Routing, Load Planning and Transportation Optimization

Introduction

Transportation decisions connect warehouse operations with customers, suppliers, production sites and transport networks. A warehouse may pick and stage orders efficiently, yet still incur high cost or poor service if vehicles are poorly routed, loads are unstable, delivery windows are missed, or return movements are empty. Routing, load planning and transportation optimization therefore form an integrated decision domain rather than three separate activities.

Routing determines the sequence in which locations are visited. Load planning determines which shipments are assigned to which vehicle, how freight is arranged, and whether vehicle weight, volume, axle and handling constraints are respected. Transportation optimization evaluates alternative designs against cost, service, capacity, risk, regulatory and sustainability objectives.

At postgraduate level, the subject should be understood as a family of decision models. Some decisions are strategic, such as selecting cross-dock locations or modal combinations. Others are tactical, such as assigning customers to delivery days and carriers. Still others are operational, such as re-optimizing routes when an order is late, a vehicle fails or traffic conditions change.

Learning Objectives

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

1. Distinguish routing, load-building, scheduling and network-design decisions.
2. Formulate and interpret vehicle-routing and transportation-assignment models.
3. Calculate cube utilization, weight utilization, route distance and transportation cost.
4. Apply practical approaches to time windows, multi-stop delivery, backhaul and multimodal planning.
5. Evaluate trade-offs among cost, capacity, service, risk, carbon emissions and operational feasibility.
6. Assess how warehouse management systems (WMS), transportation management systems (TMS), optimization software and AI support these decisions.

1. The Transportation Decision System

A transportation plan converts demand into movements. Its principal inputs are customer locations, shipment quantities, product characteristics, delivery requirements, vehicle characteristics, driver and regulatory constraints, rates, travel times and available inventory. Its outputs include vehicle assignments, stop sequences, departure times, loading instructions, carrier bookings and exception plans.

The decisions are interdependent. A route that minimizes distance may require a vehicle with insufficient cube. A load that maximizes weight utilization may create excessive handling or violate axle limits. A multimodal option may reduce line-haul cost but increase transfer risk and transit time. Consequently, optimization should not treat distance or cost as the only objective.

A useful hierarchy is:

The WMS supplies order, inventory, product and staging information. The TMS combines this information with transport master data, rates, carrier capacity, maps, traffic, time windows and execution events. Integration is essential: an optimized route is not useful if the warehouse cannot pick, stage and load it in the required sequence.

2. Core Terminology and Performance Measures

A stop is a customer, supplier or facility visit. A leg is movement between two consecutive points. A route is an ordered set of legs performed by a vehicle. A tour commonly denotes a route beginning and ending at a depot. A shipment is freight moved under one transport requirement; an order may contain several shipments. A load is the freight assigned to a vehicle movement.

Important performance measures include:

• Distance utilization: productive or revenue distance divided by total distance.
• Weight utilization: loaded weight divided by legal or rated payload.
• Cube utilization: loaded volume divided by usable vehicle volume.
• Vehicle utilization: a combined evaluation of weight, cube, time and route productivity.
• On-time delivery: deliveries completed within the contractual or promised window.
• Cost per shipment, order, case or kilometre: unit measures used for benchmarking.
• Empty kilometres: kilometres travelled without revenue freight.
• Dwell time: time spent waiting, loading, unloading or processing rather than travelling.

For shipment i with weight wᵢ and volume vᵢ, and vehicle k with payload capacity Wₖ and usable volume Vₖ, the basic utilization measures are:

The sums apply only to freight assigned to vehicle k. These ratios are meaningful only when weight and volume are measured in consistent units and when the vehicle's stated capacities reflect operationally usable capacity. For example, pallets may leave unusable gaps, and a legal payload may be lower than the manufacturer's rated payload.

3. Load Planning: From Orders to Feasible Loads

3.1 Load-building logic

Load planning begins by consolidating compatible orders. A feasible load must satisfy at least four conditions:

1. Capacity: weight, volume, pallet positions and compartment capacity are not exceeded.
2. Compatibility: products can legally and safely travel together.
3. Service: each delivery can be reached within its time window.
4. Handling: loading and unloading sequences are practical.

Additional constraints may include temperature zones, dangerous-goods segregation, security, stackability, orientation, fragility, value, customs status and customer-specific equipment requirements.

