Ethics, Governance and the Future of Logistics Analytics

Introduction

Logistics analytics has moved from retrospective reporting to predictive, prescriptive and increasingly autonomous decision-making. Organisations now use data to forecast demand, allocate inventory, price transport capacity, select carriers, sequence warehouse work, monitor driver safety and optimise last-mile delivery. These capabilities can reduce cost, emissions and delivery uncertainty. They can also create material risks: surveillance of workers, discriminatory allocation of opportunities, unsafe automation, cyber-physical disruption, opaque decisions, unlawful data processing and excessive dependence on vendors.

The central managerial challenge is therefore not simply to obtain more data or deploy more advanced algorithms. It is to establish conditions under which analytical systems are lawful, secure, explainable, contestable and aligned with organisational and social objectives. Governance must cover the whole analytical life cycle: data collection, preparation, modelling, deployment, monitoring, intervention and retirement.

In logistics, this requirement is especially important because decisions occur across organisational boundaries. A shipment may involve a shipper, freight forwarder, carrier, warehouse operator, customs authority, technology provider, driver and customer. Each party may control different data and bear different risks. A model that appears efficient for one firm may transfer congestion, workload, environmental harm or financial risk to another party.

Learning Objectives

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

1. Explain ethical principles relevant to logistics analytics.
2. Design a governance structure for data, algorithms, cybersecurity and accountability.
3. Evaluate privacy, fairness, transparency, safety and regulatory-compliance risks.
4. Assess the effects of analytics and automation on workers, communities and the environment.
5. Apply a risk-based framework to autonomous, predictive and self-optimising logistics systems.
6. Recommend controls, metrics and escalation mechanisms for responsible deployment.

1. The Ethical Character of Logistics Decisions

An analytical decision is ethical when it can be justified not only by efficiency but also by respect for rights, legitimate expectations and fair treatment. Four ethical perspectives are particularly useful.

1.1 Consequentialism

Consequentialism evaluates an action by its effects. A routing algorithm may be justified if it reduces total vehicle kilometres, emissions and late deliveries. However, aggregate benefits are insufficient if the system imposes disproportionate noise, pollution or traffic on a particular neighbourhood. Consequential analysis must therefore examine distribution, not only total performance.

1.2 Rights and duties

A rights-based approach asks whether the system respects privacy, due process, safety, property and freedom from unjustified discrimination. For example, a company may have a legitimate interest in monitoring vehicle location, but that interest does not automatically justify continuous collection of a worker’s personal location outside working hours.

1.3 Justice and fairness

Justice concerns how benefits, burdens and opportunities are distributed. A carrier-selection model trained on historical performance may systematically exclude smaller carriers because they have fewer historical observations, not because they are less capable. Fairness does not always mean identical treatment; it may require adjustments for unequal starting conditions and data scarcity.

1.4 Virtue and professional responsibility

Virtue ethics focuses on the character and judgement of decision-makers. Professional integrity requires analysts to disclose uncertainty, resist pressure to manipulate performance measures and escalate unsafe or unlawful uses. A technically accurate model can still be irresponsibly deployed.

These perspectives are complementary. A governance review should ask:

• What outcomes will the system create?
• Which rights and duties are engaged?
• Who gains and who bears the risks?
• Does the organisation demonstrate honesty, care and professional competence?

2. Governance as an Operating System for Analytics

Analytics governance is the allocation of decision rights, responsibilities, controls and escalation routes for analytical assets. It is broader than data governance. Data governance concerns the quality, ownership, access and use of data; model governance adds control over assumptions, validation, performance and change; cybersecurity governance protects confidentiality, integrity, availability and safety.

A useful governance architecture has five layers.

2.1 Accountability and the three-lines model

A practical allocation of responsibility is:

• First line: operational and analytics teams own risks in day-to-day work.
• Second line: risk, compliance, privacy and security specialists set standards and challenge decisions.
• Third line: internal audit independently evaluates governance effectiveness.

The board or executive committee remains accountable for material risk. A data scientist should not be the sole owner of a system that affects employment, safety or legal rights. Every high-impact system should have a named business owner, technical owner, data owner, security owner and accountable executive.

2.2 The governance register

A governance register should record each material analytical use case. Minimum fields include purpose, affected stakeholders, data categories, legal basis, model type, decision impact, owner, vendor dependencies, known limitations, performance thresholds, review date and retirement conditions. The register makes invisible systems visible and prevents “shadow analytics” from operating without oversight.

