12. Campaign Measurement, Attribution, and Experimentation

Introduction: From Media Delivery to Business Evidence

Media planning is incomplete when it ends with the delivery of impressions, reach, frequency, clicks, or completed views. These indicators describe what media did, but they do not necessarily establish whether advertising changed consumer behaviour or created business value. Advanced media planning therefore requires a measurement system that connects investment and exposure with intermediate responses, brand outcomes, and commercial results.

Campaign measurement is the disciplined process of defining objectives, collecting evidence, analysing performance, and using findings to improve decisions. Attribution attempts to allocate observed outcomes among media channels, touchpoints, campaigns, and other influences. Experimentation goes further by asking a more demanding question: what would have happened without the campaign? This counterfactual question is central to incrementality, because observed conversions may have occurred even in the absence of advertising.

The measurement problem is difficult because marketing outcomes are produced by multiple and interacting causes. Price, distribution, seasonality, competitor activity, economic conditions, product availability, customer service, brand reputation, and media exposure may all influence sales. A sophisticated measurement framework must therefore combine descriptive metrics, causal methods, statistical modelling, and managerial judgement.

This chapter develops such a framework. It examines KPI design, measurement architecture, dashboards, attribution models, marketing-mix modelling, brand-lift research, geo-testing, incrementality experiments, econometrics, and post-campaign evaluation. The emphasis is not on selecting one universally superior method, but on matching the method to the decision, data, campaign objective, and level of causal certainty required.

Learning Objectives

After studying this chapter, students should be able to:

1. Construct a measurement framework linking media inputs to business outcomes.

2. Distinguish output, outcome, and business-value metrics.

3. Evaluate the strengths and weaknesses of attribution models.

4. Explain the role of marketing-mix modelling and econometric analysis.

5. Design brand-lift, geo-testing, and incrementality experiments.

6. Interpret dashboards without confusing correlation with causation.

7. Identify sources of bias, data-quality problems, privacy risks, and measurement error.

8. Conduct a rigorous post-campaign evaluation and translate findings into planning decisions.

The Measurement Chain: Inputs, Responses, and Outcomes

A campaign measurement system should represent the logic through which advertising is expected to create value. A useful chain contains five levels.

Inputs are resources placed into the campaign: budget, creative assets, audience definitions, bids, inventory, distribution, and media weight. Inputs answer the question, “What did the organisation deploy?”

Outputs are delivery results, such as impressions, reach, frequency, viewability, completed views, clicks, or cost per thousand impressions. These indicate whether the campaign was delivered as planned. They do not prove that people noticed, remembered, or acted upon the advertising.

Exposure and response indicators describe what happened between delivery and commercial action. Examples include attention, video completion, website visits, search activity, ad recall, message association, consideration, product-page engagement, or store visits.

Business outcomes include leads, purchases, subscriptions, revenue, margin, retention, customer lifetime value, and market share. These are usually closer to managerial objectives but may be influenced by many factors beyond media.

Strategic and social outcomes may include brand equity, customer trust, cultural relevance, accessibility, or responsible communication. These are sometimes less immediate and more difficult to quantify, but they can determine long-term performance.

The chain should not be interpreted as a simple linear funnel. People may encounter several touchpoints, move backwards and forwards between stages, or purchase without measurable digital interaction. Nevertheless, the chain helps planners avoid a common error: treating a highly measurable intermediate event, such as a click, as a complete representation of business impact.

A campaign objective should therefore be translated into a measurement hierarchy. For a new product launch, reach, awareness, brand association, trial, and distribution-weighted sales may be appropriate. For a mature e-commerce brand, incremental orders, contribution margin, repeat purchase, and customer quality may matter more. For a public-service campaign, behavioural compliance or service uptake may be more meaningful than revenue.

Designing KPIs That Support Decisions

A key performance indicator is useful only when it is connected to an objective, has a clear definition, and can influence a decision. A large collection of metrics may create the appearance of analytical sophistication while obscuring what matters.

