Advertising Technology, AI and Future Developments

Introduction

Advertising has moved from buying broad media audiences toward managing data-rich, automated and increasingly adaptive communication systems. A contemporary campaign may use software to select an impression, predict a user’s likely response, assemble an advertisement, deliver it across devices, measure an outcome and adjust future bids within milliseconds. These capabilities create important opportunities for brand growth, but they also introduce new responsibilities. Brand managers must understand not only what technologies can do, but also how they make decisions, what data they depend on, where errors arise and when human judgment must intervene.

This chapter examines programmatic advertising, marketing automation, predictive analytics, dynamic creative, generative artificial intelligence (AI), synthetic media, immersive media and connected commerce. It also evaluates algorithmic bias, transparency, privacy, copyright and governance. The central argument is that advertising technology is not a substitute for strategy. Technology scales decisions; it does not decide what a brand should mean, whom it should serve or what conduct is ethically acceptable.

Learning Objectives

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

1. Explain the architecture and operation of programmatic advertising.
2. Distinguish marketing automation, predictive analytics, dynamic creative and generative AI.
3. Evaluate advertising technology using performance, brand, privacy and fairness criteria.
4. Identify sources of algorithmic bias and recommend governance controls.
5. Explain copyright, disclosure and human-oversight issues associated with synthetic media.
6. Assess immersive media, connected commerce and emerging capabilities in relation to brand strategy.
7. Design a responsible technology-enabled advertising workflow.

1. From Media Buying to Advertising Systems

Traditional advertising often involved a planned media schedule: a brand selected a publication, television programme, radio station or outdoor location, purchased space and evaluated reach or frequency. Digital advertising introduced addressable impressions, behavioural signals, real-time measurement and rapid experimentation. The unit of decision became the individual impression or interaction rather than only the media vehicle.

This transformation can be understood as a movement across four levels:

The four levels may coexist. A campaign can have a strategic television plan, programmatic digital buying, automated bidding and dynamically assembled creative. The role of the brand manager is to connect these layers to a coherent customer proposition.

2. Programmatic Advertising

Programmatic advertising is the automated buying and selling of digital advertising inventory through software and data-enabled marketplaces. It is not synonymous with targeted advertising: programmatic refers primarily to the method of transaction and decision-making, whereas targeting refers to the selection of audiences or contexts.

2.1 Main participants and components

The principal components are:

• Advertiser or brand: provides the objective, budget, audience, creative assets and value rules.
• Demand-side platform (DSP): enables an advertiser or agency to bid for impressions across publishers and exchanges.
• Supply-side platform (SSP): helps publishers offer inventory and manage yield.
• Ad exchange: facilitates automated transactions between demand and supply.
• Publisher: provides the digital environment in which an advertisement appears.
• Data and identity services: provide contextual, demographic, behavioural or authenticated signals, subject to law and consent.
• Verification and measurement providers: assess viewability, invalid traffic, brand safety, reach and outcomes.

2.2 How a real-time auction works

When a user opens a page or application containing an advertising opportunity, the publisher or SSP sends an opportunity request. The request may include contextual information, format, approximate location, device characteristics and permitted audience signals. Potential buyers evaluate the opportunity and submit bids. An auction mechanism selects an eligible bid, applies contractual and policy rules, and returns the winning creative for display.

The entire process may occur in a fraction of a second. The speed does not make the process neutral. Decisions depend on the data supplied, the optimisation objective, the auction design, platform incentives and exclusion rules. A campaign optimised for low-cost clicks may reach people who click frequently but rarely purchase. A campaign optimised for conversions may under-invest in new audiences and reinforce existing customer patterns.

A simplified expected-value rule is:

Here, p(a) is the predicted probability of the chosen action a, V is the monetary value assigned to that action, and C is the cost of the impression. This is a decision rule, not a universal pricing law. It assumes that the probability estimate is calibrated, the value is appropriately defined, and the cost and value are measured on comparable monetary bases. In practice, platforms may also include quality, pacing, reach, privacy and brand-safety constraints.

