14. AI Strategy, Digital Transformation and Value Realization

Introduction and Context

Artificial intelligence creates value only when it is connected to important organizational decisions, redesigned work, reliable data, and an operating model capable of sustaining change. The presence of a sophisticated model, cloud platform, or generative-AI assistant does not by itself constitute an AI strategy. An AI strategy explains where artificial intelligence should be used, why it matters, what capabilities are required, how risks will be controlled, and how benefits will be measured over time.

This distinction is important because many organizations have moved from experimentation to investment without developing a disciplined method for prioritizing use cases. A collection of disconnected pilots may produce impressive demonstrations but little enterprise value. Conversely, organizations that treat AI as part of digital transformation can use it to redesign customer journeys, improve operational decisions, automate routine work, strengthen risk management, and create new products and services.

AI strategy therefore sits at the intersection of business strategy, information systems, data governance, organizational design, finance, risk management, and workforce development. It requires both technical judgment and managerial reasoning. Leaders must evaluate not only whether a model can be built, but also whether it should be built, whether people will use it, whether the data is adequate, and whether the resulting benefit justifies the total cost and risk.

Learning Objectives

After studying this chapter, students should be able to:

1. Explain the relationship between AI strategy, digital transformation, and value realization.

2. Construct and prioritize an organizational AI use-case portfolio.

3. Assess feasibility, data readiness, economics, risk, and organizational readiness.

4. Design approaches for piloting, scaling, and governing AI solutions.

5. Evaluate operating models and human–AI collaboration arrangements.

6. Analyze implementation failures and recommend corrective action.

7. Apply responsible-AI, workforce, and change-management principles to strategic decisions.

From AI Interest to Strategic Intent

A business strategy defines how an organization creates and captures value. An AI strategy should support that logic rather than exist as an independent technology agenda. If a bank competes through trusted advice and rapid service, relevant AI priorities may include fraud detection, relationship-manager support, and faster credit assessment. If a manufacturer competes through reliability and low cost, predictive maintenance, quality inspection, and production scheduling may be more appropriate. The same technology can therefore have different strategic importance in different organizations.

Three questions help establish strategic intent:

• Where can AI improve or redefine the organization’s value proposition?
• Which capabilities must be developed to exploit those opportunities?
• What boundaries are necessary to protect customers, employees, the organization, and society?

Strategic intent should be expressed in business terms. “Deploy a large language model” is an implementation statement. “Reduce the time required to resolve complex customer requests while preserving service quality and compliance” is a strategic objective. The latter provides a basis for selecting technology, measuring outcomes, and deciding whether automation should be full, partial, or advisory.

Digital transformation is broader than AI adoption. It involves coordinated changes to processes, customer experiences, organizational structures, data practices, platforms, and culture. AI can be a powerful component of transformation, but it cannot compensate for fragmented processes, poor ownership, or unreliable information. In many cases, the most valuable AI project begins with process simplification and data standardization rather than model development.

The AI Value Chain and the Meaning of Value Realization

AI value is produced through a chain of linked activities:

1. A business problem is identified and translated into a decision or task.

2. Relevant data is collected, governed, and prepared.

3. A model or AI-enabled workflow generates a prediction, recommendation, classification, or piece of content.

4. A person or system acts on the output.

5. The action changes an operational, financial, customer, or societal outcome.

6. The outcome is measured and used to improve the system.

A model may be accurate while the overall value chain fails. For example, a demand-forecasting model can achieve strong statistical performance, yet generate no benefit if planners do not trust it, inventory policies cannot respond to its recommendations, or the forecast is delivered after procurement decisions have been made.

Value realization is the disciplined conversion of expected benefits into observable and sustained outcomes. It differs from value estimation, which occurs before implementation. A complete value-realization plan identifies the baseline, target outcome, accountable owner, measurement period, adoption assumptions, and mechanisms for correcting underperformance.

Value may be financial or non-financial. Financial benefits include increased revenue, reduced cost, lower losses, improved asset utilization, and reduced working capital. Non-financial benefits include improved safety, faster service, better employee experience, regulatory resilience, and enhanced decision quality. These outcomes should nevertheless be connected to organizational priorities and measured using credible indicators.