3. Data Governance and Responsible Data Use

3.1 Data lifecycle

The data lifecycle comprises collection, ingestion, storage, preparation, use, sharing, archiving and deletion. Controls should apply at each stage. Data that are accurate for one purpose may be inappropriate for another. A telematics stream collected for vehicle maintenance should not automatically be repurposed for individual worker discipline.

Important data-governance principles include:

• Purpose limitation: collect and use data for specified legitimate purposes.
• Data minimisation: collect the minimum information necessary.
• Quality and provenance: document source, time, transformations and uncertainty.
• Access limitation: grant access according to role and business need.
• Retention limitation: delete or anonymise data when the legitimate need ends.
• Integrity: prevent unauthorised alteration and preserve audit trails.
• Interoperability: use consistent definitions and documented interfaces.

3.2 Data quality and uncertainty

Data quality is multidimensional. Accuracy, completeness, timeliness, consistency, uniqueness and validity should be assessed separately. Missing delivery scans may not be random: they may occur more often in rural areas, during system outages or among subcontractors. Treating missingness as harmless can produce biased forecasts and unfair performance comparisons.

A useful quality score should not conceal dimensions through an unexplained average. Instead, a dashboard can report each dimension with thresholds and confidence statements. Analysts should distinguish measured facts from inferred values and communicate the uncertainty introduced by imputation, sampling or aggregation.

3.3 Privacy and consent

Logistics systems commonly process personal data such as names, addresses, signatures, vehicle identifiers, location traces, biometric access records and worker performance indicators. Privacy compliance requires more than placing a notice on a website. Organisations should identify the purpose, lawful authority, affected persons, retention period, recipients and rights of individuals.

Consent may be inappropriate where workers have limited ability to refuse. Other lawful bases may exist depending on jurisdiction, but legitimate interest or contractual necessity still requires proportionality and safeguards. Privacy impact assessments are appropriate for systematic monitoring, sensitive data, large-scale profiling and technologies that materially affect individuals.

3.4 Data sharing across supply chains

Data-sharing agreements should specify permitted purposes, security controls, breach notification, retention, onward transfers, audit rights, intellectual property, deletion and responsibility for data-subject requests. A party receiving data should not assume that the original collector had authority to use it for every secondary purpose.

4. Algorithmic Governance and Model Risk

An algorithm is not neutral merely because it is mathematical. It embodies choices about the target variable, training data, objective function, constraints and acceptable error. Governance should therefore begin before model training.

4.1 Model development controls

A model card or equivalent record should state:

• intended use and prohibited use;
• population and operating conditions;
• training-data sources and time period;
• features and transformations;
• performance measures and uncertainty;
• fairness and robustness tests;
• human decision points;
• known failure modes;
• version, approval and change history.

Validation must be independent of model development where the risk justifies it. Random train-test splits can be misleading in logistics because data are temporal and spatial. A demand model should normally be tested on later periods, and a routing or ETA model should be tested across locations, seasons, weather conditions and operational regimes.

4.2 Performance measures

A forecast may be evaluated with mean absolute error:

Here, yᵢ is the observed value, ŷᵢ is the forecast, and n is the number of observations. MAE has the same units as the target, such as parcels or minutes, and is easy to interpret. It gives equal weight to absolute errors.

Mean squared error is:

MSE is expressed in squared units and penalises large errors more strongly. Neither measure alone establishes that a model is fit for purpose. A model can have a low average error but fail dangerously in peak periods or for a particular region.

For binary classification, precision is:

Recall is:

Here, TP, FP and FN denote true positives, false positives and false negatives. The appropriate balance depends on the decision. In cargo-security screening, missing a genuine threat may be more serious than creating an additional inspection; in customer-service triage, excessive false alarms may waste scarce capacity.

4.3 Fairness assessment

Fairness must be defined in relation to the decision and affected groups. Possible assessments include comparison of error rates, selection rates, calibration and access to appeal. Statistical parity may be inappropriate when groups have different legitimate risk distributions, while equal error rates may be difficult to satisfy simultaneously with calibration. Therefore, fairness is a socio-technical judgement, not a single score.

Analysts should investigate proxy variables. A postcode may encode wealth, ethnicity or access to transport; a historical delay measure may encode previous discrimination or infrastructure inequality. Removing a sensitive attribute does not remove discrimination if correlated variables remain.

