Digital Planning, Advanced Analytics and AI

Introduction

Aggregate production planning (APP) traditionally coordinates demand, capacity, workforce, inventory, subcontracting and production rates over a medium-term horizon. Digital technologies have transformed this activity from a periodic spreadsheet exercise into a continuously informed decision process. Enterprise resource planning (ERP) systems provide transactional records; advanced planning and scheduling (APS) systems represent constraints and alternatives; cloud platforms connect distributed operations; digital twins simulate changing conditions; predictive analytics estimate what is likely to happen; and artificial intelligence (AI) supports recommendations, optimization and adaptation.

These technologies do not eliminate the fundamental managerial problem of APP. They change the speed, scale and sophistication with which the problem can be analysed. A mathematically elegant recommendation remains poor if master data are wrong, demand signals are biased, constraints are omitted, or managers cannot explain the decision. Digital planning is therefore both a technical and governance challenge.

This chapter evaluates the principal technologies used in digitally enabled aggregate planning and examines data quality, explainability, cybersecurity, human oversight and model risk. The emphasis is not on adopting technology for its own sake, but on designing a decision system that produces feasible, economically sound and accountable plans.

Learning Objectives

After studying this chapter, students should be able to:

1. Distinguish the roles of ERP, APS, cloud platforms, digital twins, predictive analytics and AI in aggregate production planning.
2. Formulate and interpret a basic digitally supported aggregate planning model.
3. Evaluate data quality, model performance and implementation readiness.
4. Explain the importance of explainability, cybersecurity, human oversight and model-risk governance.
5. Design an appropriate analytics architecture and assess its managerial implications.
6. Apply these ideas to a realistic production-planning case.

1. From Periodic Planning to Digital Decision Systems

A conventional APP process may use a monthly demand forecast, a spreadsheet model and a meeting in which planners revise the result. A digital process can ingest orders, point-of-sale data, inventory transactions, machine status, labour availability and supplier information at frequent intervals. The planning engine can then produce scenarios or recommendations, while execution data provide feedback.

The essential distinction is between a system of record and a system of decision support. An ERP system is principally a system of record: it stores transactions and executes business processes. An APS system is principally a system of decision support: it evaluates alternative plans against time, capacity, material and policy constraints. A digital planning environment links both.

Digital planning should be judged against five outcomes:

• Feasibility: the plan respects material, labour, machine and calendar constraints.
• Economic quality: the plan balances service, inventory, overtime, hiring, layoffs, subcontracting and changeover costs.
• Responsiveness: the plan can be revised quickly when conditions change.
• Transparency: users can understand the data, assumptions and logic behind recommendations.
• Resilience: the process remains reliable under disruption, cyberattack, missing data or model failure.

A higher level of automation does not necessarily improve any of these outcomes. Automation can simply make erroneous decisions faster and at greater scale.

2. The Digital Planning Architecture

A useful architecture has six interacting layers:

1. Operational data layer: orders, forecasts, inventory, bills of material, routings, work centres, calendars, labour records, supplier schedules and cost data.
2. Integration layer: interfaces, application programming interfaces (APIs), event streams and extract-transform-load processes that move data between systems.
3. Analytical layer: statistical forecasts, optimization models, simulation, machine learning and scenario analysis.
4. Planning application layer: demand planning, supply planning, rough-cut capacity planning, master production scheduling and scheduling.
5. Execution and feedback layer: production reporting, warehouse movements, quality results, maintenance events and delivery performance.
6. Governance and security layer: access controls, data ownership, audit trails, validation, model monitoring and incident response.

The architecture should support a closed loop: observe, forecast, plan, execute, measure and learn. The loop is not automatically self-correcting. Feedback must be defined carefully because execution data may reflect previous planning decisions rather than independent evidence. For example, a stockout may reduce observed sales, causing a naïve forecasting model to underestimate true demand.

