14. Digital, AI and Autonomous Planning

Introduction: From Periodic Planning to Intelligent Coordination

Supply chain planning has traditionally depended on periodic forecasts, manually maintained spreadsheets, enterprise resource planning (ERP) transactions, and meetings in which planners reconcile conflicting information. These practices remain important, but they are increasingly insufficient. Supply networks now contain more products, suppliers, channels, markets, and decision points than human planners can monitor continuously. Customer orders may change within hours, transport capacity can disappear unexpectedly, and operational data is generated by sensors, platforms, applications, and external information services.

Digital planning addresses this complexity by connecting data, models, systems, and people. Artificial intelligence (AI) extends this capability by identifying patterns, generating predictions, recommending actions, and, under controlled conditions, executing decisions. Autonomous planning goes further: it creates a decision system that can sense changes, evaluate alternatives, take approved actions, and learn from outcomes with limited human intervention.

Autonomy, however, is not the same as the absence of people. Supply chain decisions involve commercial priorities, contractual commitments, safety, sustainability, labour, regulation, and reputation. These dimensions cannot be delegated automatically merely because a mathematical model produces a recommendation. The central managerial challenge is therefore to design an effective relationship between technology and professional judgement. This chapter examines that relationship, the digital architecture that supports it, and the governance required to make AI-enabled planning reliable and responsible.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the evolution from ERP-based planning to cloud, AI-enabled, and autonomous planning.

2. Distinguish ERP, advanced planning and optimisation (APS), cloud platforms, IoT, digital twins, demand sensing, and generative AI.

3. Evaluate how digital technologies improve forecasting, inventory, capacity, replenishment, and sales and operations planning.

4. Apply a structured framework for designing human-in-the-loop autonomous planning.

5. Assess cybersecurity, privacy, model risk, fairness, explainability, and accountability issues.

6. Develop a practical transformation roadmap for a logistics or supply chain organisation.

The Digital Planning Landscape

A planning system can be understood as a combination of data, decision logic, workflows, users, and execution interfaces. Digital transformation occurs when these elements are integrated rather than operated as isolated tools. A forecasting application that cannot access reliable inventory or order data may produce sophisticated but unusable forecasts. Conversely, an ERP with accurate transactions may still provide weak decisions if it cannot model uncertainty or optimise competing objectives.

ERP as the transactional backbone

An ERP system records business transactions and provides a common data environment for finance, procurement, production, inventory, sales, and order management. In planning, ERP data commonly includes material masters, bills of material, purchase orders, sales orders, inventory balances, lead times, suppliers, and production information.

ERP systems provide control and traceability. They establish the official record of what was ordered, received, produced, shipped, or invoiced. Their traditional limitation is that they are primarily designed to record and execute transactions, not necessarily to explore thousands of future scenarios. Data may also be stored at a level of aggregation unsuitable for advanced forecasting, and master-data errors can spread rapidly across dependent processes.

Advanced planning and optimisation

APS applications complement ERP by using forecasting, finite-capacity planning, inventory optimisation, network design, scheduling, and optimisation algorithms. An APS may answer questions such as: Which plant should supply a region? How much safety stock is appropriate for a service target? What production sequence minimises changeover costs? Which orders should receive scarce capacity?

Optimisation differs from simple automation. Automation follows a predefined rule, whereas optimisation evaluates alternatives against an objective function and constraints. A model may minimise total cost subject to service, capacity, shelf-life, labour, and transportation constraints. The quality of its answer depends on the accuracy of the data, the realism of the constraints, and the appropriateness of the objective function.

Cloud planning platforms

Cloud platforms provide shared computing infrastructure, software services, data storage, and integration capabilities through networks rather than exclusively through locally installed systems. Their value lies not merely in reducing hardware ownership. Cloud environments can support rapid deployment, elastic computing, collaborative access, frequent software updates, and integration with external data and AI services.

A cloud planning platform may connect suppliers, manufacturers, logistics providers, retailers, and customers through common workflows. However, cloud adoption creates important questions regarding data residency, access rights, vendor dependence, service availability, integration costs, and exit arrangements. Moving a poor process to the cloud does not make it a good process.

IoT and event-driven visibility

The Internet of Things (IoT) refers to connected devices that collect and transmit data. In logistics, sensors may monitor vehicle location, temperature, humidity, shock, engine condition, energy use, or warehouse movement. Radio-frequency identification, telematics, barcode systems, and automated equipment generate additional event data.