Effective KPI design begins with the business question. “Did the campaign perform well?” is too vague. More useful questions include: Did advertising increase consideration among the priority audience? Did paid search generate incremental demand or merely capture existing intent? Did television add reach beyond online video? Did the campaign acquire profitable customers? Did investment in one region produce greater incremental sales than investment in another?

KPIs can be classified into four groups:

• Delivery KPIs: reach, frequency, impressions, viewability, completion rate, effective frequency, and cost metrics.
• Engagement or response KPIs: clicks, landing-page sessions, search lift, interactions, app installs, enquiries, and store visits.
• Brand KPIs: awareness, recall, consideration, preference, perceived differentiation, message takeout, and purchase intent.
• Commercial KPIs: incremental conversions, revenue, margin, return on ad spend, customer acquisition cost, retention, and lifetime value.

A KPI should have an operational definition. For example, “conversion” must specify the event, attribution window, deduplication rule, eligibility criteria, and whether the number is observed or estimated. “Reach” must state the population, time period, identity resolution method, and whether it is person-based or device-based.

Leading indicators are available early and may support optimisation, but they are often imperfect predictors of final outcomes. Lagging indicators are closer to business value but may arrive too late for in-flight action. A balanced framework uses both. A campaign may monitor delivery and attention daily, brand response weekly or through periodic survey waves, and sales or profitability over a longer evaluation period.

Metrics also require guardrails. A media team that optimises only to cost per acquisition may favour cheap but low-value customers, over-target existing buyers, or sacrifice brand-building reach. Guardrails can include minimum new-customer share, profitability thresholds, frequency limits, quality scores, and brand-safety requirements.

Measurement Architecture and Dashboard Design

A measurement architecture defines how data moves from collection to interpretation. At the foundation are data sources: ad servers, demand-side platforms, social platforms, web analytics, customer-relationship systems, point-of-sale records, research panels, surveys, and external economic or market data. These sources differ in scope, identity, time resolution, definitions, and reliability.

The next layer is data governance. Teams must establish common naming conventions, campaign taxonomies, time zones, currency rules, event definitions, consent status, and ownership. Without these controls, a dashboard may combine incompatible measurements. One platform may report clicks on an event-time basis, another on a conversion-time basis, and a sales system on a transaction-date basis.

The analytical layer transforms raw data into indicators and models. It should preserve distinctions among observed, modelled, and experimentally estimated values. A dashboard should clearly label whether a number is a platform-reported conversion, a deduplicated conversion, or an estimated incremental conversion.

A decision-oriented dashboard normally answers four questions: Was delivery achieved? Did the intended audience receive meaningful exposure? Did brand or behavioural responses change? What action should follow? Visual simplicity is valuable, but simplicity should not conceal uncertainty. Confidence intervals, sample sizes, data freshness, and known limitations should be visible where relevant.

Dashboards are most effective when they are designed around decision cycles. An operational dashboard may support daily budget and pacing decisions. A campaign dashboard may support weekly optimisation. An executive scorecard may review strategic outcomes monthly or quarterly. These should not be treated as interchangeable products.

Attribution: Allocating Credit Without Confusing Credit with Causality

Attribution assigns observed conversions or revenue to marketing touchpoints. It is useful for organising data and understanding customer journeys, but attribution does not automatically establish that a touchpoint caused the outcome. A person who clicks a retargeting advertisement shortly before purchase may already have been highly likely to buy.

Common Attribution Approaches

Single-touch attribution assigns all credit to one event. In first-touch attribution, credit goes to the first recorded interaction; in last-touch attribution, it goes to the final interaction. These models are simple and easy to implement, but they ignore other influences and are highly sensitive to tracking conventions.

Rule-based multi-touch attribution distributes credit according to a predetermined rule. Linear attribution gives equal credit to each touchpoint. Position-based attribution gives greater weight to the first and last interactions. Time-decay attribution gives more credit to recent interactions. These rules can be transparent, but their weights are assumptions rather than proof of causal contribution.

