Transport operations are increasingly managed through connected digital systems rather than isolated spreadsheets, telephone calls, and paper records. A contemporary fleet may contain vehicles equipped with global navigation satellite system (GNSS) receivers, engine sensors, cameras, electronic logging devices, temperature probes, tyre-pressure monitors, and driver-facing applications. These devices generate data that can support dispatching, compliance, maintenance, safety, customer service, emissions management, and strategic planning.

Digitalisation, however, is not equivalent to purchasing software. Value arises when reliable data are integrated with operational processes, decision rights, human expertise, and governance. A poorly designed system can automate bad decisions, expose sensitive information, or create false confidence in artificial-intelligence (AI) recommendations. Graduate-level fleet management therefore requires both technical understanding and critical evaluation.

This chapter examines transport management systems (TMS), fleet management systems (FMS), positioning and telematics, Internet of Things (IoT) architectures, electronic logging, digital freight platforms, control towers, and digital twins. It then evaluates applications of AI in forecasting, routing, maintenance, safety, and anomaly detection, together with issues of bias, explainability, privacy, cybersecurity, regulation, and organisational adoption.

Learning objectives

1. The digital fleet-management architecture

Layer	Main function	Typical examples
Physical assets	Execute transport work	Vehicles, trailers, depots, roads, cargo
Sensing and identification	Observe assets and events	GNSS, CAN bus, RFID, cameras, temperature sensors
Connectivity and data platform	Transfer, store, and standardise data	Cellular networks, APIs, message brokers, cloud databases
Applications and analytics	Support decisions and workflows	TMS, FMS, dashboards, optimisation, AI models
Governance and human control	Set rules and accountability	Access controls, audit, safety procedures, compliance

The layers are interdependent. A sophisticated optimisation model cannot compensate for inaccurate vehicle positions or incomplete delivery timestamps. Conversely, high-quality data have little value if dispatchers cannot act on recommendations.

TMS and FMS

A TMS manages transport demand and execution across orders, shipments, carriers, routes, rates, appointments, documentation, and settlement. It is usually shipment- and network-oriented. An FMS manages the vehicle and driver resource: asset location, utilisation, fuel or energy, maintenance, inspections, driver behaviour, and compliance. The systems overlap. For example, a TMS may require vehicle capacity and estimated arrival time, while an FMS supplies current location and vehicle availability.

Integration should establish a common event model. Events such as “loaded,” “departed,” “arrived at customer,” and “proof of delivery received” need consistent definitions, timestamps, location references, and responsible systems. Otherwise, performance metrics become contradictory. A useful design principle is to identify a system of record for each data object while allowing other applications controlled access through APIs.

GPS, GNSS, telematics, and IoT

GPS is one satellite-based positioning service; GNSS is the broader category that includes GPS and other satellite constellations. Position estimates may be degraded by urban canyons, tunnels, atmospheric conditions, multipath reflections, or deliberate interference. A location point should therefore be treated as an observation with uncertainty, not as unquestionable truth.

Telematics combines telecommunications with monitoring or informatics. A telematics unit may receive GNSS position, vehicle-bus data, accelerometer readings, diagnostic trouble codes, and sensor measurements, then transmit them to a platform. IoT is a wider concept covering connected physical objects, identification, sensing, communication, processing, and actuation. A refrigerated trailer sensor, for example, may transmit temperature, door status, battery level, and location at configured intervals.

The business value of a sensor depends on its decision use. Sampling every second may be justified for collision reconstruction but wasteful for a low-risk asset whose location is needed only every fifteen minutes. Data frequency should reflect operational volatility, safety requirements, battery constraints, communications cost, and storage capacity.

Electronic logging and compliance technology

Electronic logging devices record driving, on-duty, off-duty, and rest statuses, normally using vehicle motion and authenticated driver inputs. Requirements differ by jurisdiction; therefore, organisations must not assume that a device approved in one country satisfies another jurisdiction’s rules. Compliance systems should preserve tamper-evident records, support authorised corrections, maintain an audit trail, and distinguish automatically inferred events from driver-confirmed events.

