Distribution Network Design

Introduction

Distribution network design determines how products move from suppliers and manufacturing sites to customers through facilities such as distribution centres, fulfilment centres, cross-docks, wholesalers and retail outlets. It is a strategic decision because facility locations and capacities normally remain in place for many years, while demand, transport costs, service expectations, technology and regulations continue to change.

A well-designed network balances five objectives: low total landed cost, acceptable customer service, sufficient capacity, resilience and strategic flexibility. The lowest transport-cost network is not necessarily the best network. A small number of large facilities may produce economies of scale but increase delivery distances and disruption exposure. Conversely, many local facilities may improve responsiveness while increasing inventory, labour, rent and coordination costs.

Distribution network design therefore requires integrated analysis of facility numbers, locations, capacities, market coverage and product flows.

Learning Objectives

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

1. Explain the strategic role and major configurations of distribution networks.
2. Evaluate facility numbers, locations and capacities using total landed cost.
3. Apply centre-of-gravity and load–distance methods appropriately.
4. Develop and compare network scenarios using cost-to-serve and service criteria.
5. Interpret trade-offs involving inventory, transport, responsiveness, resilience and sustainability.
6. Recommend a network design while recognising modelling assumptions and governance risks.

1. Foundations of Distribution Network Design

A distribution network consists of nodes and links. Nodes include suppliers, plants, ports, warehouses, consolidation centres, stores and customers. Links represent transportation movements, information flows and, increasingly, reverse logistics. The design question is not simply “Where should a warehouse be located?” It is “Which facilities should exist, where should they be located, what should each handle, and how should demand be assigned to them?”

Strategic decisions

The principal decisions are:

• Facility number: one central facility, several regional facilities, or a hybrid structure.
• Facility location: geographic position, access to markets, suppliers, ports, labour and infrastructure.
• Facility role: storage, fulfilment, cross-docking, postponement, returns processing or emergency stockholding.
• Capacity: storage positions, throughput, dock doors, labour and peak-period capability.
• Market allocation: which facility serves each customer zone.
• Product allocation: which products are stocked or processed at each facility.
• Transportation architecture: direct shipment, milk runs, hub-and-spoke, parcel networks or multimodal flows.
• Time and risk policy: service promises, safety stock, redundancy and recovery capability.

Common configurations

A centralised network uses relatively few facilities. It often reduces inventory duplication and benefits from scale, but may create long delivery distances and a single point of failure. A decentralised network places inventory closer to customers. It can shorten lead times and reduce final-mile distance, but increases fixed cost, duplicated stock and management complexity. A hybrid network combines national or regional stock with local fulfilment or cross-docking.

A hub-and-spoke network routes flows through consolidation hubs. It can improve vehicle utilisation, although additional handling and hub congestion may offset the benefit. In direct shipping, suppliers or plants ship directly to customers; this is attractive for large, predictable orders but less efficient for fragmented demand. Cross-docking transfers inbound goods to outbound vehicles with little or no storage, requiring accurate forecasts, synchronised schedules and reliable information.

2. The Total-Landed-Cost Perspective

Facility decisions should be evaluated using the total economic effect rather than one cost category. A general annual network-cost model is:

where:

• TLC = total landed cost per planning period, in currency units;
• F = facility fixed cost, including rent, depreciation and property charges;
• H = handling and warehousing operating cost;
• I = inventory carrying cost;
• T_in = inbound transportation cost;
• T_out = outbound transportation cost;
• T_x = transfer, cross-docking or interfacility transportation cost;
• O = order-processing and information-system cost;
• R = reverse-logistics cost;
• E = expected disruption, shortage or expediting cost;
• C_s = sustainability or compliance cost, where material.

The model is additive only when all components are measured over the same period and scope. Taxes, duties and product purchase cost may be included when comparing sourcing alternatives; if they are identical across alternatives, including them does not affect the ranking. “Landed cost” should therefore be defined explicitly before analysis.

Inventory carrying cost can be approximated by:

where C_I is annual carrying cost, I_avg is average inventory in physical units, v is inventory value per unit, and r is the annual carrying-rate percentage expressed as a decimal. The rate may include capital cost, insurance, obsolescence, shrinkage and taxation. The formula assumes a reasonably stable average inventory and a carrying rate proportional to value.

For a simple cycle-stock approximation under a constant demand rate, average cycle inventory is approximately half the replenishment quantity:

This excludes safety stock, pipeline inventory, seasonal stock and display stock. It must not be used as the complete inventory estimate when demand or lead time is highly variable.