Algorithmic or data-driven attribution uses statistical or machine-learning methods to estimate the association between touchpoints and outcomes. Such systems may model removal effects, conversion probability, or path patterns. They can process complex journeys, but results depend on data quality, identity resolution, model specification, and the availability of suitable controls. A sophisticated algorithm can still produce biased conclusions if exposure is not randomly assigned.

Platform attribution is reported by an advertising or analytics platform using its own tracking, identity graph, conversion window, and modelling. Platform numbers are valuable for operational management, but they are not necessarily comparable across platforms. Each platform has an incentive and a methodological reason to claim credit for conversions it can observe.

Attribution Windows and Deduplication

An attribution window defines how long after exposure a conversion may be associated with the campaign. A short window may understate delayed effects; a long window may claim conversions that would have occurred independently. View-through attribution is particularly sensitive because exposure may be recorded without an active click, and the causal link may be weak.

Deduplication is essential when several platforms claim the same conversion. A unified event structure should establish a hierarchy of authoritative records and rules for resolving multiple identifiers. Where deterministic identification is unavailable, the organisation may need modelled estimates rather than pretending to have exact individual-level truth.

Attribution is most useful for questions such as journey description, operational optimisation, creative comparison within a controlled environment, and identifying under-measured touchpoints. It is less reliable as the sole basis for reallocating an entire media budget, especially when channels differ in their ability to observe users.

Marketing-Mix Modelling and Econometric Reasoning

Marketing-mix modelling (MMM) estimates the relationship between business outcomes and multiple explanatory variables, commonly using aggregated time-series or regional data. A model may relate sales to television spend, digital video, search, promotions, price, distribution, seasonality, competitor activity, weather, and macroeconomic conditions.

A simplified formulation is:

Sales_t = baseline_t + media effects_t + promotional effects_t + external effects_t + error_t

Media effects may incorporate adstock, which represents the persistence of advertising influence over time, and saturation, which represents diminishing returns as exposure increases. A channel may therefore have both a delayed effect and a nonlinear response curve.

MMM is valuable because it can include offline media, estimate longer-term effects, and operate without user-level tracking. It is particularly useful for strategic budget allocation and scenario planning. However, it requires sufficient variation in investment and outcomes. If a channel is always increased when demand is already expected to rise, the model may incorrectly attribute demand to media. This is a form of endogeneity.

Econometric quality depends on model specification, data granularity, variable selection, treatment of trends, seasonality, autocorrelation, multicollinearity, and validation. A model should not be accepted merely because it produces a high goodness-of-fit measure. Predictive fit is not the same as causal validity. Teams should examine residuals, conduct holdout validation, test alternative specifications, and compare estimates with experiments where possible.

MMM and attribution answer different questions. Attribution often operates at a more granular level and focuses on observed journeys. MMM estimates aggregate contribution and can capture channels that are poorly tracked. A mature organisation uses them as complementary evidence rather than forcing one method to replace the other.

Brand Lift and the Measurement of Intermediate Effects

Brand-building campaigns may influence memory, attitudes, and future choice before they influence sales. Brand-lift studies measure changes in outcomes such as awareness, ad recall, message association, consideration, favourability, or purchase intent among exposed and comparison groups.

A typical design surveys individuals who were exposed to a campaign and compares them with a control or unexposed group. The difference in change between the groups is often expressed as a lift estimate. For example:

Incremental lift = (Post-exposure outcome − Pre-exposure outcome) in the exposed group − (Post-exposure outcome − Pre-exposure outcome) in the comparison group.

The method is stronger when exposure is assigned or accurately measured, groups are comparable, samples are adequate, and survey questions are validated. It is weaker when people self-select into exposure, when exposed respondents are more engaged for reasons unrelated to advertising, or when the survey is conducted immediately after an unusually salient event.

