Data Strategy, Analytics and Decision Intelligence

Introduction

Digital transformation depends less on the volume of data an organisation possesses than on its ability to convert data into reliable decisions and coordinated action. Organisations commonly invest in cloud platforms, data lakes, dashboards and artificial intelligence while continuing to experience inconsistent definitions, duplicated records, weak ownership and low trust. These are not merely technical problems. They are strategic and organisational problems involving operating models, accountability, capabilities, ethics and value creation.

A data strategy establishes how data will support organisational objectives. It specifies which data matters, who is accountable for it, how it is acquired and governed, which analytical capabilities are required, and how insights will influence decisions. Analytics extends this strategy by transforming data into evidence. Decision intelligence goes further: it designs the complete relationship among data, models, human judgement, decisions, actions and outcomes.

For managers, the central question is not simply “What happened?” but “Which decision should be made, by whom, under what assumptions, with what expected consequences, and how will we learn whether it worked?”

Learning Objectives

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

1. Explain the strategic role and components of an enterprise data strategy.
2. Distinguish data architecture, data quality, ownership, master data and governance.
3. Design practical approaches to data literacy and data operating models.
4. Apply descriptive, diagnostic, predictive and prescriptive analytics to managerial problems.
5. Evaluate experimentation, causal inference, dashboards and key performance indicators.
6. Explain how decision intelligence connects analytical insight with action.
7. Assess ethical, regulatory, security and organisational risks in data-driven management.

1. Data as a Strategic Asset

Data is a recorded representation of observations, transactions, entities or events. Information is data interpreted in context. Knowledge is an organised understanding of relationships, mechanisms and implications. A decision is a committed choice among alternatives, while an action is the implementation of that choice. The frequently used “data–information–knowledge–wisdom” hierarchy is useful pedagogically, but it should not be treated as a strict natural law: valuable decisions may also depend on experience, values, tacit knowledge and stakeholder judgement.

Data becomes strategically valuable when it improves a material organisational outcome—for example, customer retention, operating efficiency, risk control, innovation or regulatory compliance. Its value is conditional. A dataset may be accurate but irrelevant, timely but biased, comprehensive but inaccessible, or analytically rich but collected without legitimate authority.

A robust data strategy aligns five elements:

A useful strategic test is traceability: every major data investment should be traceable from an organisational objective to a decision, a required data product, a responsible owner, a control environment and an outcome metric.

2. Designing an Enterprise Data Strategy

2.1 Strategy formulation

Data strategy should begin with business priorities rather than technology procurement. A practical formulation sequence is:

1. Identify value pools and decisions. Map decisions such as pricing, credit approval, workforce planning, maintenance, inventory allocation or customer service.
2. Assess decision pain points. Examine delay, uncertainty, inconsistency, avoidable cost and weak accountability.
3. Map required data. Identify entities, events, measures, time horizons, granularity and lineage.
4. Assess current maturity. Evaluate architecture, quality, governance, skills, adoption and delivery performance.
5. Choose a target operating model. Decide which responsibilities are centralised, federated or embedded in business units.
6. Prioritise a portfolio. Select initiatives by value, feasibility, risk reduction and learning potential.
7. Define measures. Monitor both data performance and business outcomes.

A data strategy is not a one-time document. It is a portfolio of choices revised as business models, regulations, technologies and risks change.

2.2 Operating models

In a centralised model, a central data function controls standards, platforms and most delivery. This promotes consistency and economies of scale, but may create distance from business needs. In a federated model, business domains retain substantial responsibility while complying with enterprise standards. This increases contextual relevance but requires strong coordination. A data mesh extends the federated idea by treating domain-oriented data products, self-service platforms and federated computational governance as design principles. It is not simply a tool or an excuse to decentralise without accountability.

The appropriate model depends on regulatory exposure, organisational size, data interdependence, analytical maturity and the need for local responsiveness. Many enterprises use a hybrid model: central standards and platforms combined with accountable domain ownership.

3. Data Architecture and Data Products

Data architecture describes how data is collected, stored, integrated, processed, secured and made available. It includes sources, interfaces, pipelines, storage, metadata, identity management, analytical environments and consumption channels.

3.1 Architectural patterns

A data warehouse usually stores structured, curated data for consistent reporting and analysis. A data lake stores large volumes of raw or semi-structured data, often supporting exploratory work and machine learning. A lakehouse seeks to combine flexible storage with warehouse-like management, reliability and performance. These labels do not determine quality; architecture must be evaluated against workload, governance, cost, latency, security and skills.