A simple one-vehicle capacity condition is:

where xᵢₖ = 1 if shipment i is assigned to vehicle k and 0 otherwise; wᵢ and vᵢ are shipment weight and volume; and Wₖ and Vₖ are usable vehicle capacities. These constraints are necessary but not sufficient: a mathematically feasible load may still be physically impossible because of axle distribution, shape, loading sequence or incompatibility.

3.2 Cube, weight and the binding constraint

A shipment may be cube-constrained, weight-constrained or constrained by both. Consider a vehicle with 26 pallet positions, 10,000 kg payload and 65 m³ usable volume. A load of 24 pallets weighing 9,100 kg and occupying 57 m³ has:

The load is feasible under these stated constraints. Weight is the tighter capacity measure because it has the higher utilization. However, if the goods are bulky and light, cube may become binding before weight. TMS planning should therefore preserve both measures and, where relevant, pallet positions, linear metres and axle loads.

3.3 Loading sequence and physical feasibility

For multi-stop routes, freight for the last stop is generally loaded first, subject to safety and product constraints, while freight for the first stop is positioned for easy access. This is not an absolute rule: cross-docking, side-loading, mixed-temperature equipment and customer unloading practices may reverse or modify the sequence.

A practical load plan records:

• vehicle and compartment;
• shipment, pallet or handling-unit identifier;
• destination and stop number;
• weight, dimensions and orientation;
• loading position;
• special handling and restraint instructions;
• scan confirmation and exception status.

Three-dimensional loading algorithms can improve cube utilization, but their recommendations require operational validation. A theoretically dense arrangement may increase unloading time or create unsafe manual handling.

4. Routing Models

4.1 The travelling salesperson problem

The travelling salesperson problem (TSP) seeks the least-cost tour visiting each required location once and returning to the depot. It is useful for a single vehicle with one visit per customer, but real distribution commonly requires multiple vehicles, capacities and time windows.

Let cᵢⱼ be the cost of travelling from node i to node j, and xᵢⱼ equal 1 when the arc is used. A basic objective is:

Each customer must have one incoming and one outgoing arc. Additional subtour-elimination constraints are needed; otherwise the solution may contain disconnected small cycles rather than one depot-connected tour.

4.2 Capacitated vehicle-routing problem

The capacitated vehicle-routing problem (CVRP) extends the TSP to multiple vehicles with capacity limits. Let qᵢ be demand at customer i, Qₖ vehicle capacity, and yₖ indicate whether vehicle k is used. A simplified capacity condition is:

The complete model also needs flow-conservation, depot, customer-coverage and subtour-elimination constraints. If qᵢ represents weight, the constraint controls weight; if freight is cube-constrained, a second volume constraint is required. CVRP is computationally difficult as problem size grows, which explains the extensive use of heuristics and metaheuristics in commercial systems.

4.3 Vehicle-routing problem with time windows

In the vehicle-routing problem with time windows (VRPTW), customer i must be served between earliest time aᵢ and latest time bᵢ. If service begins at time tᵢ and service duration is sᵢ, then:

For an arc from i to j, a sequencing constraint can be represented as:

Here τᵢⱼ is travel time and M is a sufficiently large constant. The constraint is active when xᵢⱼ = 1. The value of M should be as tight as practical; an unnecessarily large value can weaken computational performance and create numerical instability.

Time windows may be hard or soft. A hard window makes violation infeasible. A soft window permits violation at a penalty, which can be appropriate where customers tolerate a late delivery but service-level penalties or loss of goodwill must be represented.

4.4 Pickup-and-delivery and backhaul routing

Pickup-and-delivery routing requires a pickup and its associated delivery to be linked, often with precedence: pickup must occur before delivery. Backhaul planning uses return freight to reduce empty kilometres. It can involve supplier collections, customer returns, packaging retrieval or interfacility transfers.

Backhaul is attractive when the return shipment has compatible equipment, timing and destination. It may be rejected when the pickup causes a delivery delay, requires cleaning or segregation, creates excess dwell, or carries unacceptable damage or contamination risk. The correct objective is not “maximize backhaul” but “maximize profitable compatible backhaul subject to service and safety constraints.”

