand and lead times. Scenario modelling and sensitivity analysis expose the assumptions that drive recommendations. Decision trees quantify sequential uncertainty, and real-options thinking values flexibility and staged commitment.

The most defensible decision is not always the deterministic cost minimum. Managers should validate data, challenge assumptions, examine near-optimal alternatives, test uncertainty, and define triggers for future action. Robust network design combines analytical discipline with practical feasibility, ethical governance, regulatory compliance, and accountable managerial judgment.

Key Terms

Landed cost: Total cost of acquiring and delivering a product to a specified destination.

Total cost to serve: Full economic cost of serving a customer or market, including relevant operating and strategic cost elements.

Network design: Strategic determination of facilities, suppliers, flows, capacities, and market assignments.

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

Sourcing: Choice of suppliers, production sites, contracts, and procurement allocations.

Capacity: Maximum feasible output, handling, storage, or transport capability under stated conditions.

Effective capacity: Usable capacity after considering utilisation, availability, changeovers, and operating constraints.

Mixed-integer linear programme: Optimisation model containing continuous variables and discrete binary or integer decisions.

Flow conservation: Requirement that material entering an intermediate node is reconciled with material leaving and inventory change.

Service constraint: Model restriction representing lead time, delivery, fill-rate, or customer-service requirements.

Simulation: Computational representation of system behaviour under repeated sampled conditions.

Scenario modelling: Analysis of coherent alternative future conditions rather than a single forecast.

Sensitivity analysis: Examination of how changes in assumptions affect results.

Decision tree: Sequential representation of decisions, uncertain events, probabilities, and outcomes.

Expected monetary value: Probability-weighted average value across possible outcomes.

Regret: Difference between an alternative’s outcome and the best outcome available after the scenario is known.

Robustness: Ability of a design to perform acceptably across uncertain conditions.

Postponement: Delaying final configuration or differentiation until demand or destination is clearer.

Real option: Value arising from the ability to defer, expand, contract, switch, or abandon a decision.

Digital twin: Data-connected model used to represent and analyse the behaviour of a physical supply-chain system.

Review Questions

Short Answer Questions

1. What is the difference between landed cost and supplier purchase price?
2. Why does a facility-location model require binary decision variables?
3. Distinguish optimisation from simulation.
4. What is the purpose of sensitivity analysis?
5. Explain why effective capacity may be lower than nominal capacity.

Long Answer Questions

1. Explain the main dimensions of global network design and their interdependence.
2. Develop and discuss a landed-cost framework for comparing international and domestic sourcing.
3. Compare deterministic optimisation, scenario modelling, and Monte Carlo simulation.
4. Explain how decision trees and real-options thinking improve strategic network decisions.
5. Discuss the governance, ethical, regulatory, and technology risks of analytics-led network design.

Analytical / Application Questions

1. A supplier price is $30 per unit, origin cost is $1, freight is $4, insurance is $0.20, destination cost is $1.30, brokerage is $0.50, inland transport is $2, quality cost is $0.40, and expected returns are $0.60. A 5% tariff applies to supplier price plus origin cost and freight. Calculate landed cost per unit.
2. A company can choose a plant with expected net benefits of $18 million in a high-demand scenario with probability 0.55 and $3 million in a low-demand scenario with probability 0.45. Investment is $10 million. Compare its expected net value with importing, whose expected net value is $2 million.
3. Design a sensitivity-analysis plan for a two-facility global network. Identify at least five uncertain variables, explain how each could affect the recommendation, and propose one managerial trigger for revisiting the design.

Multiple Choice Questions with Answers

Question 1

Which item is generally included in landed cost when it is non-recoverable?

A. Recoverable value-added tax

B. Import tariff

C. Sunk research expenditure

D. Shareholder dividend

Correct Answer: B. Import tariff

Question 2

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

A. The quantity shipped on a lane

B. The demand forecast for a market

C. Whether a facility is opened

D. The unit transport cost

Correct Answer: C. Whether a facility is opened

Question 3

What is the principal purpose of simulation in network design?

A. To guarantee the mathematically cheapest network

B. To test system performance under variable conditions

C. To eliminate the need for operational data

D. To convert all uncertain inputs into fixed values

Correct Answer: B. To test system performance under variable conditions

Question 4

If nominal capacity is 100,000 units, utilisation is 80%, and availability is 90%, what is effective capacity?

A. 72,000 units

B. 80,000 units

C. 90,000 units

D. 180,000 units

Correct Answer: A. 72,000 units

Question 5

Which method is most appropriate when scenario probabilities are unreliable and management seeks protection against a very poor relative outcome?

A. Minimax regret

B. Simple average costing

C. First-in, first-out allocation

D. Break-even accounting only

Correct Answer: A. Minimax regret

Question 6

What does a service constraint in a network model typically restrict?

A. Only the supplier’s financial statements

B. Whether a flow meets requirements such as lead time

C. The historical exchange rate

D. The number of model variables

Correct Answer: B. Whether a flow meets requirements such as lead time

Question 7

A plant has a benefit of $20 million in a high-demand state with probability 0.6 and $4 million in a low-demand state with probability 0.4. Investment is $12 million. What is its expected net value?

A. $1.6 million

B. $4.0 million

C. $12.0 million

D. $13.6 million

Correct Answer: A. $1.6 million

Question 8

Which practice best reflects robust network design?

A. Selecting the lowest-cost solution under one forecast only

B. Ignoring implementation constraints until after approval

C. Comparing alternatives across plausible scenarios

D. Treating nominal capacity as guaranteed output

Correct Answer: C. Comparing alternatives across plausible scenarios

Question 9

What is postponement?

A. Increasing safety stock at every echelon

B. Delaying final product configuration until demand is clearer

C. Eliminating all regional facilities

D. Using only air freight

Correct Answer: B. Delaying final product configuration until demand is clearer

Question 10

Which is an important governance requirement for AI-supported network analytics?

A. Removing all human review

B. Using undocumented assumptions to improve speed

C. Maintaining data lineage and accountable validation

D. Allowing the system to bypass sanctions controls

Correct Answer: C. Maintaining data lineage and accountable validation