ERP Systems

ERP systems integrate functions such as sales, procurement, inventory, production, finance, quality and maintenance. For APP, their major contributions include:

• a common item and location master;
• inventory balances and transaction history;
• bills of material and routings;
• work-centre calendars and planned capacity;
• purchase orders, production orders and sales orders;
• cost and accounting information; and
• traceability and authorization controls.

ERP data are indispensable but not automatically accurate. A recorded inventory balance may differ from physical stock because of unposted movements, scrap, counting errors or timing differences. A routing may contain nominal processing time but omit cleaning, setup or inspection. ERP implementation therefore requires data stewardship, not merely software configuration.

Advanced Planning and Scheduling Systems

APS systems extend ERP by explicitly modelling constraints and alternative decisions. Typical functions include demand prioritization, supply allocation, rough-cut capacity planning, finite-capacity scheduling, material synchronization, distribution planning and what-if analysis.

An APS model may distinguish between infinite-capacity planning, which identifies material and demand requirements without fully enforcing capacity, and finite-capacity planning, which schedules work only where capacity is available. Finite planning is generally more realistic but requires better data and greater computational effort.

APS outputs should be interpreted as conditional solutions. A recommendation is valid only under the stated forecast, capacity, cost, policy and constraint assumptions. Managers should therefore inspect the binding constraints and the trade-offs, not only the proposed production quantities.

Cloud Platforms

Cloud planning platforms provide shared access, elastic computing capacity, centralized updates and integration across factories and partners. They are particularly valuable for geographically dispersed operations and collaborative planning. They can also support large-scale simulation and machine-learning workloads.

The main risks include dependence on network availability, vendor lock-in, unclear data residency, integration complexity and shared-responsibility failures. In a cloud arrangement, the provider may secure physical infrastructure, while the customer remains responsible for identities, configurations, interfaces, data classification and business continuity. Contractual service-level agreements should specify availability, recovery time, recovery point, portability and audit rights.

Digital Twins

A digital twin is a maintained digital representation of a physical asset, process, facility or supply network, connected to relevant operational data. A static simulation model is not necessarily a digital twin. The distinguishing feature is an ongoing relationship between the physical system and its digital representation.

For APP, a twin can represent demand, inventory, production capacity, material flows, workforce availability and disruption states. It can evaluate scenarios such as a machine failure, supplier delay, demand surge or overtime restriction. The value of a twin depends on model fidelity. Excessive detail can make the model slow and difficult to maintain; insufficient detail can produce misleading recommendations.

3. Quantitative Foundations for Digital Aggregate Planning

Digital technologies do not replace aggregate-planning logic. They implement, extend and accelerate it. A basic linear aggregate-planning model illustrates the decision structure.

Let periods be indexed by t = 1, 2, ..., T. Let:

• D_t = demand in period t, measured in units;
• P_t = regular production;
• O_t = overtime production;
• S_t = subcontracted production;
• I_t = ending inventory;
• B_t = ending backlog;
• W_t = workforce level;
• H_t and F_t = hires and separations;
• c_p, c_o, c_s = unit costs of regular, overtime and subcontracted production;
• h = inventory holding cost per unit per period;
• b = backlog cost per unit per period;
• c_H, c_F = hiring and separation costs.

The net balance equation is:

All flow variables must use the same unit and period basis. If inventory and backlog cannot coexist for the same item and period, an additional policy or formulation is needed; the simple equation alone permits both positive inventory and positive backlog.

A possible objective is:

This objective minimizes relevant operating cost over the planning horizon. It excludes fixed costs, revenue and capital costs unless they are explicitly added. It also assumes linear, additive costs and known parameters.

Workforce continuity may be represented as:

If regular production depends on workforce productivity, a capacity constraint may be written as:

where a is regular production capacity per worker per period. A more realistic model may include machine capacity, skills, shifts, maintenance, setup losses, minimum batch sizes and labour regulations. The simple formulation is therefore a conceptual foundation rather than a complete factory model.

