Logistics Network Design and Location Analytics

Introduction

Logistics network design determines how products, information, returns and sometimes services move through a supply chain. It answers strategic questions such as: Which facilities should be opened? Where should they be located? Which customers should each facility serve? What capacity should be installed? Should inventory and order fulfilment be centralized or distributed? How should transport, service, risk and sustainability be balanced?

These decisions are strategic because facilities may operate for decades, require substantial capital and constrain subsequent operational choices. A poor network can create permanently high transport costs, excessive inventory, weak delivery performance and exposure to disruption. A well-designed network aligns customer requirements, demand geography, production economics, transport infrastructure, tax and regulatory conditions, labour availability, resilience and environmental objectives.

Network design is not simply a search for the lowest-cost map. It is a structured decision process under uncertainty. Analytical models provide discipline, but managerial judgement remains essential because many relevant factors—such as geopolitical exposure, community acceptance and future flexibility—are difficult to quantify precisely.

Learning objectives

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

1. Explain the strategic role and principal components of logistics network design.
2. Formulate facility-location, allocation, capacity and flow decisions mathematically.
3. Compare centralized, decentralized and hybrid network structures.
4. Use scenario and sensitivity analysis to evaluate uncertainty.
5. Assess cost, service, risk and sustainability trade-offs.
6. Critically evaluate data, modelling assumptions, technology and governance issues in network decisions.

1. The architecture of a logistics network

A logistics network consists of nodes and links. Nodes include suppliers, plants, ports, cross-docks, distribution centres, fulfilment centres, stores, customers and return-processing facilities. Links represent transport routes, information connections or material flows. Products may move through several echelons—for example, supplier → plant → regional distribution centre → retailer → customer.

The principal network decisions are:

• Facility location: geographic position of plants, warehouses, hubs or service centres.
• Facility opening: which candidate sites should be activated.
• Capacity: the maximum throughput, storage or production capability of each facility.
• Allocation: which customers, markets or demand zones are assigned to which facilities.
• Flow: quantities shipped on each origin–destination link and through each period.
• Mode and route design: selection of road, rail, air, sea, pipeline or multimodal services.
• Inventory positioning: determination of where cycle stock, safety stock and postponement capacity should be held.
• Resilience configuration: use of backup suppliers, alternative facilities, spare capacity and multiple routes.

A network has several interacting layers. The physical layer contains facilities and transport assets. The informational layer supports demand visibility, orders, inventory records and shipment tracking. The financial layer captures fixed investment, operating cost, working capital, duties and taxes. The governance layer defines ownership, contracts, regulatory compliance and decision rights.

Figure: A multi-echelon logistics network with alternative paths to markets and customers.

2. Centralization, decentralization and postponement

A centralized network serves a broad market from relatively few facilities. Its advantages commonly include inventory pooling, lower duplicated fixed cost, greater purchasing leverage and simpler control. Its disadvantages may include longer average delivery distances, greater dependence on individual facilities, exposure to congestion and weaker responsiveness to local demand.

A decentralized network uses more geographically dispersed facilities. It can reduce last-mile distance and delivery time, improve local market knowledge and provide some geographic redundancy. However, it normally increases facility, labour and systems costs and fragments inventory across locations.

A hybrid network combines national or regional distribution with local fulfilment, cross-docking, stores or third-party logistics providers. The correct structure depends on product characteristics, demand density, order frequency, delivery promise, transport economics and risk tolerance.

Inventory pooling is a major argument for centralization. If independent demand streams are combined, variability may partially offset. Under a simplified model with independent demands, the standard deviation of total demand is:

Here σᵢ is the standard deviation of demand at location i, measured in units per planning period. The formula assumes statistical independence. If demands are correlated, the more general expression is:

where ρᵢⱼ is the correlation between demands i and j. Positive correlation weakens the pooling benefit; negative correlation strengthens it. Pooling does not automatically reduce total inventory because transport batch sizes, service targets, replenishment lead times and facility constraints also matter.

Postponement offers a further design alternative. Generic products are held centrally and customized, labelled or configured near the customer. This can reduce SKU-specific inventory while retaining reasonable responsiveness. It is effective only when product architecture, process technology and information systems support late differentiation.