5. Objective Functions and Trade-offs

A transportation model can minimize a weighted objective such as:

where:

• Cₜ = travel and fuel cost;
• Cᵥ = vehicle, driver and carrier cost;
• Cₛ = service-violation penalties;
• Cₑ = emissions or environmental cost;
• Cᵣ = risk, handling and other relevant cost.

The terms must be expressed in comparable monetary units or incorporated through a clearly defined multi-objective method. A weighted objective is not automatically neutral: the chosen weights encode managerial priorities. An alternative is lexicographic optimization—for example, first ensure regulatory and critical service feasibility, then minimize cost, then minimize emissions.

Cost should include more than the line-haul rate. Relevant components may include fixed vehicle cost, distance cost, time-based driver cost, tolls, accessorial charges, loading and unloading, waiting, detention, failed delivery, claims, inventory-in-transit and carbon cost. Rates should be compared on equivalent service definitions and lanes.

A basic route cost can be represented as:

where F is fixed trip cost, D is distance, r_d is cost per distance unit, H is hours, r_h is cost per hour, S is stops, r_s is stop cost, and A is accessorial cost. The formula is a planning approximation; actual carrier tariffs may be piecewise, zone-based, minimum-charge, dimensional or contract-specific.

6. Practical Routing Heuristics

Exact optimization can produce high-quality solutions for small problems but may be too slow for large or rapidly changing operations. Common approaches include:

• Nearest neighbour: repeatedly visit the closest unserved stop. It is fast but can produce poor final legs and ignores broader structure.
• Savings heuristic: begin with separate depot-to-customer routes and combine routes when the distance saving is attractive and constraints remain feasible. For customers i and j, the classical distance saving is:

where 0 denotes the depot. The formula assumes a symmetric or appropriately defined cost matrix; if costs are directional, savings must respect arc direction.

• Insertion heuristics: insert a stop into the position causing the smallest incremental cost while maintaining feasibility.
• Local search: improve a current route through 2-opt, 3-opt, relocate and exchange moves.
• Metaheuristics: tabu search, simulated annealing, genetic algorithms and adaptive large-neighbourhood search explore broader solution spaces.
• Rolling-horizon optimization: optimize a near-term operational window, execute committed decisions, and re-optimize the remaining horizon as information changes.

Heuristics should be evaluated by solution quality, runtime, stability, interpretability and operational feasibility—not merely by theoretical distance. A route that changes dramatically every hour may be mathematically efficient but operationally disruptive.

7. Multi-Stop Planning and Time Windows

Multi-stop routes create a trade-off between consolidation and complexity. More stops can reduce cost per delivery, but they increase travel variability, unloading time, damage exposure and risk of cascading lateness. The planning process should estimate total route time as:

All terms must use the same time unit. Driving time should reflect realistic travel conditions, not only free-flow map distance. Service time may vary by product, dock, customer, quantity, vehicle type and appointment procedure.

A route should include a practical buffer for uncertainty, but excessive buffer reduces vehicle productivity. Historical travel-time distributions, appointment adherence, weather and congestion data can support differentiated buffers rather than one universal percentage.

Time-window planning should also distinguish arrival time from service-start time. A vehicle may arrive early and wait, or arrive late and incur a penalty. If the customer requires a fixed appointment, the model should represent the appointment duration and site capacity; otherwise, it may schedule multiple vehicles at one dock simultaneously in a mathematically feasible but operationally impossible plan.

8. Network, Multimodal and Intermodal Optimization

Transportation optimization is not limited to road routes. A shipment may use road, rail, inland waterway, ocean, air or pipeline, with transfers between modes. Intermodal transport generally uses more than one mode under an integrated movement, often with standardized loading units. Multimodal is a broader term for using multiple modes, potentially under separate arrangements.

A mode comparison should evaluate:

• total door-to-door cost;
• transit time and variability;
• shipment size and density;
• service frequency and cutoff times;
• transfer and damage risk;
• customs and security requirements;
• emissions and energy use;
• network resilience and disruption alternatives.

For a path containing legs l, total transit time is approximately:

The additive expression assumes the legs occur sequentially. If activities overlap, the critical-path logic rather than simple addition is required. Similarly, total cost may include mode-specific charges, terminal handling, documentation, drayage, inventory carrying cost and risk premiums.