For a conversion campaign, if the predicted conversion probability is 0.004, the assigned conversion value is £80 and the impression cost is £0.20, the expected value is £0.32 − £0.20 = £0.12. The impression appears economically attractive under this simplified rule. However, the calculation does not prove causation: the person may have converted without seeing the advertisement, and the prediction may be inaccurate.

2.3 Benefits and limitations

Programmatic systems can improve operational efficiency, enable controlled experiments, extend reach across many publishers and adjust spending in response to performance. They can also support frequency management, geographic adaptation and context-sensitive delivery.

Limitations include opaque platform logic, invalid traffic, poor-quality inventory, accidental placement beside harmful content, excessive frequency and dependence on platform-reported data. In addition, automated optimisation may favour measurable short-term actions over long-term brand effects. A technically efficient campaign can therefore be strategically weak.

3. Marketing Automation and Customer Journeys

Marketing automation uses software to trigger, coordinate and measure communications according to customer data, rules or inferred events. Examples include welcome sequences, abandoned-cart reminders, lead-nurturing emails, service notifications and post-purchase recommendations.

Automation is most effective when it supports a clearly designed customer journey. A typical journey contains:

1. Entry condition: a permitted event such as subscription, product viewing or purchase.
2. Audience qualification: determination of whether the person is eligible and has consented.
3. Decision rule: selection of a message, channel, timing or suppression condition.
4. Execution: delivery through email, application, SMS, web, advertising or a sales system.
5. Measurement: assessment of response, incremental value, complaints and longer-term effects.
6. Exit or escalation: removal from the sequence, movement to another stage or referral to a human employee.

Automation should not be confused with personalisation. A generic message can be automated, while a personalised message may be created manually. Nor does more personalisation always improve customer experience. A highly specific message based on sensitive or unexpected data may appear intrusive, even if technically relevant.

Effective automation requires data quality, clear contact policies, frequency limits, preference management and a process for correcting inaccurate profiles. It should also include suppression rules, such as stopping promotional messages after a purchase, during a complaint or when a customer has withdrawn consent.

4. Predictive Analytics

Predictive analytics uses historical and current data to estimate future outcomes. In advertising, common applications include propensity scoring, churn prediction, customer lifetime value estimation, media response modelling and fraud detection.

A prediction is not a fact and is not necessarily causal. If a model predicts that a customer is likely to buy, it does not demonstrate that advertising caused the purchase. The customer may already have strong purchase intent. This distinction is central to evaluating advertising effectiveness.

4.1 Common predictive applications

• Propensity modelling: estimates the likelihood of an action, such as purchase or response.
• Look-alike modelling: identifies people or contexts statistically similar to an existing audience.
• Churn prediction: estimates the risk that a customer will stop using a service.
• Lifetime value modelling: estimates expected future contribution from a customer.
• Forecasting: predicts demand, traffic or response over time.
• Anomaly detection: identifies unusual activity, such as click fraud or account abuse.

A basic response-rate calculation is:

The numerator and denominator must be defined consistently. If 240 people respond after 20,000 eligible exposures, the response rate is 240 ÷ 20,000 = 0.012, or 1.2%. This descriptive measure does not establish incremental impact. For that purpose, advertisers may compare a treatment group with a properly selected control group.

An incremental response estimate can be expressed as:

If the treatment rate is 1.2% and the control rate is 0.8%, estimated lift is 0.4 percentage points. The validity of this estimate depends on randomisation or a credible causal design, comparable groups, adequate sample size and consistent measurement.

4.2 Model risks

Models can reproduce historical inequalities. If previous advertising excluded certain postcodes, a model trained on past conversions may learn that exclusion and continue it. Proxy variables can create similar effects: postcode, language, device type or browsing pattern may indirectly encode protected characteristics. Optimisation can also create a feedback loop: audiences receiving more impressions generate more observed data, which causes the system to target them again.