3. Cost and service structure

A network model should represent the relevant total cost rather than focusing only on transport. A practical annual cost decomposition is:

TC is total annual cost; F is facility opening and fixed operating cost; P is production or procurement cost; T_in and T_out are inbound and outbound transport costs; H is handling and processing cost; I is inventory-related cost; C_r is expected disruption or recovery cost; C_e is environmental or compliance cost; and C_risk represents other monetized risk exposures. All terms must be expressed in a common currency and time basis.

A common inventory holding approximation is:

where h is the annual holding-rate fraction, V is the monetary value per unit and Q_avg is average inventory in units. The formula assumes a reasonably stable valuation and an appropriate annual holding rate that includes capital, space, insurance, obsolescence and handling. It should not be used as a substitute for a detailed inventory policy when lead times, service levels or perishability are material.

Service must be modelled explicitly. Relevant measures include order-cycle time, on-time-in-full percentage, fill rate, response time, geographic coverage and probability of meeting a delivery promise. A network can be cheaper but strategically unacceptable if it fails a service constraint. Conversely, a high service target may require more facilities, capacity, inventory or premium transport.

4. Facility-location models

4.1 The weighted-centre heuristic

For an initial screening of a single facility, the centre-of-gravity estimate is:

xᵢ and yᵢ are coordinates of demand or supply point i, and wᵢ is a weight such as annual shipment volume. The result is a geometric screening point, not necessarily a feasible site. It ignores road topology, land availability, taxes, congestion, service zones and nonlinear transport rates. The method is therefore useful for generating candidates, not for making the final decision.

4.2 The capacitated facility-location model

Let candidate facilities be indexed by j and customers by i. Define:

A deterministic mixed-integer linear programming formulation is:

subject to:

The first constraint requires all demand to be served. The second prevents shipments from an unopened facility and enforces capacity. The third defines non-negativity and the binary opening decision. The model assumes known demand, linear variable cost, divisible flow and a valid candidate set. If demand may be unmet, a shortage variable with a penalty cost should be added rather than silently allowing infeasibility.

The model can be extended with facility throughput limits, minimum utilisation, customer-facility eligibility, multiple products, time periods, inventory balance, emissions, fixed transport charges and service constraints. With several products, xᵢⱼp denotes product p flow and capacity may be measured in weight, volume, pallets or labour hours. Unit consistency is essential: demand and capacity must use compatible units, while cost coefficients must correspond to the same flow unit.

4.3 Flow conservation in multi-echelon networks

For a transshipment node j, flow conservation requires inbound flow plus external supply minus outbound flow to equal inventory change. For period t:

I is ending inventory, x is shipment flow, S is external supply and D is local demand. This equation assumes all flows are measured in the same product units and that losses, shrinkage and processing yields are either negligible or represented by additional parameters. For a pure transshipment node with no inventory, Iⱼ,t = Iⱼ,t−1 = 0.

5. Worked example: selecting distribution capacity

An illustrative company must serve three demand zones. It is considering two candidate distribution centres. Annual demand is 10,000 units in Zone A, 8,000 in Zone B and 7,000 in Zone C. Candidate North has capacity 18,000 units and fixed annual cost £240,000. Candidate South has capacity 20,000 units and fixed annual cost £220,000. The company requires all 25,000 units to be served.

Because either centre alone lacks sufficient capacity, both facilities must be opened unless outsourcing or unmet demand is permitted. Combined capacity is 38,000 units, leaving 13,000 units of spare capacity. The fixed cost is:

£240,000 + £220,000 = £460,000 per year.

Suppose the allocation is North: 10,000 units to A and 8,000 units to B, totalling 18,000; South: 7,000 units to C. Both facilities satisfy capacity constraints. If variable transport and handling cost for this allocation is £3.20 per unit, its annual variable cost is:

25,000 × £3.20 = £80,000.

Total modelled annual cost is therefore £540,000, excluding inventory, emissions and risk costs. This is not proof that the allocation is optimal: alternative allocations must be evaluated using their applicable lane costs. The example demonstrates why opening decisions and allocation decisions should be solved together. A facility with a lower fixed cost may be inferior if it creates expensive or infeasible flows.

6. Scenario and sensitivity analysis

A single optimal solution can be misleading because demand, fuel prices, exchange rates, labour costs, service requirements and disruption probabilities are uncertain. Scenario analysis evaluates coherent alternative futures, such as:

• baseline demand and current transport rates;
• high-growth demand concentrated in one region;
• fuel-price escalation and tighter emissions regulation;
• port closure or facility outage;
• e-commerce growth requiring faster home delivery;
• nearshoring or tariff changes.

