Responsible Innovation and Future Directions

Introduction

Design thinking is often presented as a human-centred approach to opportunity discovery and solution development. Yet an innovation can be desirable to some users while excluding others, damaging the environment, manipulating behaviour, or producing harms that were not visible during early prototyping. Responsible innovation extends design thinking by asking not only whether a solution is useful and viable, but also whether it is fair, accessible, sustainable, explainable, governable and compatible with human rights.

For entrepreneurs, responsibility is not an optional public-relations activity. It influences market access, investor confidence, regulatory exposure, customer trust, employee legitimacy and long-term resilience. A product that depends on hidden consent, excessive data collection or planned obsolescence may achieve short-term growth but create legal, ethical and reputational liabilities. Conversely, inclusive and sustainable design can reveal underserved markets, reduce resource dependence and create durable competitive advantage.

This chapter examines responsible innovation across the design process. It addresses accessibility, inclusive design, sustainability, circularity, consent, dark patterns, algorithmic harm and innovation governance. It then considers emerging practices including speculative design, synthetic users, immersive prototyping and AI-assisted product discovery.

Learning Objectives

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

1. Distinguish accessibility, inclusive design, universal design and equity-oriented design.
2. Apply sustainability and circularity principles to an entrepreneurial opportunity.
3. Identify failures of consent, dark patterns and algorithmic systems.
4. Construct a proportionate governance approach for an innovation project.
5. Evaluate speculative design, synthetic users, immersive prototyping and AI-assisted discovery.
6. Make responsible design decisions under uncertainty and conflicting stakeholder interests.

1. From Human-Centred to Responsible Innovation

Human-centred design begins with the experiences, needs and aspirations of people. Responsible innovation broadens the field of concern. It includes direct users, non-users, workers, communities, future generations, ecosystems and institutions affected indirectly by a product or service.

A useful distinction is:

Responsible innovation is commonly associated with four practical commitments: anticipation, reflection, inclusion and responsiveness. Anticipation explores possible consequences before they become irreversible. Reflection examines assumptions, values and power. Inclusion brings affected parties into relevant decisions. Responsiveness enables the organisation to change direction when evidence or circumstances change.

These commitments do not guarantee a harmless innovation. They create a disciplined process for recognising uncertainty and acting accountably. In entrepreneurial settings, this is especially important because early ventures often make irreversible choices about data architecture, supply chains, pricing, defaults and target markets before they have formal governance capacity.

2. Accessibility, Inclusion and Equity

2.1 Key concepts

Accessibility refers to the extent to which a product, service, environment or communication can be used by people with disabilities. Disability may be physical, sensory, cognitive, speech-related or psychosocial. Accessibility is not synonymous with a special feature added after launch; it is a quality of the whole experience.

Inclusive design aims to accommodate a wide range of people, abilities, languages, identities, cultures, ages, economic circumstances and contexts. Universal design seeks products usable by as many people as possible without adaptation, while recognising that no single design can meet every need. Reasonable accommodation remains necessary where individual requirements cannot be addressed through general design.

Equity differs from equality. Equality gives everyone the same provision; equity recognises that different people may need different resources or modes of participation to achieve comparable opportunity. For example, a voice interface may improve access for a person with limited motor control, but it may be unsuitable for a person with a speech impairment or for someone in a noisy workplace.

2.2 Designing for variation

A responsible design process should:

• identify exclusion risks during research rather than at final testing;
• recruit participants with varied abilities and social positions;
• provide accessible research materials and compensation;
• test with assistive technologies such as screen readers, captions and keyboard navigation;
• support multiple modes of input, output and authentication;
• avoid relying solely on colour, sound, fine motor control or memory;
• provide plain language, translation and adjustable presentation;
• document known limitations and offer human support or alternative channels.

Accessibility standards can provide useful technical guidance. For digital products, the Web Content Accessibility Guidelines organise accessibility around perceivable, operable, understandable and robust principles. Compliance, however, is not identical to inclusion. A technically conforming website may still be culturally inappropriate, unaffordable or unusable in low-bandwidth settings.

2.3 Inclusive research and power

Participation is not automatically inclusive. A workshop may invite community members but give decision-making power only to the sponsoring organisation. Responsible participation makes the purpose, influence and limits of involvement clear. Participants should know whether their contributions can change the product, who owns resulting materials and how risks will be handled.