Brand lift should be interpreted as evidence of a change in an intermediate outcome, not automatically as proof of incremental sales. It is especially valuable when the campaign objective is awareness, repositioning, differentiation, or message communication. It can also diagnose creative problems: a campaign may achieve high recall but weak message association, or strong message association without improving consideration.

Geo-Testing and Incrementality Experiments

Incrementality asks what additional outcome occurred because of advertising. The ideal experiment randomly assigns eligible units to treatment and control groups, exposes the treatment group to advertising, withholds or reduces advertising for the control group, and compares outcomes.

In digital environments, user-level randomised holdout tests may be possible. In other situations, the experimental unit may be a household, store, postcode, city, sales territory, or region. Geo-testing is useful when media can be varied geographically and outcomes are available at the same geographic level.

A geo-experiment should begin with a clear hypothesis, outcome metric, treatment period, control strategy, and power assessment. Treatment and control regions should be similar in historical performance, customer mix, media availability, and economic conditions. The design may use matched markets, synthetic controls, or a pre-period model to estimate the counterfactual path for the treated region.

The basic treatment effect can be expressed as:

Incremental effect = Outcome in treatment − Counterfactual outcome without treatment.

A difference-in-differences design compares changes over time in treatment and control areas. Its credibility depends on the parallel-trends assumption: absent the campaign, treatment and control areas would have followed sufficiently similar trends. Analysts should inspect pre-period trends and test for contamination, spillover, concurrent promotions, and unequal measurement quality.

Experiments can estimate incremental conversions, revenue, profit, or customer quality. They may also reveal that a channel produces substantial attributed conversions but little incrementality, or that a channel with modest last-click credit creates valuable additional demand.

The cost of experimentation must be considered. Holding out advertising may sacrifice some short-term outcomes, and regional tests may require operational coordination. Nevertheless, learning can be more valuable than a small amount of unmeasured delivery, particularly when a decision involves a large recurring budget.

Post-Campaign Evaluation as a Management Process

Post-campaign evaluation should not be a retrospective collection of platform screenshots. It should return to the original decision and assess whether the campaign achieved its intended purpose.

The process begins by documenting the evaluation question, objectives, target audience, markets, flight dates, planned investment, creative strategy, and success criteria. Actual delivery is then reconciled against the plan. Variances should be explained rather than merely reported.

The second stage examines delivery quality: reach, frequency, viewability, brand safety, placement quality, audience composition, and cost efficiency. The third examines response and brand outcomes. The fourth estimates incremental business effects using the strongest available evidence, such as experiments, calibrated attribution, MMM, sales analysis, or triangulation across methods.

Evaluation should separate three conclusions:

1. What happened? This is descriptive performance.

2. What was caused by the campaign? This is the causal or incremental estimate.

3. What should be changed? This is the managerial recommendation.

The final report should include uncertainty, data limitations, and implications for the next planning cycle. Recommendations may concern budget allocation, audience strategy, creative sequencing, frequency management, channel roles, measurement design, or operational processes.

Illustrative Case: Measuring a Multi-Channel Subscription Campaign

Background

The following is an illustrative case. A national streaming service launches a campaign for a new annual subscription. The media plan includes connected television, online video, paid social, paid search, display retargeting, audio, and partnerships. The business objective is not simply to increase sign-ups, but to acquire subscribers who remain active and profitable after six months.

Situation and Problem

After eight weeks, platform reports show strong performance. Paid search claims the lowest cost per acquisition, retargeting reports a high conversion rate, and online video reports substantial view-through conversions. However, the finance team observes that total subscriptions are below the forecast in several regions. Senior management must decide whether to shift more budget toward search and retargeting or preserve investment in video and connected television.

Analysis

The planning team first defines a measurement hierarchy. Delivery KPIs include incremental reach, effective frequency, completion rate, and viewability. Brand KPIs include awareness of the annual plan, consideration, and message association. Commercial KPIs include new subscriptions, six-month retention, contribution margin, and incremental cost per retained subscriber.

