Human-Centered Design in the Behavioral Economy

How Digital Business Models Are Changing the Boundary Between User Welfare and Behavioral Optimization

8/16/202625 min read

Independent Research Report | August 2026

Executive Summary

Human-centered design has become one of the defining operating philosophies of the digital economy. Its foundational proposition—that products, services, and systems should be designed around the needs, capabilities, limitations, and contexts of real people rather than around technical or institutional convenience—has materially improved usability, accessibility, service delivery, and product quality across industries. The expansion of digital instrumentation, however, has changed the economic environment in which human-centered design operates. Organizations can now observe behavior continuously, test alternative interfaces at population scale, identify the moments at which users hesitate or abandon a process, personalize subsequent interactions, and connect small behavioral changes directly to conversion, engagement, retention, data acquisition, and revenue. User understanding has consequently evolved from an input into product design into a form of behavioral intelligence with measurable commercial value.

This transformation does not establish that human-centered design has become inherently exploitative. The evidence supports a more precise conclusion. The same capabilities that allow organizations to identify and remove unnecessary human friction can also be used to identify and overcome friction that represents deliberation, reluctance, resistance, or a decision to disengage. In digital environments, therefore, the distinction between designing for human behavior and optimizing human behavior for the system has become economically significant. The method may remain observational research, usability testing, experimentation, personalization, or journey redesign; the decisive variable is increasingly the objective function into which that knowledge is incorporated.

Evidence from consumer-protection research indicates that this tension is no longer theoretical. The OECD describes “dark commercial patterns” as practices embedded in digital choice architecture that can steer, deceive, coerce, or manipulate consumers into decisions that are often contrary to their interests.[1] In 2024, an international review coordinated through the U.S. Federal Trade Commission, the International Consumer Protection and Enforcement Network, and the Global Privacy Enforcement Network examined 642 subscription websites and mobile applications across 26 countries. Nearly 76 percent of the reviewed services displayed at least one potential dark pattern and nearly 67 percent displayed multiple potential dark patterns. The authorities explicitly did not determine that every identified practice violated applicable law, so these findings should not be interpreted as a universal estimate of unlawful behavior. They nevertheless demonstrate that potentially manipulative choice architecture is sufficiently widespread in the examined digital markets to constitute a material consumer-governance issue.[2] (Federal Trade Commission)

The economic mechanism underlying these practices is broader than any one category of dark pattern. Digital firms increasingly possess information about how users respond to defaults, salience, sequence, pricing presentation, notifications, recommendations, social proof, time pressure, cancellation friction, and alternative interface structures. Individual users generally encounter only the final choice architecture and cannot observe the experiments, counterfactual interfaces, ranking rules, or population-level behavioral data informing it. The organization therefore enters the interaction with an informational advantage not simply about the product, but increasingly about how the user's own behavior is likely to change under alternative presentations of that product. The resulting asymmetry can be used constructively—for example, to reduce complexity or improve accessibility—but it also increases the institution's capacity to influence choice.

Consumer finance illustrates how rapidly this architecture can scale. The U.S. Consumer Financial Protection Bureau reported that five large Buy Now, Pay Later providers originated approximately 180 million loans worth more than US$24 billion in 2021, nearly ten times the lending volume reported for 2019. The same research recognized genuine consumer advantages, including zero-interest structures, accessibility, and relatively simple repayment arrangements, but also identified risks involving autopay, data harvesting, incremental-sales incentives, and borrower overextension.[3] More recent CFPB research found that 21.2 percent of consumers with a credit record used BNPL in 2022, that approximately 63 percent of BNPL borrowers originated multiple simultaneous loans at some point during the year, and that one-third borrowed from multiple BNPL providers. These figures do not establish that BNPL causes financial distress—the Bureau itself notes important pre-existing differences between users and non-users—but they illustrate why apparently frictionless product design cannot be assessed solely through adoption or conversion.[4] (Consumer Financial Protection Bureau)

The same distinction is important in health, wellness, and workplace design. WHO estimates that 15 percent of working-age adults had a mental disorder in 2019, while depression and anxiety are associated with approximately 12 billion lost working days and US$1 trillion in lost productivity each year. WHO simultaneously identifies excessive workloads, low job control, job insecurity, discrimination, and inequality as workplace mental-health risks.[5] The implication is that a human-centered intervention focused entirely on helping individuals manage stress can generate real benefits while leaving the system producing that stress largely unchanged. Human-centeredness therefore cannot be inferred simply from whether an intervention demonstrates empathy or improves an individual's ability to adapt. The organization must also examine whether it is redesigning the upstream conditions causing the problem. (World Health Organization)