Entrepreneurs should distinguish representation from consultation. Representation involves meaningful influence over decisions; consultation may only collect opinions. A community advisory group, paid co-design partnership or participatory evaluation can provide stronger accountability than a one-time focus group.

3. Sustainability and Circularity

Sustainability concerns the capacity to meet present needs without undermining the ability of future generations to meet theirs. In innovation, it requires attention to environmental, social and economic dimensions across the product or service life cycle. A life-cycle perspective considers raw materials, manufacturing, logistics, use, maintenance and end-of-life treatment.

Circularity challenges the conventional linear model of “take, make, use and dispose”. Circular strategies include reducing material use, designing for durability and repair, reusing components, refurbishing products, remanufacturing, sharing underused assets and recovering materials. Recycling is useful but usually lower in the hierarchy than prevention, reuse and repair because it often involves energy loss and material degradation.

A circular business model may involve product-as-a-service, leasing, take-back schemes, modular upgrades, resale platforms or industrial symbiosis. However, circularity must not be assumed automatically. A subscription can increase transport and packaging; a sharing platform can intensify use and replacement; and a “smart” product may create electronic waste through software incompatibility.

3.1 Life-cycle reasoning

A preliminary life-cycle assessment can compare impact hotspots even before a formal assessment is possible. The design team should ask:

1. What materials and energy enter the system?
2. Which processes generate emissions, waste or hazardous exposure?
3. How long does the product remain useful?
4. Can it be repaired, upgraded, reused or safely disassembled?
5. Who bears environmental and social costs?
6. What evidence supports environmental claims?

Environmental claims should be specific, comparable and substantiated. Statements such as “eco-friendly” or “carbon neutral” without a defined boundary can mislead customers. Entrepreneurs should state whether a claim concerns manufacturing, use, transport, packaging or the complete life cycle, and should avoid presenting offsetting as equivalent to eliminating emissions.

3.2 Design tensions

Responsible innovation involves trade-offs. A heavier device may be more durable but increase transport emissions. Biodegradable packaging may require industrial composting unavailable to users. Data-intensive personalisation may improve service relevance while increasing energy consumption and surveillance. Such tensions should be made explicit rather than hidden behind a single sustainability score.

One practical method is a consequence matrix. For each design option, assess benefits and harms across users, workers, communities, ecosystems and future users. Record the evidence, uncertainty, reversibility and mitigation plan. The matrix is not a precise moral calculator; it is a device for making assumptions visible and supporting informed judgement.

4. Consent, Privacy and Data Responsibility

Consent is ethically meaningful only when it is informed, voluntary, specific, understandable and revocable. A pre-ticked box, bundled agreement or confusing interface may produce formal acceptance without genuine choice. Consent should not be treated as a one-time event when data practices, purposes or risks change.

A responsible consent flow should explain:

• what data are collected;
• why they are needed;
• how long they will be retained;
• who will receive them;
• whether automated decisions are involved;
• what happens if consent is refused;
• how consent can be withdrawn.

Data minimisation means collecting only what is reasonably necessary for a defined purpose. Purpose limitation prevents data gathered for one service from being silently repurposed. Privacy by design incorporates protective defaults, access controls, retention limits and security from the beginning rather than after a breach.

Consent is not always sufficient. A person may agree to data collection under economic pressure, lack a realistic alternative or be unable to understand complex consequences. For children and vulnerable populations, stronger safeguards and age-appropriate communication are required. Organisations should also assess impacts on people who never consented but may be inferred, observed or affected by others’ data.

5. Dark Patterns and Behavioural Manipulation

Dark patterns are interface choices that steer people toward decisions they might not otherwise make. Examples include difficult cancellation, misleading button labels, hidden fees, forced continuity, confirm-shaming, disguised advertising, privacy-invasive defaults and repeated interruptions. Persuasive design is not inherently unethical; the issue is whether influence is transparent, proportionate and consistent with the user’s interests.

A useful test asks:

1. Is the choice architecture symmetrical, or is one option deliberately easier?
2. Can users understand the consequence before acting?
3. Is refusal as easy as acceptance?
4. Does the design exploit urgency, fear, fatigue or vulnerability?
5. Would the organisation defend the interface publicly to an affected user?