For scenarios s with probability pₛ, expected cost is:

This measure is useful only when probabilities are credible. If probabilities are speculative, presenting the range of outcomes may be more honest than calculating a false precision.

Sensitivity analysis changes one input or a controlled group of inputs to identify decision thresholds. Important questions include:

• At what demand level does a second facility become economic?
• How high must fuel cost rise before rail or a different location becomes preferable?
• How much service deterioration is caused by closing one site?
• What capacity is required to survive a specified disruption?
• Which cost or demand assumptions have the greatest influence on the decision?

A useful managerial output is a break-even value. If alternative A has fixed cost F_A and variable cost v_A per unit, while B has F_B and v_B, the quantity at which their costs are equal is:

This expression applies when v_A ≠ v_B and both alternatives have comparable service and feasible capacity. It should not be interpreted outside the range where both designs can actually serve demand.

Robust design may favour a slightly higher expected cost in exchange for lower downside exposure. A risk-adjusted objective can be represented conceptually as:

where R is a selected risk measure and λ reflects management’s risk aversion. R might be variance, maximum regret, conditional value at risk or expected disruption loss. The choice must be disclosed because different risk measures can produce different networks.

7. Risk, resilience and sustainability

Network risk has several dimensions: single-source dependence, geographic concentration, political instability, extreme weather, cyberattack, labour disruption, infrastructure failure and financial distress of logistics providers. Resilience options include dual sourcing, geographically separated facilities, emergency inventory, flexible contracts, alternate routes, modular capacity and visibility systems.

Redundancy is not free. Two facilities may improve continuity but duplicate fixed cost and inventory. Resilience should therefore be evaluated against plausible disruption scenarios, recovery time objectives and the business value of continuity. A critical pharmaceutical product may justify redundancy that would be uneconomic for a low-margin commodity.

Sustainability should be integrated rather than treated as an afterthought. Relevant measures include transport emissions, facility energy, packaging, land use, water, waste, congestion and social effects. If emissions per unit on lane i–j are eᵢⱼ, total transport emissions are:

The units of eᵢⱼ must match the flow units—for example, kilograms of CO₂-equivalent per pallet. A carbon constraint can be written as E_transport ≤ E_max, or emissions can be assigned a shadow price and included in the objective. Analysts must avoid double counting when carbon cost is already embedded in fuel, regulatory or supplier-cost data.

Social and ethical issues include worker safety, fair labour standards, land acquisition, community displacement, traffic impacts and unequal service access. A mathematically optimal location may be unacceptable if it depends on exploitative labour or violates environmental obligations. Governance should include documented assumptions, stakeholder consultation, auditability and compliance with planning, customs, competition, data-protection and environmental law.

8. Technology, data and AI

Network models depend on reliable master data: facility coordinates, capacities, demand history, product dimensions, lane distances, rates, lead times, duties and service commitments. Data cleansing is often more valuable than sophisticated algorithms. Common errors include mixing annual and weekly demand, confusing gross and net weight, using straight-line distance instead of road distance and omitting return flows.

Geographic information systems support spatial visualisation, drive-time analysis, congestion assessment and site screening. Optimization solvers handle mixed-integer models, while simulation can test queueing, variability and disruption behaviour that deterministic models cannot represent well. Digital twins may connect a network model to operational data for continuous monitoring.

AI and machine learning can improve demand forecasting, estimated travel times, anomaly detection and scenario generation. They do not eliminate the need for optimization or governance. Black-box recommendations may encode historical geographic bias, overlook rare disruptions or exploit correlations that do not remain stable. Human review, explainability, version control, data lineage and stress testing are required before high-impact location decisions are automated.

9. Applied case: Northstar Medical Distribution

This is an illustrative case. Northstar Medical Distribution supplies hospitals across a large country. It currently operates one national centre near the capital. The centre provides low pooled inventory but has experienced congestion, long delivery distances and vulnerability to a single-site outage. Hospital customers require 98% on-time delivery for essential products and next-day delivery in major cities.

Background and problem

Demand has grown unevenly, with the greatest growth in the north and west. Products include standard consumables and temperature-sensitive medicines. The national centre has adequate average capacity but limited peak throughput. Management is considering: (1) retaining one centre; (2) opening two regional centres; or (3) retaining the national centre for slow-moving products while adding one regional fulfilment centre for fast-moving products.