IoT changes planning from periodic reporting to event-driven management. A temperature excursion may trigger a quality review; a delayed vehicle may cause a replanning event; a machine vibration pattern may indicate an impending failure. The planning challenge is to distinguish meaningful signals from noise. High-frequency data is not automatically high-quality data, and excessive alerts can produce decision fatigue.

Digital twins

A digital twin is a dynamic digital representation of a physical asset, process, facility, or network that is updated with relevant data. A supply chain digital twin can represent suppliers, lanes, facilities, inventory, capacities, lead times, demand, and policies. It allows managers to test scenarios without immediately disrupting physical operations.

A useful twin is not simply a visual dashboard. It contains relationships and behavioural assumptions that support simulation or decision analysis. For example, it may estimate how a port delay affects production schedules, customer service, working capital, and alternative transport costs. The twin is only as credible as its data, assumptions, update frequency, and validation against actual outcomes.

From Forecasting to Demand Sensing

Conventional statistical forecasting uses historical demand to estimate future demand. Methods may include moving averages, exponential smoothing, regression, intermittent-demand models, and time-series techniques. These remain valuable because they are transparent, efficient, and often effective for stable demand.

Demand sensing supplements the baseline forecast with near-real-time signals. These may include point-of-sale data, online searches, promotions, weather, mobility, social activity, competitor information, order cancellations, inventory availability, and transportation events. The objective is not necessarily to replace the statistical forecast but to identify short-term deviations that matter operationally.

For example, a baseline forecast may indicate stable weekly demand for a consumer product, while retail sell-through data shows an unexpected regional acceleration. A demand-sensing model can adjust the near-term forecast, allowing replenishment and transport plans to respond earlier. Yet demand sensing can also amplify temporary noise, promotional distortions, or biased data. It should therefore use confidence thresholds, exception rules, and human review for material changes.

Forecast accuracy must be assessed at the level at which decisions are made. Measures such as mean absolute error, weighted absolute percentage error, forecast bias, and service performance are useful, but no single metric is sufficient. A highly accurate aggregate forecast may conceal unacceptable errors for critical stock-keeping units. Similarly, improving forecast accuracy is not valuable if it increases inventory or reduces customer service through poor execution. Digital planning should connect forecast metrics with business outcomes.

Artificial Intelligence in Planning

AI in supply chain planning includes machine learning, deep learning, probabilistic modelling, optimisation, natural-language processing, computer vision, and generative AI. Machine-learning models identify relationships in data and can forecast demand, estimate lead-time risk, classify exceptions, detect anomalies, and predict equipment failure. Optimisation models select actions under constraints. These approaches are complementary rather than interchangeable.

A planning system may use machine learning to predict demand, an optimisation engine to allocate inventory, and a rules engine to enforce safety or regulatory constraints. The architecture should make these roles clear. A model that predicts a late shipment does not, by itself, determine whether the organisation should expedite, reroute, substitute, or inform the customer.

Generative AI and the planning profession

Generative AI creates text, summaries, explanations, code, scenarios, and other content from prompts and contextual information. In planning, it can summarise exceptions, explain forecast changes, prepare meeting briefs, translate supplier communications, assist with data queries, draft scenario narratives, and provide a natural-language interface to planning systems.

Its greatest near-term value may be augmentation rather than unsupervised decision execution. A planner could ask, “Which products are at risk of stockout in the next fourteen days, and what are the lowest-cost mitigation options?” The system can retrieve data, apply approved models, and present alternatives. The planner remains responsible for validating assumptions and authorising action.

Generative AI can produce fluent but inaccurate responses. It may misinterpret ambiguous terminology, rely on incomplete data, expose confidential information, or present an invented explanation with excessive confidence. Retrieval from approved enterprise data, source citation, access controls, output validation, and logging are therefore essential. A language model should not be treated as an authoritative planning database merely because its response is persuasive.

