Business Problem Framing and Decision Models

Introduction

Business analytics creates value only when analysis is connected to a real decision. A technically sophisticated dashboard, prediction model, or optimization routine may still be useless if it addresses the wrong problem, measures an irrelevant outcome, or ignores organizational constraints. Business problem framing is the disciplined process of translating a managerial concern into a clearly bounded analytical problem. Decision modelling then represents the available choices, uncertainties, objectives, constraints, and consequences in a form that supports action.

For example, “sales are declining” is a managerial observation, not yet an analytical question. It might lead to several different investigations: identifying which customer segments are leaving, estimating the effect of price changes, forecasting future demand, or deciding how to allocate a limited marketing budget. Each interpretation requires different data, methods, stakeholders, and success criteria.

This chapter develops a practical framework for moving from managerial concern to analytical specification. It explains stakeholder analysis, problem statements, objectives, hypotheses, decision variables, constraints, key performance indicators (KPIs), outcomes, decision trees, expected value, and governance. The central principle is simple: analytics should improve a decision, not merely produce information.

Learning Objectives

By the end of this chapter, you should be able to:

1. Distinguish symptoms, problems, analytical questions, and decisions.
2. Translate managerial concerns into precise problem statements and measurable objectives.
3. Identify stakeholders, decision owners, affected groups, and information users.
4. Formulate hypotheses and specify the data needed to test them.
5. Define decision variables, constraints, assumptions, KPIs, and measurable outcomes.
6. Apply structured decision models, including weighted objectives and expected value.
7. Evaluate the scope, feasibility, ethics, and organizational value of an analytics project.
8. Communicate an analytical brief that is useful to both technical and managerial audiences.

1. From Managerial Concern to Analytical Problem

A managerial concern is an observation, tension, or opportunity that requires attention. A business problem is the underlying condition that prevents an organization from achieving a desired state. An analytical problem is a structured representation of that condition that can be investigated with evidence. A decision is a choice among alternatives under particular objectives and constraints.

These concepts should not be treated as interchangeable.

A useful transformation is:

Concern → desired state → gap → decision → analytical question → evidence → action → outcome

Suppose a retailer reports declining profit. Profit decline may result from lower volume, lower prices, higher product costs, increased returns, or an unfavorable product mix. “Analyze profit” is too broad. A better framing might be: “Which combination of price and promotional changes can increase quarterly contribution margin without reducing unit sales below the required threshold?” This version identifies an objective, decision variables, and a constraint.

Symptoms and root causes

A symptom is visible but may not be causal. Low employee productivity, for instance, may be associated with inadequate training, poor scheduling, unreliable systems, unclear targets, or low morale. An analyst should avoid assuming that the most visible measure represents the root cause.

Root-cause analysis can use process mapping, interviews, segmentation, historical comparisons, and causal reasoning. Correlation can identify useful patterns, but it does not by itself establish that changing one variable will produce a desired outcome. The distinction is especially important when the proposed intervention is costly or affects employees, customers, or citizens.

2. Stakeholder Analysis and Scope Definition

Stakeholders are people or groups who influence, use, are affected by, or are accountable for an analytical initiative. A project can fail even when its analysis is accurate if the decision owner does not trust it, operational teams cannot implement it, or affected groups were excluded from its design.

Typical stakeholder roles include:

• Decision owner: Has authority to act on the recommendation.
• Executive sponsor: Provides strategic support, resources, and escalation.
• Subject-matter expert: Explains operational context and process exceptions.
• Data owner or steward: Controls access, quality, definitions, and permitted use.
• Analyst or data scientist: Designs the measurement and analysis.
• Technology team: Provides systems, integration, deployment, and security.
• Frontline users: Apply the recommendation in daily work.
• Affected parties: Experience the consequences of the decision, whether or not they use the analysis.
• Regulator or auditor: May require evidence of compliance and explainability.

A stakeholder matrix can classify participants by influence and impact.

Stakeholder interviews should explore five questions:

1. What decision is currently difficult or delayed?
2. What outcome would indicate improvement?
3. What alternatives are genuinely available?
4. What constraints cannot be violated?
5. What evidence would make the recommendation credible?

Scope should specify the population, geography, period, process stage, decision owner, deliverable, and exclusions. “Improve customer service” is not a sufficient scope statement. “Recommend peak-hour staffing levels for the urban contact centre during the next six months, using service-level, labour-cost, and abandonment-rate measures” is more actionable.