Worked Example: Interpreting a Digital Planning Recommendation

Suppose a planner receives the following one-period recommendation: regular production 800 units, overtime 100 units, subcontracting 0 units, beginning inventory 150 units and demand 900 units. There is no beginning backlog. The ending balance is:

I₁ = 150 + 800 + 100 - 900 = 150 units.

The plan is materially feasible for this period if capacity permits 800 regular and 100 overtime units. If regular capacity is only 700 units and overtime capacity is 50 units, the recommendation is infeasible even though the inventory equation balances. This illustrates a central digital-planning principle: balance feasibility and capacity feasibility are different tests.

If regular production costs £12 per unit, overtime costs £18, and holding cost is £2 per ending unit, the period’s variable and holding cost is:

12(800) + 18(100) + 2(150) = £12,100.

The calculation does not establish that the plan is optimal because alternative plans, future periods, backlog costs and workforce decisions may change the total horizon cost.

4. Predictive Analytics and AI-Assisted Optimization

Descriptive, Diagnostic, Predictive and Prescriptive Analytics

Descriptive analytics reports what happened, such as production attainment or inventory turnover. Diagnostic analytics investigates why it happened, such as identifying a supplier delay or recurring setup loss. Predictive analytics estimates what is likely to happen, such as demand, failure probability or lead time. Prescriptive analytics recommends what should be done, often using optimization or simulation.

Forecasting and optimization should not be conflated. A forecast estimates uncertain conditions; an optimization model selects decisions given conditions, costs and constraints. A highly accurate demand forecast can still lead to a poor plan if capacity or policy constraints are incorrectly represented. Conversely, a robust optimization model may perform acceptably with an imperfect forecast if it explicitly considers uncertainty.

Forecasting Models

Time-series methods may use level, trend, seasonality and explanatory variables. Machine-learning methods can model nonlinear relationships among price, promotions, weather, market signals and customer characteristics. However, complex models can overfit historical data and fail when the operating environment changes.

Forecast accuracy should be measured against a suitable baseline and evaluated over time. Common measures include mean absolute error:

where y_i is the observed value, ŷ_i is the forecast and n is the number of observations. MAE is expressed in the same units as the demand measure and treats all absolute errors proportionally.

Mean absolute percentage error is:

It is unsuitable or unstable when actual demand is zero or very small. Forecast evaluation should therefore consider bias, service impact, intermittent demand, asymmetric costs and forecast value added—not just one numerical score.

Forecast bias can be represented as:

The sign convention must be stated. In this definition, positive bias means systematic over-forecasting.

AI-Assisted Optimization

AI may support APP in several ways:

• predicting demand, yield, lead time and equipment failure;
• classifying disruptions and recommending escalation;
• generating candidate schedules or repair moves;
• learning cost-to-go approximations for complex sequential decisions;
• identifying patterns in planner overrides; and
• using natural-language interfaces to query plans and assumptions.

Optimization remains essential where constraints and trade-offs must be respected. A practical architecture often combines machine learning for uncertain inputs with mathematical programming, constraint programming, heuristics or simulation for decisions. An AI-generated recommendation should be passed through feasibility checks and policy rules before execution.

Reinforcement learning may be attractive for sequential planning, but it requires a carefully designed state representation, action space, reward function and safe exploration policy. Historical data may encode poor decisions or abnormal conditions. A model trained to imitate historical schedules can reproduce systematic inefficiency rather than discover improvement.

5. Data Quality and Information Governance

Data quality is multidimensional. Important dimensions include:

Common APP data failures include obsolete bills of material, incorrect yield factors, unrecorded scrap, inaccurate inventory, unavailable labour skills, unrealistic supplier lead times and calendars that ignore maintenance. The remedy is not simply a data-cleaning project. Data ownership, validation rules, physical audits and feedback mechanisms must be embedded in the operating process.