A Framework for Autonomous Planning

Autonomous planning can be represented as a closed-loop process:

1. Sense: collect internal transactions, external signals, sensor data, and execution events.

2. Understand: cleanse, reconcile, contextualise, and interpret the data.

3. Predict: estimate demand, supply, lead times, risks, and likely outcomes.

4. Decide: compare feasible actions against objectives and constraints.

5. Act: release orders, modify schedules, allocate inventory, or initiate approved workflows.

6. Learn: compare results with predictions, investigate deviations, and improve models and policies.

This loop should not be viewed as a licence for unrestricted automation. Each decision requires an autonomy level. Low-risk, repetitive decisions may be executed automatically. Medium-risk decisions may be recommended by the system and approved by a planner. High-risk decisions should require senior or cross-functional authorisation.

A practical autonomy matrix considers four factors: financial impact, operational reversibility, customer or safety consequence, and regulatory sensitivity. For example, automatically adjusting a replenishment quantity within an approved range may be acceptable. Automatically reallocating scarce medical products or changing a regulated product’s handling conditions requires stronger controls.

The human-in-the-loop design should specify who reviews an exception, what evidence is displayed, how much time is available, what thresholds trigger escalation, and who can override the system. Human involvement must be meaningful. If planners receive thousands of alerts or are expected to approve recommendations without adequate explanation, the organisation has created rubber-stamping rather than oversight.

Transformation Roadmap

Digital transformation should begin with a business problem, not a technology catalogue. A structured roadmap can proceed through the following stages.

Establish the planning ambition

The organisation should define whether it seeks improved forecast accuracy, lower inventory, faster response, better resilience, reduced manual work, or integrated partner collaboration. Objectives should be measurable and connected to service, cost, working capital, risk, and sustainability.

Assess process and data readiness

Teams should map current planning processes, decision rights, systems, interfaces, manual workarounds, and pain points. Data profiling should examine completeness, timeliness, consistency, accuracy, uniqueness, and lineage. Particular attention is needed for item masters, location codes, units of measure, calendars, lead times, supplier attributes, and inventory status.

Build a governed data and integration foundation

APIs, event streams, data warehouses, data lakes, and master-data management may be used to connect systems. Integration design should distinguish systems of record from analytical copies and should preserve data lineage. A data owner should be responsible for important data domains, while stewardship roles manage quality and definitions.

Pilot a bounded use case

A good pilot has a clear decision, measurable value, manageable risk, and accessible data. Examples include demand sensing for a selected product family, predictive ETA for a transport lane, or inventory optimisation for a regional network. The pilot should use a baseline, a control comparison where possible, and a defined process for human review.

Scale through operating-model change

Scaling requires more than installing software. Planners need training in data interpretation, model limitations, exception management, and ethical use. Roles may shift from manually preparing forecasts to investigating exceptions and managing policies. Performance measures should avoid penalising planners for appropriately challenging an algorithm.

Institutionalise governance and continuous improvement

A model register, approval process, monitoring dashboard, incident procedure, and periodic review should be established. Models must be retrained or recalibrated when demand patterns, product portfolios, suppliers, regulations, or business strategies change. Retirement is also part of governance; obsolete models should not remain active simply because no one owns them.

Risk, Security, Privacy, and Responsible AI

Digital planning expands the attack surface of the supply chain. Threats include compromised credentials, ransomware, malicious software updates, manipulated sensor data, insecure APIs, supplier access abuse, denial of service, and theft of commercially sensitive information. A cyberattack on planning data can create physical consequences, including incorrect replenishment, unsafe production schedules, or disrupted deliveries.

Security should follow defence-in-depth principles: identity and access management, least-privilege permissions, multi-factor authentication, network segmentation, encryption, secure development, vulnerability management, backups, incident response, and supplier assurance. Critical planning decisions should have continuity procedures for system outages. Organisations should know which activities can continue manually and for how long.

Privacy arises when data can identify individuals or reveal sensitive behaviour. Driver location, employee performance, customer purchasing, and household-level demand information may require lawful collection, limited use, retention controls, and appropriate transparency. Data minimisation is important: a model should not collect personal information merely because it is technically available.

Model governance addresses the lifecycle of analytical and AI systems. It includes documented purpose, data sources, assumptions, performance measures, validation, approval, monitoring, change control, and accountability. Important controls include:

• Accuracy and robustness: Does the model perform adequately across relevant conditions?
• Explainability: Can users understand the main factors behind a recommendation?
• Fairness: Does the system systematically disadvantage suppliers, regions, customers, or worker groups without legitimate justification?
• Human accountability: Is a named decision-maker responsible for outcomes?
• Auditability: Are inputs, outputs, overrides, and system versions recorded?
• Contestability: Can affected users challenge or correct a decision?