Ethical design uses clear language, balanced options, visible costs, easy withdrawal and confirmation before consequential actions. Metrics such as conversion rate should be balanced with complaint rates, cancellation friction, accessibility outcomes and user comprehension. Optimising a narrow metric can reward manipulation even when the broader experience deteriorates.

6. Algorithmic Harm and Responsible AI

Algorithms can reproduce or intensify social inequality through biased data, unsuitable objectives, proxy variables, feedback loops and unequal error costs. Harm may arise even without malicious intent. A credit model may disadvantage a group because historical lending decisions encoded exclusion. A recruitment system may learn patterns associated with past hiring rather than actual job competence. A recommendation system may amplify sensational or discriminatory content because engagement is used as a proxy for value.

Responsible assessment should examine:

• the purpose and legitimacy of the system;
• data provenance, quality and representativeness;
• performance across relevant demographic and contextual groups;
• false-positive and false-negative consequences;
• explainability appropriate to the decision’s importance;
• human oversight and appeal mechanisms;
• security, monitoring and model drift;
• whether automation is necessary at all.

Fairness is not one universal statistical condition. Different fairness criteria can conflict, especially where groups have different base rates. Therefore, teams must specify the decision context, protected interests, acceptable error distribution and governance authority. A model can perform similarly across groups yet still support an unjust objective.

Synthetic data and generative AI require particular caution. Synthetic records can reduce exposure of real personal data, but they may preserve biases from the source data, create unrealistic correlations or give a false impression of privacy. Generative systems may fabricate evidence, reproduce copyrighted material, disclose confidential inputs or generate inaccessible and culturally inappropriate content. Human review, traceability and controlled use are essential.

7. Innovation Governance

Governance is the system through which an organisation makes, documents, reviews and corrects innovation decisions. It should be proportionate: a low-risk internal prototype does not require the same controls as a medical, financial, educational or public-sector system. Nevertheless, every project should identify responsibility rather than assuming that ethics belongs to an unnamed future department.

A practical governance architecture includes:

Governance should cover the complete innovation life cycle: discovery, framing, prototyping, pilot, launch, scaling and retirement. A product that was acceptable at pilot scale may create new harms when deployed across different populations or integrated with other systems. Sunset criteria are therefore as important as launch criteria.

Useful governance instruments include an algorithmic impact assessment, data protection impact assessment, accessibility conformance review, supplier code, model card, data sheet, red-team exercise and independent ethics review. These documents are valuable only when connected to decisions, budgets and authority. Documentation without power to change the project becomes compliance theatre.

8. Emerging Directions in Responsible Ideation

8.1 Speculative design

Speculative design explores possible, preferable or undesirable futures through scenarios, prototypes and artefacts. It can make invisible assumptions tangible and provoke public debate. A speculative service that shows a future city where access to housing is algorithmically rationed may help stakeholders discuss surveillance, inequality and governance before such a system exists.

Its strength is not prediction. Speculative design creates distance from immediate commercial constraints and asks what kind of future is being normalised. Its limitation is that provocative artefacts can confuse audiences, aestheticise suffering or remain accessible only to elite participants. Designers should identify the scenario’s assumptions, invite affected communities into interpretation and connect discussion to actionable decisions.

8.2 Synthetic users and research proxies

Synthetic users are AI-generated personas, simulated agents or artificial respondents used to explore hypotheses. They can help a team generate edge cases, test conversation flows or rehearse a research protocol when real participants are unavailable. They are not substitutes for human research. A synthetic user reflects patterns encoded in its data and prompt; it does not possess lived experience, consent, accountability or genuine preferences.

The correct role is auxiliary: use synthetic users to expand questions, not to validate market demand or represent marginalised communities. All claims based on them should be labelled as simulation. High-stakes decisions require engagement with real people and, where appropriate, compensation for their expertise.

8.3 Immersive prototyping

Virtual, augmented and mixed-reality prototypes allow teams to experience spatial layouts, training environments, product ergonomics and service journeys before physical construction. They are valuable where scale, movement, embodiment or social presence matters. They can reveal accessibility issues that a flat interface prototype conceals.

Immersive research also introduces motion sickness, sensory overload, exclusion of people without equipment, biometric data collection and emotional intensity. Participants should be briefed, able to pause or withdraw, and protected from unnecessary recording. Researchers should not infer real-world behaviour solely from behaviour in a simulated environment.

