Logistics, Transportation and Network Design Analytics

Introduction

Logistics and network design determine how products, information and associated services move from suppliers to customers. In supply-chain management, these decisions are not merely operational. They shape cost, responsiveness, working capital, asset utilisation, sustainability and resilience for many years. A warehouse location, transport mode, delivery promise or consolidation policy can therefore create structural advantages—or persistent sources of inefficiency.

Network design analytics applies quantitative models, data and managerial judgement to decisions such as:

• where to locate plants, warehouses, cross-docks and fulfilment centres;
• which customers and suppliers should be served by each facility;
• which transport modes and carriers should be used;
• how shipments should be consolidated and routed;
• how last-mile operations should be designed; and
• how cost, service, capacity, carbon and disruption risk should be balanced.

The central managerial challenge is that the lowest transport cost is rarely the lowest total supply-chain cost. A geographically dispersed network may reduce delivery distance but increase inventory, facility and coordination costs. Conversely, a highly centralised network may achieve economies of scale but produce longer lead times and greater exposure to disruption.

Learning Objectives

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

1. distinguish major logistics-network decisions and their planning horizons;
2. calculate and interpret total landed cost and logistics performance measures;
3. formulate basic facility-location, allocation, transportation and routing models;
4. evaluate consolidation, mode selection and last-mile alternatives;
5. compare network options using service, capacity, carbon and resilience criteria; and
6. explain the governance, data and implementation requirements of analytics-led network design.

1. The Logistics Network as a System

A logistics network consists of nodes and links. Nodes include suppliers, factories, ports, terminals, warehouses, cross-docks, stores, parcel hubs and customers. Links represent road, rail, ocean, air, inland waterway, pipeline or digital coordination channels. Material flows may be forward, reverse or circular.

Three decision horizons are useful:

A good design recognises interactions among decisions. For example, adding a regional distribution centre may reduce outbound distance but duplicate safety stock. Changing from road to rail can lower carbon emissions but require more inventory because of longer or less reliable transit. The appropriate unit of analysis is therefore total system performance rather than an isolated freight rate.

Important terminology

A hub consolidates or transfers flows; a spoke connects a hub to an origin or destination. A cross-dock receives goods and transfers them rapidly to outbound vehicles with little or no storage. Milk runs collect from or deliver to multiple points on a planned route. Backhauling uses a vehicle's return trip for additional freight. Landed cost is the total cost of making a product available at the relevant destination. Last mile is the final movement to the customer, often the most expensive and operationally variable segment per unit delivered.

2. Measuring Logistics and Total Landed Cost

A practical landed-cost model should include all relevant cost categories, not only the supplier price and freight charge. For shipment or product class i, a general expression is:

where:

• P_i = purchase or production cost;
• F_i = international and domestic freight;
• D_i = duties, tariffs and customs charges;
• H_i = handling, warehousing and terminal charges;
• I_i = inventory-carrying cost caused by pipeline and cycle stock;
• C_i = insurance and compliance cost;
• R_i = expected cost of damage, loss, delay and returns; and
• E_i = relevant environmental or emissions cost.

All terms must be expressed in the same monetary unit and time basis. The formula is additive only when costs are defined consistently and double counting is avoided. For example, a carrier invoice should not be included in both freight and handling.

For a shipment with annual volume Q, unit transport cost can be written as F/Q. When inventory cost is calculated from average inventory, a common approximation is:

where h is the annual holding-rate fraction, V is unit value, and \bar{I} is average units held. This approximation assumes a stable holding rate and that average inventory adequately represents exposure. It should be supplemented by insurance, obsolescence and financing effects when material.

Worked example: comparing transport modes

A firm must move 10,000 units annually. Each unit has a value of $80 and weighs 4 kg. Mode A costs $0.90 per kg, has a 5-day transit time and requires 1,000 units of additional average pipeline inventory. Mode B costs $0.55 per kg, has a 14-day transit time and requires 2,500 units of pipeline inventory. The annual holding rate is 24%. Ignore duties and damage.

Mode A freight cost is:

10,000 × 4 × 0.90 = $36,000.