A network model represents facilities and transport links as nodes and arcs. Flow-conservation requires inbound flow plus supply to equal outbound flow plus demand at each relevant node. Capacity, service and compatibility restrictions then determine feasible paths. Scenario analysis is valuable because the lowest-cost path under normal conditions may be fragile under port closure, labour disruption or fuel-price volatility.

9. Worked Example: Integrated Route and Load Assessment

An illustrative distributor must serve four customers from one depot. The vehicle has a usable payload of 5,000 kg and usable cube of 32 m³. The proposed route is Depot–A–B–C–D–Depot. The shipment data are:

The resulting utilizations are:

Suppose driving time for the complete route is 190 minutes and planned waiting/buffer time is 35 minutes. Then:

The load is feasible on the stated weight and cube measures. However, feasibility still requires checking the delivery windows and loading sequence. If customer C has the earliest appointment but is placed third, the route may fail despite acceptable capacity utilization. The TMS should therefore test alternative sequences, including the additional service and distance cost of changing the order. The WMS should stage freight so that unloading at A, B, C and D is safe and does not require unnecessary re-handling.

10. Illustrative Case: Regional Grocery Distributor

Background

The following is a fictional illustrative case. A regional grocery distributor operates two warehouses, 85 delivery vehicles and approximately 1,100 store deliveries per week. The company experiences 18% average empty kilometres, frequent late-afternoon delivery congestion and poor visibility of refrigerated vehicle capacity.

Problem

Routes are planned by experienced dispatchers using spreadsheets. Orders are released in waves, and late changes are communicated by telephone. The organization measures transport cost per trip but does not consistently measure cube utilization, waiting time, missed windows or return-load performance.

Analysis

A six-week data study reveals three causes:

1. Delivery territories were based on historical practice rather than current demand density.
2. Vehicle selection used pallet count but ignored actual cube and temperature-compartment requirements.
3. Supplier pickups were booked after outbound routes had been fixed, preventing profitable backhaul.

The analysis also identifies a governance issue: planners routinely override customer time windows without recording the reason, making service performance difficult to audit.

Approach and solution

The distributor implements an integrated WMS–TMS process. The WMS sends confirmed handling units, dimensions, temperatures and release times. The TMS creates routes using capacity, time-window, refrigeration and driver-hour constraints. A savings-based initial solution is improved through local search. A planner reviews exceptions rather than rebuilding every route manually.

The company introduces a hierarchical objective: first satisfy food safety, legal driving and critical customer windows; second minimize total transport cost; third reduce empty kilometres and emissions. Supplier pickups are considered only when they preserve delivery feasibility and product compatibility. A load-scanning process confirms the planned sequence at dispatch.

Outcome

After a controlled eight-week pilot, management reports lower empty distance, improved cube visibility and fewer failed deliveries. The exact numerical benefit is intentionally not generalized because results depend on lane structure, fleet, rates and demand volatility. The important outcome is the changed decision process: route quality is evaluated through cost, service, safety, utilization and execution stability rather than distance alone.

Learning points

• Data quality is a prerequisite for optimization.
• A TMS does not replace operational judgement; it reallocates judgement toward exceptions and trade-offs.
• Refrigeration, food safety and loading sequence must be explicit constraints.
• Backhaul should be evaluated on net contribution, not gross revenue.
• Performance measurement must compare planned, executed and financially settled movements.

11. Technology, AI and Data Governance

Modern TMS platforms provide geocoding, map-based routing, carrier tendering, rate management, appointment scheduling, electronic proof of delivery, ETA prediction, freight audit and analytics. Telematics supplies vehicle position, temperature, harsh braking, fuel and engine data. APIs connect carriers, marketplaces, ports and visibility platforms.

AI and machine learning are useful for travel-time prediction, demand forecasting, stop-service-time estimation, anomaly detection and dynamic ETA. They can also assist planners by explaining why a route is infeasible or identifying likely missed appointments. However, predictive accuracy does not guarantee an operationally good plan. Models may reproduce biased historical decisions, fail under unprecedented disruption or optimize a proxy such as distance while neglecting driver welfare and customer service.

Good governance requires:

• documented master-data ownership;
• controlled geocodes, dimensions and service times;
• versioned optimization parameters;
• human approval for material exceptions;
• audit trails for overrides and carrier selection;
• cybersecurity controls for connected vehicles and APIs;
• privacy protection for driver and customer data;
• validation of model performance across locations and seasons.