A facility may reduce outbound transport while increasing inventory. If demand is divided among n identical facilities, each facility may require its own cycle stock and safety stock. Risk-pooling benefits often favour centralisation, whereas service distance and local responsiveness favour decentralisation. The precise inventory effect depends on demand correlation, replenishment policy, lead time and service targets; it should not be inferred from facility count alone.

Capacity and fixed-cost implications

A candidate facility must have enough capacity for normal and peak demand. A basic utilisation ratio is:

where U is peak utilisation, D_peak is peak workload, and K_available is practical capacity in the same workload units. Practical capacity is less than theoretical capacity because of maintenance, absenteeism, congestion, slotting constraints and operating calendars. A utilisation ratio above 1 indicates infeasibility; a ratio very close to 1 may be operationally fragile.

Capacity must be measured consistently. Pallet positions measure storage, while cases per hour, orders per day and dock movements measure throughput. A warehouse can have adequate pallet capacity but insufficient picking or dock capacity.

3. Locating Facilities: Quantitative Methods

3.1 Centre-of-gravity method

The centre-of-gravity method gives an initial geographic location based on weighted demand or shipment volume. For customer zones with coordinates (x_i, y_i) and weights w_i, the weighted coordinates are:

The weights may represent annual tonnes, pallets, cases, deliveries or revenue, depending on the decision purpose. Coordinates must use a meaningful distance system, preferably projected coordinates or a locally valid approximation. Latitude and longitude should not be averaged over large geographic areas without appropriate geospatial treatment.

The method assumes that transportation cost is broadly related to distance and that a continuous location is possible. It ignores land price, roads, legal restrictions, labour, taxes, facility capacity, congestion, service-time limits and nonlinear freight rates. Consequently, the result is a screening point, not an automatic site recommendation.

Worked example. Three customer zones have the following annual shipment weights and coordinates:

Total weight is 200 tonnes. Therefore:

x* = (100×10 + 60×40 + 40×70) ÷ 200 = 30 km

y* = (100×20 + 60×50 + 40×30) ÷ 200 = 32 km

The initial centre is (30, 32). A site near this point should then be tested against actual roads, land availability, labour and service constraints.

3.2 Load–distance analysis

The load–distance method compares candidate sites using weighted distance:

where LD_j is the load–distance score for candidate site j, w_i is demand or shipment weight for zone i, and d_ij is the distance from site j to zone i. Lower scores are preferred when transport cost per unit distance is comparable.

If rates differ by lane, a cost-weighted version is more appropriate:

where q_i is shipment quantity, d_ij is route distance, and r_ij is transportation cost per unit of quantity-distance. This formula assumes a linear rate and does not capture minimum charges, vehicle-capacity steps or route consolidation unless those are modelled separately.

Road distance or travel time should replace straight-line distance when service performance depends on the transport network. A site with a lower geometric distance may have higher actual travel time because of mountains, congestion or poor road access.

3.3 Facility-location optimisation

For more advanced analysis, a capacitated facility-location model can be used. Let y_j equal 1 if candidate facility j is opened and 0 otherwise; let x_ij be quantity shipped from facility j to customer zone i; F_j be fixed cost; c_ij be variable distribution cost; D_i be demand; and K_j be facility capacity.

subject to:

The first constraint serves all demand. The second prevents shipments from a closed facility and enforces capacity. The model is a simplified mixed-integer formulation. Practical models may add maximum delivery times, multiple products, inbound flows, inventory, emissions, minimum throughput, labour availability and disruption scenarios. A mathematically optimal result is not necessarily implementable if its data or assumptions are weak.

4. Market Coverage, Service and Product Flows

Network design must translate customer expectations into measurable constraints. Relevant service measures include order-cycle time, delivery reliability, fill rate, product availability, cut-off time, returns response and geographic coverage.

A coverage rule may require every customer zone to be within a maximum travel time of an eligible facility. If a_ij equals 1 when facility j can serve zone i within the promised limit, then a simplified coverage constraint is:

This means each zone must have at least one open facility capable of serving it. Multiple coverage can be required for resilience by replacing the right-hand side 1 with 2 or another target. The rule is binary and does not represent the probability of on-time delivery; that requires a more detailed reliability model.

Product flows should be designed by product characteristics, not only by geography. Fast-moving products may be stocked regionally, slow-moving products centrally, and customised products postponed until an order is received. Temperature-sensitive, hazardous, regulated or high-value goods require specialised facilities and controls. Returns, repair, recycling and disposal flows should be included because reverse logistics can materially change facility requirements and transport cost.