Data may be processed through:

• Batch processing, appropriate when periodic updates are sufficient.
• Streaming processing, appropriate when events must be acted upon rapidly.
• Operational integration, connecting systems that run transactions.
• Analytical integration, preparing data for reporting, modelling and decision support.

Architecture should distinguish a system of record from a system of insight. A transaction system may be authoritative for a sale, while a governed analytical environment is authoritative for a monthly performance measure. This distinction prevents arguments based on vague claims that one database is universally “the truth.”

3.2 Data products and metadata

A data product is a deliberately designed, reusable data asset with defined users, purpose, quality expectations, interface, documentation, owner and support arrangements. Examples include a customer master, a daily inventory position or a credit-risk feature set.

Metadata describes data. Business metadata explains meaning and ownership; technical metadata describes schemas, formats and pipelines; operational metadata records freshness, failures and usage; lineage shows how data moved and changed from source to output. Without metadata, self-service often becomes uncontrolled interpretation.

4. Data Quality, Ownership and Master Data

4.1 Dimensions of quality

Data quality is fitness for a defined purpose, not an abstract property. Important dimensions include:

• Accuracy: correspondence with the relevant real-world value.
• Completeness: presence of required values or records.
• Validity: conformity to permitted formats, ranges or rules.
• Consistency: absence of unjustified contradictions across records or systems.
• Timeliness: availability within the required time window.
• Uniqueness: absence of unintended duplicate records.
• Integrity: preservation of relationships and referential constraints.

For a measurable quality dimension, a rate can be defined as:

The numerator and denominator must be explicitly defined. For example, a completeness rate for mandatory telephone numbers should count only records in scope and should specify whether an empty string, an invalid number or an unknown value fails the rule. A high rate does not prove that the rule itself is appropriate.

Quality improvement follows a lifecycle: define critical data elements, profile current data, identify root causes, assign remediation, prevent recurrence at source, monitor continuously and escalate material exceptions. Fixing defects only in a reporting layer may improve a dashboard while leaving the underlying process defective.

4.2 Ownership and stewardship

The data owner is accountable for a domain or data product, including meaning, access decisions, risk and quality expectations. A data steward performs operational coordination: documenting definitions, resolving issues, monitoring controls and facilitating cross-functional agreement. A custodian manages technical storage, availability and security on behalf of the owner. These roles may be performed by different people or, in smaller organisations, combined—but accountability must remain explicit.

A RACI matrix can clarify whether a role is responsible for doing work, accountable for the outcome, consulted before action or informed afterward. Ownership should attach to business domains such as customer, product, supplier, employee and finance—not only to IT systems.

4.3 Master and reference data

Master data represents relatively stable business entities, such as customers, products, locations and suppliers. Reference data consists of permitted classifications or codes, such as country codes, currencies and order statuses. Transactional data records events involving those entities, such as invoices or shipments.

Master data management (MDM) establishes identification, matching, survivorship, hierarchy and distribution rules. A common challenge is that the same customer appears under multiple identifiers. Matching may use deterministic rules, such as an exact government-approved identifier, or probabilistic rules based on several attributes. Probabilistic matching requires thresholds, review procedures and false-match controls; an incorrect merge can be more damaging than an unresolved duplicate.

5. Data Governance, Risk and Responsible Use

Data governance is the exercise of authority and control over data assets. It includes decision rights, policies, standards, accountabilities, controls and escalation mechanisms. Governance should enable responsible use, not merely create committees.

A practical governance framework covers:

1. Purpose and scope: what data and decisions are governed.
2. Policies and standards: definitions, retention, classification, access and quality.
3. Decision rights: who approves definitions, access, sharing and exceptions.
4. Control mechanisms: logging, monitoring, review, validation and audit.
5. Issue management: prioritisation, remediation and escalation.
6. Assurance: testing whether controls operate effectively.

Data classification commonly distinguishes public, internal, confidential and highly restricted information, although categories should reflect actual organisational risk. Access should follow least privilege: users receive only the access necessary for an authorised purpose. Encryption, key management, segregation of duties, backup, incident response and vendor controls complement—not replace—sound governance.

Privacy requires attention to lawful basis, purpose limitation, data minimisation, transparency, retention, individual rights and cross-border processing. Requirements differ by jurisdiction; managers must consult applicable law rather than assume that one global rule applies. De-identification reduces risk but is not automatically irreversible, especially when datasets can be linked.