Artificial intelligence is likely to intensify the issue because it lowers the marginal cost of individualized behavioral adaptation. Conventional digital optimization typically tests variants across population segments. AI systems can increasingly alter content, sequencing, recommendations, language, interaction style, and potentially persuasive strategy dynamically. This increases the potential value of personalization, but it also increases the consequences of misaligned objectives. A system that becomes dramatically better at optimizing conversion has not necessarily become dramatically better at optimizing user welfare. NIST's AI Risk Management Framework explicitly treats AI risk as a socio-technical governance problem requiring organizations to govern, map, measure, and manage risk rather than assuming technical performance is sufficient evidence of trustworthy operation.[6] (NIST)

The central finding of this report is therefore that human-centered design is ethically underdetermined. Deep knowledge of a user does not specify whose objective should govern the interaction. Empathy produces information. Experimentation produces behavioral evidence. Optimization produces increasingly effective interventions. Incentives determine what those interventions are selected to accomplish, while governance determines which interventions remain permissible. Human-centered methods can therefore generate substantial shared value when organizational and user interests remain aligned; where those interests diverge, the same methods can increase the institution's power over the user.

The strategic implication is that the next generation of human-centered design requires a broader definition of success. Task completion, usability, conversion, retention, engagement, and satisfaction remain important, but they are not sufficient. High-consequence digital systems increasingly need to evaluate comprehension, effective agency, reversibility, entry–exit symmetry, downstream welfare, behavioral burden, information asymmetry, and the legitimacy of the objective being optimized. A genuinely human-centered system should not simply understand human behavior more precisely. It should demonstrate that this understanding is governed in ways that prevent human limitations from becoming an unrestricted commercial optimization surface.

1. Human-Centered Design Created a New Operating Logic for Product Development

Human-centered design developed across several intellectual traditions, including ergonomics, human factors, participatory design, industrial design, cognitive psychology, usability engineering, and human-computer interaction. Although these traditions differed in method and terminology, they challenged a common assumption: that a technically functional system should be considered successful even if the human being using it struggled to understand, operate, or adapt to it. The emerging alternative was to study people in context, identify needs and constraints, prototype potential solutions, test them with users, and iteratively redesign the system until the interaction became more consistent with real human capability. This represented a substantive shift in organizational decision-making because the user ceased to be treated primarily as the endpoint of a production process and became an input into the process by which products and services were constructed.

The contribution remains economically significant. Human-centered methods can reveal that a service has been organized according to the internal structure of the institution rather than the mental model of the customer; that an application requires information users do not possess at the moment it is requested; that an accessibility barrier excludes a meaningful population; or that a complex process generates errors because the organization assumes levels of attention, memory, or expertise that real users do not have. Correcting these mismatches can simultaneously improve user experience and organizational performance. Better design can reduce errors, service contacts, abandonment, operational cost, and unnecessary complexity. There is therefore no empirical or conceptual basis for treating human-centered design itself as inherently extractive.

The strategic problem emerges one level above the methodology. Human-centered research tells an organization more about the user, but it does not determine what the organization should do with that knowledge. A research team may discover that users hesitate at a particular point in a purchasing process because the long-term financial commitment becomes salient. One intervention could clarify that commitment further and support better deliberation. Another could reduce the visibility of the information producing hesitation and thereby increase conversion. Both interventions might be informed by the same high-quality research. The distinction lies in the institutional objective governing the use of the insight, not the sophistication of the research that generated it.

This distinction has become increasingly important as design has moved from isolated product creation toward continuous product optimization. The modern designer often operates inside organizations where interfaces are instrumented, product performance is measured in real time, and teams are accountable for commercially meaningful metrics. Human-centered design has therefore become intertwined with analytics, growth, experimentation, personalization, behavioral science, product management, and increasingly machine learning. The result is not the disappearance of HCD but the expansion of its economic context.

2. Digital Instrumentation Converted Human Behavior Into a Continuously Measurable Asset

The most consequential transformation in contemporary design is not visual or methodological. It is the expansion of observability. Digital organizations can increasingly observe behavior at a resolution and scale unavailable to earlier generations of designers. They can identify where people pause, what they ignore, which option they select, which interface causes abandonment, which message causes them to return, how often they attempt to cancel, how they respond to price changes, and which sequence of interactions produces the highest probability of completing a transaction. Millions of interactions can then be aggregated into patterns and tested experimentally.

This infrastructure materially changes the economic value of user research. Historically, information about human needs primarily informed product development. In modern digital platforms, behavioral information can also become part of an ongoing feedback system. Design changes behavior; resulting behavior becomes data; data informs another design change; and the cycle repeats. The system therefore moves from design informed by research toward behavioral optimization informed by continuous measurement.