Brand managers should ask:

• What population does the data represent?
• Which variables are used, and which act as proxies?
• Are error rates similar across relevant groups?
• Does optimisation reward short-term action at the expense of fairness?
• Can a person challenge or correct an automated decision?

5. Dynamic Creative Optimisation

Dynamic creative optimisation (DCO) assembles an advertisement from approved components such as headline, image, product, price, call to action and background. The system selects combinations according to audience, context, inventory, location, weather, device, stage in the journey or predicted response.

DCO differs from generative AI. DCO typically selects and combines pre-approved assets within defined rules. Generative AI creates new text, images, audio or video, although the two systems may be integrated.

A responsible DCO system requires:

• a clear brand asset library;
• design rules for hierarchy, contrast, logo use and accessibility;
• legal approval for prices, claims and disclosures;
• limits on sensitive personalisation;
• testing of combinations across devices and languages;
• a process for withdrawing defective assets;
• monitoring for inconsistent or misleading outputs.

Dynamic messages should be evaluated at three levels: the individual asset, the assembled advertisement and the campaign pattern. An individual advertisement may be accurate while the total campaign systematically excludes a particular audience or overexposes another.

6. Generative AI and Synthetic Media

Generative AI produces new content in response to prompts, examples or structured instructions. In advertising it may assist with copy variation, storyboards, image concepts, translation, voice production, video editing, customer-service responses and audience insight synthesis. Synthetic media is the broader category of artificially generated or substantially altered text, images, audio and video.

The value of generative AI lies in speed, variation and accessibility. It can help a small brand create multiple language versions or allow a creative team to explore concepts quickly. Its limitations include factual error, stereotypical representation, inconsistent brand voice, security risks, hidden training-data issues and difficulty determining authorship or rights.

A useful operating model is human-in-the-loop production:

1. A human defines the objective, audience, proposition and constraints.
2. The system generates or transforms candidate material.
3. A trained reviewer checks accuracy, cultural suitability, accessibility, brand fit and legal risk.
4. Rights and provenance are recorded.
5. The material is tested and monitored after publication.
6. Complaints, corrections and takedown procedures remain available.

Human oversight must be substantive rather than ceremonial. Clicking an approval button without examining outputs does not constitute meaningful governance.

6.1 Copyright and intellectual property

Copyright questions may arise from both inputs and outputs. An advertiser must consider whether training, uploading or transforming material is permitted; whether a generated output resembles a protected work; whether a voice or likeness has been used with permission; and who owns or can license the resulting material under applicable law and contract.

Legal rules differ across jurisdictions and continue to develop. Therefore, organisations should maintain records of asset sources, licences, model terms, prompts where relevant, approvals and alterations. They should not assume that an output is free of rights concerns merely because a machine generated it.

6.2 Disclosure and authenticity

Synthetic media can mislead audiences when it presents a person, event, endorsement or product experience as real. Disclosure should be prominent, understandable and appropriate to the medium. It is particularly important where a realistic synthetic voice, likeness, testimonial or news-like scene could affect consumer judgment.

Disclosure does not cure every problem. A deceptive claim remains deceptive even when labelled as AI-generated. Likewise, consent, privacy, dignity and defamation concerns can remain after disclosure.

7. Algorithmic Bias, Transparency and Accountability

Algorithmic bias is a systematic pattern of unequal or unjustified outcomes associated with data, design, implementation or use. It can arise through:

• Representation bias: some groups are missing or underrepresented in training data.
• Measurement bias: the selected outcome is an imperfect or unequal measure of value.
• Historical bias: past discrimination is reproduced in the data.
• Aggregation bias: one model is applied to groups with different patterns.
• Deployment bias: a model is used in a context for which it was not designed.
• Feedback bias: automated decisions influence the data used for later decisions.