4.4 Explainability and contestability

Explainability has at least three audiences: engineers need technical explanations, managers need operational explanations, and affected persons need understandable reasons and routes of appeal. A global feature-importance chart does not explain every individual decision. Explanations should identify relevant factors without exposing security-sensitive information or implying false certainty.

A contestable system permits a person to challenge a decision, obtain review by a competent human and have errors corrected. Contestability is especially important when analytics affects employment, access to services, insurance, credit, carrier opportunities or regulatory treatment.

5. Cybersecurity and Cyber-Physical Resilience

Logistics increasingly links enterprise systems with operational technology: warehouse-control systems, automated storage and retrieval, robots, scanners, vehicles, ports, refrigeration and building controls. A cyber incident can therefore become a physical safety or continuity event.

The core security objectives are:

• Confidentiality: prevent unauthorised disclosure.
• Integrity: prevent unauthorised or undetected alteration.
• Availability: maintain service when needed.
• Authenticity: verify identities and message sources.
• Safety: prevent digital actions from causing unacceptable physical harm.
• Resilience: continue essential operations and recover quickly.

Controls should include network segmentation, multifactor authentication, least privilege, secure software updates, vulnerability management, encryption where appropriate, immutable backups, supplier assurance, logging and rehearsed incident response. Operational technology requires special care: a patch that is safe in an office environment may interrupt a conveyor or automated crane.

A risk assessment can use expected loss as a simple prioritisation tool:

Here, pⱼ is the estimated probability of scenario j, and Iⱼ is its impact, measured in monetary, operational, safety or social terms. This is not a prediction of actual loss and should not be used to monetise dignity or life mechanically. Low-probability catastrophic events may require controls even when their numerical expected value appears modest.

Incident response should define detection, containment, communication, recovery and post-incident learning. Contracts must require suppliers to report incidents promptly and preserve forensic evidence.

6. Regulatory and Institutional Compliance

Regulatory obligations vary by jurisdiction and application. Relevant domains commonly include data protection and privacy, employment law, consumer protection, competition law, transport safety, product liability, customs and trade controls, records management, cybersecurity and environmental reporting. Artificial-intelligence regulation increasingly uses risk-based categories, with stronger requirements for systems affecting safety or fundamental rights.

Compliance should be integrated into design rather than treated as a final legal check. A compliance-by-design process asks:

1. What jurisdictions and regulators apply?
2. What data and decisions are involved?
3. Is the system high impact or safety critical?
4. What documentation, notice, human oversight and assessment are required?
5. Which party is responsible when a vendor supplies the model?
6. How will changes trigger reassessment?

Certification can provide evidence of control but does not transfer accountability. Contractual language such as “the vendor is responsible for compliance” cannot eliminate the customer’s duties toward workers, customers or regulators.

7. Workforce, Social and Environmental Impacts

Analytics can improve work by reducing repetitive tasks, supporting safer routes and identifying equipment hazards. It can also intensify work through unrealistic targets, continuous monitoring, algorithmic scheduling and reduced discretion. The relevant unit of analysis is not only the organisation but the wider labour and community system.

7.1 Workforce transition

Automation changes task composition rather than simply eliminating or preserving jobs. New roles may emerge in data stewardship, robot supervision, cybersecurity, model validation and exception management. Effective transition programmes combine skills analysis, paid training, transparent selection for redeployment and worker participation in system design.

Human oversight must be meaningful. A human who can only approve an algorithmic recommendation under severe time pressure is not exercising genuine oversight. The person must have authority, competence, adequate information and protection from retaliation when escalating a concern.

7.2 Social distribution

Optimisation can externalise costs. A model minimising delivery time may increase neighbourhood congestion; a warehouse productivity model may shift injury risk to temporary workers; dynamic pricing may disadvantage remote communities. Impact assessments should examine distribution by geography, employment status, gender, disability, income and other relevant dimensions while respecting privacy.

7.3 Environmental governance

Analytics can reduce fuel use, empty running, spoilage and energy consumption. Yet computational infrastructure, increased delivery frequency and rebound effects may offset gains. Environmental metrics should therefore use a system boundary and distinguish absolute reduction from intensity reduction. For example, emissions per parcel may fall while total emissions rise if shipment volume increases substantially.