The distinction is economically important because small behavioral effects can become material at scale. A minor increase in conversion, retention, disclosure, advertising interaction, or purchasing frequency may be financially significant when applied across millions of users. The commercial incentive to understand the behavioral mechanism behind those effects consequently increases. User behavior stops functioning solely as evidence of whether a product is useful and becomes an input into optimizing the economics of the product itself.

This does not create an automatic conflict with human welfare. Where the organizational metric is closely aligned with the user's objective, continuous optimization can produce substantial shared value. A navigation product that reduces the time required to reach the correct destination, for example, can improve both user experience and product competitiveness. The governance problem becomes more difficult where the proxy being optimized can improve while the user's underlying outcome deteriorates. Retention can increase because the product is better, or because cancellation is harder. Engagement can increase because content is more relevant, or because the platform has become more difficult to disengage from. Conversion can increase because information is clearer, or because inconvenient information has become less salient.

The same behavioral metric can therefore represent materially different human outcomes.

3. The Commercialization of Behavioral Insight Changed the Role of Friction

Few concepts illustrate the tension more clearly than “friction.” In conventional product design, friction generally describes unnecessary difficulty: repeated data entry, confusing navigation, inaccessible controls, unclear language, or procedural complexity that prevents a person from accomplishing a legitimate goal. Removing this type of friction is a central achievement of good design.

But not all friction is unnecessary.

Some friction creates time for deliberation. Financial disclosures, confirmation steps before irreversible actions, identity verification, cooling-off periods, warnings, and explicit consent can all create additional effort. Their value lies precisely in preventing an action from becoming completely effortless.

The strategic distinction is between wasteful friction and protective friction.

When organizations adopt “reduce friction” as a universal objective, that distinction can disappear. A consumer's hesitation becomes a conversion obstacle. A cancellation step becomes churn. A privacy decision becomes consent-rate optimization. A warning becomes abandonment risk. The system may then remove the very moment at which the user is reassessing whether the transaction remains in their interest.

The reverse mechanism can also occur. Friction can be deliberately introduced where the organization wants to discourage behavior. Acquisition may be reduced to a small number of taps while cancellation requires navigating settings, rejecting offers, confirming intent repeatedly, or contacting customer service. The product remains formally cancellable, yet the probability and cost of exit have changed.

This is why a mature human-centered design standard cannot judge an interaction simply by how little effort it requires. It must ask whose desired behavior the friction is reducing or increasing and whether the intervention preserves legitimate user agency.

4. Dark Patterns Provide the Clearest Empirical Evidence of Misaligned Choice Architecture

The strongest evidence that interface optimization can conflict with user welfare comes from the growing literature and regulatory activity concerning dark patterns. The OECD defines dark commercial patterns as digital practices commonly embedded in user interfaces that can steer, deceive, coerce, or manipulate consumers into choices that are frequently contrary to their interests.[1] Its 2022 review synthesizes evidence concerning prevalence, effectiveness, and consumer harm and describes practices designed to extract additional money, personal data, or attention. (OECD)

The importance of the concept is that it shifts analysis away from the simplistic question of whether a user technically retained a choice. A consumer may have an available alternative while the architecture surrounding that alternative has been deliberately structured to make it harder to identify, understand, or execute. Autonomy therefore depends on the effective choice environment, not merely the formal existence of more than one button.

Regulatory evidence indicates that potentially manipulative practices are common in at least some major digital-market contexts. In early 2024, consumer and privacy authorities from 26 countries reviewed 642 websites and applications offering subscription services. Nearly 76 percent of the services reviewed contained at least one possible dark pattern and nearly 67 percent contained more than one. The review examined mechanisms including sneaking, obstruction, interface interference, forced action, and social proof. Authorities did not determine through that review whether individual instances were unlawful, so its findings should be interpreted as evidence of potentially manipulative design practices rather than as an enforcement verdict against three-quarters of the services examined.[2] (Federal Trade Commission)

The strategic importance of this evidence extends beyond dark-pattern terminology. It demonstrates that organizations already possess sufficient design sophistication to influence consumers not merely by changing the underlying offer but by changing how the offer and its alternatives are cognitively encountered. The economic value of interface architecture is therefore not hypothetical. It is measurable enough to attract optimization investment and significant enough to attract regulatory intervention.

5. The Digital Economy Has Created a Structural Information Asymmetry Between Institutions and Individuals

Traditional consumer-protection models often focus on informational asymmetry concerning the product: the seller knows more about quality, price, risk, or contract terms than the buyer. Digital environments add a second asymmetry. Organizations can increasingly know more about how the consumer is likely to respond to alternative presentations of the decision itself.

A large platform may know that one sequence of screens increases acceptance relative to another; that a particular time interval increases the probability of re-engagement; that one notification style works better for users with certain behavioral histories; or that introducing a retention offer at a particular stage of cancellation reduces churn. An individual consumer typically has no equivalent visibility into the system. They see one interface and one sequence of choices.