Automation should be explainable enough for dispatchers, customers, auditors and regulators to understand the basis of a material decision.

12. Challenges, Limitations and Professional Judgement

Optimization models simplify reality. Travel times are uncertain, demand changes after planning, dimensions may be inaccurate, and customer behaviour may not follow appointments. A mathematically optimal solution may be unacceptable if it creates excessive driver workload, unsafe loading, unstable daily schedules or inequitable allocation of difficult routes.

Common implementation failures include:

• optimizing inaccurate shipment dimensions;
• ignoring loading and unloading time;
• using average travel times without variability;
• treating all vehicles as interchangeable;
• allowing soft constraints without realistic penalties;
• omitting accessorial and waiting costs;
• measuring savings against an unrealistic baseline;
• failing to retrain planners and drivers;
• changing routes too frequently.

Professional practice therefore combines optimization with scenario testing, pilot deployment, exception management and post-execution learning. Key performance indicators should be balanced: cost, on-time-in-full service, utilization, empty kilometres, emissions, claims, driver hours, route stability and customer complaints.

13. Best-Practice Framework

A robust transportation-optimization process follows these steps:

1. Define the service promise: delivery windows, priorities, product requirements and escalation rules.
2. Clean the data: locations, dimensions, weights, calendars, travel times, rates and vehicle capacities.
3. Model true constraints: legal, safety, temperature, dock, labour and handling requirements.
4. Choose the objective hierarchy: feasibility first, then cost, service, resilience and sustainability according to policy.
5. Construct and improve routes: use exact methods for small problems and suitable heuristics for larger or dynamic problems.
6. Validate physically: review loading, axle balance, stop sequence, driver hours and warehouse readiness.
7. Execute with visibility: use scans, telematics, ETA updates and exception workflows.
8. Measure planned versus actual results: identify systematic bias in time, cost and service assumptions.
9. Continuously improve: revise parameters, contracts, territories, appointment rules and network design.

Chapter Summary

Routing determines the sequence and timing of transport movements; load planning assigns compatible freight to vehicles while respecting weight, cube, handling and regulatory constraints. TSP, CVRP and VRPTW provide foundational mathematical models, but practical transportation planning often requires pickup-and-delivery, backhaul, multimodal, uncertainty and real-time considerations.

Weight and cube utilization are necessary capacity measures, not complete measures of vehicle productivity. Time windows require realistic travel, service and waiting-time assumptions. Transportation objectives should include total cost, service performance, risk, emissions and operational stability. Exact optimization, heuristics and human review are complementary tools. WMS–TMS integration, accurate master data, auditable governance and disciplined execution determine whether a mathematically attractive plan produces real operational value.

Key Terms

Routing: Determining the sequence and timing of vehicle visits.

Load planning: Assigning and physically arranging freight in vehicles.

Vehicle-routing problem: Optimization of vehicle movements among a depot and required locations.

TSP: Single-tour routing problem visiting each location once.

CVRP: Vehicle-routing problem with vehicle-capacity constraints.

VRPTW: Vehicle-routing problem incorporating customer time windows.

Time window: Earliest and latest permissible service times.

Cube utilization: Loaded usable volume divided by vehicle usable volume.

Weight utilization: Loaded weight divided by legal or rated payload.

Backhaul: Return movement carrying freight after an outbound delivery route.

Empty kilometres: Distance travelled without revenue freight.

Stop time: Time required to perform loading, unloading or service at a location.

Subtour: Disconnected cycle in a routing model that does not connect properly to the depot.

Heuristic: Practical method that seeks a good solution without guaranteeing the global optimum.

Intermodal transport: Integrated movement using more than one transport mode.

Soft constraint: Constraint whose violation is permitted at a specified penalty.

Rolling horizon: Repeated re-optimization of the remaining planning period.

Total landed transport cost: Transport cost including relevant accessorial, handling, delay and risk components.

Transportation management system: Software supporting transport planning, execution, settlement and visibility.

Review Questions

Short Answer Questions

1. Distinguish cube utilization from weight utilization.
2. What is the main difference between the TSP and the CVRP?
3. Why must service time be included in multi-stop route planning?
4. Define a soft time-window constraint.
5. State two reasons why a backhaul may be operationally unattractive.