Its inventory cost is 1,000 × 80 × 0.24 = $19,200, giving relevant annual cost of $55,200.

Mode B freight cost is 10,000 × 4 × 0.55 = $22,000. Its inventory cost is 2,500 × 80 × 0.24 = $48,000, giving relevant annual cost of $70,000.

Although Mode B has the lower freight rate, Mode A is cheaper by $14,800 under the stated assumptions. This conclusion could change if capacity, carbon prices, reliability or cash constraints were included.

Useful service measures include on-time-in-full (OTIF), order-cycle time, delivery reliability, damage rate, fill rate, cost per shipment, cost per tonne-kilometre, vehicle utilisation and first-attempt delivery success. Measures should be defined precisely. OTIF, for instance, requires agreed definitions of “on time,” quantity tolerance and order completion.

3. Facility Location and Network Configuration

Facility-location analysis typically balances fixed facility cost, variable flow cost and service requirements. A simplified capacitated facility-location model uses binary variables y_j, equal to 1 if facility j is opened, and flow variables x_ij, representing quantity shipped from facility j to customer i.

subject to:

Here, F_j is the fixed annual cost of facility j, c_ij is the variable cost of serving customer i from facility j, d_i is customer demand and K_j is facility capacity. The model assumes known demand, linear variable cost, divisible flow and no omitted service constraint. It is a baseline model, not a complete representation of reality.

The model can be extended with:

• maximum delivery distance or time;
• multiple products and capacity dimensions;
• supplier, plant and warehouse tiers;
• minimum throughput or contractual commitments;
• taxes, duties and transfer prices;
• carbon emissions and carbon budgets;
• disruption scenarios and backup capacity; and
• multi-period investment and expansion decisions.

Centre-of-gravity analysis

For an initial geographic screen, the weighted centre of gravity is often calculated as:

where (x_i,y_i) are customer coordinates and w_i is a weight such as demand or shipment volume. This method assumes a continuous plane and usually ignores road topology, facility cost, labour, taxes, congestion and service constraints. It identifies a starting point, not a final site decision.

A more robust process is: define the service promise; cleanse demand and lane data; generate candidate sites; estimate fixed, variable and risk costs; optimise allocation; test scenarios; validate with local experts; and conduct a total-cost and implementation review.

4. Transportation Modes, Consolidation and Capacity

Mode selection is a trade-off among cost, speed, reliability, flexibility, capacity, damage exposure and emissions. Typical characteristics are shown below.

Consolidation combines shipments to obtain lower rates and better asset utilisation. It may be temporal, spatial, supplier-based, customer-based or mode-based. Consolidation is attractive when shipment fixed costs are significant, demand is sufficiently predictable and service tolerates waiting. It is harmful when delay penalties, stockout risk or product obsolescence exceed freight savings.

Vehicle utilisation should be assessed in both weight and cube. A truck can be “full” by volume but underloaded by weight, or the reverse. A basic load factor is:

Capacity must be measured in compatible units—kilograms, cubic metres, pallets or loading metres. A single average utilisation figure can conceal severe lane imbalance and empty backhauls.

5. Routing and Last-Mile Analytics

The vehicle-routing problem (VRP) determines routes for vehicles serving customers under constraints such as vehicle capacity, delivery windows, driver hours and maximum route duration. The travelling-salesperson problem is a special case involving one vehicle, one visit per customer and a closed tour.

Exact optimisation can provide high-quality solutions for small instances, but large operational problems often require heuristics or metaheuristics. Common approaches include savings algorithms, local search, tabu search, genetic algorithms, column generation and mixed-integer programming. In practice, the best solution is often a hybrid: optimisation generates a plan, while dispatchers apply operational knowledge and real-time adjustments.

A simplified routing objective is:

where x_ijk equals 1 when vehicle k travels from node i to node j, and c_ij is the relevant travel cost. A complete VRP also requires constraints ensuring each customer is visited appropriately, vehicle capacity is respected, routes begin and end at permitted depots, and time windows are satisfied.