Ethical analysis should ask: Who benefits? Who bears risk? Could the data encode historical disadvantage? Is the use understandable and contestable? Are proxies likely to reproduce protected characteristics? Responsible practice requires impact assessment, human oversight, proportionality, documentation and mechanisms for correction or appeal.

6. Data Literacy and Analytical Capability

Data literacy is the ability to read, work with, analyse and communicate data, including understanding uncertainty and limitations. It is not equivalent to knowing a spreadsheet function. Leaders need to interpret distributions, rates, trends, correlation, causation, sampling, measurement error and model risk.

A capability architecture may include:

• Data users, who interpret and apply governed information.
• Analysts, who frame questions, prepare data and communicate evidence.
• Analytics translators or product managers, who connect business decisions to analytical delivery.
• Data engineers, who build reliable pipelines and platforms.
• Data scientists and statisticians, who develop and validate models.
• Governance, privacy and risk specialists, who establish controls.

Effective literacy programmes use real organisational decisions, not isolated technical courses. They should teach how to challenge a metric, inspect a definition, recognise misleading visualisation and communicate uncertainty. Incentives and leadership behaviour matter: employees will not use evidence if leaders reward confident anecdotes over transparent analysis.

7. From Descriptive Analytics to Decision Intelligence

Analytics can be classified by the question it addresses:

The categories overlap. A forecast does not prescribe an action unless the decision context, alternatives, constraints and objective are specified.

7.1 Descriptive and diagnostic analysis

Descriptive analysis requires stable definitions and appropriate aggregation. A rate is not the same as a count, and averages can conceal variation. For observations x₁, x₂, ..., x_n, the arithmetic mean is:

where n is the number of observations. The mean is sensitive to extreme values; the median may better represent a skewed distribution. Diagnostic analysis uses segmentation, cohort analysis, drill-down, process mining and comparison with a relevant baseline. “Why” requires more than finding that two variables move together.

7.2 Predictive analytics

Predictive models estimate an unknown or future outcome from observed features. A simple linear model can be written as:

where yᵢ is the outcome for observation i, xᵢ is a predictor, β₀ is the intercept, β₁ is the expected change in outcome associated with a one-unit increase in xᵢ under the model, and εᵢ is the unexplained error. This model assumes an appropriate functional form and requires attention to independence, influential observations, variance and measurement quality. A regression coefficient is not automatically a causal effect.

For binary outcomes, logistic regression models the log odds:

where p is the predicted probability of the outcome and x is a predictor. The model is suitable for probabilities between zero and one, but interpretation, calibration, class imbalance and decision thresholds must be examined.

Models should be evaluated using a time-appropriate holdout or cross-validation, avoiding leakage from future information. Depending on the use case, evaluation may include mean absolute error, root mean squared error, calibration, precision, recall, discrimination, stability and subgroup performance. No single metric is universally sufficient.

7.3 Prescriptive analytics and optimisation

Prescriptive analysis combines predictions with objectives, constraints and available actions. A general optimisation formulation is:

Here x is a vector of decision variables, f(x) is the objective function, gⱼ(x) are constraint functions, bⱼ are limits and m is the number of constraints. The formulation is meaningful only when the objective, constraints, feasible alternatives and data assumptions are credible. An optimiser can produce a mathematically valid answer to a poorly specified managerial problem.

For example, workforce scheduling may maximise expected service coverage while constraining labour hours, legal rest periods, skills and budget. Managers must also consider robustness: a solution that is optimal under one forecast may be poor when demand is uncertain.

7.4 Decision intelligence

Decision intelligence treats decisions as designed organisational systems. It identifies the decision owner, trigger, alternatives, information, rules, model, authority level, action, feedback and outcome. It distinguishes:

• Operational decisions: frequent, rule-based and often suitable for automation.
• Tactical decisions: periodic resource allocation and prioritisation.
• Strategic decisions: infrequent, uncertain and value-laden, requiring judgement and scenario analysis.

A decision record should capture the information available at the time, the assumptions made, the chosen action, confidence, responsible person and later outcome. This enables learning without judging past decisions solely by outcomes: a sound decision can have a bad result because of chance, while a poor decision can succeed accidentally.

8. Experimentation, Causality and Measurement

Managers often want to know whether an intervention caused an outcome. Observational comparisons are vulnerable to confounding, selection bias, regression to the mean and changing conditions. Randomised controlled experiments address this by assigning eligible units to treatment and control groups by chance, subject to ethical and operational constraints.