A useful distinction is between:

• Output metrics, such as number of predictions, documents processed, or chatbot interactions.
• Adoption metrics, such as active users, recommendation acceptance, or workflow completion.
• Outcome metrics, such as resolution time, defect rate, conversion, retention, or loss reduction.
• Risk metrics, such as error rates by group, privacy incidents, override frequency, or drift.

Counting AI outputs without observing outcomes creates an illusion of progress. Strategic governance gives greater weight to outcome and risk measures.

Building an AI Use-Case Portfolio

An AI use case is a defined application of AI to a business activity, decision, or customer interaction. A good use-case description identifies the user, process, decision, data, expected intervention, benefit, and risk. “Use AI in human resources” is too broad. “Assist recruiters by ranking candidates against transparent, job-related criteria, with mandatory human review” is sufficiently specific for assessment.

Organizations should discover use cases from several sources: strategic priorities, customer pain points, employee frustrations, process-performance data, regulatory obligations, technology road maps, and frontline experimentation. Top-down direction ensures strategic alignment; bottom-up discovery reveals practical opportunities that executives may not see.

Use cases can be grouped into broad value patterns:

• Efficiency and automation: reducing manual effort, cycle time, or operating cost.
• Decision augmentation: improving forecasts, recommendations, diagnosis, or prioritization.
• Risk and control: detecting anomalies, fraud, safety hazards, or compliance failures.
• Customer and employee experience: providing personalization, assistance, search, or conversational support.
• Innovation and growth: creating new products, services, channels, or revenue models.
• Knowledge leverage: making organizational knowledge easier to locate, interpret, and apply.

A portfolio is preferable to a single-project approach because benefits, risks, and dependencies differ across use cases. A portfolio can contain quick wins, strategic bets, foundational investments, and deliberate exclusions. It should not be a list of every proposed idea; it is a set of choices about where scarce resources will be concentrated.

Prioritization Framework

A practical prioritization score can combine strategic value, economic value, feasibility, readiness, and risk. For example:

Priority score = (Strategic value × Economic potential × Adoption potential × Data/technical feasibility) − Risk penalty

This is not a scientific formula. Its purpose is to make assumptions visible and enable comparison. Each factor should be assessed using defined criteria rather than intuition alone. Strategic value may include contribution to a critical objective; economic potential may reflect measurable cost or revenue impact; adoption potential may reflect workflow fit and user willingness; feasibility may include data, integration, and model complexity; risk may include legal, ethical, security, and reputational exposure.

Portfolio decisions should also consider time horizon. A document-classification pilot may produce benefit within months, whereas an AI-enabled product may require several years of platform, market, and regulatory development. A balanced portfolio prevents short-term automation projects from crowding out longer-term strategic capabilities.

Feasibility, Data Readiness, and Economic Discipline

Technical feasibility asks whether an acceptable solution can be developed and operated. It includes the availability of suitable models, infrastructure, integration interfaces, latency requirements, cybersecurity controls, and the ability to monitor performance. Feasibility is relative to the required level of accuracy and the consequences of error. A recommendation system may tolerate occasional errors; an application supporting safety-critical decisions may require much stronger assurance.

Data readiness is multidimensional. It includes:

• Availability: whether relevant data exists and can legally be accessed.
• Quality: whether it is accurate, complete, timely, consistent, and appropriately labeled.
• Representativeness: whether it reflects the populations and conditions in which the system will operate.
• Lineage: whether the origin, transformation, and ownership of data are known.
• Rights and permissions: whether use complies with contracts, privacy requirements, and intellectual-property obligations.
• Interoperability: whether data can be connected across systems without losing meaning.

Data readiness should be assessed before promising model performance. A frequent mistake is to begin with a preferred algorithm and then search for data that supports it. Strategy begins with the decision and its information requirements, not with the fashionable technology.

Economic analysis should include total cost of ownership rather than only development expenditure. Costs may include data preparation, cloud or computing resources, software licenses, model evaluation, integration, cybersecurity, human review, training, change management, monitoring, incident response, retraining, and eventual retirement. Generative-AI applications may also incur costs for retrieval systems, content filtering, usage-based inference, and evaluation of factuality.

A simplified net-value model is:

Net value = realized benefits − total cost of ownership − expected risk cost

Expected risk cost can be represented as the probability of an adverse event multiplied by its estimated impact, while recognizing that some harms cannot be captured accurately in monetary terms. ROI calculations should identify assumptions such as adoption rate, productivity improvement, error reduction, and benefit timing. Sensitivity analysis is useful: if a project is attractive only when adoption reaches an unrealistic level, it should not be treated as a secure investment.