Fairness in supply chain planning is context-dependent. An allocation model that prioritises high-margin customers may be commercially rational but could create unacceptable consequences for vulnerable communities or contractual partners. Fairness does not always require identical treatment; it requires explicit, defensible criteria and awareness of disparate effects.

Applied Case: Northstar Foods’ Intelligent Replenishment Programme

Background

Northstar Foods, an illustrative regional food manufacturer, supplies supermarkets through two distribution centres and several contract carriers. It experiences frequent forecast revisions, excess stock of slow-moving products, and stockouts during promotions. Planning is based on an ERP system, spreadsheets, and weekly meetings. Retail point-of-sale data is available but arrives with inconsistent product and location codes.

Situation and problem

Senior management proposes an AI planning platform that will automatically change replenishment quantities. The operations director supports the investment, while the supply planning manager warns that poor master data and promotional information may produce unstable recommendations. The information-security team is concerned about external access to customer and supplier data. Planners fear that automation will make their expertise irrelevant.

Analysis

A diagnostic review finds that the main problem is not the absence of an advanced algorithm. Several products have inconsistent units of measure, supplier lead times are not regularly reviewed, and promotions are recorded differently by sales teams. The ERP inventory position is reliable for finished goods but less reliable for goods in transit. Historical data also contains periods of product unavailability, meaning observed sales understate true demand.

The organisation therefore decides that a fully autonomous system would be inappropriate as an initial target. It defines three objectives: improve near-term availability for promoted products, reduce avoidable excess inventory, and create a traceable exception process.

Approach and solution

Northstar first establishes data ownership and standardises product, customer, location, and calendar definitions. It creates a controlled interface for point-of-sale data and marks periods of stockout so that the forecasting model can distinguish low demand from unavailable supply.

The pilot covers one product family and two distribution centres. A demand-sensing model produces a fourteen-day adjustment to the statistical baseline. An optimisation model recommends replenishment quantities subject to shelf-life, minimum order, supplier capacity, and service constraints. Recommendations outside predefined limits require planner approval. The system explains the principal drivers of each material change, such as a promotion, sales acceleration, delayed inbound order, or inventory imbalance.

A dashboard tracks forecast error, bias, service level, inventory days, overrides, and the proportion of recommendations accepted. Planners receive training and are evaluated partly on the quality of exception management rather than on accepting every recommendation. Cybersecurity controls restrict access by role, and all changes are logged.

Outcome

In the illustrative case, the pilot demonstrates that the largest early gains come from data standardisation, better promotional visibility, and focused exception management. The model performs well for stable products but is less reliable for newly launched items and unusual promotions. Northstar therefore retains human approval for those categories and postpones automatic execution. The company then expands the pilot gradually, adding supplier collaboration and transport-event data.

Learning points

The case illustrates five principles. First, AI cannot compensate for undefined data and processes. Second, a pilot should target a bounded decision rather than promise total transformation. Third, autonomy must be matched to risk. Fourth, adoption depends on redesigning roles and incentives. Finally, model performance must be judged by business outcomes and operational reliability, not by forecast accuracy alone.

Common Failure Modes and Managerial Implications

One common mistake is technology-first procurement: selecting a platform before defining the decision problem. Another is treating integration as a technical afterthought. If identifiers, calendars, units, and ownership are inconsistent, the system may create an appearance of precision while reproducing basic errors.

Organisations also frequently automate exceptions without understanding them. A planner may override a recommendation because of a customer commitment or a supply constraint not represented in the data. Recording the override without investigating its cause turns valuable expertise into unexplained noise. Overrides should be analysed as feedback about missing data, flawed rules, or legitimate business judgement.

Another risk is metric conflict. A model optimised for inventory reduction may damage service; a model optimised for utilisation may increase lateness; a model optimised for forecast accuracy may ignore profitability. Objectives should therefore be multi-dimensional and explicitly prioritised.

Change resistance should not be dismissed as irrational opposition. Planners may understand local conditions that the model does not capture, and they may be accountable for consequences the technology vendor does not bear. Involving users in design, testing, threshold setting, and review improves both system quality and professional acceptance.

Current Developments and Emerging Directions

Planning systems are moving toward composable architectures in which specialised services communicate through APIs and event streams. This makes it possible to combine ERP, forecasting, optimisation, logistics visibility, and AI services, but it also increases integration and governance complexity.