8. Autonomous and Self-Optimising Logistics

Autonomous logistics ranges from decision support to systems that sense conditions, choose actions and execute them with limited human intervention. Examples include autonomous mobile robots, automated ports, driver-assistance systems, dynamic warehouse control and reinforcement-learning dispatchers.

A maturity ladder is useful:

1. Descriptive: reports what happened.
2. Diagnostic: identifies why it happened.
3. Predictive: estimates what may happen.
4. Prescriptive: recommends actions.
5. Supervised autonomous: executes within human-defined boundaries.
6. Self-optimising: changes policies or parameters in response to conditions.

Risk increases when autonomy, speed, scale and irreversibility increase. A recommendation affecting one shipment can be reviewed; an autonomous policy affecting thousands of shipments may propagate an error before detection.

8.1 Safe autonomy principles

High-impact autonomous systems should have:

• explicit operating boundaries and prohibited states;
• fail-safe or graceful-degradation modes;
• independent monitoring of input drift and output anomalies;
• human override that is technically feasible and tested;
• simulation and stress testing before deployment;
• staged rollout and rollback capability;
• separation between learning and execution permissions;
• logs sufficient to reconstruct decisions;
• defined accountability when multiple agents interact.

A self-learning system should not modify its objective function or safety constraints without controlled approval. Adaptive behaviour is not a substitute for governance.

8.2 Digital twins and synthetic environments

Digital twins can test warehouse layouts, port operations and disruption scenarios before physical implementation. Their results depend on assumptions, data quality and model fidelity. A twin is not reality; it is a representation with a validity domain. Decisions should document what the twin excludes, such as informal workarounds, extreme weather or human fatigue.

9. A Responsible Analytics Implementation Framework

A practical stage-gate process is as follows.

Gate 1: Define the problem

State the operational objective, affected stakeholders, alternatives and unacceptable harms. Confirm that analytics is necessary; sometimes a process redesign is safer and cheaper.

Gate 2: Classify impact

Rate the system according to privacy, safety, employment, financial, environmental and systemic consequences. High-impact systems receive stronger review.

Gate 3: Assess data and rights

Document provenance, lawful use, data minimisation, retention, sharing and individual rights. Complete privacy and equality assessments where required.

Gate 4: Develop and validate

Use appropriate temporal and geographic tests. Assess robustness, uncertainty, fairness, security and failure modes. Have an independent reviewer challenge assumptions.

Gate 5: Pilot with safeguards

Use limited scope, monitoring, human review and a rollback plan. Engage workers, customers, carriers or communities affected by the system.

Gate 6: Deploy and monitor

Track technical metrics and real-world outcomes. Monitor drift, incidents, override rates, complaints, distributional effects and near misses.

Gate 7: Review, renew or retire

Reassess after material changes, incidents, regulatory developments or changes in operating conditions. Retire systems that no longer have a legitimate purpose or cannot meet safety and fairness requirements.

10. Illustrative Case: Algorithmic Driver Scheduling

Background

A national parcel company introduces a predictive scheduling system for 3,000 delivery drivers. The system uses historical route duration, parcel volume, traffic, customer time windows and driver availability to assign routes and estimate completion time. Management expects a 10% reduction in overtime.

Problem

After deployment, average productivity improves, but injury reports and worker complaints increase. Drivers in rural areas receive lower performance scores because travel conditions are less predictable. New drivers are assigned demanding routes because the model interprets limited experience as availability. The company also discovers that vehicle-location data are retained indefinitely and accessible to several managers.

Analysis

The system has at least four risk dimensions. First, the target—completed stops per shift—may reward speed while ignoring safety and route difficulty. Second, historical data contain structural differences between urban and rural operations. Third, continuous location monitoring creates privacy and employment-law concerns. Fourth, the model’s recommendation is treated as mandatory, so human oversight is nominal.

A governance review should compare outcomes by route type and worker group, examine false assumptions in the target variable, assess injury and near-miss rates, and verify whether data collection is proportionate. The company should consult worker representatives and investigate whether overtime reduction is being achieved by transferring workload or risk.

Approach and solution

The company redesigns the objective as a constrained multi-criteria decision rather than a single productivity target. Safety incidents, legal driving limits, realistic service windows and route complexity become hard constraints. Productivity is optimised only within those boundaries. Route difficulty is estimated using observable operational factors rather than protected or proxy attributes. New drivers receive a protected ramp-up period.