Transparency has several layers. Procedural transparency explains who is responsible and what process is followed. Technical transparency describes inputs, model purpose, validation and limitations. User transparency informs people when data or automated systems materially affect their experience. Outcome transparency reports performance and disparate effects.

Complete disclosure of source code is not always necessary or sufficient. A technically complex explanation may not help an affected person understand what happened. Meaningful transparency combines accessible explanations, documentation, auditability and avenues for appeal.

A practical governance framework includes:

Fairness cannot be reduced to one universal metric. Different fairness criteria may conflict, especially when base rates differ between groups. Advertising teams therefore need legal, ethical and contextual judgment rather than relying on a single numerical threshold.

8. Privacy, Consent and Data Stewardship

Advertising technology often uses identifiers, inferred interests, location, device signals and purchase information. Privacy governance requires more than hiding names. A profile can be sensitive even when pseudonymised, and combining apparently harmless attributes can create an intrusive inference.

Good practice includes data minimisation, purpose limitation, retention controls, security, preference management, clear notices and mechanisms for access or correction where required. Sensitive categories require particular caution and may be restricted by law or platform policy. Children and vulnerable groups require enhanced safeguards.

Consent should be informed, specific and freely given where consent is the legal basis. A pre-ticked or confusing interface may fail to provide meaningful choice. Organisations should distinguish service communications from advertising and should respect withdrawal promptly.

9. Measuring Technology-Enabled Advertising

Technology can generate many metrics, but measurement quality depends on the decision being evaluated. Common measures include impressions, reach, frequency, click-through rate, conversion rate, cost per acquisition, return on advertising spend, viewability, brand lift and incremental sales.

Return on advertising spend is commonly expressed as:

If attributed revenue is £50,000 and advertising cost is £10,000, ROAS is 5. This result depends on the attribution method and does not necessarily represent profit. It may include sales that would have occurred without advertising, omit production or technology costs, and reward channels that capture demand rather than create it.

A more complete economic assessment considers contribution margin. If each sale produces £30 of contribution after variable costs, 2,000 incremental sales create £60,000 contribution. If total campaign cost is £25,000, incremental contribution after campaign cost is £35,000. The calculation is useful only if sales are genuinely incremental and costs are comprehensively included.

Attribution models allocate credit; experiments estimate causal effects. Where feasible, advertisers should use holdout tests, geo-experiments or other credible designs alongside platform attribution. Brand effects may require longer observation periods than direct-response metrics.

10. Immersive Media and Connected Commerce

Immersive media includes augmented reality (AR), virtual reality (VR), mixed reality and interactive three-dimensional environments. Applications include virtual product trials, spatial retail, interactive storytelling, training, events and branded worlds. The strategic value of immersion is not novelty alone. It should solve a customer problem, such as helping a consumer visualise furniture in a room or understand how a product operates.

Immersive experiences create additional concerns: biometric and spatial data, motion tracking, accessibility, motion sickness, harassment, identity and the permanence of user-generated content. Brand managers should provide non-immersive alternatives and avoid collecting more data than the experience requires.

Connected commerce links advertising, content, commerce, payment, fulfilment and service. A connected television advertisement may allow product discovery; a social platform may support checkout; a smart device may reorder a product. This convergence reduces distance between persuasion and transaction, making accuracy, pricing, consent and cancellation especially important.

The strongest connected-commerce experiences are useful and predictable. They explain what will happen after a click or voice command, display total costs, protect payment information and allow the customer to pause or reverse an action where appropriate.

11. Illustrative Case: Aurora Home Appliances

Background

Aurora Home Appliances, a fictional mid-sized retailer, sells energy-efficient kitchen products through stores and an online channel. It wants to increase sales of induction cookers while improving representation among younger renters and customers outside major cities.

Problem

The company uses a DSP, automated email journeys and DCO. Its reporting shows strong click-through rates but limited evidence of incremental sales. An audit finds that the optimisation system concentrates impressions among existing website visitors, excludes some low-data regions and uses a generative tool to create product images without a complete rights record. Several advertisements also imply that installation is free when this is true only in selected areas.