This asymmetry changes the nature of market power. The institution controls not only the product but the behavioral environment in which the product is evaluated. It can continuously learn from the reactions of large populations while each user experiences only an individual interaction.

The asymmetry does not imply that personalization or experimentation is inherently harmful. Both can reduce unnecessary search costs and create significant value. But it does mean that conventional notions of informed choice become more difficult to apply. A user may understand the immediate transaction while remaining unaware that the environment has been optimized using behavioral information about thousands or millions of comparable users.

For human-centered design, this creates a governance requirement that traditional usability practice did not need to address at the same scale. The question is no longer simply whether the user can understand the interface. It is whether the architecture optimizing the interface remains aligned with the user's legitimate interests.

6. Subscription Models Expose the Tension Between Retention and Effective Exit

Subscription businesses create recurring value when customers continue receiving benefits over time. Retention is therefore a legitimate economic measure. But retention is not equivalent to loyalty. A person can remain subscribed because the service continues creating value, because they forgot they were subscribed, because cancellation is difficult, or because the system repeatedly interrupts attempts to leave.

This distinction makes exit architecture one of the most revealing tests of human-centeredness. At acquisition, customer and company incentives are generally aligned: both parties want an appropriate user to complete enrollment. At cancellation, those incentives can diverge. The company benefits economically from continued payment; the user's objective is to terminate the relationship.

A system designed around human agency should therefore apply design quality to both sides of the relationship. Joining and leaving need not be mechanically identical, but extreme asymmetry is difficult to reconcile with a strong interpretation of human-centeredness unless additional friction serves a legitimate protective purpose.

The wider implication is that conventional journey design can become incomplete when the organization defines the journey according to commercial objectives. Acquisition, onboarding, activation, engagement, and retention receive substantial design attention because they contribute to growth. Cancellation, withdrawal of consent, account deletion, and disengagement often receive less. Yet those “negative” journeys may provide the most revealing evidence about whether the organization treats customer agency as an operating constraint or primarily as a source of behavioral data.

7. Consumer Finance Demonstrates Why Convenience and Welfare Cannot Be Assumed to Be Equivalent

Buy Now, Pay Later provides a particularly useful case because the products combine strong convenience benefits with potentially material financial consequences. CFPB's 2022 market study found that the five providers examined originated approximately 180 million loans totaling more than US$24 billion during 2021, representing nearly tenfold growth from 2019. The Bureau explicitly recognized benefits including no-interest structures, relatively simple repayment plans, and easy access, while also identifying potential risks including mandatory autopay, data harvesting, incremental-sales incentives, and borrower overextension.[3] (Consumer Financial Protection Bureau)

The market continued to expand. CFPB research published in 2025 found that 21.2 percent of consumers with a credit record used BNPL in 2022, up from 17.6 percent in 2021. Approximately 63 percent of borrowers had multiple simultaneous BNPL loans at some point in the year, while 33 percent borrowed from multiple providers. Nearly two-thirds of BNPL loans in the Bureau's matched sample went to borrowers with subprime or deep-subprime credit scores.[4] (Consumer Financial Protection Bureau)

These statistics should not be interpreted as proving that BNPL creates financial distress. CFPB's own work notes that users and non-users differed before BNPL use and that many consumers use the products without visible indications of financial stress. Earlier analysis found that BNPL borrowers were, on average, more highly indebted and more likely to use several forms of credit, but many of those differences preceded BNPL adoption. (Consumer Financial Protection Bureau)

The design implication is nevertheless important. A product can reduce transaction friction while increasing the distance between the experience of consumption and the full salience of its future financial obligation. The correct human-centered metric is therefore not simply whether the user can complete the purchase easily. It includes whether total obligations remain understandable, whether cumulative borrowing is visible, whether users can dispute transactions effectively, and whether the design supports informed decisions rather than merely completed ones.

8. Attention and Engagement Metrics Are Useful Proxies but Incomplete Measures of Human Value

The rise of attention-based digital businesses created another structural tension. Engagement metrics are valuable because they reveal whether people use a product, find content relevant, return to a service, or interact with other users. They are also commercially significant because repeated engagement can generate advertising inventory, transactions, subscription retention, or data.

The measurement problem is that engagement is an observable behavior, while value is a broader construct.

A user may spend more time on an educational platform because the content is genuinely useful. Another may spend more time in a social system because conflict, uncertainty, or intermittent reinforcement repeatedly reactivates attention. The metric records increased use in both cases.

This does not justify describing all engagement optimization as addictive or harmful. Such a claim would exceed the evidence and conflate several distinct behavioral and clinical phenomena. It does, however, indicate that firms need stronger evidence before claiming that higher engagement demonstrates greater user welfare.