Automation does not eliminate responsibility. Dispatch practices, unrealistic schedules, poor training, or pressure to override alerts may still create fatigue risk. Compliance data should be used as part of a safety-management system, not merely as a punitive surveillance tool.

2. Digital freight platforms and control towers

A digital freight platform connects shippers, carriers, brokers, drivers, and sometimes financial or customs services. Core functions may include tendering, capacity discovery, instant quotation, booking, tracking, document exchange, exception management, and payment. Platforms can reduce search and transaction costs, but network effects may also concentrate bargaining power and create dependency on a dominant intermediary.

A control tower is an organisational and analytical capability that provides end-to-end visibility, exception management, and coordinated intervention across a transport network. It is not simply a large screen. Effective control towers combine:

Visibility is not the same as control. A dashboard may show that a shipment is late without identifying the decision, owner, or feasible remedy. Control-tower design should therefore begin with failure modes: missed appointments, temperature excursions, border delays, vehicle breakdowns, capacity shortages, and unsafe driving. Each exception requires a threshold, a responsible role, a response playbook, and a closure record.

Useful measures include data completeness, event latency, percentage of exceptions resolved within service-level targets, avoidable detention, on-time-in-full delivery, empty kilometres, and user adoption. A control tower that generates many alerts but few effective interventions may increase workload rather than improve performance.

3. Digital twins and decision models

A digital twin is a continuously updated digital representation of a physical asset, process, or network, linked to its real-world counterpart through data and models. A static simulation or dashboard is not necessarily a digital twin. A credible twin needs a defined physical scope, state variables, update mechanism, model assumptions, and a purpose such as prediction, diagnosis, optimisation, or scenario analysis.

For a fleet network, state variables might include vehicle position, load, remaining driving time, battery state of charge, maintenance condition, appointment status, and traffic conditions. The twin can test scenarios such as adding a depot, changing charging schedules, closing a road, or reallocating vehicles. Its outputs remain conditional on data quality and model validity. A twin should expose uncertainty and allow comparison between predicted and observed outcomes.

Digital twins are particularly valuable when experiments in the physical system are expensive or unsafe. They are less valuable when the process is poorly measured or when managers lack authority to implement recommended changes.

4. Data governance and information quality

Fleet data are heterogeneous. They vary in time resolution, coordinate systems, units, ownership, reliability, and legal sensitivity. Governance should define data ownership, permitted use, retention periods, quality responsibilities, access rights, incident response, and deletion procedures.

Here, C is completeness, n_present is the number of required records containing the field, and n_required is the total number of required records. The measure assumes that all required records have equal importance; weighted completeness is preferable when missing a safety-critical event is more serious than missing a low-value administrative field.

Data lineage should show where a value originated, how it was transformed, and which model or decision used it. Time synchronisation is essential. Differences between vehicle time, mobile-device time, server time, and local time zones can distort dwell, driving, and arrival calculations.

5. AI applications in fleet operations

AI includes statistical learning, machine learning, optimisation, computer vision, natural-language systems, and hybrid approaches. The appropriate method depends on the decision, available data, consequences of error, and need for explanation. AI should not be selected merely because it is fashionable.

Forecasting

Forecasts may concern shipment volume, lane demand, travel time, fuel or energy consumption, maintenance events, or no-show probability. A time-series forecast can be represented generally as:

ŷ_t+h is the forecast for horizon h, y is the observed target, x represents explanatory variables, and f is the fitted forecasting function. Assumptions depend on the selected model. Historical patterns may fail after a network redesign, strike, regulation change, or extreme weather event.

n is the number of observations, yᵢ the actual value, and ŷᵢ the forecast. MAE has the same units as the target and is often easier for managers to interpret than squared-error measures. Percentage errors can be misleading when actual demand is zero or very small.

Forecast bias should also be monitored. A model that is slightly less accurate but consistently conservative may be preferable when stockouts or missed appointments are costly; the choice should be linked to the organisation’s loss function.