Analysis

The campaign has four distinct problems. First, attribution may be overstating the value of retargeting. Second, the data pattern may create geographic and socioeconomic bias. Third, the generated assets create copyright and accuracy risk. Fourth, dynamic claims are not sufficiently connected to regional fulfilment conditions.

Approach and solution

Aurora establishes a cross-functional review involving marketing, legal, data protection, merchandising, accessibility and customer service. It defines the primary objective as incremental contribution rather than clicks. A regional holdout test compares areas receiving the revised campaign with statistically comparable areas receiving the usual communication.

The company then:

• limits retargeting frequency and allocates budget to prospecting;
• removes unnecessary sensitive and proxy variables;
• tests reach and response across age, region and device groups;
• replaces unsupported generated product images with licensed photography and clearly documented synthetic assets;
• creates a rules-based feed for price, stock and installation claims;
• adds human approval for every new claim and high-risk creative combination;
• provides an accessible explanation of data use and preference controls;
• monitors complaints, unsubscribe rates, invalid traffic and incremental sales.

Outcome

In this illustrative case, the revised campaign produces fewer low-quality clicks but a more credible increase in incremental sales. The company also identifies that the best-performing creative is not the most visually complex version; it is a clear demonstration of installation, energy use and delivery timing. The business therefore treats technology as a means of reducing uncertainty and improving relevance, not as a reason to maximise automation.

Learning points

1. Platform metrics should not be accepted without examining causality.
2. Accuracy includes operational claims such as delivery and installation.
3. Inclusion requires testing outcomes, not merely adding diverse images.
4. Rights management must cover generated and transformed assets.
5. Human review is most important where consequences, sensitivity or reputational risk are high.

12. Future Industry Capabilities

Future advertising systems are likely to combine multimodal generative models, agentic software, privacy-preserving measurement, contextual intelligence, connected commerce and immersive interfaces. AI agents may plan media, negotiate inventory, generate variants and recommend budget changes. Retail media networks may connect transaction data with media inventory. Synthetic research environments may allow rapid concept testing, although simulated responses should not be treated as substitutes for real consumers.

Several capabilities deserve careful attention:

• Privacy-enhancing measurement: methods that reduce exposure of individual-level data while enabling aggregate analysis.
• On-device intelligence: processing signals locally rather than transferring raw data to a central system.
• Provenance systems: records that identify sources, alterations and approvals of media.
• Semantic contextual targeting: selecting content based on meaning rather than relying solely on personal profiles.
• Agentic optimisation: software that takes multi-step actions toward a defined goal.
• Retail-media integration: advertising linked to retailer audiences, inventory and transactions.
• Spatial commerce: product discovery and purchasing within three-dimensional environments.

The future capability that matters most is not maximum automation but accountable adaptability. Brands will need systems that can explain important decisions, preserve human agency, protect privacy, identify uncertainty and stop safely when conditions change.

13. Best-Practice Framework for Brand Managers

A brand manager assessing an advertising technology should follow a disciplined sequence:

1. Start with the brand and customer problem. Define the desired customer value before selecting a tool.
2. Specify the decision. Identify what the system will choose: audience, bid, message, timing or offer.
3. Assess data necessity. Use the least intrusive data that can achieve the purpose.
4. Set boundaries. Define prohibited segments, claims, contexts, frequency and uses.
5. Require evidence. Test performance against a control or credible benchmark.
6. Review distributional effects. Examine who receives, misses or bears the consequences of the system.
7. Document provenance and rights. Record sources, licences, transformations and approvals.
8. Maintain human escalation. Assign people who can pause, correct and explain the system.
9. Monitor after launch. Watch for drift, fraud, complaints, bias and unexpected creative combinations.
10. Review strategic fit. Confirm that optimisation supports long-term brand meaning as well as short-term response.