Long Answer Questions

1. Explain how WMS and TMS integration affects load planning and route execution.
2. Discuss the principal constraints that should be included in a VRPTW model.
3. Compare exact optimization, heuristics and metaheuristics for large transportation problems.
4. Explain how total transportation cost should be constructed for managerial decisions.
5. Evaluate the benefits and governance risks of AI-supported dynamic routing.

Analytical / Application Questions

1. A vehicle has 8,000 kg payload and 48 m³ usable cube. A proposed load weighs 6,800 kg and occupies 39 m³. Calculate both utilization percentages and identify the tighter capacity constraint. What additional physical checks are required?
2. Four customers have incompatible delivery windows with one vehicle. Develop a decision process for determining whether to split the route, use a second vehicle, negotiate a window, or apply a late-delivery penalty.
3. A distributor is considering rail plus road instead of direct road transport. Construct a comparison framework covering cost, door-to-door time, variability, transfer risk, emissions, inventory effects and resilience.

Multiple Choice Questions with Answers

Question 1

Which constraint directly limits the total shipment weight assigned to vehicle k?

A. ∑ᵢ wᵢ xᵢₖ ≤ Wₖ

B. ∑ᵢ vᵢ xᵢₖ ≥ Vₖ

C. tⱼ ≤ tᵢ − τᵢⱼ

D. sᵢⱼ = cᵢⱼ − c₀ᵢ − c₀ⱼ

Correct Answer: A. ∑ᵢ wᵢ xᵢₖ ≤ Wₖ

Question 2

A vehicle carries 4,200 kg on a usable payload of 5,000 kg. What is its weight utilization?

A. 76%

B. 80%

C. 84%

D. 90%

Correct Answer: C. 84%

Question 3

Which problem extends the TSP by incorporating multiple vehicles and capacity limits?

A. Assignment problem

B. CVRP

C. Shortest-path problem

D. Inventory-routing-free model

Correct Answer: B. CVRP

Question 4

In a time-window model, what do aᵢ and bᵢ generally represent?

A. Vehicle weight and volume

B. Fixed and variable cost

C. Earliest and latest service times

D. Pickup and delivery quantities

Correct Answer: C. Earliest and latest service times

Question 5

What is the principal purpose of a subtour-elimination constraint?

A. To increase vehicle payload

B. To prevent disconnected cycles that do not form valid depot-connected routes

C. To calculate fuel consumption

D. To set customer service duration

Correct Answer: B. To prevent disconnected cycles that do not form valid depot-connected routes

Question 6

For customers i and j, the classical savings value sᵢⱼ equals:

A. cᵢⱼ + c₀ᵢ + c₀ⱼ

B. c₀ᵢ + c₀ⱼ − cᵢⱼ

C. cᵢⱼ − c₀ᵢ − c₀ⱼ

D. c₀ᵢ × c₀ⱼ ÷ cᵢⱼ

Correct Answer: B. c₀ᵢ + c₀ⱼ − cᵢⱼ

Question 7

A load occupies 23 m³ in a vehicle with 32 m³ usable cube. Approximately what is cube utilization?

A. 62.5%

B. 71.9%

C. 79.0%

D. 91.0%

Correct Answer: B. 71.9%

Question 8

Which statement best describes a soft time window?

A. The delivery can never occur outside the window

B. The window is ignored during optimization

C. Violation is allowed but incurs a specified penalty

D. The window applies only to vehicle weight

Correct Answer: C. Violation is allowed but incurs a specified penalty

Question 9

Which is an appropriate reason to reject an apparently profitable backhaul?

A. It uses a compatible vehicle and preserves delivery windows

B. It reduces empty kilometres without extra handling

C. It causes a critical outbound delivery to miss its appointment

D. It improves vehicle utilization

Correct Answer: C. It causes a critical outbound delivery to miss its appointment

Question 10

Why should AI-supported routing be governed with audit trails and human review?

A. Optimization models never use data

B. Predictive accuracy guarantees operational feasibility

C. Models can reproduce data bias or fail under unusual disruptions

D. Human review always eliminates transport cost

Correct Answer: C. Models can reproduce data bias or fail under unusual disruptions