The team identifies three measurement problems. First, platform attribution windows overlap, allowing several channels to claim the same subscription. Second, search and retargeting mainly reach people who have already visited the website. Third, subscription quality differs by source; a cheap initial sign-up is not necessarily a profitable customer.

The team conducts a regional geo-test. Matched regions are assigned to a treatment condition with increased video and connected-television weight and a control condition with the existing media mix. Paid search is held at a stable level in both groups to reduce confounding. Historical subscription trends are examined to confirm that the regions had parallel pre-campaign patterns.

A brand-lift survey is also conducted among the priority audience. It measures awareness, annual-plan consideration, understanding of the principal value proposition, and intention to subscribe. Website analytics and customer records are linked through privacy-compliant aggregate reporting rather than unrestricted sharing of individual-level data.

Results and Decision

The evaluation finds that retargeting has a strong attributed conversion rate but produces limited incremental subscriptions when reduced in the holdout test. Search produces genuine incremental demand, but its marginal efficiency declines after a certain investment level. Video and connected television generate modest direct attribution but increase awareness and branded search activity. The treatment regions record a stronger increase in subscriptions and better six-month retention than the matched control regions.

Management therefore rejects a simple last-click reallocation. It maintains a reach-and-demand-generation role for video and connected television, funds search up to its estimated marginal-return threshold, reduces overreliance on retargeting, and introduces a retained-subscriber KPI. The next campaign includes a larger pre-registered experiment and a clearer separation between prospecting and existing-customer communication.

Learning Points

The case demonstrates that attributed efficiency is not identical to incremental efficiency. It also shows why brand measures, experiments, and customer-quality outcomes should be combined. The most useful result is not a single winning channel, but a clearer understanding of the role each channel plays in the customer journey and the level at which additional investment remains productive.

Challenges, Risks, and Common Errors

Measurement systems face several recurring problems. Selection bias occurs when people who are more likely to buy are also more likely to be exposed. Survivorship bias can arise when analysis includes only users who remain trackable or active. Confounding occurs when another variable changes alongside media. Leakage occurs when control groups are exposed through other channels. Double counting occurs when platforms claim the same event.

Other errors include optimising to cheap rather than valuable conversions, treating viewability as attention, interpreting correlation as causation, ignoring delayed effects, using inconsistent conversion definitions, and evaluating campaigns only at the end of the purchase funnel. Small samples can produce unstable lift estimates, while large samples can make trivial differences appear statistically significant.

Measurement can also be distorted by platform incentives. Independent validation, transparent methodology, and triangulation are important safeguards. Analysts should distinguish statistical significance from managerial significance and report practical effect sizes, uncertainty ranges, and assumptions.

Ethics, Privacy, and Governance

Campaign measurement uses information about people, households, devices, locations, and behaviour. Ethical measurement requires lawful collection, meaningful consent where required, purpose limitation, data minimisation, secure storage, controlled access, and appropriate retention periods. Identity graphs and probabilistic matching should not be treated as infallible representations of individuals.

Organisations should avoid sensitive targeting or measurement practices that create unfair exclusion, discriminatory outcomes, or intrusive profiling. Governance should specify who may access raw data, how vendors are audited, how consent signals are respected, and how model outputs are challenged. Privacy-enhancing methods, including aggregation, clean rooms, contextual measurement, and carefully designed experiments, can reduce dependence on unrestricted individual tracking.

Professional responsibility also includes honest reporting. Analysts should not suppress inconvenient control results, present modelled numbers as observed facts, or claim causal certainty unsupported by the design. A credible measurement culture values learning over the appearance of success.

Technology, Automation, and Emerging Practice

Modern measurement increasingly uses cloud data platforms, clean rooms, automated taxonomy checks, identity resolution, causal-inference software, and machine-learning models. Automation can improve pacing alerts, anomaly detection, data validation, and dashboard refreshes. Machine learning can support prediction and segmentation, but predictive accuracy does not by itself establish incrementality.