Generative AI copilots are likely to become more common in planning workspaces. Their useful applications include conversational analysis, automated meeting preparation, scenario explanation, and guided workflow execution. The key distinction is between generating a recommendation and authorising a decision. Organisations will increasingly require provenance, confidence indicators, policy checks, and approval records.

Digital twins are also becoming more operational as organisations connect simulation with live events. Their value will depend on calibration and decision relevance, not visual sophistication. In parallel, edge computing may permit some sensor-based decisions closer to warehouses, vehicles, or production equipment, reducing latency and dependence on central connectivity.

Sustainability is becoming a planning objective rather than a separate reporting activity. Carbon emissions, energy use, waste, packaging, and circular flows can be included in network, inventory, and transportation decisions. However, sustainability metrics must be transparent about boundaries and trade-offs; a system should not claim optimisation of environmental impact without defining how that impact is measured.

Best-Practice Principles

Effective digital and autonomous planning is guided by several principles:

1. Start with decisions and outcomes, not fashionable technologies.

2. Treat data quality, master data, and integration as strategic capabilities.

3. Use the simplest model that performs adequately and can be governed.

4. Match automation to risk, reversibility, and consequence.

5. Keep humans responsible for policy, exceptions, and high-impact decisions.

6. Make recommendations explainable, traceable, and contestable.

7. Monitor drift, bias, security events, overrides, and business outcomes.

8. Design for failure, including outages, bad data, cyberattacks, and model degradation.

9. Involve planners, operators, suppliers, and affected stakeholders in implementation.

10. Review the system continuously as products, markets, regulations, and strategies change.

Chapter Summary

Digital planning integrates transactional systems, analytical models, connected devices, cloud services, and human decision-making. ERP provides the transactional backbone, while APS supports forecasting, optimisation, scheduling, and scenario analysis. Cloud platforms improve collaboration and scalability; IoT creates event-driven visibility; digital twins support simulation; demand sensing updates forecasts using short-term signals; and generative AI assists with explanation, interaction, and workflow support.

Autonomous planning operates through a sense–understand–predict–decide–act–learn loop. Its success depends on assigning appropriate autonomy levels, maintaining meaningful human oversight, and connecting recommendations to approved policies and constraints. Transformation should progress from business objectives and data readiness through bounded pilots, operating-model redesign, and institutionalised governance.

Cybersecurity, privacy, fairness, explainability, auditability, and accountability are not optional additions. They are conditions for trustworthy planning. The most effective organisations use AI to augment professional judgement while retaining clear responsibility for consequential decisions.

Key Terms

Advanced Planning and Optimisation (APS): Software that supports forecasting, optimisation, scheduling, inventory planning, and supply network decisions.

Autonomous planning: A planning capability that senses conditions, generates decisions, executes approved actions, and learns with limited human intervention.

Cloud planning platform: A network-delivered planning environment providing shared computing, storage, applications, and collaboration services.

Data lineage: A record of where data originated, how it was transformed, and where it is used.

Demand sensing: The use of near-real-time signals to adjust a baseline demand forecast.

Digital twin: A dynamic digital representation of a physical asset, process, facility, or supply network.

ERP: An integrated system that records and coordinates core organisational transactions and resources.

Explainability: The extent to which users can understand the reasons for a model’s output or recommendation.

Generative AI: AI that creates new text, scenarios, code, summaries, or other content from instructions and contextual data.

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

Internet of Things (IoT): Connected devices and sensors that collect and exchange operational data.

Model drift: Declining model performance caused by changes in data, behaviour, markets, or operating conditions.

Model governance: Policies and controls covering the development, validation, use, monitoring, modification, and retirement of models.

Optimisation: The selection of the best feasible decision according to defined objectives and constraints.

Predictive analytics: The use of data and models to estimate future events, outcomes, or risks.

Responsible AI: The development and use of AI with attention to fairness, privacy, security, transparency, accountability, and social impact.

Supply chain control loop: The recurring process of sensing conditions, analysing them, deciding, acting, and learning from results.

Systems of record: Authoritative systems that maintain official transactional data.