8.4 AI-assisted product discovery

AI systems can synthesise interview transcripts, identify themes, generate alternative concepts, search technical literature, simulate workflows and produce early interface variants. They can accelerate divergent thinking, but speed increases the risk of unexamined assumptions. Generated concepts may converge on familiar patterns, reproduce stereotypes, obscure provenance or violate confidentiality.

A responsible workflow separates assistance from authority:

1. Define the research question and evidence boundary.
2. Obtain permission before entering sensitive material into a tool.
3. Use AI to generate hypotheses, alternatives and coding suggestions.
4. Verify outputs against primary evidence and stakeholder input.
5. Record prompts, sources, model limitations and human decisions.
6. Test the resulting concept with real users and affected non-users.
7. Monitor performance and harm after deployment.

AI should not silently replace discovery. If it summarises interviews, the team should retain access to original evidence and inspect whose experiences were omitted. If it generates personas, those personas should be treated as provisional representations rather than facts.

9. An Integrated Responsible Innovation Process

The following process integrates design thinking with responsibility:

Frame. Define the social or market problem without prematurely defining the solution. State who defines the problem and who may contest it.

Map. Identify users, non-users, workers, suppliers, communities, ecosystems and future stakeholders. Map benefits, burdens, dependencies and power.

Anticipate. Develop intended-use, misuse, non-use and failure scenarios. Consider second-order effects, reversibility and uncertainty.

Co-design. Involve affected people through accessible, compensated and meaningful participation. Make influence and limits explicit.

Prototype. Test not only desirability and usability but also accessibility, privacy, sustainability, safety and governance assumptions.

Evaluate. Use evidence from technical tests, lived experience, life-cycle reasoning and impact assessment. Examine distribution of benefits and harms.

Decide. Record who approves the next step, what conditions apply and what evidence would cause a pause or termination.

Deploy and learn. Monitor outcomes, publish relevant limitations, provide remedy and revise the product when evidence changes.

This process changes the definition of success. Growth, adoption and revenue remain important, but they are evaluated alongside dignity, agency, inclusion, ecological integrity, safety and accountability.

10. Illustrative Case: TransitEase Mobility Platform

Background

TransitEase is a fictional early-stage venture developing a mobile platform that combines public transport, shared bicycles and demand-responsive vans. Its proposed value proposition is to reduce journey time for commuters in a large city.

Problem

Initial research focused on smartphone users travelling to central business districts. The prototype used location tracking, dynamic pricing and an automated route-ranking model. Review revealed several risks: wheelchair users could not distinguish accessible vehicles, cancellation was difficult, the lowest-cost option required extensive personal data, and the route model consistently deprioritised peripheral neighbourhoods because historical demand was lower.

Analysis

The team mapped stakeholders beyond paying riders: people with mobility and sensory disabilities, drivers, residents of underserved areas, public transport authorities, pedestrians and communities affected by traffic. An accessibility review identified missing keyboard support, inadequate colour contrast and no text alternative for map animations. A privacy review found that continuous location collection was not necessary for all features. A fairness review showed that optimising average travel time could worsen access for low-demand neighbourhoods.

The team also examined the business model. Dynamic pricing might allocate scarce vehicles efficiently, but could make essential travel unaffordable during emergencies. A subscription model could reduce per-trip friction but create forced continuity if cancellation remained difficult.

Approach and solution

TransitEase introduced several changes:

• an accessible booking route with screen-reader compatibility, captions and voice-independent interaction;
• an explicit accessibility filter showing vehicle constraints before payment;
• a guest mode using reduced location data;
• clear, equal-weight accept and decline choices for data sharing;
• one-step cancellation and no hidden renewal;
• a service obligation for coverage in underserved areas;
• fairness monitoring by neighbourhood, disability-related access needs and trip type;
• human appeal for route or fare disputes;
• a community advisory panel with paid participation;
• modular software and vehicle maintenance requirements to extend asset life.

The team did not claim that the algorithm was “bias-free”. Instead, it defined acceptable service thresholds, published limitations and created a process for reviewing complaints and model drift.

Outcome

In the pilot, the platform attracted fewer rapid sign-ups than the original design because it removed manipulative prompts and required clearer choices. However, customer comprehension improved, cancellation complaints fell and public-sector partners considered the system more credible. The venture also discovered a new market among employers seeking accessible commuting services. The case illustrates that responsible innovation can involve rejecting short-term optimisation in favour of legitimacy, resilience and broader value creation.