A useful data-governance arrangement assigns a data owner accountable for definition and policy, a data steward responsible for quality maintenance, and technical custodians responsible for storage and access. Critical data should have validation thresholds and exception queues. For example, a new standard lead time may require approval if it changes by more than a defined percentage or conflicts with supplier evidence.

Data lineage is particularly important for AI. Users should be able to determine which source systems, transformations, time windows and features contributed to a forecast or recommendation. Without lineage, an apparently precise output cannot be properly challenged.

6. Explainability, Human Oversight and Model Risk

Explainability

Explainability means that relevant users can understand the basis, limitations and consequences of a model output. It does not necessarily require revealing proprietary source code. For an APP recommendation, useful explanations include:

• the demand and capacity assumptions used;
• the constraints that were binding;
• the cost or service trade-off selected;
• the major factors influencing a forecast;
• the confidence or uncertainty range; and
• the effect of changing a key assumption.

A simple linear optimization model is often intrinsically interpretable. A complex machine-learning model may require post-hoc tools such as feature importance, partial dependence or local explanations. These tools are aids, not proofs of causality. A feature may correlate with an outcome without causing it.

Human-in-the-Loop Design

Human oversight should be designed according to risk rather than added symbolically. Low-risk recommendations, such as a replenishment suggestion within established limits, may be automatically executed. High-risk decisions, such as reducing safety stock for a critical medical component or reallocating scarce capacity among strategic customers, require human approval.

A robust workflow includes:

1. automated data and feasibility checks;
2. recommendation generation;
3. explanation of assumptions and exceptions;
4. planner review and documented override;
5. execution with authorization controls; and
6. monitoring of results and model performance.

Human oversight can also fail. Planners may suffer from automation bias and accept an output because it appears scientific. Alternatively, they may reject every recommendation because the system does not explain itself. Training, clear escalation rules and audit trails are therefore necessary.

Model Risk

Model risk is the possibility that a model produces harmful decisions because of incorrect specification, poor data, unsuitable assumptions, implementation errors or changing conditions. It includes risks in forecasts, optimization models, simulations and AI systems.

A model-risk framework should cover:

• model inventory and ownership;
• documented purpose, scope and assumptions;
• independent validation before deployment;
• testing on out-of-sample and stress scenarios;
• performance and drift monitoring;
• version control and change approval;
• fallback procedures; and
• periodic retirement or redevelopment.

Validation should test both statistical performance and operational usefulness. A model may have low forecast error but create excessive overtime because it ignores capacity. Conversely, a model with modest forecast accuracy may improve service by correctly representing uncertainty and priorities.

7. Cybersecurity and Operational Resilience

Digitally integrated planning expands the attack surface. Threats include credential theft, ransomware, malicious master-data changes, denial of service, supply-chain compromise and manipulation of production or inventory records. A false inventory balance can be as operationally damaging as a machine failure.

Key controls include least-privilege access, multifactor authentication, network segmentation, encryption, secure APIs, backups, immutable logs, vulnerability management and tested incident-response procedures. Privileged changes to bills of material, routings, supplier bank details or production calendars should require stronger authorization and monitoring.

Cybersecurity must be integrated with production continuity. A highly secure system that cannot be recovered during a plant outage is not resilient. Business-continuity plans should define manual planning procedures, offline contact lists, recovery priorities and acceptable degraded-service levels. Recovery-point objective specifies the maximum tolerable amount of data loss measured in time; recovery-time objective specifies the target time for restoring service. Both should be agreed with operational stakeholders.

AI introduces additional concerns, including adversarial inputs, data poisoning, prompt injection in natural-language interfaces and leakage of confidential information. Organizations should prevent unrestricted transmission of sensitive production, customer or supplier data to external AI services.