Human-centered measurement therefore needs to connect behavioral proxies with outcomes that matter beyond the immediate session. Those outcomes vary by product: learning, successful completion, financial stability, meaningful social connection, improved health, reduced administrative burden, or simply enjoyable entertainment. The central governance principle is that the commercial metric should not silently become the definition of the human outcome.

9. Wellness and Workplace Design Illustrate the Risk of Treating Structural Problems as Individual Adaptation Problems

The rapid expansion of workplace mental-health and wellbeing interventions reflects a genuine social and economic problem. WHO estimates that approximately 15 percent of working-age adults had a mental disorder in 2019 and that depression and anxiety contribute to approximately 12 billion lost working days each year, representing around US$1 trillion in lost productivity.[5] WHO also identifies excessive workloads, low control, job insecurity, discrimination, inequality, and other working conditions as material mental-health risks. (World Health Organization)

This evidence matters because it demonstrates that many wellbeing problems cannot be represented adequately as individual deficits. An employee can benefit from mindfulness training, digital therapy, coaching, or productivity support while still working inside an environment characterized by excessive workload or insufficient control. The individual intervention may therefore be useful without being sufficient.

The human-centered-design failure occurs when an organization treats the individual's difficulty adapting as the principal design problem while leaving the organizational architecture producing that difficulty substantially unchanged. The visible symptom becomes the target of intervention; the structural cause remains outside scope.

A stronger HCD approach would operate at both levels. It would provide appropriate support to the individual while examining workload, scheduling, managerial behavior, autonomy, staffing, role clarity, incentives, and recovery capacity. The central principle is that helping people tolerate a system is not necessarily equivalent to redesigning the system around people.

10. Personalization Increases Both User Value and Institutional Behavioral Power

Personalization is one of the clearest examples of a technology whose benefits and governance risks arise from the same mechanism. A system that understands preferences can reduce irrelevant information, improve search, adapt recommendations, simplify repeated tasks, and potentially increase accessibility. Greater contextual knowledge can make digital services substantially more useful.

The same contextual knowledge increases the institution's ability to predict and influence behavior.

The appropriate analytical distinction is therefore not personalized versus non-personalized design. It is accountable versus unaccountable personalization.

A user should not need access to the entire recommendation algorithm to participate meaningfully in a digital service. But high-impact personalization increasingly raises questions about which data were used, whether sensitive attributes were inferred, what alternatives were withheld, whether the personalization primarily benefits the individual or the provider, and whether a user can meaningfully change or contest the resulting treatment.

The asymmetry becomes particularly important when personalization operates on vulnerability. A system inferring that an individual is price-sensitive, lonely, financially stressed, highly engaged, or particularly susceptible to a particular message is qualitatively different from a system remembering a preferred language or clothing size. Both are personalization, but their governance consequences are not equivalent.

This suggests that the depth of behavioral insight should be matched by the depth of governance. More precise knowledge of a person's vulnerabilities should create more stringent limits on use, not automatically broader permission to optimize.

11. Artificial Intelligence Is Likely to Industrialize Behavioral Adaptation

Artificial intelligence can significantly lower the cost of generating and testing individualized interventions. Traditional digital optimization frequently relies on predesigned variants deployed to defined user segments. AI systems can increasingly generate content dynamically, alter explanation, vary conversational style, summarize personalized options, predict likely responses, and choose among alternative interaction strategies.

The potential benefit is substantial. Interfaces can become more accessible, relevant, adaptive, and responsive. Complex services can be explained differently to users with different levels of expertise. Information overload can be reduced. Customer support can become faster and more contextual.

The same capability also increases behavioral leverage.

A commercially optimized AI assistant could theoretically adapt persuasion to individual users at substantially lower cost than a human sales organization. A recommendation system could combine behavioral history with real-time conversational information. An autonomous interface could experiment continuously with language and sequence rather than waiting for periodic redesign.

The correct governance question is consequently not whether AI is persuasive. Almost all communication can influence behavior. The question is whether the objective being optimized and the constraints surrounding that objective remain legitimate.

NIST's AI RMF is relevant because it rejects the assumption that technical capability alone determines trustworthy use. The framework is designed to help organizations identify and manage risks throughout the lifecycle of AI systems and explicitly frames trustworthy AI as an organizational and socio-technical challenge.[6] (NIST)

For human-centered design, the implication is straightforward: greater behavioral optimization capability makes objective governance more important, not less. A system that becomes twice as effective at optimizing an incomplete proxy can potentially magnify the incompleteness of the proxy.

12. The Core Governance Problem Is the Misalignment Between Empathy and Incentive

Empathy occupies a central place in human-centered design because understanding the user's experience can reveal needs that conventional business analysis misses. But empathy itself does not specify a moral or economic objective.