Routing and dispatching

Routing combines constraints, objectives, and uncertainty. Constraints may include vehicle capacity, driver hours, time windows, compatibility, road restrictions, depot hours, charging requirements, and service priorities. Objectives may minimise distance, cost, emissions, lateness, or a weighted combination.

where D is distance, T travel or service time, L lateness, and E emissions or energy use; c_d, c_t, c_l, and c_e are compatible cost coefficients. The terms must be converted to a common monetary or utility scale. This expression is illustrative, not a complete vehicle-routing formulation. Feasibility constraints must be checked separately.

AI can predict travel times or infer likely service durations, while operations-research algorithms generate feasible routes. In practice, hybrid systems often outperform a claim that a single “AI route” can solve every operational issue. Dispatchers need the ability to lock stops, impose priorities, inspect constraint violations, and override recommendations with a recorded reason.

Predictive maintenance

Predictive maintenance estimates failure risk or remaining useful life from fault codes, vibration, temperature, oil condition, mileage, operating severity, and repair history. The maintenance decision should consider both predicted risk and consequence. A low-probability failure on a critical vehicle may merit immediate inspection, whereas a higher-probability failure on a redundant asset may be scheduled later.

Classification performance must account for class imbalance. If only 1% of trips contain a safety-critical failure, a model that predicts “no failure” every time has 99% accuracy but no practical value. Precision, recall, false-negative cost, calibration, and lead time are more informative. For a binary event:

TP is true positives, FP false positives, and FN false negatives. Precision answers how many alerts were correct; recall answers how many actual events were detected. The threshold should reflect safety, downtime, inspection capacity, and cost.

Safety analytics and anomaly detection

Telematics can identify harsh braking, speeding, distraction indicators, seat-belt events, unstable driving, and route deviations. These measures are proxies, not complete measures of driver competence. A harsh-braking event may represent dangerous driving, an unavoidable emergency, or an inaccurate accelerometer reading. Fair assessment therefore requires context, exposure adjustment, coaching, and an appeals process.

Anomaly detection identifies observations that differ from expected behaviour. Methods include statistical control limits, clustering, isolation-based methods, autoencoders, and rule-based checks. A robust workflow distinguishes data errors from operational anomalies and malicious activity. For example, an impossible jump in location may indicate poor satellite reception, device malfunction, or spoofing. The system should preserve the raw record and confidence score rather than silently replacing it.

6. Bias, explainability, privacy, and cybersecurity

Bias and fairness

Bias can enter through sampling, labels, measurement, historical decisions, or deployment context. A driver-risk model trained on one depot may penalise another depot because of different terrain, traffic, vehicle mix, or route allocation. If historical maintenance records reflect unequal inspection practices, a model may reproduce those inequalities.

Evaluation should compare performance across relevant groups and operating contexts, while recognising that fairness metrics can conflict. Organisations should document the target, features, exclusions, training population, validation design, known limitations, and appeal process. Sensitive attributes may sometimes be needed for fairness auditing, even when they are excluded from operational prediction.

Explainability and human oversight

Explainability is not a single technical property. A fleet manager may need a local explanation for one alert, a global description of model behaviour, and evidence that the model remains calibrated over time. Explanations should be accurate, stable, and understandable to the decision-maker. A visually attractive explanation that misrepresents the model is harmful.

High-impact decisions should have human review, clear accountability, and a documented override process. Human oversight is ineffective if the operator cannot understand the recommendation, lacks time to review it, or is punished for rejecting it.

Privacy and worker data

Location, driving behaviour, biometric video, and work schedules can reveal sensitive information. Lawful and ethical processing normally requires a defined purpose, data minimisation, proportionality, security, retention limits, transparency, and mechanisms for access or correction where applicable. Consent may be weak in an employment relationship; organisations should not treat a nominal acceptance click as sufficient ethical justification.

Policies should distinguish operational use from secondary uses such as productivity ranking, disciplinary action, insurance sharing, or commercial data sales. Privacy impact assessments are appropriate before deploying cameras, facial recognition, continuous location tracking, or driver-scoring systems.