Privacy changes and the decline of some third-party identifiers are encouraging a shift toward first-party data, consented relationships, contextual signals, aggregated measurement, media-mix models, and experiment-based validation. Cross-media measurement is also developing as organisations seek comparable reach and frequency across television, streaming, audio, social, search, and out-of-home media.

The emerging direction is not a single replacement for cookies or attribution. It is a portfolio approach in which operational analytics, experiments, MMM, brand research, and customer-value analysis are calibrated against one another. The most advanced organisations treat measurement as an organisational capability rather than a reporting task.

Best-Practice Principles

A robust measurement programme should:

• Begin with a decision and business objective rather than with available data.
• Define metrics, conversion events, attribution windows, and ownership before launch.
• Separate delivery, response, brand, and incremental business measures.
• Use experiments whenever a material causal decision can be tested.
• Combine user-level, aggregate, survey, and financial evidence appropriately.
• Report uncertainty, data limitations, and assumptions.
• Evaluate customer quality and profitability, not only acquisition volume.
• Protect privacy and maintain governance over identities and sensitive information.
• Avoid allowing any platform to be the sole judge of its own performance.
• Carry learning from every campaign into the next planning cycle.

Chapter Summary

Campaign measurement connects media investment with brand and business outcomes. A strong framework distinguishes inputs, delivery outputs, intermediate responses, brand effects, and commercial results. KPIs must be tied to decisions, clearly defined, and balanced across short- and long-term objectives.

Attribution helps describe and allocate observed conversions, but it does not automatically prove causality. Attribution windows, platform incentives, identity limitations, and selection bias can distort results. Marketing-mix modelling provides an aggregate, strategic perspective and can include offline and long-term effects, while brand-lift studies measure changes in awareness and attitudes. Geo-tests and other incrementality experiments provide stronger evidence about what advertising caused by estimating the counterfactual outcome.

Post-campaign evaluation should distinguish what happened, what the campaign caused, and what management should do next. Ethical governance, privacy protection, transparent assumptions, and triangulation are essential. The most mature measurement systems combine dashboards, attribution, econometrics, brand research, and experimentation into a coherent learning process.

Key Terms

Attribution: The allocation of observed conversions or value among marketing touchpoints.

Incrementality: The additional outcome caused by advertising compared with what would have occurred without it.

Counterfactual: The estimated outcome that would have occurred under an alternative condition, usually without treatment.

KPI: A defined indicator used to evaluate progress toward an objective and support decisions.

Adstock: A representation of the delayed and persistent effect of advertising over time.

Saturation: The diminishing marginal effect of increasing media investment or exposure.

Marketing-mix modelling: Aggregate statistical modelling that estimates the contribution of marketing and other factors to business outcomes.

Econometrics: The application of statistical methods and economic reasoning to analyse relationships in observed data.

Brand lift: A measured change in brand awareness, attitudes, or intentions associated with campaign exposure.

Geo-test: An experiment that varies media treatment across geographic areas to estimate incremental effects.

Holdout group: A group deliberately withheld from a treatment to provide a comparison condition.

Difference-in-differences: A method comparing changes over time between treatment and control groups.

Selection bias: Distortion caused when exposed and unexposed groups differ systematically before treatment.

Confounding: A situation in which another factor influences both exposure and the measured outcome.

View-through conversion: A conversion attributed to an impression that was not clicked before conversion.

Reach: The number or proportion of a defined audience exposed at least once.

Frequency: The number of exposures received by an audience member or population on average.

Customer lifetime value: The expected economic value generated by a customer over the relationship period.

Data governance: Policies and controls governing data definitions, access, quality, security, and use.

Triangulation: The use of multiple methods or data sources to strengthen interpretation.