From Pilot to Scaled Capability

A pilot is a controlled test of a business hypothesis, not a small version of full deployment. Its purpose is to reduce uncertainty about value, technical performance, user behavior, risk, and operating requirements. A pilot should therefore have a defined hypothesis, baseline, target measures, user group, duration, decision rights, and exit criteria.

A strong pilot usually proceeds through four stages. First, the organization frames the problem and establishes the current-state baseline. Second, it develops or configures a minimum viable solution using representative data. Third, it tests the solution in a controlled environment with realistic users and exception cases. Fourth, it evaluates results and makes a scale, redesign, pause, or stop decision.

Pilots fail when they are evaluated only through demonstrations. A realistic evaluation includes data-quality testing, performance under unusual conditions, user acceptance, workflow impact, security assessment, fairness analysis where relevant, and comparison with the existing process. For generative systems, evaluation should address unsupported claims, inappropriate content, prompt sensitivity, information leakage, and human verification behavior.

Scaling is not simply replicating a model. It requires production engineering, identity and access management, model and data versioning, monitoring, incident management, service-level expectations, support arrangements, procurement, and financial controls. It may also require process redesign. A pilot that works because a few experts manually correct outputs may fail when deployed to thousands of users unless that correction work is explicitly designed and resourced.

The transition from pilot to scale should be governed by evidence. Important questions include:

• Has the solution improved a meaningful outcome against a credible baseline?
• Do users adopt it in the intended workflow?
• Are errors understood and manageable?
• Are controls effective for the relevant level of risk?
• Can the organization afford and operate it continuously?
• Is ownership clear after the project team disbands?

Operating Models and Accountability

Organizations commonly use one of three operating approaches. In a centralized model, a central AI or data team owns most expertise, platforms, standards, and delivery. This supports consistency and efficient investment but may create distance from business processes. In a federated model, business units develop applications within enterprise standards, allowing stronger domain alignment but creating risks of duplication and inconsistent controls. A hybrid model combines central platforms, governance, and specialist capabilities with distributed product ownership.

The appropriate model depends on scale, regulatory exposure, technical maturity, and the diversity of business needs. Regardless of structure, accountability must be explicit. A business owner should be responsible for the outcome and process. A product or service owner should manage the solution lifecycle. Data owners should address data quality and access. Technical teams should maintain reliability and security. Risk, legal, compliance, and internal audit functions should provide proportionate challenge rather than becoming involved only after deployment.

AI systems require lifecycle governance. Relevant activities include use-case approval, data assessment, model selection, testing, documentation, deployment authorization, monitoring, change control, incident response, and retirement. A model card, system record, or equivalent documentation can describe purpose, data, limitations, evaluation results, intended users, prohibited uses, and escalation procedures.

Human–AI Collaboration and Workforce Implications

The central workforce question is not whether AI replaces people in the abstract, but how tasks, roles, authority, and skills will change. Work can be decomposed into perception, prediction, generation, judgment, relationship, physical execution, and accountability. AI may automate some components while increasing the importance of others.

Three broad collaboration patterns are useful. In human-in-the-loop arrangements, a human must review or approve the AI output before action. In human-on-the-loop arrangements, the system acts within defined boundaries while people monitor performance and intervene when necessary. In human-out-of-the-loop arrangements, action is automated with limited routine intervention; this is appropriate only where consequences are limited, controls are strong, and performance is demonstrably reliable.

Human review is not automatically an effective safeguard. Reviewers may experience automation bias, become overconfident in system outputs, or face workloads that make meaningful checking impossible. Effective collaboration therefore requires clear authority, explainable interfaces where appropriate, training in system limitations, escalation channels, and performance measures that do not reward uncritical acceptance.

Workforce planning should examine role redesign, reskilling, redeployment, hiring, and transition support. Employees who understand the process often possess knowledge essential for defining exceptions and detecting failures. Excluding them can reduce adoption and weaken system quality. At the same time, organizations should communicate honestly about potential displacement and avoid presenting automation as a purely beneficial change.