Analysis

The design team first separated products by service requirement and temperature-control needs. It validated demand by hospital, road travel times, facility capacity, inventory policy and carrier performance. It then formulated a multi-period mixed-integer model with facility-opening variables, product flows, capacity, inventory balance and service constraints. A disruption scenario removed each candidate facility for four weeks. Sustainability analysis estimated transport emissions using actual road distances and vehicle-load assumptions.

The single-centre design had the lowest fixed cost and the highest inventory-pooling benefit, but it failed the next-day requirement for western hospitals under peak demand. The two-regional-centre design improved service and disruption resilience but duplicated temperature-controlled inventory and required additional specialist staff. The hybrid design preserved central pooling for slow movers while locating fast movers near high-demand markets. It required more complex replenishment and information coordination, but its service performance was less sensitive to a national-centre outage.

Approach and solution

Northstar selected the hybrid design subject to quarterly review. It established minimum emergency stock for essential medicines, qualified a backup carrier, and reserved capacity at a third-party facility in the west. The model was implemented with scenario dashboards showing cost, service, emissions and capacity utilisation. Temperature-sensitive flows received stricter feasibility rules than ordinary products.

Outcome and learning points

The case illustrates that the lowest fixed-cost network is not necessarily the lowest total-risk design. Product segmentation, postponement, pooled inventory and selective regionalization can provide a better balance than choosing between complete centralization and complete decentralization. It also shows that resilience is a design property that should be tested explicitly, not inferred from average utilisation.

10. Best-practice decision process and limitations

A disciplined network study normally follows these stages:

1. Define the strategic horizon, service promises, products, scenarios and decision rights.
2. Build and audit a common dataset with consistent units and geographic definitions.
3. Generate feasible candidate sites using operational, legal, environmental and labour filters.
4. Formulate the base optimization model and validate it with domain experts.
5. Run alternative scenarios, sensitivity tests and disruption simulations.
6. Compare cost, service, risk, emissions, social effects and implementation complexity.
7. Conduct site due diligence, stakeholder consultation and financial appraisal.
8. Approve a staged implementation with monitoring triggers and contingency plans.

Limitations should be stated clearly. Linear models may misrepresent volume discounts, congestion and economies of scale. Annual averages can conceal seasonal peaks. Deterministic demand can produce fragile solutions. Travel distance may not represent actual travel time. Candidate-site lists may exclude innovative alternatives. Monetary valuation of social or environmental effects can be contentious. Finally, optimization can identify an attractive structure but cannot guarantee successful execution, carrier availability or organisational acceptance.

Chapter Summary

Logistics network design integrates facility location, opening, capacity, allocation, flow, inventory, transport and resilience decisions. Centralized networks benefit from pooling and simplified control, whereas decentralized networks may provide faster and more geographically resilient service; hybrid and postponement designs often combine their strengths. Capacitated facility-location and multi-echelon flow models provide rigorous representations of strategic choices, but their results depend on valid data, units and assumptions. Scenario and sensitivity analysis reveal break-even points and vulnerability to uncertain demand, costs and disruptions. Sound decisions evaluate total cost together with service, risk, sustainability, ethics and implementation feasibility. GIS, optimization, simulation and AI can strengthen analysis, but governance, explainability and expert judgement remain essential.

Key Terms

Facility location: Selection of geographic sites for logistics or production facilities.

Allocation: Assignment of customers, markets or demand zones to facilities.

Capacity: Maximum feasible throughput, storage or production capability.

Centralization: Concentration of activities in relatively few facilities.

Decentralization: Distribution of activities across multiple geographic facilities.

Hybrid network: Network combining centralized and regional or local structures.

Postponement: Delaying product differentiation until demand or destination is known.

Inventory pooling: Combining demand streams to reduce the effect of independent variability.

Mixed-integer linear programming: Optimization with linear relationships and integer or binary decisions.

Flow conservation: Accounting identity requiring inbound, outbound, supply and inventory flows to balance.

Candidate site: Location eligible for evaluation in a network model.

Service constraint: Model requirement specifying an acceptable delivery or fulfilment performance.

Scenario analysis: Evaluation of a model under coherent alternative futures.

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

Resilience: Capacity to withstand, respond to and recover from disruption.

Post-disruption recovery time: Time required to restore an acceptable level of service.

Carbon intensity: Emissions associated with a unit of logistics activity or flow.