Cybersecurity

Connected fleets enlarge the attack surface: telematics units, mobile devices, charging systems, APIs, cloud services, supplier portals, and vehicle control networks may all be targeted. Core controls include asset inventories, secure configuration, encryption in transit and at rest, strong authentication, least privilege, network segmentation, patch management, vulnerability disclosure, backups, logging, supplier assurance, and incident exercises.

Safety and cybersecurity are connected. A compromised dispatch account may create fraudulent loads; manipulated location data may conceal theft; interference with vehicle systems may create physical danger. Security risk assessments should therefore consider confidentiality, integrity, availability, safety, and recovery—not only data theft.

7. Implementation and performance management

A balanced scorecard should combine outcome, process, data, model, people, and risk indicators. Possible measures include cost per kilometre, empty distance, on-time delivery, fuel or energy intensity, collision rate, unplanned downtime, alert precision, forecast error, data completeness, system availability, user adoption, privacy incidents, and cyber response time.

A business case should include total cost of ownership: devices, connectivity, integration, licences, cloud processing, training, support, cybersecurity, replacement, and change management. Benefits should be risk-adjusted and separated into realised savings, avoided losses, service improvements, and strategic options. A claimed reduction in fuel, for example, should be tested against changes in payload, weather, route mix, fuel prices, and driver composition.

Applied case: Northstar Cold Chain (illustrative)

Background

Northstar Cold Chain operates 180 refrigerated vehicles serving hospitals and food manufacturers. It has an FMS for vehicle locations and a separate TMS for orders. Temperature records are downloaded after delivery, while dispatchers rely on telephone calls for major disruptions.

Problem

The company experiences late deliveries, unnecessary emergency maintenance, and occasional temperature excursions. Management proposes a control tower, continuous temperature telemetry, predictive maintenance, and an AI driver-risk score.

Analysis

An initial data audit finds inconsistent vehicle identifiers, unsynchronised clocks, missing door-status events, and different definitions of “arrival.” Temperature sensors are not calibrated uniformly. Historical maintenance data contain many repairs but few records of inspections that found no fault, making supervised failure modelling unreliable. The proposed driver-risk score would also compare urban and rural routes without adjusting for exposure or traffic conditions.

Approach and solution

Northstar first creates a master vehicle and trailer register, standardises event definitions, synchronises device clocks, and calibrates sensors. It defines a temperature-excursion playbook: the control tower receives an alert only when temperature exceeds a product-specific limit for a specified duration, with an uncertainty band and sensor-health check. Dispatchers can contact the driver, inspect door status, divert the load, or approve a documented exception.

For maintenance, the company begins with rules based on diagnostic codes and service intervals, then collects labelled inspection outcomes before deploying a predictive model. For safety, it uses exposure-adjusted coaching indicators rather than an automatic disciplinary score. Routing software uses predicted travel time but leaves critical hospital stops under dispatcher control. The pilot covers 30 vehicles for twelve weeks and compares results with a similar non-pilot group.

Outcome

The pilot reports fewer unresolved temperature alerts and faster exception closure, but no statistically credible maintenance reduction during the short period. Several false alerts are traced to sensor placement and intermittent connectivity. Driver acceptance improves after Northstar publishes data-use rules, allows event review, and separates coaching data from payroll decisions. The company postpones full AI scaling until the maintenance labels and sensor reliability improve.

Learning points

The case demonstrates that integration and governance precede advanced AI. It also shows why a control tower requires response authority, why safety metrics need contextual interpretation, and why pilot evaluation should distinguish technical performance from business outcomes.

Contemporary developments and critical evaluation

Edge computing allows preliminary filtering and inference inside vehicles, reducing latency and communications demand. Electric fleets introduce battery state estimation, charging coordination, thermal management, and route feasibility under energy uncertainty. Computer vision is expanding into damage inspection, loading verification, and driver assistance, but visual data create heightened privacy and liability concerns. Generative AI may assist with exception summaries and natural-language queries, yet generated explanations must be grounded in authoritative records and reviewed where decisions affect safety or contractual rights.