Challenges and Limitations

Advertising technology can create false precision. A dashboard with many decimal places may conceal uncertain attribution, incomplete data or unstable predictions. Automation can also produce scale without understanding: it may deliver millions of impressions while failing to communicate a distinctive proposition.

Other limitations include vendor dependency, fragmented measurement, changing privacy rules, model drift, cybersecurity threats and unequal access to advanced tools. Small organisations may lack specialist staff to audit platforms, while large organisations may have complex accountability gaps. Finally, cultural interpretation cannot be fully automated. A message that performs well statistically may still be inappropriate, stereotypical or damaging in a particular community.

Technology should therefore be evaluated as a sociotechnical system: people, data, software, institutions, incentives and audiences interact to produce outcomes. Responsible advertising requires competence in all of these dimensions.

Chapter Summary

Advertising technology has transformed media buying, customer journeys, creative production and measurement. Programmatic advertising automates the purchase of impressions, while marketing automation coordinates communications across customer journeys. Predictive analytics estimates likely outcomes but does not by itself establish causal impact. Dynamic creative assembles approved assets, whereas generative AI creates or transforms new content and therefore introduces additional accuracy, rights and authenticity concerns.

Algorithmic systems can reproduce historical inequalities through unrepresentative data, proxy variables, unsuitable objectives and feedback loops. Meaningful governance requires purpose definition, data stewardship, validation, documentation, transparency, human oversight and post-launch monitoring. Privacy, consent, copyright, synthetic-media disclosure and accessibility are strategic responsibilities rather than merely legal afterthoughts.

Immersive media and connected commerce can create useful, interactive experiences when they solve customer problems and protect agency. Future advertising capabilities will increasingly combine AI agents, privacy-enhancing measurement, provenance, contextual intelligence and commerce integration. The effective brand manager will therefore combine technological literacy with ethical judgment, causal measurement, creative leadership and accountability.

Key Terms

Programmatic advertising: Automated software-mediated buying and selling of digital advertising inventory.

Demand-side platform (DSP): Technology used by buyers to manage and bid for digital advertising opportunities.

Supply-side platform (SSP): Technology used by publishers to offer and manage advertising inventory.

Real-time bidding: Automated auctioning of an individual advertising impression.

Marketing automation: Software-supported triggering, coordination and measurement of customer communications.

Predictive analytics: Use of data and models to estimate future outcomes.

Propensity model: A model estimating the likelihood that an individual or group will take an action.

Incremental lift: The difference in outcomes between an exposed treatment group and a comparable control group.

Dynamic creative optimisation: Automated assembly or selection of approved creative components.

Generative AI: Artificial intelligence that produces new content from prompts, examples or structured inputs.

Synthetic media: Artificially generated or substantially altered text, image, audio or video.

Algorithmic bias: Systematic and unjustified differences in automated outcomes associated with data, design or use.

Transparency: Provision of understandable information about system purpose, operation, responsibility and effects.

Human-in-the-loop: A process in which people meaningfully review, approve or intervene in automated decisions.

Attribution: Allocation of credit for an outcome among advertising contacts or channels.

Causal measurement: Evaluation designed to estimate whether advertising caused an outcome.

Immersive media: Interactive augmented, virtual or mixed-reality communication environments.

Connected commerce: Integration of advertising, content, transaction, payment, fulfilment and service.

Data minimisation: Limiting data collection and use to what is necessary for a specified purpose.

Provenance: Recorded information about the source, history, alteration and approval of content.

Review Questions

Short Answer Questions

1. What is the difference between programmatic advertising and targeted advertising?
2. Why does a predictive conversion model not necessarily prove that advertising caused a purchase?
3. Distinguish dynamic creative optimisation from generative AI.
4. Identify two ways in which algorithmic bias can arise in advertising.
5. Why are copyright and provenance records important for synthetic media?