For a simple experiment, the difference in average outcomes is:

where ȳ_T is the average outcome for the treatment group and ȳ_C is the average outcome for the control group. Interpretation requires a valid randomisation process, a clearly defined treatment, stable measurement, sufficient exposure and attention to interference, attrition and non-compliance.

A confidence interval quantifies sampling uncertainty under a specified statistical procedure; it does not guarantee that the estimate is practically important or that all sources of bias are absent. Managers should predefine the primary outcome, decision threshold, duration and stopping rules where feasible. Multiple testing, novelty effects and spillovers can produce misleading results.

When randomisation is not possible, quasi-experimental methods—such as difference-in-differences, regression discontinuity or matched comparisons—may be useful, but each depends on assumptions that must be tested and explained. Causal claims should therefore be proportionate to design strength.

9. Dashboards and Managerial Communication

A dashboard is a decision interface, not a collection of charts. Its design should begin with the decision and audience. A useful dashboard specifies the metric definition, owner, period, comparison, target, data freshness, uncertainty and expected action.

Good dashboard practice includes:

• limiting the number of prominent measures;
• distinguishing leading indicators from lagging outcomes;
• showing denominators and relevant populations;
• using consistent scales and meaningful baselines;
• enabling controlled drill-down without overwhelming users;
• displaying data freshness and quality warnings;
• separating monitoring from exploration;
• documenting definitions and escalation thresholds.

A KPI can be expressed as a measure of performance relative to a target, but targets may distort behaviour if they become the sole objective. Goodhart’s law—when a measure becomes a target, it can cease to be a good measure—highlights the need for balanced measures, qualitative review and controls against gaming.

10. Worked Managerial Example

A retailer wants to prioritise customers for a retention intervention. It has a predictive model estimating the probability of churn and an estimated treatment effect for each customer. The decision is not simply to contact customers with the highest churn probability. The firm should prioritise customers for whom intervention is likely to change behaviour, subject to budget, consent and contact-frequency constraints.

Suppose three customers have the following estimated values:

A simplified expected incremental value is:

For A, the value is 0.20 × 120 − 5 = 19. For B, it is 0.35 × 80 − 5 = 23. For C, it is 0.10 × 200 − 5 = 15. Under this simplified model, B should be prioritised before A and C, despite B having neither the highest churn probability nor the highest margin.

The calculation assumes that the incremental probability estimate is causal or sufficiently decision-relevant, the margin is correctly measured, contact cost is comparable, customers have consented, and contacting one customer does not affect another. In practice, the policy should be tested through a controlled experiment, monitored for unequal impact and revised when response behaviour changes.

11. Illustrative Case: Northstar Health Supplies

Background

Northstar Health Supplies, a fictional national distributor, operated separate systems for sales, inventory, procurement and delivery. Executives received weekly reports showing conflicting revenue and stock figures. Branch managers maintained spreadsheets to correct customer and product records.

Problem

The company wanted to reduce stock-outs while maintaining working capital. However, product codes differed across systems, inventory balances were not synchronised, and no business owner was accountable for the definition of “available stock.” Analysts spent most of their time reconciling data rather than evaluating decisions.

Analysis

A diagnostic review found three root causes. First, the same product was represented by multiple identifiers. Second, inventory events were recorded at different times and with inconsistent status codes. Third, dashboards reported a single stock number without distinguishing physical stock, reserved stock, damaged stock and stock in transit.

The executive team reframed the initiative around a decision: which replenishment action should be taken for each location and product each day? This definition clarified the necessary data and owners.

Approach and solution

Northstar established a product master with a governed identifier, hierarchy and matching process. The supply-chain director became accountable for inventory definitions, while data stewards in procurement and operations managed issue resolution. A data product combined orders, receipts, reservations, adjustments and transfers, with lineage and freshness indicators.

The organisation introduced a dashboard showing available-to-promise stock, forecast demand, supplier lead time, stock-out risk and recommended replenishment. Recommendations were subject to minimum order quantities, budget limits and service-level priorities. Planners retained authority to override recommendations but had to record the reason.

A pilot compared branches using the new decision process with similar branches using the existing process. The evaluation plan specified stock-out rate, inventory value, service level, override rate and data-quality exceptions before launch.