8. Selecting and Implementing Digital Planning Technologies

Technology selection should begin with planning pain points, not product features. A structured assessment can examine:

Implementation is usually more successful when staged. A possible sequence is: establish master-data governance; stabilize ERP transactions; introduce visibility and baseline dashboards; implement constrained planning for a selected product family; add predictive models; then automate low-risk decisions. A pilot should have a control or baseline process where possible, a defined evaluation period and measures such as service level, inventory, schedule adherence, planner effort and expedite frequency.

Digital transformation also changes roles. Planners increasingly act as exception managers, scenario analysts and policy owners. They need skills in data interpretation, constraint analysis, model interrogation and communication—not only system navigation.

9. Illustrative Case: Northstar Appliances

Background

Northstar Appliances, a fictional manufacturer, produces three families of domestic water heaters at two plants. Demand is seasonal and promotional. The company uses an ERP system for orders, inventory and production reporting, but each plant maintains separate spreadsheets for capacity and overtime. Senior management experiences frequent stockouts during promotions, excess finished goods after the season and disagreements about the accuracy of inventory data.

Problem

The planning team proposes an AI forecasting tool and a cloud APS platform. However, an initial diagnostic reveals that 14% of active items have inconsistent units of measure, several routings omit setup time, and inventory adjustments are posted weekly rather than daily. The company also has no documented model-approval process.

Analysis

A cross-functional team maps the data lineage from customer order to production recommendation. It finds that observed sales are censored during stockouts; using sales alone would underestimate demand. It also finds that one plant’s stated capacity assumes two shifts, although labour agreements permit only one shift during part of the year.

The team defines a baseline using the existing spreadsheet process. It then builds a pilot model for one product family. The demand model uses historical orders, promotion indicators and stockout flags. The APS model includes inventory balance, finite work-centre capacity, overtime limits, minimum production quantities and customer-priority rules. The system reports binding constraints and allows planners to compare normal, high-demand and supplier-delay scenarios.

Approach and Solution

Northstar first assigns owners for item master data, routings, calendars and inventory accuracy. Physical cycle counts are increased for high-value and high-variability items. The company adopts automated checks for negative inventory, missing routings, impossible lead times and capacity exceeding approved calendars.

The cloud platform is integrated with ERP through controlled interfaces. Recommendations cannot be released automatically when they affect critical products, exceed overtime thresholds or rely on stale data. Planners receive an explanation panel showing demand assumptions, bottlenecks, expected inventory and the cost of major alternatives. All overrides require a reason code.

Cybersecurity controls include multifactor authentication, role-based access, segregated development and production environments, encrypted interfaces and tested backup restoration. A fallback spreadsheet and manual approval procedure are retained for temporary platform unavailability.

Outcome

After one seasonal cycle, the pilot records improved service and lower emergency expediting relative to the baseline. The improvement is not attributed solely to AI: data correction, explicit capacity modelling and better cross-functional review contribute substantially. Forecast accuracy improves for promoted items, but performance remains weak for newly launched products because historical data are limited. The company therefore uses human overrides and wider uncertainty buffers for new products.

Learning Points

1. Better planning resulted from combining data governance, optimization and organizational change.
2. Forecast accuracy alone was not an adequate success measure.
3. A cloud platform created value only after interfaces and master data were controlled.
4. Explainability and approval thresholds made automation acceptable to planners.
5. Fallback procedures reduced dependence on a single digital service.

10. Contemporary Developments and Critical Evaluation

Several developments are shaping digital APP. Generative AI can provide conversational access to planning data, summarize exceptions and translate business questions into analytical queries. It should not be allowed to alter production plans without deterministic validation and authorization. Industrial Internet of Things devices provide higher-frequency equipment and process data, but more data do not automatically mean more useful information.

Probabilistic forecasting and robust optimization are increasingly important where demand, yield and lead times are uncertain. Instead of optimizing against one point forecast, planners may evaluate a range of scenarios or protect performance against adverse conditions. The choice between expected-cost optimization, robust optimization and service-protection policies depends on the organization’s risk preferences and the consequences of failure.