A useful flow-design sequence is:

1. Segment customers by service need, order profile and geography.
2. Segment products by demand velocity, value, cube, risk and handling requirements.
3. Estimate demand by zone and time period, including seasonality.
4. Generate feasible facility and transportation alternatives.
5. Assign product flows subject to capacity and service constraints.
6. Calculate total landed cost, risk and environmental effects.
7. Test peak, growth and disruption scenarios.
8. Select a robust design and define implementation milestones.

5. Scenario Analysis and Cost-to-Serve

A scenario is a coherent combination of assumptions, such as “one national centre,” “three regional centres,” or “two regional centres plus outsourced fulfilment.” Scenario analysis is preferable to a single forecast because strategic uncertainty cannot be eliminated by more decimal places.

Each scenario should specify facility count, location, capacity, customer allocation, transport modes, inventory policy, labour model, demand assumptions, service policy and implementation timing. The analyst should compare both expected performance and sensitivity to changes.

Cost-to-serve

Cost-to-serve attributes supply-chain cost to a customer, channel, product or order. A simplified customer cost model is:

where each component is the relevant cost for customer or segment k. Profitability after service cost can be expressed as:

where R_k is revenue and C_product,k is product or purchase cost. Costs should be assigned using operational drivers, such as order lines, cartons, kilometres, delivery stops, handling minutes and return units, rather than broad averages wherever practical.

Cost-to-serve frequently reveals that a high-revenue customer is not necessarily highly profitable. Small, frequent, urgent or geographically remote orders may consume disproportionate resources. The result should support differentiated service policies, not unfair discrimination or arbitrary withdrawal of service.

Scenario comparison

A decision matrix may include the following:

Weighted scoring can be used when non-cost factors matter, but weights must be documented and sensitivity-tested. It should complement, not conceal, a cost model. If scenario rankings reverse when a reasonable weight changes slightly, the decision is not robust.

6. Worked Network-Design Scenario

The following is an illustrative case. A packaged-food distributor serves four market zones. Management is comparing a single central distribution centre (Scenario A) with two regional centres (Scenario B). The annual figures are simplified and expressed in thousands of currency units.

The totals are obtained by adding the six components. Scenario B reduces outbound transport by 680 and expected shortage cost by 110, but increases facility, handling, inbound and inventory costs by 570 in total. Its net disadvantage is 40 thousand per year under the stated assumptions.

However, Scenario B may still be preferable if its service improvement creates benefits greater than 40 thousand, or if fuel prices, demand growth or disruption risk changes. Suppose management estimates an additional annual benefit of 100 thousand from improved availability and customer retention. The adjusted economic comparison becomes:

• Scenario A net cost: 5,830 thousand.
• Scenario B net cost after service benefit: 5,870 − 100 = 5,770 thousand.

Under that assumption, Scenario B is preferred by 60 thousand. This conclusion is conditional, not universal. The service benefit must be supported by evidence, and the model should test demand, fuel, labour, inventory and disruption assumptions.

The capacity test is also essential. If each regional facility has practical peak capacity of 1,200 pallet-equivalent units and its assigned peak workload is 1,080 units, utilisation is:

U = 1,080 ÷ 1,200 = 0.90, or 90%

Ninety percent may be feasible but leaves little allowance for variability, maintenance or promotional peaks. Management may need temporary overflow, flexible labour or additional capacity. A design that is cheaper on average can fail operationally at peak periods.

7. Resilience, Sustainability and Governance

Traditional models often optimise expected cost under normal conditions. Modern design must also consider disruption. Sources of disruption include floods, earthquakes, strikes, cyberattacks, supplier failure, border restrictions, pandemics and energy shortages. Resilience mechanisms include dual sourcing, alternative transport lanes, geographically separated facilities, emergency inventory, reciprocal fulfilment agreements and documented recovery procedures.

Redundancy has a cost, and complete protection is impossible. Scenario analysis should estimate consequences under credible disruptions. Expected disruption cost may be represented conceptually as:

where p_s is the probability of disruption scenario s and C_s is its resulting cost. Probabilities must be evidence-based or explicitly treated as subjective estimates. This expected-value approach can hide severe low-probability losses, so worst-case and service-recovery measures should also be reported.

Sustainability analysis should include transport emissions, building energy, refrigeration, packaging, empty running and returns. A basic activity-based emissions estimate is:

where A_l is activity on lane or operation l and e_l is the corresponding emissions factor. Activity might be tonne-kilometres, vehicle-kilometres, kilowatt-hours or pallet-handling events. Emissions factors and system boundaries must be stated; otherwise comparisons may be misleading.