Outcome

The pilot produced more consistent definitions and reduced manual reconciliation. Managers reported greater confidence because the dashboard displayed assumptions and exceptions rather than presenting a single unexplained number. The company did not treat the pilot result as proof that the model would work permanently; it continued monitoring supplier disruption, forecast drift and branch behaviour.

Learning points

1. The initiative succeeded because it began with a decision and not a dashboard.
2. Master data and ownership were prerequisites for useful analytics.
3. A recommendation requires constraints, authority and an override process.
4. Data quality should be measured at the point where errors originate.
5. Evaluation must include business outcomes, adoption and unintended effects.

12. Challenges and Limitations

Data strategy fails when organisations confuse availability with usefulness. Common failure modes include collecting data without a decision use, implementing tools before agreeing definitions, centralising all work and creating bottlenecks, decentralising without standards, optimising local metrics at the expense of enterprise outcomes, and treating model accuracy as the same as business value.

Analytics also has epistemic limits. Historical data may not represent future conditions. Models can amplify measurement error, institutional bias and missing populations. A forecast becomes less reliable when behaviour changes in response to the forecast itself. A dashboard can create false precision when definitions are unstable. Automation can scale a flawed policy faster than a human process would.

Technology introduces additional risks: cloud concentration, vendor lock-in, cyberattack, insecure interfaces, uncontrolled copies, model supply-chain risk and unsustainable computational cost. Contemporary generative AI increases the need for source grounding, access controls, prompt and output logging, human review, intellectual-property safeguards and testing for hallucination and confidential-data leakage. Generative systems can assist exploration and communication, but they do not eliminate the need for governed source data or accountable decisions.

13. Best Practices and Contemporary Direction

A mature organisation generally follows these principles:

• tie every major data initiative to a decision and measurable outcome;
• define critical data elements and business terms before scaling consumption;
• assign accountable domain owners and operational stewards;
• build quality controls into source processes and pipelines;
• treat metadata, lineage and documentation as product features;
• use the least data and least privilege necessary for a legitimate purpose;
• validate models on realistic, time-appropriate data;
• monitor drift, calibration, fairness, adoption and outcome impact;
• separate prediction from causal claims and prescription;
• test important interventions rather than relying solely on correlation;
• design dashboards around action, not visual novelty;
• maintain human accountability for consequential decisions;
• review data and AI systems throughout their lifecycle.

Emerging practice is moving toward real-time decisioning, event-driven architecture, privacy-enhancing technologies, synthetic data, knowledge graphs, decision automation and AI copilots. These developments are valuable only when they improve a clearly defined decision while preserving traceability, security, fairness and organisational learning.

Chapter Summary

Data strategy connects business priorities with data assets, capabilities, governance and adoption. Its effectiveness depends on beginning with important decisions rather than beginning with technology. Data architecture provides the structural foundation; data products, metadata and lineage make information discoverable and interpretable. Quality must be defined for a purpose and managed through ownership, stewardship and prevention. Master data management creates consistent representations of core entities, while governance establishes decision rights, controls, privacy and accountability.

Analytics progresses from describing and diagnosing events to predicting outcomes and prescribing actions. Predictive accuracy alone does not establish causality or business value. Prescriptive analytics requires explicit objectives, constraints and feasible alternatives. Decision intelligence integrates data, models, human judgement, action and feedback into a learning system. Experiments and quasi-experimental designs strengthen causal understanding, while dashboards should function as disciplined decision interfaces. Ethical, regulatory, security and organisational considerations are integral to responsible data use, particularly as AI expands the scale and speed of analytical work.

Key Terms

Data strategy: A coordinated plan for using data to achieve organisational objectives.

Data architecture: The structures and technologies used to collect, store, integrate, process and use data.

Data product: A governed, reusable data asset designed for defined users and decisions.

Metadata: Information describing data’s meaning, structure, origin, use and operation.

Data lineage: A record of data’s movement and transformation from source to output.

Data quality: Fitness of data for a specified purpose.

Data owner: The person accountable for a data domain or product and its appropriate use.

Data steward: The role responsible for operational data coordination and quality management.

Master data: Consistent records of core entities such as customers, products or suppliers.

Reference data: Controlled codes or classifications used to describe data.

Data governance: Authority, policies, standards and controls governing data.

Data literacy: The ability to understand, analyse, communicate and use data responsibly.

Descriptive analytics: Analysis of what has happened.

Predictive analytics: Estimation of likely future or unknown outcomes.

Prescriptive analytics: Analysis recommending actions under objectives and constraints.

Decision intelligence: Design of the relationship between information, models, decisions, actions and feedback.