Empathy is principally an information-generating capability.

It can reveal that people are confused, anxious, rushed, uncertain, overwhelmed, socially isolated, price-sensitive, or uncomfortable with a decision. That knowledge can support a protective intervention, such as clearer information or additional control. It can also support a commercially useful intervention intended to overcome the user's resistance.

The design method cannot resolve the conflict.

The conflict is institutional.

A useful way to represent the problem is therefore not “ethical designers versus unethical designers,” but empathy–incentive misalignment. Designers and researchers can correctly understand a user while working inside an organization whose metrics reward outcomes that conflict with the user's longer-term interests.

This explains why better user research alone cannot solve the problem. An organization can possess world-class research capability and still deploy design patterns that reduce effective agency. The missing variable is not insight but governance: decision rights, performance metrics, escalation mechanisms, regulatory constraints, professional standards, and leadership willingness to reject interventions that improve a commercial metric at unacceptable human cost.

13. The Evidence Supports a Bounded “Monetization of Fragility” Thesis, Not a Universal One

The phrase “monetization of fragility” captures an important structural possibility but requires careful classification. The strongest supported version of the thesis is that digital systems can identify predictable human limitations or vulnerable conditions, modify the architecture surrounding a decision, and derive commercial value from resulting behavioral changes. Dark-pattern research provides direct evidence that mechanisms capable of impairing autonomous choice exist and are deployed at meaningful scale. Consumer-finance research demonstrates how convenience, data, and deferred financial consequences can coexist inside rapidly scaling product categories. Workplace evidence demonstrates that interventions directed at individuals can coexist with structural conditions contributing to distress.

The broader claim that contemporary HCD as a discipline has systematically transformed into a machinery for exploiting vulnerability is not established by these findings. Many human-centered interventions demonstrably reduce burden, increase accessibility, improve safety, and align organizational performance with user outcomes. The thesis should therefore remain bounded to particular mechanisms, incentive environments, and evidence.

At least three competing explanations need to remain visible when harmful-looking patterns appear. In some cases, an organization may deliberately adopt an interface precisely because it increases conversion or retention despite a known cost to user agency. In others, no individual intends a harmful result; local teams optimize incomplete metrics and the aggregate system produces an undesirable outcome emergently. In still other cases, additional friction or constraint may have a legitimate purpose such as security, fraud prevention, accidental-action prevention, or regulatory compliance.

The absence of perfect information about intent does not eliminate the possibility of governance. Systems can be evaluated against observable properties: whether material information is hidden; whether entry and exit are asymmetrical; whether a default disproportionately benefits one party; whether refusal is reasonably available; whether users understand financial consequences; whether vulnerable states are used in targeting; and whether the organization monitors downstream outcomes.

This shifts governance from trying to prove motives toward evaluating mechanisms and consequences.

14. The Regulatory Direction Is Moving From Disclosure Toward Choice Architecture

Traditional consumer regulation often assumed that markets would function adequately if firms disclosed relevant information. Digital interface research has complicated that assumption because the manner in which information is presented can materially affect whether it is noticed, understood, or acted upon. A disclosure buried in a flow can be formally available yet behaviorally ineffective.

The OECD's work on dark commercial patterns reflects this broader shift by treating interface architecture itself as a consumer-policy issue rather than focusing solely on underlying contract terms.[1] The FTC's international review similarly examines design practices that may manipulate purchases or privacy decisions rather than limiting analysis to whether the information technically existed somewhere in the interface.[2] (Federal Trade Commission)

The implication for companies is significant. Compliance with disclosure rules may increasingly be necessary but insufficient. Organizations need to consider effective comprehension, not only legal availability. The strategic advantage will increasingly belong to firms capable of demonstrating that their design systems support both commercial performance and defensible consumer outcomes.

AI is likely to accelerate this regulatory evolution because static disclosure becomes less meaningful when the interface itself can change dynamically. Adaptive interfaces will require stronger auditability, experiment governance, and evidence concerning how optimization objectives affect different user groups.

15. Human-Centered Design Requires a New Definition of Performance

The current generation of product metrics is strongest where digital systems are easiest to instrument. Conversion, active users, retention, engagement, abandonment, revenue, and completion can be measured continuously. Human outcomes such as understanding, autonomy, regret, trust, downstream financial health, dependency, or longer-term wellbeing are more difficult to measure and often materialize later.

This creates an optimization bias toward what is immediately observable.

A next-generation human-centered measurement system should therefore not discard commercial KPIs but complement them with indicators capable of detecting whether the metric is being achieved through acceptable mechanisms.