Location data are minimised, access is role-based, and retention is limited to operational and legal needs. Drivers receive an explanation of route factors, a channel to report inaccurate conditions and a right to human review. The pilot is conducted in two regions with independent monitoring. The model is not allowed to update itself during the pilot; changes require approval.

Outcome

The pilot produces a smaller overtime reduction than the original forecast but lowers severe schedule overruns and improves worker acceptance. Rural performance comparisons become more meaningful after route complexity is incorporated. Management also identifies that some overtime was caused by unrealistic customer time windows, a process issue that the original algorithm had concealed.

Learning points

1. The objective function expresses organisational values.
2. Productivity metrics can create unsafe incentives.
3. Fairness requires examination of context and opportunity, not merely removal of sensitive variables.
4. Privacy and security controls must cover operational data after collection.
5. Worker participation improves both legitimacy and model quality.
6. A slower deployment with credible oversight can produce better long-term value.

11. Metrics for Governance Effectiveness

Governance should be measured through a balanced set of indicators.

Metrics should not become new targets detached from purpose. Gaming is possible whenever a measure is rewarded without considering its side effects. Governance committees should therefore review leading indicators, lagging outcomes and qualitative evidence together.

Challenges and Limitations

Responsible governance faces persistent limitations. Laws differ across countries and may not clearly address new forms of algorithmic decision-making. Fairness criteria can conflict. Explanations may be incomplete for complex models. Small carriers and subcontractors may lack resources to meet sophisticated assurance demands. Cybersecurity is never absolute, and supply-chain dependencies make attribution difficult. Excessive approval processes can suppress beneficial innovation, while weak controls can produce harm at scale.

There are also epistemic limits. Historical data describe previous operations, not all possible futures. A model trained during stable conditions may fail during pandemics, geopolitical shocks or extreme weather. Human judgement is not automatically unbiased, but replacing it with automation can make errors less visible and harder to contest. Governance must therefore preserve humility, experimentation under controlled conditions and the ability to stop a system.

Best-Practice Checklist

Before approving a high-impact logistics analytics system, an organisation should be able to answer “yes” to the following:

• Is the purpose legitimate, specific and necessary?
• Are affected stakeholders identified and consulted?
• Are data provenance, quality, access and retention documented?
• Has privacy, fairness, safety and cybersecurity risk been assessed?
• Are performance and limitations tested across time, geography and operating conditions?
• Is there a named accountable owner and independent challenge?
• Can affected persons obtain understandable reasons and human review?
• Are human override, rollback and incident-response procedures tested?
• Are vendors contractually and technically governed?
• Are workforce, community and environmental effects monitored?
• Are changes, drift and retirement conditions defined?

Chapter Summary

Ethical logistics analytics requires more than accurate prediction. It requires legitimate purpose, respect for rights, fair distribution of benefits and burdens, professional integrity and accountable human judgement. Governance should integrate data quality and provenance, privacy, algorithmic validation, fairness assessment, cybersecurity, operational safety, regulatory compliance and stakeholder participation.

High-impact systems should be managed through a lifecycle of classification, assessment, independent validation, controlled piloting, continuous monitoring and retirement. Performance measures must be interpreted in context because an apparently efficient target can intensify work, shift risk or externalise environmental harm. Autonomous and self-optimising systems require explicit boundaries, fail-safe behaviour, meaningful human oversight, auditability and rollback capability. The future of logistics analytics is therefore not determined by technical capability alone; it depends on whether organisations can make innovation trustworthy, contestable and socially sustainable.

Key Terms

Algorithmic accountability: Responsibility for the design, operation, effects and correction of an algorithmic system.

Algorithmic fairness: Evaluation of whether an analytical system produces unjustified disparities in treatment or outcomes.

Audit trail: A chronological record that permits actions, data changes and decisions to be reconstructed.

Contestability: The ability of an affected person or organisation to challenge and obtain review of a decision.

Data minimisation: Limiting data collection to what is necessary for a specified legitimate purpose.

Data provenance: Information about the origin, transformations, custody and use of data.

Digital twin: A digital representation of a physical asset, process or system used for analysis and simulation.

Drift: Change in data, relationships or operating conditions that reduces model validity.

Explainability: The ability to communicate how or why an analytical result was produced.

Fairness assessment: A structured examination of differential treatment, errors and outcomes across relevant groups.

Human oversight: Competent and authorised human involvement capable of understanding, challenging or overriding a system.

Model governance: Policies and controls governing model development, validation, deployment, monitoring and retirement.