Sustainability is also entering APP models. Carbon emissions, energy peaks, water use, waste and circular-material availability may be represented as constraints or additional objective terms. Multi-objective planning requires transparent priority rules because economic, service and environmental objectives can conflict.

Critical evaluation remains necessary. Digital systems can centralize power, intensify surveillance of workers, embed historical bias and encourage decisions that optimize measurable indicators while damaging less visible objectives. Data collected for productivity may be repurposed for individual worker evaluation without adequate consultation. Responsible implementation requires proportional data collection, privacy protection, non-discrimination, accessibility and clear accountability.

Best-Practice Framework

A mature digital APP capability follows these principles:

• define the decision and its success measures before selecting technology;
• maintain one governed set of critical master-data definitions;
• separate forecasts, assumptions, constraints and decisions;
• combine predictive methods with constraint-aware optimization;
• test recommendations against operational and disruption scenarios;
• expose uncertainty, binding constraints and trade-offs;
• apply risk-based human approval and override procedures;
• monitor data drift, model performance and business outcomes;
• secure identities, interfaces, configurations and backups;
• retain auditable versions and a practical fallback process; and
• review ethical, workforce and sustainability effects periodically.

Chapter Summary

Digital aggregate production planning integrates ERP records, APS capabilities, cloud infrastructure, digital twins, predictive analytics and AI-assisted optimization. ERP provides transactional evidence, while APS and optimization convert demand, capacity, inventory and policy information into feasible alternatives. Cloud platforms improve collaboration and scalability but introduce dependency, integration and security concerns. Digital twins support scenario analysis when their physical relationships and assumptions are maintained.

Predictive analytics estimates uncertain conditions; optimization selects actions subject to constraints. Their combination is powerful but depends on reliable data, realistic models and appropriate performance measures. Data quality must address accuracy, completeness, timeliness, consistency, validity, uniqueness and lineage. Explainability should reveal assumptions, binding constraints, trade-offs and uncertainty. Human oversight must be risk-based, and model-risk governance must include validation, monitoring, version control and fallback procedures.

Cybersecurity is an operational requirement because manipulated planning data can disrupt production as seriously as physical failures. Successful implementation is socio-technical: it requires governance, skills, process redesign and responsible use, not merely software installation. The central managerial principle is that digital planning should make decisions more informed, feasible, resilient and accountable—not simply more automated.

Key Terms

ERP system: Integrated system of record supporting enterprise transactions and processes.

Advanced planning and scheduling (APS): Software that plans and schedules using demand, resources and constraints.

Cloud platform: Internet-accessible computing and software infrastructure delivered by a provider.

Digital twin: A maintained digital representation of a physical asset, process or network.

Predictive analytics: Methods that estimate future or unknown outcomes from data.

Prescriptive analytics: Analysis that recommends actions under objectives and constraints.

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

Master data: Persistent definitions such as items, locations, routings, calendars and suppliers.

Data lineage: Traceable record of data origin, transformation and use.

Explainability: The ability to communicate how and why a model produces an output.

Model risk: Potential harm arising from incorrect model specification, data, implementation or use.

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

Finite-capacity planning: Planning that explicitly limits production by available capacity.

Forecast bias: Systematic tendency for forecasts to be above or below actual outcomes.

Mean absolute error: Average absolute difference between observed and forecast values.

Constraint: A rule or limitation restricting feasible decisions.

Data drift: Change in the statistical properties of input data over time.

Recovery-time objective: Target time for restoring a service after disruption.

Least privilege: Security principle granting only the access necessary for a role.

Digital resilience: Ability to continue, recover and adapt when digital systems or data are disrupted.

Review Questions

Short Answer Questions

1. What is the principal difference between an ERP system and an APS system in aggregate planning?
2. Why can an inventory-balance equation be satisfied while a plan remains infeasible?
3. Identify four important dimensions of data quality.
4. What information should an explainable planning recommendation provide?
5. Define model risk in the context of digital aggregate planning.