Review Questions

Short Answer Questions

1. What is the principal difference between ERP and APS systems in supply chain planning?

2. Define demand sensing and explain how it complements a baseline forecast.

3. What is meant by a human-in-the-loop planning system?

4. Identify two cybersecurity risks associated with digital planning platforms.

5. Why is data lineage important in AI-enabled planning?

Long Answer Questions

1. Explain how ERP, cloud platforms, IoT, digital twins, and AI can be integrated into a digital planning architecture.

2. Describe the sense–understand–predict–decide–act–learn framework for autonomous planning.

3. Discuss the major stages of a digital planning transformation roadmap.

4. Evaluate the role of generative AI in planning and explain its principal limitations.

5. Discuss how model governance, privacy, fairness, explainability, and accountability support responsible AI in supply chains.

Analytical / Application Questions

1. A distributor has accurate ERP transactions but frequent stockouts during promotions. Design a bounded digital pilot, specifying data inputs, model outputs, human controls, and performance measures.

2. An optimisation model reduces inventory by prioritising high-margin customers during a shortage. Analyse the commercial, ethical, contractual, and fairness issues before authorising automatic execution.

3. A planning model’s forecast accuracy declines after a major product redesign and supplier change. Recommend a diagnostic and governance response, including data checks, model review, and operational continuity measures.

22. Multiple Choice Questions with Answers

Question 1

Which capability most directly distinguishes an APS from a conventional ERP system?

A. Recording invoices and financial transactions

B. Exploring constrained planning and optimisation alternatives

C. Maintaining employee payroll records

D. Storing historical purchase orders

Correct Answer: B. Exploring constrained planning and optimisation alternatives

Question 2

A demand-sensing system is most appropriately used to:

A. Replace all strategic forecasting permanently

B. Adjust a baseline forecast using recent demand and operational signals

C. Eliminate the need for inventory policies

D. Authorise every supply-chain decision without review

Correct Answer: B. Adjust a baseline forecast using recent demand and operational signals

Question 3

Which sequence best represents the autonomous planning control loop described in the chapter?

A. Act, sense, predict, learn, decide, understand

B. Sense, understand, predict, decide, act, learn

C. Predict, act, sense, decide, learn, understand

D. Understand, learn, act, sense, decide, predict

Correct Answer: B. Sense, understand, predict, decide, act, learn

Question 4

Why should a fully autonomous replenishment policy usually be avoided for a newly launched product?

A. New products never require forecasts

B. Historical data and behavioural patterns may be insufficient or unstable

C. ERP systems cannot record new products

D. Optimisation cannot include constraints

Correct Answer: B. Historical data and behavioural patterns may be insufficient or unstable

Question 5

Which practice provides the strongest protection against excessive human rubber-stamping of AI recommendations?

A. Increasing the number of alerts sent to planners

B. Removing explanation features to speed approval

C. Using risk-based thresholds and meaningful exception review

D. Measuring planners only by recommendation acceptance

Correct Answer: C. Using risk-based thresholds and meaningful exception review

Question 6

In the Northstar Foods case, why was data standardisation undertaken before scaling the AI pilot?

A. Because cloud platforms cannot process clean data

B. Because inconsistent identifiers and units could distort recommendations

C. Because human planners were prohibited from using data

D. Because demand sensing does not use transaction data

Correct Answer: B. Because inconsistent identifiers and units could distort recommendations

Question 7

Which control is most directly concerned with recording model inputs, outputs, overrides, and versions?

A. Auditability

B. Elastic computing

C. Demand sensing

D. Finite scheduling

Correct Answer: A. Auditability

Question 8

A generative-AI planning assistant produces a convincing but unsupported explanation. Which control is most relevant?

A. Source grounding, output validation, and logging

B. Increasing the model’s autonomy immediately

C. Removing access controls

D. Ignoring data provenance

Correct Answer: A. Source grounding, output validation, and logging

Question 9

What is the principal purpose of a digital twin in supply-chain planning?

A. To replace all physical operations with virtual operations

B. To simulate and evaluate scenarios using a dynamic representation of the network

C. To store only accounting transactions

D. To guarantee that every forecast is accurate

Correct Answer: B. To simulate and evaluate scenarios using a dynamic representation of the network

Question 10

A model reduces cost but systematically deprioritises a vulnerable customer group. Which responsible-AI concern is most directly raised?

A. Fairness and disparate impact

B. Cloud elasticity

C. Data compression

D. Sensor calibration only

Correct Answer: A. Fairness and disparate impact