A scope boundary prevents scope creep. It also identifies questions that may be important but cannot be answered in the current project because of time, data, authority, or ethical limits.

3. Objectives, Questions, and Hypotheses

Objectives

An objective states what the analysis is intended to achieve. Strong objectives are specific, measurable, achievable within the project’s authority, relevant to the organization, and time-bounded. This is commonly expressed as a SMART objective, although the framework should be used thoughtfully rather than mechanically.

An objective should distinguish between:

• Business objective: Improve profitable retention by the end of the financial year.
• Analytical objective: Estimate which customer characteristics and service experiences predict renewal.
• Operational objective: Deploy a prioritization rule for service teams.
• Measurement objective: Track renewal, complaint resolution, cost, and fairness after implementation.

The analytical objective is not necessarily the business objective. Predicting churn accurately is useful only if the organization can intervene effectively and economically.

Analytical questions

Analytical questions should specify the unit of analysis, outcome, explanatory factors, time period, and intended use. Examples include:

• Descriptive: What happened to monthly conversion rates by channel?
• Diagnostic: Which process stages explain the increase in failed applications?
• Predictive: Which accounts have a high probability of late payment in the next 30 days?
• Prescriptive: Which accounts should receive collection support, subject to a fixed staffing capacity?
• Causal: What is the estimated effect of a free-delivery offer on repeat purchase rate?

The last question is stronger than “Are customers who receive free delivery more loyal?” because it asks about the effect of an intervention, not merely an association.

Hypotheses

A hypothesis is a testable proposition about a relationship, difference, or mechanism. A good hypothesis is specific and falsifiable.

Example:

• H1: Reducing average response time from 24 hours to 12 hours increases the probability of customer renewal.
• H0: Reducing average response time from 24 hours to 12 hours does not change renewal probability.

A hypothesis should not be included merely to make a project appear scientific. Some projects are primarily descriptive or operational. Where causal claims are intended, the design must support them through appropriate comparisons, experiments, quasi-experimental methods, or carefully qualified observational analysis.

4. Decision Objectives, Variables, and Constraints

Decision models describe what can be chosen, what the organization is trying to achieve, and what limits apply.

Decision variables

A decision variable represents a controllable choice. Examples include:

• Number of employees assigned to each shift.
• Quantity ordered from each supplier.
• Price or discount level.
• Marketing budget allocated to each channel.
• Whether an application receives manual review.
• Delivery route selected for each shipment.

A forecast variable is not automatically a decision variable. Demand may be forecast, but it is not controlled directly. The distinction prevents analysts from recommending actions based on factors that managers cannot change.

Parameters and assumptions

Parameters are known or estimated quantities used by a model, such as unit cost, capacity, average processing time, or estimated response probability. Assumptions state conditions under which the model is considered reasonable. Examples include stable customer definitions, reliable cost estimates, or an assumption that available staffing can be reassigned.

Assumptions should be documented because changing them may change the recommendation.

Constraints

Constraints define what cannot be violated. Common constraints include:

• Budget limits.
• Labour or machine capacity.
• Legal and regulatory requirements.
• Minimum service levels.
• Inventory availability.
• Risk tolerances.
• Privacy and data-use restrictions.
• Equity or non-discrimination requirements.

A recommendation that maximizes revenue but violates a legal restriction is not a valid recommendation. Constraints are part of the problem definition, not an afterthought.

Objective functions

When a decision has one principal measurable objective, it may be represented as an objective function. For example, contribution margin can be expressed as:

Revenue and variable costs must be measured over the same period and scope, and both must use consistent monetary units. Fixed costs are excluded from contribution margin because the measure focuses on the amount available to cover fixed costs and profit.

A simple budget allocation problem can be written as:

Here, xᵢ is the amount allocated to option i, rᵢ is expected return per unit allocation, cᵢ is cost per unit allocation, B is the total budget, and Z is expected return. This model assumes returns are represented adequately by rᵢ, costs are additive, and interactions among options are ignored. Those assumptions may be unrealistic if channels compete for the same customers or show diminishing returns.

Multiple objectives

Business decisions often involve conflicting objectives. A hospital may seek to reduce waiting time while controlling cost and preserving equitable access. A retailer may seek to increase profit while maintaining customer satisfaction.

One approach is a weighted score:

where Score(j) is the score of alternative j, wₖ is the weight assigned to criterion k, sⱼₖ is the standardized score of alternative j on criterion k, and the weights usually satisfy ∑wₖ = 1. Scores must be standardized before combining different units such as dollars, hours, and satisfaction points. The method assumes that decision makers can express trade-offs through weights and that the criteria are sufficiently distinct to avoid double counting.