Last-mile design choices include own fleet versus third-party delivery, parcel lockers, pickup points, micro-fulfilment centres, crowd-shipping, delivery windows and dynamic routing. Important performance measures include cost per stop, stops per route-hour, miles per delivery, failed-delivery rate, customer wait time and emissions per parcel.

Urban delivery analytics must account for curb access, congestion, parking, delivery density and regulatory restrictions. Increasing delivery density usually reduces cost per stop, while narrow time windows and failed attempts increase it. A promised service level should therefore be designed jointly with the delivery network, not added after routing decisions are made.

6. Carbon, Resilience and Multi-Criteria Decisions

Transport emissions can be estimated using activity data and an emission factor:

For freight, activity may be tonne-kilometres, vehicle-kilometres or litres of fuel. The emission factor must match the activity definition and boundary. If a shipment weighs 8 tonnes, travels 500 km and has an emission factor of 0.08 kg CO₂e per tonne-kilometre, estimated emissions are 8 × 500 × 0.08 = 320 kg CO₂e. This is an estimate; load factor, empty running, fuel type and allocation method can materially change the result.

Resilience concerns the ability to prepare for, absorb, recover from and adapt to disruption. Relevant design variables include supplier and facility diversification, alternative modes, safety stock, spare capacity, geographic dispersion, visibility and recovery time. Redundancy increases cost, so its value should be assessed against disruption probability and impact.

A multi-criteria model may minimise a weighted objective:

where C, S, E and R are normalised cost, service penalty, emissions and risk measures, and weights sum to 1. Because units differ, normalisation is essential. Weights represent preferences, not objective truth; sensitivity analysis should show whether the preferred network changes when weights vary.

A stronger approach uses constrained optimisation, such as minimising cost subject to a maximum emissions budget and a service-level requirement. This is often easier to explain to executives than arbitrary weights.

7. Applied Case: NorthStar Home Appliances

The following is an illustrative case.

Background

NorthStar sells 120,000 small appliances annually to retailers and direct consumers in three regions: North, Central and South. Products are imported through a coastal port. Management operates one national warehouse near the port and is considering a second inland fulfilment centre.

Problem

The national warehouse provides low fixed cost and efficient inbound handling, but average delivery time to the South is four days, compared with the promised two days. Expedited shipments have increased, and customers report failed last-mile deliveries. Management must decide whether to open an inland facility, how to allocate demand, and whether to use parcel lockers in dense urban areas.

Analysis

The project team first cleansed twelve months of order, weight, cube, postcode, carrier, promised-date and actual-delivery data. It found that 42% of southern orders could be served within two days from an inland location, while only 11% met the promise from the national warehouse. However, the inland option required an annual fixed cost of $1.8 million and duplicated safety stock.

The team modelled three alternatives:

The figures are scenario estimates, not accounting results. The locker option reduces failed delivery attempts and route kilometres but requires retailer contracts and customer adoption. The team also tested a port closure, a carrier capacity reduction and a 20% demand increase. Under the port-closure scenario, the second facility alone offered limited protection because both facilities depended on the same inbound gateway. A secondary port contract was therefore included in the resilience plan.

Approach and solution

Management selected the inland facility and a limited locker rollout in high-density cities. Southern demand was allocated to the inland site when service or transport savings justified the extra inventory. A dual-port option was retained as a contingency rather than used as the primary flow. A control tower monitored order backlog, carrier capacity, delivery reliability and exception events.

Outcome

The projected network increased annual logistics cost by $0.8 million relative to the base design but improved two-day service by 20 percentage points and reduced the emissions index by 16%. The investment case was accepted because the service improvement protected strategic retail contracts and reduced expedited freight. Actual performance was to be reviewed after six months using pre-agreed measures.

Learning points

1. Facility decisions must include inventory, service and risk effects.
2. A scenario can appear resilient while sharing a common port or carrier dependency.
3. Last-mile alternatives should be evaluated using failed attempts and customer convenience, not only distance.
4. Pilot implementation and post-implementation measurement are essential because model estimates are uncertain.