Experimentation: Controlled intervention used to estimate effects.

Dashboard: A visual decision interface presenting selected measures and context.

Model drift: Deterioration in model behaviour as data, environments or relationships change.

Review Questions

Short Answer Questions

1. Why should a data strategy begin with business decisions rather than technology platforms?
2. Distinguish a data owner, data steward and data custodian.
3. What is the difference between master data and transactional data?
4. Why is predictive accuracy insufficient for evaluating a managerial model?
5. Identify four dimensions of data quality.

Long Answer Questions

1. Explain how data architecture, metadata, lineage and data products contribute to trustworthy analytics.
2. Compare centralised, federated and hybrid data operating models, including their advantages and risks.
3. Explain the progression from descriptive analytics to decision intelligence, using a managerial example.
4. Discuss the role of governance, privacy, fairness and human accountability in analytical decision-making.
5. Design a data strategy for an organisation seeking to improve customer retention. Include ownership, quality, analytics, experimentation and adoption.

Analytical / Application Questions

1. A bank’s churn model identifies customers with the highest probability of leaving, but its retention offers are expensive and effective only for some customers. Design a decision rule that combines predicted risk, incremental treatment effect, value and constraints. State the assumptions that must be tested.
2. A manufacturer reports that a new forecasting system reduced stock-outs by 10% after implementation. Propose an evaluation design that distinguishes causal impact from seasonal variation, supplier changes and regression to the mean.
3. An organisation has three conflicting definitions of “active customer.” Develop a governance and data-product process to establish a trusted definition, assign accountability, document lineage and manage legitimate use of alternative definitions.

Multiple Choice Questions with Answers

Question 1

Which statement best describes a data product?

A. A raw database copy with no designated users

B. A reusable, governed data asset designed for defined users and decisions

C. A dashboard containing only financial measures

D. A software licence for storing data

Correct Answer: B. A reusable, governed data asset designed for defined users and decisions

Question 2

Which role is primarily accountable for the meaning, appropriate use and quality expectations of a data domain?

A. Data owner

B. Database operator

C. Dashboard viewer

D. External auditor

Correct Answer: A. Data owner

Question 3

A retailer has 800 records in scope, of which 760 satisfy a completeness rule. What is the quality rate?

A. 5%

B. 40%

C. 95%

D. 105%

Correct Answer: C. 95%

Question 4

Which analytical question is primarily predictive?

A. What was last quarter’s revenue?

B. Why did delivery delays increase?

C. Which customers are likely to churn next month?

D. Which staffing plan maximises service subject to constraints?

Correct Answer: C. Which customers are likely to churn next month?

Question 5

In the simple model yᵢ = β₀ + β₁xᵢ + εᵢ, what does β₁ represent under the stated model interpretation?

A. The intercept

B. The unexplained error for observation i

C. The expected change in outcome associated with a one-unit predictor increase

D. The number of observations

Correct Answer: C. The expected change in outcome associated with a one-unit predictor increase

Question 6

A campaign produces an incremental retention probability of 0.25, a contribution margin of 100 monetary units and a contact cost of 8. What is the simplified expected incremental value?

A. 17 monetary units

B. 25 monetary units

C. 33 monetary units

D. 108 monetary units

Correct Answer: A. 17 monetary units

Question 7

Which feature is essential to a meaningful prescriptive analytics problem?

A. A colourful visualisation

B. An objective, feasible alternatives and constraints

C. A very large raw dataset regardless of purpose

D. A prediction with no available action

Correct Answer: B. An objective, feasible alternatives and constraints

Question 8

What is the principal purpose of random assignment in a controlled experiment?

A. To guarantee a positive result

B. To eliminate the need for measurement

C. To reduce systematic differences between treatment and control groups

D. To make the sample larger than the population

Correct Answer: C. To reduce systematic differences between treatment and control groups

Question 9

Which dashboard practice most directly reduces misleading interpretation?

A. Removing denominators from rate measures

B. Showing metric definitions, freshness and relevant baselines

C. Using as many colours as possible

D. Displaying only favourable outcomes

Correct Answer: B. Showing metric definitions, freshness and relevant baselines

Question 10

Why should a model be evaluated on time-appropriate holdout data?

A. To introduce future information into training

B. To estimate performance under conditions resembling intended use

C. To avoid defining the target variable

D. To guarantee that the model is causal

Correct Answer: B. To estimate performance under conditions resembling intended use