Governance, Ethics, and Responsible Investment

Strategic AI governance addresses legality, legitimacy, and accountability. Privacy requires purpose limitation, data minimization, appropriate access, retention controls, and protection against unauthorized disclosure. Fairness requires attention to unequal error rates, exclusion, proxy variables, and the impact of automated decisions on protected or vulnerable groups. Transparency may require informing people that AI is being used, explaining relevant factors, and providing routes for correction or appeal.

Security threats include data poisoning, prompt injection, model extraction, adversarial manipulation, unauthorized use, and leakage of confidential information. Intellectual-property questions may arise when training or retrieval data contains licensed, confidential, or personal material. Governance must therefore include technical controls, contractual safeguards, documentation, and ongoing monitoring.

Risk should be proportionate to context. An internal tool that summarizes low-sensitivity documents does not require the same controls as a system influencing employment, credit, healthcare, public benefits, or safety. However, low-risk classification should be evidence-based and revisited when the system, data, users, or purpose changes.

Responsible AI is not only a compliance function. It is a design discipline that can improve trust, adoption, reliability, and long-term value. A system that is technically impressive but unacceptable to affected stakeholders is strategically weak.

Illustrative Case: Meridian Health Insurers’ Claims Transformation

Background

Meridian Health Insurers, a fictional regional insurer, processes medical claims through a mixture of automated rules and specialist review. Processing delays were increasing, and employees spent substantial time locating policy documents and summarizing case histories. Senior management proposed a generative-AI claims assistant to “automate claims decisions.”

Situation and Problem

An initial vendor demonstration produced fluent summaries and appeared to reduce handling time. However, claims leaders identified serious concerns. The historical records contained inconsistent terminology, scanned documents, and missing fields. Some claims involved vulnerable customers and complex medical judgments. The legal team also questioned whether external processing of claim information was permissible. A narrow technology-led pilot could have created unacceptable risk.

Analysis

The transformation team reframed the objective from autonomous claims adjudication to safer decision support. It mapped the claims process and separated low-risk information-retrieval tasks from decisions requiring professional judgment. It established a baseline for average handling time, rework, escalation, error rates, and customer complaints. Data profiling revealed that document quality and policy-version control were more urgent than model selection.

A portfolio approach was adopted. The first use case was retrieval of approved policy clauses with citations. The second was generation of a draft case summary for trained examiners. Autonomous approval of claims was excluded from the first phase. The team created a risk register, restricted access, required source citations, logged user interactions, and established an escalation process for uncertain or contradictory records.

Approach and Solution

The pilot used a controlled knowledge base containing approved policy documents and historical records with sensitive fields appropriately protected. Users were trained to verify citations and report errors. The evaluation compared AI-assisted examiners with a control group using existing tools. Measures included handling time, factual accuracy, citation correctness, rework, user adoption, and customer-impact indicators.

The pilot showed meaningful time savings for retrieval and summarization, but it also exposed weaknesses. Performance declined when records contained poor scans or conflicting policy versions. Rather than scaling immediately, Meridian invested in document digitization, metadata standards, policy-version governance, and improved exception handling. A later deployment retained mandatory human approval and prohibited the system from making final eligibility decisions.

Outcome

The organization realized value through shorter administrative processing, more consistent access to policy information, and reduced search effort. It did not claim that the assistant replaced claims professionals. The project also avoided a potentially costly failure by identifying data and governance problems before autonomous decision-making was attempted.

Learning Points

The case demonstrates that strategic reframing can preserve value while reducing risk. It illustrates the importance of process decomposition, baseline measurement, data readiness, human accountability, and scale decisions based on evidence rather than demonstrations. It also shows that foundational data work can be a direct enabler of AI value, not an unrelated preliminary activity.

Common Failure Modes and Corrective Practices

Several recurring mistakes undermine AI transformation. The technology-first mistake begins with a model or platform and searches for a problem later. It should be corrected by starting with strategic outcomes and user needs. The pilot trap produces many experiments without product ownership, funding, or scale criteria. A portfolio board should require explicit continuation or termination decisions.

The accuracy-only mistake treats model performance as equivalent to business success. Evaluation should include workflow outcomes, adoption, fairness, security, and cost. The data-afterthought mistake discovers late that data is incomplete, inaccessible, or unauthorized. Data readiness assessment belongs in early investment decisions.

The automation-bias mistake assumes that a human reviewer will catch every error. Interfaces, training, workload, sampling audits, and escalation procedures must support active review. The benefit-overstatement mistake reports theoretical productivity as realized savings. Benefits should be attributed conservatively and linked to changes in staffing, capacity, revenue, quality, or risk.