Digital twin: Data-connected representation used to analyse or monitor a physical system.

Shadow price: Implied value of relaxing a binding constraint by one unit.

Review Questions

Short Answer Questions

1. Why is logistics network design considered a strategic decision?
2. State two advantages and two disadvantages of a centralized network.
3. What is the purpose of a capacity constraint in a facility-location model?
4. Distinguish scenario analysis from sensitivity analysis.
5. Why must demand and capacity use compatible units in an optimization model?

Long Answer Questions

1. Compare centralized, decentralized and hybrid logistics networks, including inventory, service, cost and resilience implications.
2. Explain the capacitated facility-location model and interpret each major constraint.
3. Discuss how sustainability and social considerations can be incorporated into network design.
4. Explain why deterministic optimization alone may produce fragile strategic recommendations.
5. Evaluate the role and limitations of GIS, simulation, optimization and AI in network planning.

Analytical / Application Questions

1. A firm has three candidate facilities with different fixed costs, capacities and customer-lane costs. Design a mathematical formulation that determines which facilities open and how customer demand is allocated. State the assumptions required.
2. A centralized network has lower expected cost but fails during a four-week facility outage. Propose a scenario and sensitivity analysis that compares it with a two-site design, specifying relevant metrics and decision thresholds.
3. A company must reduce emissions while preserving a 98% on-time service target. Recommend how to modify its network model and explain the likely cost, inventory, transport and resilience trade-offs.

Multiple Choice Questions with Answers

Question 1

Which decision is a facility-location model primarily designed to support?

A. Selecting employee training methods

B. Choosing facility sites and associated allocations

C. Setting product colour standards

D. Auditing financial statements

Correct Answer: B. Choosing facility sites and associated allocations

Question 2

In the capacitated facility-location formulation, what does yⱼ = 1 usually indicate?

A. Customer j is out of stock

B. Facility j is closed

C. Facility j is opened

D. Shipment j is returned

Correct Answer: C. Facility j is opened

Question 3

What is a principal benefit of inventory pooling?

A. It always eliminates transport cost

B. It can reduce the effect of independent demand variability

C. It guarantees zero stockouts

D. It removes the need for forecasting

Correct Answer: B. It can reduce the effect of independent demand variability

Question 4

Which statement best describes scenario analysis?

A. Changing one coefficient while holding every strategic condition fixed

B. Evaluating coherent alternative future conditions

C. Replacing all quantitative analysis with judgement

D. Calculating only the average historical demand

Correct Answer: B. Evaluating coherent alternative future conditions

Question 5

If a facility has capacity 18,000 units and is assigned 20,000 units, which condition is violated?

A. Demand balance

B. Facility capacity

C. Binary integrality only

D. Carbon accounting

Correct Answer: B. Facility capacity

Question 6

A hybrid logistics network typically combines:

A. Only air transport and no warehousing

B. Centralized and regional or local fulfilment structures

C. Only one supplier and one customer

D. Inventory elimination and unlimited capacity

Correct Answer: B. Centralized and regional or local fulfilment structures

Question 7

For independent demand streams, the standard deviation of combined demand is calculated using:

A. The arithmetic sum of all standard deviations

B. The square root of the sum of squared standard deviations

C. The product of all means

D. The largest demand divided by capacity

Correct Answer: B. The square root of the sum of squared standard deviations

Question 8

Which measure is most directly associated with resilience?

A. Single-site dependence

B. Number of product colours

C. Office floor area

D. Invoice font size

Correct Answer: A. Single-site dependence

Question 9

If transport emissions per unit flow are eᵢⱼ and flow is xᵢⱼ, total transport emissions are represented by:

A. ∑ᵢ∑ⱼ eᵢⱼxᵢⱼ

B. ∑ᵢ∑ⱼ eᵢⱼ + xᵢⱼ

C. ∑ᵢ∑ⱼ eᵢⱼ/xᵢⱼ for all cases

D. ∑ᵢ∑ⱼ xᵢⱼ − eᵢⱼ

Correct Answer: A. ∑ᵢ∑ⱼ eᵢⱼxᵢⱼ

Question 10

Why should AI recommendations for network design be subject to governance and human review?

A. AI cannot process any numerical data

B. Black-box models may encode bias or fail under rare disruptions

C. Optimization models never use data

D. Human judgement is legally prohibited

Correct Answer: B. Black-box models may encode bias or fail under rare disruptions