For subscription businesses, retention should be interpreted alongside cancellation success, complaint rates, inactive-but-paying customers, and customer-reported value. For financial products, conversion and repayment should be considered alongside borrowing concentration, dispute outcomes, arrears, and customer understanding. For AI systems, engagement should be separated from justified reliance, user agency, and outcome quality. For workplace systems, adoption of wellbeing tools should not substitute for measures of workload, control, management quality, and underlying occupational risk.

The principle is simple but demanding: if the organization's claimed value proposition concerns human outcomes, the evidence used to validate the proposition should include human outcomes rather than only behavioral proxies.

16. The Next Generation of HCD Will Require an Integrity Layer Around Optimization

Human-centered design does not need to abandon experimentation, personalization, persuasion, growth, or commercial objectives. Modern products cannot operate effectively without many of these capabilities. What is missing is a clearer boundary around how those capabilities may be used.

A stronger HCD operating model would introduce an integrity layer around behavioral optimization. Consequential design decisions would be evaluated not only for usability and business performance but for whether users can understand the material consequence of the choice; whether the decision can be reversed at reasonable cost; whether entry and exit are reasonably symmetric; whether alternative choices remain visible; whether the system uses sensitive behavioral information to increase leverage; whether short-term behavioral metrics remain aligned with longer-term outcomes; and whether evidence claiming user benefit is sufficiently independent from the metric the organization is optimizing.

This approach changes the role of the design function. Design would no longer be responsible only for making organizational objectives easier for users to execute. It would also help identify circumstances in which those objectives need constraints.

That requires institutional authority. Designers cannot realistically protect user agency if every conflict between agency and conversion is resolved solely by revenue accountability. Human-centered governance therefore needs escalation paths, multidisciplinary review for high-consequence interventions, explicit restrictions on certain forms of targeting, and leadership metrics that recognize long-term trust and consumer outcomes.

The objective is not a frictionless world.

It is a world in which friction, persuasion, personalization, and automation are purposefully governed rather than automatically optimized.

Strategic Implications

For executives, the commercial risk is that organizations continue to optimize behavioral proxies faster than they develop mechanisms for validating whether those proxies remain connected to durable customer value. This can create regulatory exposure, reputational damage, customer mistrust, and strategic fragility even when short-term metrics improve. The strongest customer relationships are unlikely to be built by eliminating every barrier to conversion; they are more likely to be built by increasing confidence that the organization will not exploit informational advantage when incentives temporarily diverge.

For product and design leaders, the mandate is to broaden design quality from usability toward interaction integrity. The design system increasingly needs principles governing cancellation, consent, recommendation, personalization, financial commitment, automated action, and other high-consequence interactions. Teams should be able to distinguish friction removed because it does not serve the user from friction removed because it obstructs the company's preferred outcome.

For data and AI leaders, the central issue is objective design. Behavioral models should not be treated as neutral prediction infrastructure when their outputs determine how users are subsequently influenced. Increasing predictive accuracy increases the importance of governance over targeting, sensitive inference, experimentation, and feedback loops. NIST's lifecycle orientation provides a useful precedent: risk should be governed throughout design, deployment, measurement, and use rather than added as an interface disclaimer after the optimization system is complete. (NIST)

For regulators, the evidence suggests that focusing only on disclosure or individual prohibited interface elements may become progressively less effective as systems become personalized and adaptive. Mechanism-based governance—addressing obstruction, deception, coercion, material asymmetry, sensitive targeting, and contestability—offers a more durable path because it focuses on how behavioral influence operates rather than on one static interface implementation.

Research Limitations

The evidence base remains incomplete in several important respects. Regulatory sweeps such as the 2024 international subscription review provide valuable prevalence evidence within a defined sample, but they are not population-representative estimates of the entire digital economy. Studies of dark patterns use different definitions and taxonomies, making direct aggregation difficult. Organizations' internal experiment data are generally unavailable to independent researchers, limiting the ability to establish whether harmful outcomes reflect deliberate strategy, emergent metric optimization, or other operational constraints. The pace of change in generative AI further complicates longitudinal analysis because product architecture can evolve more quickly than peer-reviewed research.

Causal interpretation also requires discipline. Evidence that BNPL users have higher debt burdens does not by itself establish that BNPL caused those burdens; CFPB research documents important pre-existing differences. (Consumer Financial Protection Bureau) Evidence that workplace mental-health conditions are associated with workload or low control does not establish that every individual wellbeing intervention is ineffective; WHO's position is instead that organizational risk factors and individual support both require attention. (World Health Organization) Similarly, the presence of persuasive interface features does not prove clinically meaningful dependency, and personalization does not automatically constitute manipulation.

The strongest conclusions in this report are consequently structural rather than universal. Digital organizations possess significantly greater capacity to observe and optimize human behavior than earlier organizations; potentially manipulative design practices are empirically documented at meaningful scale; commercial objectives can diverge from user objectives; and stronger behavioral optimization increases the need for governance capable of preserving agency and downstream welfare. The magnitude of harm associated with each specific mechanism remains context-dependent and requires direct evidence.