Governance issues include data ownership, model transparency, privacy, regulatory compliance, labour standards and equitable access. Customer-location data may be commercially sensitive or personally identifiable. Automated allocation systems should be auditable and monitored for systematic service disadvantages. Facility selection must comply with zoning, environmental permits, occupational safety, food or pharmaceutical regulations, customs requirements and transport law.

8. Technology and Contemporary Practice

Geographic information systems support map-based analysis of demand, travel time, congestion, land and risk. Warehouse-management and transport-management systems provide operational data for estimating dwell time, picking productivity, route performance and capacity. Digital twins can test facility and flow configurations before physical implementation.

Artificial intelligence and machine learning can improve demand forecasting, route prediction, anomaly detection and dynamic capacity planning. They do not remove the need for managerial judgement. Forecast bias, unstable data, opaque recommendations and feedback loops can produce poor network decisions. Models should be validated out of sample, monitored after deployment and subject to human approval for high-impact decisions.

Contemporary networks increasingly use micro-fulfilment, dark stores, urban consolidation centres, parcel lockers, robotics, autonomous systems and postponement. These technologies may reduce response time, but their economics depend on order density, labour cost, product range, building constraints and maintenance capability. Technology should be selected after the service and flow problem is defined, not as a substitute for sound network analysis.

9. Implementation and Best Practices

A network study should begin with a clear decision statement and a common data dictionary. Demand, product dimensions, customer coordinates, order history, transport rates, facility costs, lead times and service targets must be reconciled. Data quality problems should be recorded rather than hidden.

Recommended practice includes:

• use a multi-year planning horizon and include growth;
• model average and peak periods separately;
• distinguish fixed, variable, step and one-time transition costs;
• use actual travel time and lane rates where available;
• include inventory, returns, taxes and handling in a defined scope;
• test sensitivity to demand, fuel, labour, exchange rates and service assumptions;
• evaluate resilience and emissions alongside cost;
• validate outputs with warehouse, transport, sales, finance and compliance experts;
• compare the model with a practical “do nothing” baseline;
• phase implementation with pilots, contingency plans and post-implementation review.

A good recommendation explains not only which scenario wins, but why it wins, what assumptions drive the result, what could make it fail and how management should monitor it.

Challenges and Limitations

Network models are limited by forecast error, incomplete cost data, aggregation, uncertain probabilities and simplifying assumptions. Average distance can conceal congestion and route structure. Annual costs can conceal hourly labour constraints. Facility capacity may be treated as a single number even though storage, picking, docking and refrigeration have different bottlenecks. Cost-to-serve may become misleading if shared costs are allocated arbitrarily.

There are also organisational limitations. A network that is optimal for logistics may conflict with sales territories, tax structures, supplier contracts or local employment commitments. Closing a facility can create social impacts that are not captured by financial models. Conversely, preserving a site for political reasons should be transparent rather than presented as a purely economic decision.

The appropriate response is not to reject quantitative modelling, but to combine models with sensitivity analysis, stakeholder review, pilot testing and explicit judgement.

Chapter Summary

Distribution network design integrates facility number, location, capacity, market coverage and product flows. Centralisation can reduce fixed, inventory and handling duplication, while decentralisation can improve responsiveness, coverage and recovery capability. The correct choice depends on total landed cost and service requirements rather than transport cost alone.

The centre-of-gravity method provides an initial weighted location; load–distance analysis compares candidate sites; and capacitated facility-location models evaluate fixed costs, flows and capacity constraints. Cost-to-serve connects operational consumption to customer or product profitability. Scenario analysis enables comparison under alternative demand, cost, service and disruption assumptions.

Sound design requires consistent units, explicit assumptions, peak-capacity testing, realistic travel data, inventory analysis, sustainability measurement and governance controls. Technology can improve analysis and execution, but it cannot compensate for poor data or unclear objectives. The strongest network recommendation is economically defensible, operationally feasible, resilient, sustainable and transparent about uncertainty.

Key Terms

Distribution network: The facilities, transportation links, information flows and processes used to move products to customers.

Centralised network: A network using relatively few facilities serving broad geographic markets.

Decentralised network: A network using multiple facilities positioned close to customer markets.

Hybrid network: A combination of central, regional and local fulfilment structures.

Total landed cost: The complete relevant cost of sourcing, moving, storing, handling and delivering products.

Centre of gravity: A demand-weighted geographic point used for preliminary facility-location analysis.

Load–distance method: A method comparing candidate sites using weighted shipment volume and distance.

Facility-location model: An optimisation model selecting facilities and allocating flows subject to constraints.

Capacity: The practical quantity of storage, throughput or labour that a facility can support.

Peak utilisation: Peak workload divided by practical available capacity.

Market coverage: The ability of facilities to serve defined customer zones within specified conditions.