Review Questions

Short Answer Questions

1. Distinguish between a delivery KPI, a brand KPI, and a commercial KPI.

2. Why does attribution not necessarily establish causality?

3. What is the purpose of an attribution window?

4. Define incrementality in the context of media planning.

5. What does adstock represent in marketing-mix modelling?

Long Answer Questions

1. Explain how a measurement hierarchy can connect media inputs with long-term business outcomes.

2. Compare single-touch, rule-based multi-touch, algorithmic, and platform attribution approaches.

3. Discuss the uses and limitations of marketing-mix modelling in strategic budget allocation.

4. Explain how brand-lift research should be designed and interpreted.

5. Describe the stages of a rigorous post-campaign evaluation.

Analytical / Application Questions

1. A retargeting campaign reports a very low cost per acquisition, but a holdout test shows little difference in total sales. Analyse the likely reasons and recommend how the advertiser should use the evidence.

2. Design a geo-test for a regional product launch, specifying the treatment, control, outcome metric, pre-period checks, and major threats to validity.

3. A campaign achieves high reach and video completion but no improvement in consideration or incremental sales. Develop a diagnostic framework covering media, creative, measurement, and market explanations.

22. Multiple Choice Questions with Answers

Question 1

Which measurement most directly addresses whether advertising caused additional conversions?

A. Last-click attribution

B. Incrementality testing

C. Impression delivery

D. Average frequency

Correct Answer: B. Incrementality testing

Question 2

What is the principal weakness of last-touch attribution?

A. It cannot report conversions

B. It ignores the final customer interaction

C. It assigns all credit to one touchpoint and ignores earlier influences

D. It can be used only for offline media

Correct Answer: C. It assigns all credit to one touchpoint and ignores earlier influences

Question 3

In marketing-mix modelling, adstock is used to represent which idea?

A. The legal ownership of advertising data

B. The persistence of advertising effects over time

C. The number of duplicate conversions

D. The proportion of invalid impressions

Correct Answer: B. The persistence of advertising effects over time

Question 4

A geo-test is most credible when treatment and control regions demonstrate similar pre-campaign trends. This condition supports which assumption?

A. The parallel-trends assumption

B. The saturation assumption

C. The identity-resolution assumption

D. The viewability assumption

Correct Answer: A. The parallel-trends assumption

Question 5

Which of the following is an intermediate brand outcome rather than a delivery output?

A. Impressions

B. Reach

C. Message association

D. Cost per thousand impressions

Correct Answer: C. Message association

Question 6

A platform reports a conversion after an impression but no click. What type of attribution is this?

A. First-touch attribution

B. View-through attribution

C. Linear attribution

D. Position-based attribution

Correct Answer: B. View-through attribution

Question 7

Why might a campaign with a low attributed cost per acquisition still be economically weak?

A. It may acquire customers who would have converted without advertising or who have low lifetime value

B. Low acquisition cost always indicates excessive reach

C. Attributed conversions cannot be recorded digitally

D. Customer quality is unrelated to profitability

Correct Answer: A. It may acquire customers who would have converted without advertising or who have low lifetime value

Question 8

Which dashboard practice most improves interpretive reliability?

A. Showing only the largest positive numbers

B. Combining all platform conversions without deduplication

C. Labelling observed, modelled, and experimentally estimated values separately

D. Removing sample sizes to simplify presentation

Correct Answer: C. Labelling observed, modelled, and experimentally estimated values separately

Question 9

What is the main purpose of a holdout group in an incrementality experiment?

A. To receive the highest possible frequency

B. To provide an estimate of the outcome without treatment

C. To increase the campaign's attributed conversions

D. To replace all brand research

Correct Answer: B. To provide an estimate of the outcome without treatment

Question 10

Which statement best reflects ethical measurement practice?

A. Use all available individual data indefinitely

B. Allow each platform to define conversions independently

C. Report causal claims even when the design supports only correlation

D. Apply data minimisation, consent controls, security, and transparent limitations

Correct Answer: D. Apply data minimisation, consent controls, security, and transparent limitations