Weights are not objective facts. They encode values and priorities, so they should be discussed with decision owners and tested through sensitivity analysis.

5. KPIs and Measurable Outcomes

A KPI is a defined measure used to monitor performance against an objective. Not every metric is a KPI. A KPI has a clear purpose, owner, calculation, frequency, target or benchmark, data source, and interpretation rule.

A KPI specification should answer:

• What is being measured?
• What is the numerator and denominator?
• Who is included and excluded?
• What is the time period?
• What data source is authoritative?
• How often is it updated?
• What target or threshold applies?
• What action follows an unacceptable result?

Common formulas include:

The retention formula assumes that the beginning and ending customer populations are defined consistently and that new customers can be identified accurately. A rate without a denominator definition can be misleading.

Outcomes should be separated into:

• Input measures: Resources used, such as staff hours.
• Activity measures: Actions completed, such as calls made.
• Output measures: Immediate products, such as applications processed.
• Outcome measures: Effects, such as reduced default or improved renewal.
• Impact measures: Longer-term organizational or social consequences.

A project that increases the number of calls made but does not improve repayment may optimize an activity rather than the desired outcome. Analysts should also identify unintended outcomes, such as increased complaints, employee overload, exclusion of vulnerable groups, or deterioration in quality.

6. Structured Decision Models Under Uncertainty

A decision model makes alternatives and consequences explicit. A basic decision tree contains decision nodes, chance nodes, possible states, probabilities, and payoffs.

When probabilities and monetary outcomes are meaningful, expected monetary value (EMV) can support comparison:

EMV(a) is the expected monetary value of action a, P(s | a) is the probability of state s given action a, and V(a,s) is the payoff from action a when state s occurs. The probabilities for all mutually exclusive and exhaustive states under an action must sum to 1. EMV is appropriate as a decision aid when the payoffs and probabilities are credible and monetary value is a reasonable representation of preference. It should not be treated as a guarantee of the outcome.

Worked example

A firm must choose between launching a standard product or a premium product. Management estimates the following outcomes:

For the standard product:

For the premium product:

The premium product has the higher EMV, although it also has greater downside risk. The analysis should therefore include risk tolerance, cash-flow capacity, strategic positioning, and the quality of the probability estimates. A risk-averse decision maker may prefer the standard product despite its lower expected value.

Sensitivity analysis

Sensitivity analysis asks whether a recommendation changes when an assumption changes. In the example, the premium option is preferred when:

Solving for p gives a threshold of approximately 0.455. Thus, if the probability of strong demand is above about 45.5%, the premium option has the higher expected monetary value under the stated payoffs. This threshold is more informative than presenting one forecast as certain.

7. A Practical Problem-Framing Process

A repeatable process improves quality without turning analysis into bureaucracy.

Step 1: Identify the decision

State the choice, decision owner, deadline, and alternatives. If no decision can be identified, the project may be exploratory research rather than decision support.

Step 2: Define the desired outcome

Specify what improvement matters and how it will be recognized. Include financial, operational, customer, employee, and social dimensions where relevant.

Step 3: Map the system and stakeholders

Document process stages, actors, inputs, outputs, dependencies, and affected groups. This often reveals that the initial problem statement is located too late or too early in the process.

Step 4: Formulate analytical questions and hypotheses

Separate descriptive, diagnostic, predictive, and prescriptive questions. State causal claims cautiously and identify what evidence would support or contradict them.

Step 5: Specify measures and data

Define the unit of analysis, population, time window, variables, KPI formulas, data sources, and quality requirements. Check for missingness, bias, inconsistent definitions, and measurement changes over time.

Step 6: State objectives and constraints

Identify the objective function or objectives, available decision variables, hard constraints, assumptions, and risk limits.

Step 7: Choose an appropriate method

A simple comparison, process analysis, or dashboard may be sufficient. More complex methods should be justified by the decision, not selected for prestige. Forecasting is useful when timing and quantity matter; classification is useful when prioritizing cases; optimization is useful when choices must be allocated under constraints.

Step 8: Define success before analysis

Specify technical validation criteria and business success criteria. A model may have strong predictive performance but produce little value if actions are ineffective or too expensive.

Step 9: Plan implementation and monitoring

Assign ownership, determine how recommendations enter workflows, and define monitoring thresholds. A decision model is incomplete until someone can act on it and detect when it no longer performs adequately.