Learning points

1. The relevant stakeholder set expands as system effects become visible.
2. Accessibility and privacy can improve the core service rather than merely constrain it.
3. Fairness depends on objectives and distribution, not only model accuracy.
4. Governance should be designed before scale, not after a crisis.
5. Responsible choices can reveal new commercial opportunities.

11. Challenges and Limitations

Responsible innovation has practical limits. Entrepreneurs may lack money, expertise, time or bargaining power over suppliers and platforms. Stakeholders may disagree about what counts as harm. Accessibility improvements can conflict with visual simplicity; environmental goals can conflict with affordability; privacy can conflict with personalisation. No framework removes these tensions.

There is also a risk of ethical overload: lengthy assessments can delay action while failing to improve decisions. Proportionality is therefore essential. Focus first on irreversible, high-impact and poorly understood risks. Use staged experimentation, narrow data collection and explicit stop conditions. Seek independent expertise when the project concerns health, children, biometric identification, critical infrastructure or vulnerable groups.

Another limitation is measurement. Some outcomes, such as dignity, cultural safety and loss of autonomy, are difficult to reduce to a single indicator. Quantitative metrics should support, not replace, qualitative judgement and affected-community testimony.

Best-Practice Checklist

Before approving an innovation, ask:

• Who benefits, who may be excluded and who bears hidden costs?
• Are participation and consent understandable, voluntary and reversible?
• Can the product be used through varied abilities, devices, languages and contexts?
• What resources, emissions and waste occur across its life cycle?
• Are defaults, pricing and cancellation practices transparent?
• What data and algorithms are necessary, and what alternatives exist?
• How will unequal errors, misuse and downstream effects be detected?
• Who has authority to pause, change or withdraw the innovation?
• What remedy is available to people who are harmed?
• What evidence would show that the original problem framing was wrong?

Chapter Summary

Responsible innovation extends human-centred design to include non-users, communities, workers, ecosystems and future generations. Accessibility enables participation by people with diverse abilities; inclusive design addresses variation in identity, context and circumstance; equity recognises that equal treatment may not produce equal opportunity. Sustainability requires life-cycle thinking, while circularity seeks to reduce extraction and waste through durability, repair, reuse, refurbishment and recovery.

Meaningful consent must be informed, specific, voluntary and revocable. Dark patterns undermine agency by making unwanted choices easier or desired choices harder. Algorithmic systems can create harm through biased data, unsuitable objectives, feedback loops and unequal error consequences. Governance therefore requires clear purpose, stakeholder participation, risk assessment, decision rights, assurance, monitoring and remedy.

Speculative design helps communities question possible futures; synthetic users can generate hypotheses but cannot replace lived experience; immersive prototypes reveal spatial and embodied issues but create new accessibility and privacy risks; and AI-assisted discovery can accelerate ideation only when evidence, confidentiality, provenance and human judgement are protected. The responsible entrepreneur treats innovation as an ongoing, governable relationship with society rather than a one-time launch event.

Key Terms

Accessibility: The extent to which a product, service or environment can be used by people with diverse abilities.

Algorithmic harm: Physical, economic, social or psychological harm caused or amplified by an algorithmic system.

Circular economy: An economic approach that reduces extraction and waste by retaining products, components and materials in use.

Consent: Voluntary, informed and specific agreement to an activity, with a meaningful ability to withdraw.

Dark pattern: An interface design that manipulates or obstructs a person’s decision for the organisation’s benefit.

Equity: Fairness that recognises different needs, circumstances and structural disadvantages.

Inclusive design: Designing for a broad range of people, identities, abilities and contexts.

Life-cycle thinking: Considering impacts from material extraction through production, use, maintenance and end of life.

Model drift: Change in system performance or data relationships after deployment.

Participatory design: Design in which affected people contribute meaningfully to decisions and artefacts.

Privacy by design: Integrating privacy protections into system architecture and processes from the outset.

Reasonable accommodation: A necessary adjustment that enables participation where general design is insufficient.

Responsible innovation: Innovation that anticipates consequences, reflects on values, includes affected parties and responds to evidence.

Speculative design: The use of imagined artefacts or scenarios to explore possible and preferable futures.

Sustainability: Meeting present needs without undermining future generations’ ability to meet their needs.