Long Answer Questions

1. Explain the main participants in a programmatic advertising ecosystem and evaluate its principal advantages and limitations.
2. Discuss how marketing automation can improve customer journeys while also creating privacy and relevance risks.
3. Explain the role of human oversight in generative-AI advertising production.
4. Compare platform attribution with causal measurement and describe when each may be useful.
5. Evaluate the strategic opportunities and governance challenges of immersive media and connected commerce.

Analytical / Application Questions

1. A campaign reports £80,000 in attributed revenue from £20,000 of advertising expenditure. Calculate ROAS and explain why this figure may not represent incremental profitability.
2. Design a governance process for a DCO campaign that uses location, weather and predicted purchase intent. Specify data, fairness, creative, approval and monitoring controls.
3. Using the Aurora Home Appliances case, recommend a measurement plan that would distinguish retargeting effects from purchases that would have occurred without advertising.

Multiple Choice Questions with Answers

Question 1

Which statement best describes programmatic advertising?

A. The manual purchase of television airtime through an agency

B. The automated buying and selling of digital advertising inventory through software

C. The creation of all advertising content by generative AI

D. The use of only demographic surveys to select media

Correct Answer: B. The automated buying and selling of digital advertising inventory through software

Question 2

A campaign receives 240 responses from 20,000 eligible exposures. What is the response rate?

A. 0.12%

B. 1.2%

C. 12%

D. 120%

Correct Answer: B. 1.2%

Question 3

What is the principal difference between DCO and generative AI?

A. DCO is used only offline, while generative AI is used only online

B. DCO generally assembles approved components, while generative AI creates or transforms content

C. DCO measures sales, while generative AI measures reach

D. DCO requires no data, while generative AI always requires personal data

Correct Answer: B. DCO generally assembles approved components, while generative AI creates or transforms content

Question 4

Why is a high attributed ROAS not necessarily evidence of advertising effectiveness?

A. ROAS cannot be calculated using revenue

B. Attribution may include purchases that would have occurred without advertising

C. ROAS is always lower than profit

D. Advertising expenditure is never included in ROAS

Correct Answer: B. Attribution may include purchases that would have occurred without advertising

Question 5

Which example illustrates historical bias?

A. A model is tested on data from a new market

B. A model reproduces past exclusion because it was trained on historically discriminatory outcomes

C. A customer changes a communication preference

D. A publisher labels an advertisement as sponsored

Correct Answer: B. A model reproduces past exclusion because it was trained on historically discriminatory outcomes

Question 6

Which practice is most consistent with meaningful human oversight?

A. Automatically approving every generated asset

B. Allowing the system to set its own objectives without review

C. Having trained reviewers check accuracy, rights, brand fit and risk before publication

D. Removing all records of prompts and approvals

Correct Answer: C. Having trained reviewers check accuracy, rights, brand fit and risk before publication

Question 7

What does incremental lift compare?

A. Treatment response with control response

B. Revenue with logo size

C. Reach with the number of employees

D. Creative length with media cost

Correct Answer: A. Treatment response with control response

Question 8

Which is an appropriate privacy practice?

A. Collecting every available signal in case it becomes useful

B. Retaining personal data indefinitely

C. Using data minimisation and purpose limitation

D. Treating pseudonymous data as risk-free

Correct Answer: C. Using data minimisation and purpose limitation

Question 9

In connected commerce, which issue becomes especially important because persuasion is close to transaction?

A. Accurate pricing, fulfilment information and payment protection

B. Eliminating customer service

C. Avoiding all measurement

D. Using only printed media

Correct Answer: A. Accurate pricing, fulfilment information and payment protection

Question 10

What is a key limitation of predictive analytics in advertising?

A. It can estimate future outcomes but does not automatically establish causation

B. It cannot use historical data

C. It always produces identical predictions for all customers

D. It eliminates the need for strategic objectives

Correct Answer: A. It can estimate future outcomes but does not automatically establish causation