Step 10: Communicate limitations

Document uncertainty, excluded populations, assumptions, possible failure modes, and conditions under which the recommendation should be revisited.

8. Illustrative Case: Reducing E-Commerce Order Cancellation

The following case is fictional and intended for instructional use.

Background

Northstar Home, an online retailer, has observed that order cancellations have risen from 6% to 10% over two quarters. Senior management initially requests a predictive model to identify likely cancellations.

Problem framing

The initial request focuses on prediction, but the decision is not yet clear. Interviews reveal that the operations team can change inventory reservations, delivery promises, and customer communication, but it cannot immediately alter supplier lead times. The decision owner is the operations director. The relevant outcome is not simply cancellation prediction; it is preventable cancellation at an acceptable intervention cost.

A refined problem statement is:

“Which operational actions should Northstar take for orders at risk of cancellation during the next quarter, subject to service capacity and a requirement that intervention costs do not exceed the expected margin preserved?”

Stakeholders

The operations director owns the decision. Customer service supplies process knowledge. The inventory team owns stock data. Marketing is consulted because promotions affect demand. Customers are affected stakeholders. The data protection officer reviews the use of customer information.

Decision variables and constraints

Possible decision variables include whether to send a proactive update, offer an alternative product, expedite delivery, or take no action. Constraints include available customer-service capacity, limited expedited-shipping budget, truthful communication requirements, and exclusion of sensitive personal characteristics from operational targeting unless a lawful and justified basis exists.

Analytical questions

1. Which observable order characteristics are associated with cancellation?
2. Which cancellations are operationally preventable?
3. What is the estimated effect of each intervention on cancellation probability?
4. Which intervention yields the greatest expected preserved margin per service unit?

The first question is predictive; the third is causal and requires stronger evidence than a simple association.

KPIs and outcomes

Primary outcomes are cancellation rate, preventable cancellation rate, preserved contribution margin, and intervention cost. Guardrail measures include complaint rate, delivery-promise accuracy, customer fairness across regions, and employee workload.

Approach

The team first audits definitions. It discovers that cancellations caused by customer preference and cancellations caused by stock-outs were combined. The data is reclassified by reason. A baseline descriptive analysis shows that stock-out-related cancellations are concentrated in a small number of products and suppliers.

The team then pilots proactive inventory messages for comparable orders. A controlled design is preferred so that the intervention group can be compared with a suitable control group. If random assignment is operationally feasible, it provides stronger evidence about the intervention effect than comparing customers who chose to receive messages with those who did not.

A prioritization model is developed only after the intervention is shown to have value. The model ranks orders by estimated preventable cancellation risk and expected margin preserved. The operations director sets a daily capacity constraint for the number of cases that can receive manual attention.

Outcome

The pilot indicates that proactive communication reduces cancellations for stock-out-risk orders but has little effect on cancellations caused by customer preference. The company therefore narrows deployment to the relevant segment rather than contacting all customers. The final monitoring plan tracks cancellation by reason, intervention group, product, supplier, region, and customer-service workload.

Learning points

The case demonstrates five principles. First, a prediction request may conceal a broader decision problem. Second, outcome definitions must distinguish causes that require different actions. Third, intervention value depends on preventability and cost, not only risk prediction. Fourth, stakeholder and data-definition work can create more value than model complexity. Fifth, monitoring must include customer and operational harms as well as the headline KPI.

9. Ethics, Governance, and Responsible Use

Problem framing determines what is measured and whose interests are represented. Ethical risks therefore arise before model development.

Important governance questions include:

• Is the data collected lawfully and used for a legitimate purpose?
• Are individuals informed when their data materially affects decisions?
• Are sensitive attributes used appropriately for fairness assessment, prediction, or neither?
• Could a proxy variable reproduce discrimination?
• Can a decision be explained to an affected person?
• Is there a human review or appeal process?
• Are retention, access, security, and deletion rules defined?
• Are model outputs treated as uncertain estimates rather than unquestionable facts?

Fairness cannot be reduced to a single universal metric. Different measures may conflict, especially when base rates differ across groups. Teams should identify the relevant harm, affected population, legal context, and decision process. A model that is accurate on average may still perform poorly for a small group or allocate opportunities unfairly.

Governance also requires version control for definitions, data, models, thresholds, and assumptions. An apparently stable KPI may change when a system, policy, or data-capture process changes. Documentation should make such changes visible.