8. Technology, Governance and Implementation

Modern logistics analytics uses transportation-management systems, warehouse-management systems, telematics, GPS, electronic proof of delivery, digital freight marketplaces and control towers. Artificial intelligence can support demand forecasting, ETA prediction, anomaly detection, dynamic pricing and route adjustment. Its value depends on data quality and operational integration; an opaque model cannot compensate for inaccurate addresses, missing timestamps or poorly defined service rules.

Governance requirements include:

• common master data for locations, products, carriers and units of measure;
• documented cost and emissions boundaries;
• access controls and cybersecurity for shipment and customer data;
• audit trails for automated decisions;
• human review for high-impact exceptions;
• monitoring for biased or systematically poor service outcomes; and
• compliance with transport, customs, labour, safety, privacy and competition regulation.

Implementation should proceed through a baseline, pilot, measurement and scale cycle. Before changing a network, managers should establish the current cost-to-serve, service distribution, carbon estimate and disruption exposure. After implementation, results should be compared with the baseline while controlling for demand, seasonality and fuel-price changes.

9. Challenges and Limitations

Network models can create false precision. Demand forecasts are uncertain, rates change, facilities have indivisibilities, and service quality is difficult to represent in a single constraint. Average distance may ignore congestion; nominal capacity may not be available when needed; and historical data may reflect a network that itself created the observed demand pattern.

Other limitations include:

• optimisation bias toward measurable costs;
• underestimation of change-management and labour constraints;
• double counting of inventory or handling costs;
• inappropriate use of centre-of-gravity results as final site recommendations;
• failure to model reverse logistics and returns; and
• neglect of extreme but plausible disruption scenarios.

Robust practice combines optimisation with sensitivity analysis, stress testing, expert validation and transparent assumptions. A model should show not only the preferred answer but also the conditions under which that answer changes.

Best-Practice Checklist

1. Define the decision, time horizon and service promise.
2. Use clean, granular demand and lane data.
3. Calculate total landed cost rather than freight cost alone.
4. Separate fixed, variable, inventory, risk and externality costs.
5. Represent capacity, time windows and operational feasibility.
6. Test demand, rate, fuel, disruption and carbon scenarios.
7. Compare alternatives using service and resilience as well as cost.
8. Validate assumptions with carriers, facility teams and customers.
9. Pilot major changes and establish post-implementation controls.
10. Document model governance, data ownership and decision rights.

Chapter Summary

Logistics, transportation and network design analytics support strategic, tactical and operational supply-chain decisions. Facility-location models balance fixed facility costs, flow costs, capacity and service requirements. Transportation analysis must consider mode characteristics, consolidation, utilisation, reliability and inventory consequences. Routing models address vehicle movement under capacity and time-window constraints, while last-mile design must include delivery density, failed attempts and customer convenience.

Total landed cost integrates purchase, freight, duties, handling, inventory, insurance, risk and relevant environmental costs. Carbon analysis uses activity data and appropriate emission factors. Resilience requires consideration of redundancy, alternative capacity, recovery time and common dependencies. Because no model captures reality perfectly, scenario analysis, sensitivity testing, governance and implementation measurement are essential. The best network is not necessarily the cheapest or fastest in isolation; it is the configuration that best supports the organisation’s overall cost, service, sustainability and risk objectives.

Key Terms

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

Network design: Configuration of nodes, links, flows, capacities and policies in a supply chain.

Total landed cost: Complete cost of making a product available at a destination.

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

Consolidation: Combining shipments to improve utilisation or reduce transport cost.

Backhauling: Carrying freight on a vehicle’s return journey.

Intermodal transport: Use of two or more transport modes in one movement.

Load factor: Actual utilised capacity divided by available capacity.

Vehicle-routing problem: Optimisation of vehicle routes subject to operational constraints.

Last mile: Final movement from a logistics node to the customer.

OTIF: Delivery that is both on time and in the correct quantity.

Centre of gravity: Demand-weighted geographic estimate for an initial facility screen.

Capacity constraint: Restriction on the quantity, weight, volume or time a resource can handle.

Pipeline inventory: Inventory in transit between supply-chain stages.