Synthetic user: An AI-generated or simulated representation used to explore behaviour or test hypotheses.

Universal design: Designing products and environments to be usable by the widest practical range of people.

Value-sensitive design: Designing technology with explicit attention to human values and ethical consequences.

Review Questions

Short Answer Questions

1. Distinguish accessibility from inclusive design.
2. Why is recycling not always the highest circularity strategy?
3. Identify four conditions of meaningful consent.
4. What is a dark pattern? Give one example.
5. Why can a technically accurate algorithm still produce social harm?

Long Answer Questions

1. Explain how anticipation, reflection, inclusion and responsiveness contribute to responsible innovation.
2. Discuss how an entrepreneur can integrate life-cycle and circularity principles into a new product.
3. Evaluate the relationship between privacy by design, consent and data minimisation.
4. Describe a proportionate governance architecture for an AI-enabled service.
5. Compare the benefits and limitations of speculative design, synthetic users and immersive prototyping.

Analytical / Application Questions

1. A food-delivery venture proposes dynamic prices, continuous location tracking and an automated driver-allocation system. Analyse the accessibility, consent, fairness, sustainability and governance risks, and propose a staged pilot plan.
2. Design a responsible innovation assessment for a digital lending product used by financially vulnerable customers. Specify stakeholders, evidence, decision rights, monitoring metrics and remedy mechanisms.
3. An AI tool summarises interviews and generates personas for a healthcare start-up. Determine what uses are acceptable, what validation is required and what information should never be treated as evidence solely because the AI produced it.

Multiple Choice Questions with Answers

Question 1

Which principle requires an innovation team to examine possible consequences before they become difficult to reverse?

A. Anticipation

B. Conversion

C. Standardisation

D. Automation

Correct Answer: A. Anticipation

Question 2

Which option best describes equity in design?

A. Providing exactly the same feature to every person

B. Giving different people the support required for fair opportunity

C. Designing only for the largest customer segment

D. Removing all forms of personalisation

Correct Answer: B. Giving different people the support required for fair opportunity

Question 3

Which circular strategy generally retains more value than recycling?

A. Increasing disposal frequency

B. Replacing repairable components with sealed units

C. Designing for repair and reuse

D. Using additional packaging

Correct Answer: C. Designing for repair and reuse

Question 4

Which is an example of a dark pattern?

A. A clearly labelled privacy setting

B. A visible and one-step cancellation option

C. Captions provided for video content

D. A subscription that is easy to start but deliberately difficult to cancel

Correct Answer: D. A subscription that is easy to start but deliberately difficult to cancel

Question 5

Why is consent alone insufficient as a complete data-governance strategy?

A. Consent can never be documented

B. People may lack a realistic alternative or may not understand wider consequences

C. Consent always prevents data minimisation

D. Consent eliminates the need for security

Correct Answer: B. People may lack a realistic alternative or may not understand wider consequences

Question 6

What is a responsible use of synthetic users?

A. Replacing all interviews with marginalised communities

B. Treating generated preferences as proof of market demand

C. Generating hypotheses and edge cases for later human validation

D. Using simulated responses as legally binding consent

Correct Answer: C. Generating hypotheses and edge cases for later human validation

Question 7

Which governance feature provides a pathway for people to challenge or correct harmful decisions?

A. An appeal and remedy mechanism

B. A higher advertising budget

C. A more complex interface

D. A shorter retention policy alone

Correct Answer: A. An appeal and remedy mechanism

Question 8

What is a major limitation of AI-assisted interview summarisation?

A. It cannot produce any text

B. It may omit or distort experiences represented in the source material

C. It always requires physical prototyping

D. It automatically creates informed consent

Correct Answer: B. It may omit or distort experiences represented in the source material

Question 9

Which statement best characterises speculative design?

A. It predicts the future with certainty

B. It explores possible or preferable futures through scenarios and artefacts

C. It is limited to usability testing of existing products

D. It replaces governance with imagination

Correct Answer: B. It explores possible or preferable futures through scenarios and artefacts

Question 10

Why should responsible innovation include monitoring after launch?

A. Deployment can reveal new harms, unequal effects and model drift

B. Pre-launch testing always has no value

C. Monitoring is needed only for marketing campaigns

D. Post-launch evidence cannot change design decisions

Correct Answer: A. Deployment can reveal new harms, unequal effects and model drift