Cost-to-serve: The operational cost attributable to serving a customer, product, channel or order.

Cross-docking: Transfer of goods from inbound to outbound vehicles with limited storage.

Risk pooling: Reduction in aggregate variability achieved by combining demand across locations.

Reverse logistics: Movement and processing of returns, repairs, recycling and disposal.

Scenario analysis: Comparison of coherent alternative assumptions and network configurations.

Resilience: The ability to withstand, respond to and recover from disruption.

Postponement: Delaying final product configuration or allocation until demand is known.

Practical capacity: Usable capacity after allowing for operational constraints and variability.

Review Questions

Short Answer Questions

1. Why is distribution network design a strategic rather than merely operational decision?
2. Distinguish a centralised, decentralised and hybrid distribution network.
3. What does the centre-of-gravity method calculate, and why is it only a preliminary tool?
4. Define cost-to-serve and identify two operational cost drivers.
5. Why should peak capacity be evaluated separately from average capacity?

Long Answer Questions

1. Explain the components of total landed cost and discuss how facility count affects each component.
2. Describe the assumptions, procedure and limitations of load–distance analysis.
3. Explain how inventory, transport and customer service create trade-offs in network design.
4. Discuss how resilience and sustainability should be incorporated into facility-location decisions.
5. Compare optimisation models with scenario analysis as approaches to distribution network design.

Analytical / Application Questions

1. A company is considering one national distribution centre versus three regional centres. Develop a structured evaluation framework covering cost, service, capacity, inventory, risk and emissions. State which data would be required.
2. Using the centre-of-gravity formula, calculate the weighted location for four customer zones of your choice, then explain why the resulting point may not be an appropriate final site.
3. A proposed facility has practical peak capacity of 18,000 cases per day and forecast peak workload of 16,800 cases per day. Evaluate its utilisation, identify operational risks and recommend at least two mitigation actions.

Multiple Choice Questions with Answers

Question 1

Which cost is most directly represented by the facility fixed-cost component of total landed cost?

A. Annual building rent or depreciation

B. Fuel consumed by delivery vehicles

C. Inventory obsolescence only

D. Customer order revenue

Correct Answer: A. Annual building rent or depreciation

Question 2

The centre-of-gravity method weights coordinates primarily by which factor?

A. Facility age

B. Customer demand or shipment volume

C. Number of warehouse employees

D. Product selling price only

Correct Answer: B. Customer demand or shipment volume

Question 3

If peak workload is 9,000 units and practical capacity is 12,000 units, peak utilisation is:

A. 25%

B. 50%

C. 75%

D. 133%

Correct Answer: C. 75%

Question 4

In a capacitated facility-location model, the constraint ∑xᵢⱼ = Dᵢ generally ensures that:

A. Every facility is opened

B. Each customer zone’s demand is served

C. All facilities have equal capacity

D. Transportation cost is zero

Correct Answer: B. Each customer zone’s demand is served

Question 5

Which situation most strongly favours a centralised network?

A. Very urgent local deliveries with dispersed demand

B. High demand pooling benefits and relatively stable service requirements

C. A requirement for two independent local recovery sites

D. Severe congestion around the central location

Correct Answer: B. High demand pooling benefits and relatively stable service requirements

Question 6

A load–distance score is generally preferred when it is:

A. Higher, because it indicates more coverage

B. Lower, when weights and distances are comparable

C. Equal to annual revenue

D. Based only on facility rent

Correct Answer: B. Lower, when weights and distances are comparable

Question 7

Which item is normally a limitation of a simple centre-of-gravity calculation?

A. It requires no demand data

B. It automatically includes legal and labour constraints

C. It may ignore road networks and facility capacity

D. It can only be used for reverse logistics

Correct Answer: C. It may ignore road networks and facility capacity

Question 8

Cost-to-serve analysis is primarily intended to identify:

A. The cost consumed by serving different customers or segments

B. The legal ownership of a warehouse

C. The number of product colours

D. The accounting value of buildings only

Correct Answer: A. The cost consumed by serving different customers or segments

Question 9

If a scenario has total cost of 5,870 thousand and an estimated service benefit of 100 thousand, its adjusted net cost is:

A. 5,770 thousand

B. 5,870 thousand

C. 5,970 thousand

D. 6,070 thousand

Correct Answer: A. 5,770 thousand

Question 10

Which practice best supports responsible use of AI in network design?

A. Accepting recommendations without validation

B. Removing all human review

C. Validating data and monitoring model performance after deployment

D. Using only historical averages regardless of change

Correct Answer: C. Validating data and monitoring model performance after deployment