Other failures arise from weak ownership, fragmented architecture, uncontrolled vendor dependency, poor change communication, and failure to monitor models after deployment. AI systems operate in changing environments; data distributions, user behavior, policies, and business processes can all shift. Continuous monitoring and periodic reauthorization are therefore strategic necessities.

Contemporary Directions

AI strategy is increasingly shaped by foundation models, generative AI, retrieval-augmented generation, multimodal systems, AI agents, and specialized smaller models. These developments expand the range of possible applications but also increase uncertainty about factuality, security, intellectual property, and cost. Organizations are consequently investing in model-routing strategies, evaluation platforms, private deployment options, retrieval controls, and usage governance.

The direction of travel is from isolated models toward AI-enabled operating systems: shared data products, reusable services, workflow orchestration, observability, identity controls, and enterprise evaluation. At the same time, regulatory expectations are moving toward risk-based governance, documentation, transparency, human oversight, and accountability. Strategic leaders must track both technical progress and institutional requirements.

Best Practices for Value-Oriented AI Strategy

A mature organization generally follows these practices:

1. Anchor AI priorities in measurable strategic outcomes.

2. Maintain a portfolio rather than an unranked list of experiments.

3. Assess data, integration, risk, and adoption before committing to development.

4. Define baselines, owners, benefits, and stop criteria at the beginning.

5. Treat pilots as tests of business and operating hypotheses.

6. Design human roles, training, and accountability alongside the model.

7. Use proportionate governance throughout the lifecycle.

8. Budget for total cost of ownership, not only initial construction.

9. Scale reusable platforms and standards while preserving domain ownership.

10. Monitor outcomes, drift, incidents, user behavior, and unintended effects after launch.

Chapter Summary

AI strategy connects artificial intelligence to organizational purpose, competitive choices, operating capabilities, and measurable outcomes. Digital transformation provides the broader context in which AI changes processes, customer experiences, data practices, and work. A disciplined organization builds a use-case portfolio, prioritizes opportunities according to value, feasibility, readiness, adoption, and risk, and evaluates total cost of ownership.

Value realization requires more than model accuracy. It depends on workflow adoption, human action, process redesign, outcome measurement, and sustained ownership. Pilots should test explicit hypotheses and use evidence-based criteria for scaling. Operating models may be centralized, federated, or hybrid, but accountability must remain clear. Human–AI collaboration requires thoughtful task design, training, oversight, and workforce transition planning. Responsible governance addresses privacy, fairness, security, transparency, intellectual property, and proportionality. The most durable AI transformations combine technical capability with strategic discipline and organizational learning.

Key Terms

AI strategy: A coordinated set of choices defining how an organization will use AI to achieve strategic objectives while managing capability and risk.

Digital transformation: Organization-wide change involving technology, processes, data, structure, culture, and stakeholder experience.

AI use case: A specific application of AI to a business task, decision, process, or interaction.

Use-case portfolio: A prioritized collection of AI opportunities managed as an interconnected investment set.

Value realization: The conversion of expected AI benefits into measured and sustained organizational outcomes.

Baseline: A description of current performance against which a proposed intervention is evaluated.

Data readiness: The degree to which data is available, usable, governed, representative, and legally accessible for a purpose.

Total cost of ownership: All direct and indirect costs incurred across an AI system’s lifecycle.

Adoption: The extent to which intended users actually incorporate an AI solution into their work.

Pilot: A controlled experiment designed to test technical, business, user, and risk assumptions.

Model drift: A decline in performance caused by changes in data, relationships, behavior, or operating conditions.

Centralized operating model: An arrangement in which a central team controls most AI expertise, platforms, and delivery.

Federated operating model: An arrangement in which business units develop solutions within shared enterprise standards.

Human-in-the-loop: A design requiring human review or approval before the system’s output produces an action.

Automation bias: The tendency to place excessive trust in automated recommendations or outputs.

Responsible AI: The design and governance of AI to address fairness, privacy, security, transparency, accountability, and social impact.

Risk-adjusted value: Expected benefit evaluated after considering implementation cost and the probability and impact of adverse outcomes.

Retrieval-augmented generation: A method that supplies a generative model with retrieved, relevant information to improve grounded responses.