Operational technology: Hardware and software that monitors or controls physical processes.

Purpose limitation: Restriction of data use to specified and legitimate purposes.

Risk appetite: The amount and type of risk an organisation is willing to accept.

Self-optimising system: A system that adapts policies, parameters or actions in response to changing conditions.

Shadow analytics: Unauthorised or unregistered analytical activity outside formal governance.

Supply-chain cyber resilience: The ability to prevent, withstand, respond to and recover from cyber incidents across connected partners.

Systemic impact: Effects that propagate across organisations, markets, communities or infrastructure.

Model card: A structured document describing a model’s purpose, data, performance, limitations and governance status.

Review Questions

Short Answer Questions

1. Why is analytics governance broader than data governance?
2. Define purpose limitation and explain its relevance to telematics data.
3. What is meant by meaningful human oversight?
4. State two reasons why a model can perform well on average but fail operationally.
5. Identify three cybersecurity objectives relevant to cyber-physical logistics systems.

Long Answer Questions

1. Explain how consequentialist, rights-based and justice-based ethics may lead to different evaluations of a routing algorithm.
2. Design a governance architecture for a high-impact warehouse automation system.
3. Discuss the relationship between data quality, model validity and fairness in carrier selection.
4. Evaluate the principal governance requirements for self-optimising logistics systems.
5. Analyse how logistics analytics may improve productivity while worsening workforce or community outcomes.

Analytical / Application Questions

1. A delivery model has low overall MAE but much higher error in rural regions. Propose an investigation, a set of fairness and operational metrics, and a deployment decision.
2. A port operator wants to connect autonomous cranes to a cloud-based optimisation platform. Develop a risk-based cybersecurity, safety and rollback plan.
3. A company uses driver location data for routing, performance management and disciplinary action. Assess the governance issues and recommend a lawful, ethical data-use design.

Multiple Choice Questions with Answers

Question 1

Which governance mechanism assigns a named executive responsibility for a high-impact analytics system?

A. A data dictionary

B. An accountability structure

C. A random train-test split

D. A network firewall

Correct Answer: B. An accountability structure

Question 2

What does purpose limitation require?

A. Using all available data to maximise accuracy

B. Retaining data indefinitely for future analysis

C. Using data only for specified legitimate purposes

D. Sharing data with every supply-chain partner

Correct Answer: C. Using data only for specified legitimate purposes

Question 3

Which metric has the same units as the forecast target?

A. Mean absolute error

B. Mean squared error

C. Variance

D. A squared correlation coefficient

Correct Answer: A. Mean absolute error

Question 4

In the formula Precision = TP ÷ (TP + FP), what does FP represent?

A. Correct positive predictions

B. Missed positive cases

C. False positive predictions

D. The total number of observations

Correct Answer: C. False positive predictions

Question 5

Why can removing a sensitive attribute fail to eliminate algorithmic discrimination?

A. Models cannot process numerical data

B. Other variables may act as proxies

C. Validation always removes fairness

D. Sensitive attributes are never correlated with outcomes

Correct Answer: B. Other variables may act as proxies

Question 6

Which control most directly supports recovery after ransomware affects a warehouse system?

A. Immutable, tested backups

B. A larger training dataset

C. A customer satisfaction survey

D. A revised data dictionary

Correct Answer: A. Immutable, tested backups

Question 7

What is a principal risk of using historical productivity as the sole target for driver scheduling?

A. It necessarily improves safety

B. It may reward unsafe speed and reproduce historical bias

C. It prevents all model drift

D. It eliminates the need for worker consultation

Correct Answer: B. It may reward unsafe speed and reproduce historical bias

Question 8

Which feature is essential for a contestable analytical decision?

A. A prohibition on human review

B. An understandable reason and an appeal route

C. Permanent retention of every personal datum

D. Automatic deployment without monitoring

Correct Answer: B. An understandable reason and an appeal route

Question 9

A forecast has observed values 100 and 120 and forecasts 90 and 130. What is its MAE?

A. 5 units

B. 10 units

C. 20 units

D. 100 units

Correct Answer: B. 10 units

Question 10

Which statement best describes a self-optimising logistics system?

A. It only reports historical performance

B. It adapts policies or parameters as conditions change

C. It cannot operate without manually changing every route

D. It is exempt from safety constraints

Correct Answer: B. It adapts policies or parameters as conditions change