The strategic question is not whether a company is “AI-enabled.” It is whether digital capabilities improve decisions while preserving resilience, fairness, privacy, security, and human accountability. Organisations should prefer auditable systems that fail safely over opaque systems that promise marginal gains without trustworthy controls.

Chapter Summary

Digital fleet management integrates physical assets, sensing, connectivity, software, analytics, and governance. TMSs coordinate shipments and transport execution, whereas FMSs focus on vehicles, drivers, maintenance, and compliance; integration requires shared identifiers, event definitions, and data ownership. Telematics and IoT provide valuable observations, but positioning, sensors, connectivity, and timestamps are imperfect.

Digital freight platforms reduce transaction friction, while control towers convert visibility into coordinated exception management. Digital twins support scenario analysis when the physical system is adequately measured and the model’s purpose is explicit. AI can forecast demand, predict travel time, support routing, estimate maintenance risk, analyse safety, and detect anomalies. Appropriate evaluation requires out-of-sample testing, operational metrics, calibration, context, and consideration of false-positive and false-negative costs.

Bias, explainability, privacy, and cybersecurity are core management concerns rather than optional technical add-ons. Successful implementation begins with a defined decision problem, reliable data, bounded pilots, human oversight, workflow integration, and measurable outcomes. In contemporary fleet operations, digital maturity is demonstrated by trustworthy decision capability—not by the volume of data or the novelty of the algorithm.

Key Terms

Artificial intelligence: Computational methods that perform tasks associated with learning, prediction, reasoning, or perception.

Control tower: A coordinated capability for network visibility, exception management, and intervention.

Digital twin: A continuously updated digital representation linked to a physical asset, process, or network.

Electronic logging device: A device that records regulated driving and duty-status information.

Fleet management system: Software for managing vehicles, drivers, utilisation, maintenance, fuel or energy, and compliance.

GNSS: Global Navigation Satellite System; the general category of satellite positioning systems.

Internet of Things: Connected physical objects capable of sensing, communicating, processing, or acting.

MAE: Mean absolute error, an average absolute difference between actual and forecast values.

Model calibration: Agreement between predicted probabilities and observed event frequencies.

Predictive maintenance: Maintenance planning based on predicted condition or failure risk.

Telematics: The integrated use of telecommunications and monitoring or informatics.

Transport management system: Software for planning, executing, tracking, and settling transport activity.

Anomaly detection: Identification of observations or events that differ from expected behaviour.

Edge computing: Processing data near the source, such as within a vehicle or depot.

IoT gateway: A device or service that aggregates, translates, and transmits sensor data.

Human-in-the-loop: A design in which people review, approve, override, or contextualise automated outputs.

Review Questions

Short Answer Questions

Long Answer Questions

Analytical / Application Questions

Multiple Choice Questions with Answers

Question 1

Which system is primarily focused on shipment planning, carrier execution, appointments, and freight settlement?

Question 2

What is the main characteristic that distinguishes a control tower from a simple visibility dashboard?

Correct Answer: B. It links exceptions to owners, workflows, and interventions

Question 3

A model produces 80 true positives and 20 false positives. What is its precision?

Question 4

Which measure answers the question: what proportion of actual failures did the model detect?

Question 5

Why can a model with 99% accuracy still be ineffective for rare maintenance failures?

Question 6

B. A continuously updated digital representation linked to a physical counterpart

Correct Answer: B. A continuously updated digital representation linked to a physical counterpart

Question 7

Correct Answer: C. Define purpose, minimise data, secure access, and provide review mechanisms

Question 8

Correct Answer: A. It can conceal theft or create false operational decisions

Question 9

Why should a routing objective convert distance, time, lateness, and emissions into compatible units or costs?

Correct Answer: A. To make the objective mathematically and managerially comparable

Question 10

C. Define the decision, audit data, pilot, validate outcomes, and scale with governance

Correct Answer: C. Define the decision, audit data, pilot, validate outcomes, and scale with governance

Chapter 13 Digital Fleet Management, Telematics, and AI

Digital Fleet Management, Telematics, and AI

Introduction