Review Questions

Short Answer Questions

1. Why is AI strategy broader than selecting an AI model or platform?

2. Distinguish output metrics, adoption metrics, and outcome metrics.

3. Identify four dimensions of data readiness.

4. What is the purpose of a pilot’s exit criteria?

5. Explain automation bias in human–AI collaboration.

Long Answer Questions

1. Explain how an organization can construct and prioritize an AI use-case portfolio.

2. Discuss the relationship between digital transformation and AI value realization.

3. Compare centralized, federated, and hybrid AI operating models.

4. Explain why total cost of ownership is essential when evaluating an AI investment.

5. Describe the governance requirements for scaling an AI system from pilot to production.

Analytical / Application Questions

1. A retailer has accurate demand forecasts, but store managers rarely use them. Diagnose the value-realization problem and propose an intervention plan.

2. A bank proposes an AI system to rank loan applicants. Develop an assessment covering data readiness, economic value, human oversight, fairness, and scale criteria.

3. An organization has thirty AI pilots but no production deployments. Analyze the likely portfolio and operating-model failures and recommend a recovery strategy.

22. Multiple Choice Questions with Answers

Question 1

Which statement best describes AI value realization?

A. Increasing the number of models deployed across the organization

B. Converting expected benefits into measured and sustained outcomes

C. Improving model accuracy without changing business processes

D. Replacing all manual work with automated systems

Correct Answer: B. Converting expected benefits into measured and sustained outcomes

Question 2

Which use case is most strategically well-defined?

A. Use AI in customer service

B. Deploy a large language model across the enterprise

C. Reduce complex-request resolution time using an assistant with human review

D. Modernize the organization through intelligent technology

Correct Answer: C. Reduce complex-request resolution time using an assistant with human review

Question 3

Which of the following is an adoption metric rather than an output metric?

A. Number of documents summarized

B. Number of predictions generated

C. Percentage of recommendations accepted by intended users

D. Number of model API calls

Correct Answer: C. Percentage of recommendations accepted by intended users

Question 4

Why should total cost of ownership be included in AI investment analysis?

A. Because model accuracy is determined by accounting costs

B. Because lifecycle expenses include integration, monitoring, training, and support

C. Because it eliminates the need for benefits measurement

D. Because it guarantees that a project will be profitable

Correct Answer: B. Because lifecycle expenses include integration, monitoring, training, and support

Question 5

What is the primary purpose of a pilot?

A. To create a public demonstration regardless of business results

B. To test technical, business, user, and risk assumptions under controlled conditions

C. To avoid establishing a baseline until after deployment

D. To guarantee immediate enterprise-wide scaling

Correct Answer: B. To test technical, business, user, and risk assumptions under controlled conditions

Question 6

A forecasting model performs well, but planners receive its recommendations after procurement decisions are made. Which part of the AI value chain is failing most directly?

A. Model training only

B. Data storage only

C. The connection between AI output and organizational action

D. The definition of artificial intelligence

Correct Answer: C. The connection between AI output and organizational action

Question 7

Which operating model combines enterprise-wide standards and platforms with business-unit product ownership?

A. Purely centralized

B. Purely federated

C. Hybrid

D. Unmanaged

Correct Answer: C. Hybrid

Question 8

Which practice best reduces automation bias?

A. Hiding uncertainty so users remain confident

B. Requiring meaningful review, training users on limitations, and providing escalation routes

C. Allowing the system to make all high-impact decisions

D. Measuring only how frequently users accept recommendations

Correct Answer: B. Requiring meaningful review, training users on limitations, and providing escalation routes

Question 9

In the illustrative Meridian case, why was autonomous claims approval excluded from the initial phase?

A. The organization had no interest in reducing processing time

B. The task involved complex judgments, sensitive data, and unresolved readiness issues

C. Generative AI cannot summarize documents

D. Human review is never useful in insurance

Correct Answer: B. The task involved complex judgments, sensitive data, and unresolved readiness issues

Question 10

Which formula most appropriately represents a simplified risk-adjusted value perspective?

A. Net value = model accuracy + number of users

B. Net value = realized benefits − total cost of ownership − expected risk cost

C. Net value = pilot duration − governance effort

D. Net value = technology expenditure + theoretical productivity

Correct Answer: B. Net value = realized benefits − total cost of ownership − expected risk cost