10. Technology, AI, and Contemporary Practice

Modern analytics platforms can accelerate problem discovery through natural-language querying, automated dashboards, and generative AI assistance. These tools are useful for summarizing data, suggesting hypotheses, drafting documentation, and identifying possible anomalies. They do not replace decision ownership, domain validation, causal reasoning, privacy review, or accountability.

Generative AI may produce plausible but unsupported explanations, confuse correlation with causation, or expose confidential information if used inappropriately. Analysts should verify generated content against authoritative data, restrict sensitive inputs, record material use of AI, and ensure that a qualified person reviews recommendations.

The growth of real-time analytics and automated decisions increases the importance of defining intervention thresholds, fallback procedures, monitoring drift, and assigning responsibility. Faster decisions are not necessarily better decisions if the underlying problem is poorly framed.

Challenges and Limitations

Problem framing involves judgment and cannot eliminate uncertainty. Common limitations include:

• Stakeholders may disagree about the true objective.
• Available data may measure a proxy rather than the desired outcome.
• Historical data may reflect previous policies and biases.
• Correlation may be mistaken for causation.
• Probabilities and payoffs may be estimated imprecisely.
• Multiple objectives may be difficult to combine without hiding value judgments.
• Organizational capacity may prevent implementation of the analytically optimal choice.
• KPIs may encourage gaming or harmful short-term behaviour.
• A model may degrade when customers, competitors, regulations, or processes change.

These limitations do not make analytics futile. They require explicit assumptions, sensitivity analysis, stakeholder review, pilot testing, and continuous monitoring.

Best-Practice Checklist

Before approving an analytics project, confirm that:

1. A specific decision and decision owner have been identified.
2. The problem is distinguished from its symptoms.
3. The population, period, process, and exclusions are defined.
4. Stakeholders and affected groups have been consulted.
5. Objectives, hypotheses, decision variables, and constraints are documented.
6. KPIs have precise definitions, denominators, owners, and targets.
7. Data quality, privacy, security, and access requirements are assessed.
8. The chosen method is proportionate to the decision.
9. Success criteria and guardrail measures are specified before implementation.
10. Assumptions, uncertainty, ethical risks, and limitations are communicated.
11. A monitoring and review plan is assigned to named owners.

Chapter Summary

Business analytics begins with a decision, not a dataset. Effective problem framing converts a managerial concern into a bounded problem statement, analytical questions, hypotheses, measurable objectives, and an explicit decision context. Stakeholder analysis identifies decision owners, data stewards, subject-matter experts, users, and affected groups. Decision models distinguish controllable decision variables from forecasts and parameters, while constraints represent budget, capacity, legal, ethical, and operational limits.

KPIs must be defined with clear numerators, denominators, populations, periods, sources, owners, and actions. Structured decision models can compare alternatives using objectives, weighted criteria, expected monetary value, and sensitivity analysis, but their assumptions and value judgments must remain visible. The fictional Northstar Home case showed why prediction alone is insufficient: value depends on whether an intervention is effective, feasible, affordable, and responsible.

Finally, analytics projects require governance throughout their lifecycle. Privacy, fairness, explainability, security, monitoring, and accountability are part of problem definition. A well-framed problem improves the likelihood that analysis will lead to a decision that is useful, implementable, and aligned with organizational and stakeholder needs.

Key Terms

Analytical question: A precisely stated question that can be investigated using evidence.

Business problem: A condition or opportunity requiring organizational action.

Constraint: A limit that a feasible decision must satisfy.

Decision owner: The person or group with authority to act on a recommendation.

Decision variable: A controllable choice represented in a decision model.

Expected monetary value: The probability-weighted average payoff of an action across possible states.

Guardrail metric: A measure used to detect unacceptable side effects or harms.

Hypothesis: A specific, testable proposition about a relationship, difference, or mechanism.

Impact measure: A longer-term consequence of an activity or intervention.

Key performance indicator: A defined measure used to monitor progress toward an important objective.

Objective function: A formal representation of what a model seeks to maximize or minimize.

Outcome: A result produced by an action or intervention.

Parameter: A known or estimated quantity used within a model.

Problem framing: The process of translating a managerial concern into a structured analytical and decision problem.

Root cause: An underlying condition that contributes to an observed problem.

Sensitivity analysis: Examination of how conclusions change when assumptions or inputs vary.

Stakeholder: A person or group that influences, uses, is affected by, or is accountable for a project.

Symptom: An observable indication of a problem that may not reveal its cause.