Carbon intensity: Emissions associated with a defined logistics activity or output.

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

Control tower: Coordinated platform for visibility, exception management and decisions.

Service-level constraint: Requirement limiting delivery time, reliability or availability performance.

Review Questions

Short Answer Questions

1. What is the difference between strategic, tactical and operational logistics decisions?
2. Why can a lower freight rate produce a higher total landed cost?
3. Define consolidation and identify one condition under which it is beneficial.
4. What does a vehicle load factor measure?
5. Why is a centre-of-gravity calculation only an initial location-screening tool?

Long Answer Questions

1. Explain the structure of a capacitated facility-location model and discuss its assumptions.
2. Compare road, rail, ocean, air and intermodal transport using cost, speed, flexibility and sustainability criteria.
3. Discuss how last-mile design affects cost, service quality and customer experience.
4. Explain how carbon and resilience can be incorporated into network-design decisions.
5. Evaluate the role of data governance and human judgement in AI-supported logistics analytics.

Analytical / Application Questions

1. A company must choose between a low-freight-cost mode with long transit time and a high-freight-cost mode with short transit time. Construct a total-cost comparison including pipeline inventory and state the assumptions required.
2. Design a scenario-analysis framework for deciding whether to open a regional distribution centre. Identify demand, capacity, service, cost, carbon and disruption scenarios.
3. A delivery network has high failed-delivery rates despite short average distances. Diagnose possible causes and recommend a set of operational and analytical interventions.

Multiple Choice Questions with Answers

Question 1

Which cost is most likely to be omitted when a manager evaluates only the freight invoice?

A. Purchase price

B. Pipeline inventory carrying cost

C. Customer demand

D. Facility location coordinates

Correct Answer: B. Pipeline inventory carrying cost

Question 2

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

A. The quantity shipped on a lane

B. Whether a facility is opened

C. The weight of a product

D. The delivery time for a customer

Correct Answer: B. Whether a facility is opened

Question 3

A shipment weighs 8 tonnes, travels 500 km, and has an emission factor of 0.08 kg CO₂e per tonne-kilometre. What are its estimated emissions?

A. 32 kg CO₂e

B. 320 kg CO₂e

C. 3,200 kg CO₂e

D. 40,000 kg CO₂e

Correct Answer: B. 320 kg CO₂e

Question 4

Which condition generally supports freight consolidation?

A. Highly urgent delivery requirements

B. Predictable demand and tolerance for waiting

C. Severe product obsolescence

D. Zero shipment fixed costs

Correct Answer: B. Predictable demand and tolerance for waiting

Question 5

What is the principal purpose of a vehicle-routing model?

A. To determine product selling price

B. To assign vehicles and routes subject to operational constraints

C. To calculate customs duty rates

D. To forecast annual GDP

Correct Answer: B. To assign vehicles and routes subject to operational constraints

Question 6

What does OTIF measure?

A. Transport emissions only

B. On-time and complete delivery performance

C. Warehouse labour productivity only

D. Average inventory value

Correct Answer: B. On-time and complete delivery performance

Question 7

Why should centre-of-gravity results not be treated as final facility decisions?

A. They cannot use demand weights

B. They ignore all geographic information

C. They usually omit roads, fixed costs, capacity and service constraints

D. They always locate facilities at ports

Correct Answer: C. They usually omit roads, fixed costs, capacity and service constraints

Question 8

Which action most directly improves resilience through redundancy?

A. Using one carrier for every lane

B. Removing backup capacity

C. Establishing qualified alternative facilities or suppliers

D. Reducing visibility data

Correct Answer: C. Establishing qualified alternative facilities or suppliers

Question 9

If a vehicle carries 12 tonnes with an available capacity of 20 tonnes, its weight load factor is:

A. 40%

B. 60%

C. 80%

D. 120%

Correct Answer: B. 60%

Question 10

Why is normalisation needed in a weighted multi-criteria objective?

A. The criteria may have different units and scales

B. It eliminates the need for data

C. It guarantees the cheapest solution

D. It removes managerial preferences

Correct Answer: A. The criteria may have different units and scales