Long Answer Questions

1. Evaluate the benefits and limitations of cloud-based aggregate planning for a multi-plant manufacturer.
2. Explain how predictive analytics and mathematical optimization can be combined in an AI-assisted planning architecture.
3. Discuss the role of digital twins in disruption analysis and identify the conditions required for their effective use.
4. Develop a risk-based human-oversight framework for automated production-planning decisions.
5. Explain why cybersecurity, data governance and business continuity should be treated as parts of production planning rather than separate IT concerns.

Analytical / Application Questions

1. A factory has accurate demand forecasts but regularly violates labour and machine limits. Diagnose the likely modelling and governance failures and propose a corrective architecture.
2. Design a pilot study to evaluate whether an AI-supported APS system improves performance. Specify the baseline, metrics, evaluation period and controls for confounding factors.
3. Using the Northstar Appliances case, determine which decisions should be automated, which should require planner approval and which should remain fully manual. Justify your classification using operational and model-risk criteria.

Multiple Choice Questions with Answers

Question 1

Which system is primarily a system of record for transactions such as inventory movements, purchase orders and production orders?

A. ERP system

B. Digital twin

C. Optimization solver

D. Forecast-monitoring dashboard

Correct Answer: A. ERP system

Question 2

In the aggregate balance equation, if beginning inventory is 150 units, production is 900 units and demand is 800 units, with no beginning backlog, what is ending inventory?

A. 50 units

B. 150 units

C. 250 units

D. 1,850 units

Correct Answer: C. 250 units

Question 3

Which measure is expressed in the same units as the observed demand?

A. MAPE

B. MAE

C. Forecast percentage bias only

D. Coefficient of variation

Correct Answer: B. MAE

Question 4

What is a major limitation of MAPE?

A. It cannot compare forecasts across time periods

B. It is unstable when actual demand is zero or very small

C. It always penalizes under-forecasting more than over-forecasting

D. It requires a finite-capacity model

Correct Answer: B. It is unstable when actual demand is zero or very small

Question 5

Which feature most clearly distinguishes a digital twin from a static simulation model?

A. It contains a graphical user interface

B. It uses only historical data

C. It maintains an ongoing connection with the physical system and its data

D. It excludes operational constraints

Correct Answer: C. It maintains an ongoing connection with the physical system and its data

Question 6

Which practice best supports explainability of an APS recommendation?

A. Concealing all assumptions to prevent user interference

B. Reporting binding constraints, assumptions and major trade-offs

C. Replacing all mathematical models with natural-language output

D. Allowing recommendations without version records

Correct Answer: B. Reporting binding constraints, assumptions and major trade-offs

Question 7

Which cybersecurity principle grants users only the access necessary for their responsibilities?

A. Least privilege

B. Data duplication

C. Open authorization

D. Infinite capacity

Correct Answer: A. Least privilege

Question 8

In the chapter’s model, positive forecast bias means that forecasts are:

A. Systematically below actual values

B. Exactly equal to actual values

C. Systematically above actual values

D. Unavailable for evaluation

Correct Answer: C. Systematically above actual values

Question 9

Why should a high-risk recommendation, such as reducing safety stock for a critical component, normally require human approval?

A. Because optimization cannot represent costs

B. Because the consequence of error is material and requires accountable judgment

C. Because cloud systems cannot process inventory data

D. Because forecasts are never useful

Correct Answer: B. Because the consequence of error is material and requires accountable judgment

Question 10

Which sequence best represents a closed-loop digital planning process?

A. Execute, ignore, forecast, delete, plan

B. Observe, forecast, plan, execute, measure and learn

C. Forecast, outsource, encrypt, hire and stop

D. Optimize, replace ERP, automate every decision and archive

Correct Answer: B. Observe, forecast, plan, execute, measure and learn