Unit of analysis: The entity about which observations and conclusions are made.

Review Questions

5 Short Answer Questions

1. How does a business problem differ from an analytical question?
2. Why must a decision owner be identified before an analytics project begins?
3. Distinguish a decision variable from a forecast variable.
4. What information should be included in a KPI definition?
5. Why is sensitivity analysis useful in decision modelling?

5 Long Answer Questions

1. Explain the process of converting a managerial concern into a well-defined analytical problem, using an original example.
2. Discuss the role of stakeholder analysis in defining scope, data requirements, and implementation feasibility.
3. Explain how objectives, decision variables, assumptions, and constraints interact in a decision model.
4. Compare descriptive, diagnostic, predictive, and prescriptive analytical questions. Give one business example of each.
5. Discuss the ethical and governance issues that should be considered before deploying an automated business decision.

3 Analytical / Application Questions

1. A university wants to reduce student withdrawal. Frame the problem by identifying the decision owner, analytical questions, possible decision variables, constraints, primary KPIs, and at least two guardrail measures.
2. A company has a $100,000 marketing budget and must choose among channels with different expected returns, costs, and audience restrictions. Design a decision model and explain what assumptions and sensitivity analyses would be necessary before allocating the budget.
3. A bank’s model predicts that some applicants are likely to default. Explain why prediction alone does not determine the lending decision. Identify stakeholders, possible interventions, ethical risks, outcome measures, and a suitable monitoring plan.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes a managerial concern from an analytical question?

A. A concern is always quantitative, while an analytical question is always qualitative.

B. A concern describes an observed issue, while an analytical question specifies what evidence should be established.

C. A concern identifies the final decision, while an analytical question identifies the budget.

D. A concern is produced by a model, while an analytical question is produced by a stakeholder.

Correct Answer: B. A concern describes an observed issue, while an analytical question specifies what evidence should be established.

Question 2

Which of the following is a decision variable in a staffing problem?

A. Forecast customer arrivals

B. Historical average waiting time

C. Number of employees assigned to each shift

D. Probability of a service complaint

Correct Answer: C. Number of employees assigned to each shift

Question 3

A project measures completed calls but the real objective is improved repayment. Completed calls are best classified as a(n):

A. Impact measure

B. Activity measure

C. Constraint

D. Decision variable

Correct Answer: B. Activity measure

Question 4

Using the chapter’s retention formula, a firm starts with 1,000 customers, gains 200 new customers, and ends with 900 customers. What is the retention rate?

A. 70%

B. 80%

C. 90%

D. 110%

Correct Answer: B. 80%

Question 5

Which condition is required for a basic expected monetary value calculation?

A. Every payoff must be positive.

B. All alternatives must have identical costs.

C. The possible states must be mutually exclusive and exhaustive, with credible probabilities.

D. The decision maker must be risk neutral in every context.

Correct Answer: C. The possible states must be mutually exclusive and exhaustive, with credible probabilities.

Question 6

Why should scores be standardized before combining cost, time, and satisfaction in a weighted model?

A. To eliminate all managerial judgment

B. To place measures with different units on a comparable scale

C. To guarantee that the highest-cost alternative wins

D. To convert all measures into probabilities

Correct Answer: B. To place measures with different units on a comparable scale

Question 7

Which stakeholder typically controls data access, definitions, and permitted use?

A. Data owner or steward

B. External customer

C. Frontline competitor

D. Marketing respondent

Correct Answer: A. Data owner or steward

Question 8

What is the main purpose of a guardrail metric?

A. To replace the primary business objective

B. To detect unacceptable side effects or harms

C. To increase the number of model variables

D. To guarantee causal identification

Correct Answer: B. To detect unacceptable side effects or harms

Question 9

A prediction model identifies customers likely to cancel, but no effective intervention is available. What is the most important framing concern?

A. Prediction accuracy automatically creates business value.

B. The model may not support an actionable decision or measurable outcome.

C. The project should ignore cancellation definitions.

D. The decision variable is the historical cancellation rate.

Correct Answer: B. The model may not support an actionable decision or measurable outcome.

Question 10

Which statement about generative AI in analytics is most appropriate?

A. It removes the need for stakeholder validation.

B. Its outputs should be accepted when they sound plausible.

C. It can assist analysis, but outputs require verification, privacy protection, and human accountability.

D. It should always be used instead of descriptive analysis.

Correct Answer: C. It can assist analysis, but outputs require verification, privacy protection, and human accountability.