Conclusion

Human-centered design has not become obsolete. It has become more powerful than its original governance model anticipated.

The field developed to correct systems that ignored human capability. Digitalization subsequently made human capability, limitation, preference, hesitation, attention, and behavior increasingly measurable. Continuous experimentation made those measurements actionable. Personalization made interventions more precise. Artificial intelligence is now reducing the cost of adapting those interventions at increasingly granular levels.

The resulting opportunity is substantial. Organizations can build products that are more accessible, intuitive, personalized, responsive, and useful than previous generations of systems.

The corresponding risk is equally structural.

Once human behavior becomes an optimization surface, knowing the user better does not guarantee serving the user better.

The evidence reviewed in this report supports a clear but bounded conclusion. Dark commercial patterns are empirically documented and materially present in digital markets. An international review found potential dark patterns in nearly 76 percent of 642 subscription services examined. Consumer-finance markets demonstrate how frictionless design, behavioral data, and future financial obligations can coexist at enormous scale. WHO evidence demonstrates why apparently individual problems can reflect structural conditions. AI risk-management frameworks increasingly recognize that technology must be governed as part of a wider socio-technical system rather than evaluated only through technical performance. (Federal Trade Commission)

The broader claim that modern human-centered design has universally transformed from care into the monetization of human fragility goes beyond this evidence. The more defensible conclusion is more useful: the institutional environment surrounding human-centered design now makes both outcomes possible.

Empathy can support care or increase leverage.

Personalization can reduce burden or increase asymmetry.

Friction removal can improve accessibility or compress deliberation.

Retention can demonstrate value or conceal impaired exit.

Engagement can represent usefulness or simply successful capture of attention.

Artificial intelligence can improve relevance or industrialize persuasion.

The distinguishing factor is therefore no longer the sophistication of the design methodology.

It is the governance of the objective being optimized.

The next generation of human-centered design should consequently be judged by a higher standard than whether the system is easy to use. A truly human-centered system should demonstrate that people can understand material consequences, retain meaningful choice, reverse important decisions, leave without unreasonable obstruction, contest consequential personalization, and benefit from behavioral insight rather than merely become easier to influence because of it.

Human-centeredness, in this emerging environment, is not proven by how deeply an organization understands people.

It is proven by what the organization is willing not to do with that understanding when commercial incentives and human interests diverge.

References

[1] OECD. Dark Commercial Patterns. OECD Digital Economy Papers, No. 336. OECD Publishing, Paris, 2022. The report develops a working definition of dark commercial patterns and reviews evidence concerning prevalence, effectiveness, consumer harms, and potential policy responses. (OECD)

[2] U.S. Federal Trade Commission. FTC, ICPEN, GPEN Announce Results of Review of Use of Dark Patterns Affecting Subscription Services, Privacy. July 2024. The international review covered 642 subscription websites and mobile applications; nearly 76 percent contained at least one possible dark pattern and nearly 67 percent contained multiple possible dark patterns. (Federal Trade Commission)

[3] Consumer Financial Protection Bureau. Buy Now, Pay Later: Market Trends and Consumer Impacts. September 2022. The CFPB reviewed data from five major providers and identified both competitive benefits and risks including data harvesting, incremental-sales incentives, autopay-related harms, and borrower overextension. (Consumer Financial Protection Bureau)

[4] Consumer Financial Protection Bureau. Consumer Use of Buy Now, Pay Later and Other Unsecured Debt. January 2025. The analysis matched BNPL originations from six large providers with de-identified credit records to examine persistence, multiple simultaneous loans, and broader consumer-debt positions. (Consumer Financial Protection Bureau)

[5] World Health Organization. Mental Health at Work. Updated September 2024. WHO estimates that 15 percent of working-age adults had a mental disorder in 2019 and that depression and anxiety account for approximately 12 billion lost working days and US$1 trillion in lost productivity annually. (World Health Organization)

[6] Tabassi, E. / National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1, January 2023. The voluntary framework addresses trustworthy and responsible AI across design, development, deployment, and use. (NIST)

[7] Consumer Financial Protection Bureau. Consumer Use of Buy Now, Pay Later: Insights from the CFPB Making Ends Meet Survey. March 2023. The analysis reports higher indebtedness and use of other credit among BNPL borrowers while emphasizing that many differences preceded BNPL adoption. (Consumer Financial Protection Bureau)

[8] WHO and International Labour Organization. Mental Health at Work: Policy and Implementation Materials. 2022–2024. The organizations identify heavy workloads and other working conditions as relevant mental-health risks and call for organizational as well as individual interventions. (World Health Organization)