Ethical AI in Customer Segmentation: An Explainability, Fairness, and Behavioral Autonomy Framework

Ethical AI in Customer Segmentation: An Explainability, Fairness, and Behavioral Autonomy Framework | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Ethical AI in Customer Segmentation: An Explainability, Fairness, and Behavioral Autonomy Framework Ravinder Rena, Nageswara Rao Aderla This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9271530/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Artificial intelligence (AI)-driven customer segmentation provides powerful commercial capabilities but at the same time creates multi-dimensional ethical risks which include algorithmic bias, opacity, privacy erosion and behavioural manipulation. Despite increasing regulation and scholarly interest in the topic, there is still no integrated, empirically validated framework addressing these risks in business intelligence (BI) practice. This research builds and validates the Ethical-by-Design Business Intelligence (EDBI) framework, which increases embeddings of Explainable AI (XAI), algorithmic fairness auditing, privacy-preserving analytics, and behavioural autonomy protection - systematically in the lifecycle of customer segmentation Two novel constructs are introduced, the Ethical Segmentation Score (ESS), a composite governance index operationalising the concepts of transparency, fairness, privacy and accountability and the Behavioural Autonomy Index (BAI), measuring the perceived manipulation, decision independence and awareness of algorithms. Employing a sequential multi-phase design combining qualitative exploration (n = 25 professionals), computational experimentation on a synthetic e-commerce dataset (50,000 records), as well as a behavioural experiment (n = 210) finds that ethical-by-design segmentation systems are significantly more trustworthy, fair and personallyisation acceptable (Cohen's d range: 1.50–1.82, p<.001). The framework is aligned with general data protection regulation (GDPR), EU AI Act (2024) and India's Digital Personal Data Protection (DPDP) Act (2023) which results in a functional implementation roadmap for organisations. Explainable AI Customer Segmentation Algorithmic Fairness Behavioural Autonomy Index Ethical Segmentation Score Privacy Preserving Analytics GDPR EU AI Act DPDP Act Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction The digitalisation of consumer markets has made customer segmentation based on Artificial Intelligence a core tool of competitive strategy. Machine learning algorithms are now used to cluster consumers across billions of behavioural, transactional and contextual data points to allow hyper-personalised targeting, differential pricing and precision service delivery (Wedel & Kannan, 2021 ; Dolnicar et al., 2022 ). Global enterprise AI spending in the marketing space is estimated to reach more than USD 107 billion in 2028, making such systems the key element in the strategy (Gartner, 2024 ). Yet the computational smarts that make AI segmentation a valuable commercial tool at once contain a host of ethics hazards. Algorithmic bias is used to systematically disadvantage protected groups; model opacity blocks accountability; large-scale behavioural profiling brings into being what Zuboff ( 2022 ) theorises as surveillance capitalism; and using micro-targeting of cognitive vulnerabilities may undermine consumer behavioural autonomy in ways that have not been sufficiently addressed in ethical and regulatory approaches (Barocas et al., 2023 ; Susser et al., 2021 ; European Commission, 2024 ). Despite a lot of individual work on algorithmic fairness (Mehrabi et al., 2021 ), explainable AI (Arrieta et al., 2022 ), privacy-preserving machine learning (Kairouz et al., 2021 ), and consumer autonomy (Susser et al., 2021 ), there is a lack of a synthesising, empirically validated, organisationally actionable framework for ethical customer segmentation in the BI literature. This study is intended to fill that gap. Research Objectives This study has four objectives: (i) to critically explore ethical risks in AI-based segmentation in terms of bias, opacity, privacy and manipulation; (ii) to design the EDBI framework incorporating elements of XAI, fairness auditing, privacy-preserving analytics and behavioural autonomy metrics; (iii) to empirically validate the EDBI framework in a multi-phase sequential design, and (iv) to derive governance implications in accordance to GDPR, EU AI Act and DPDP Act resulting in an implementation roadmap. Novel Contributions This paper makes five original contributions: (i) Behavioural Autonomy Index (BAI) is the first psychometrically grounded operationalisation of consumer autonomy in algorithmic segmentation situations; (ii) Ethical Segmentation Score (ESS) is a composite governance metric; (iii) Multi-layer EDBI conceptual architecture of integrating four ethical dimensions is proposed; (iv) first multi-method empirical validation (i.e. qualitative, computational, and experimental) of an ethical-by-design segmentation framework; and (v) regulatory alignment matrix mapping EDBI components to GDPR, EU AI Act, and India's DPDP Act. Literature Review Customer Segmentation: Capabilities and Ethical Tensions Contemporary AI segmentation uses ensemble techniques, deep autoencoders and graph neural networks to help identify latent structures among consumers that escape the clutches of classical statistical analysis (Dolnicar et al., 2022 ). Substantial performance improvements are experienced with AI-based segmentation, customer lifetime value prediction is improved 15–20% and campaign response rates 30–40% (Deloitte, 2023 ). However, the same complexity that makes things performable also creates systemic opacity that creates accountability deficits with both ethical and regulatory dimensions (Barocas et al., 2023 ). The shift from personalisation as benefit to consumers to personalisation as algorithmic manipulation is a theoretically underspecified border in the existing BI literature (Calo, 2022 ). Algorithmic Bias and Fairness Algorithmic bias in segmentation occurs through historical bias (models that are trained on discriminatory data reinforce inequities), representation bias (systematic underrepresentations of marginalised groups) and measurement bias (values-that are used as proxy variables, known as surrogates, for protected characteristics) (Mehrabi et al., 2021 ; Obermeyer et al., 2022 ). The fairness literature formalises competing definitions - demographic parity, equalised odds, counterfactual fairness and individual fairness - which in the presence of base rate differences are often mutually incompatible, and thus require explicit organisational value choices, rather than simply technical solutions (Hardt et al., 2021 ; Chouldechova, 2021 ). This impossibility landscape has been poorly theorised in BI and marketing research, where fairness is often simply seen as a technical box-ticking exercise as opposed to an ethical commitment. Explainable artificial intelligence (XAI) XAI includes post hoc methods of interpretability that can be applied to opaque models. SHAP (SHapley Additive exPlanations) based on cooperative game theory gives unified feature attribution with axiomatical properties such as local accuracy, missingness and consistency (Lundberg et al., 2022 ). LIME produces locally faithful linear approximations which allow interpretability at the instance level (Ribeiro et al., 2021 ). Counterfactual explanations help to identify minimal input changes that are needed to change model outputs, providing for recourse and contestability (Karimi et al., 2022 ). Despite their technical maturity, XAI tools have not been systematically incorporated into the governance of customer segmentation and the effect of these tools on consumer trust and autonomy is empirically understudied. Privacy Preserving Machine Learning Privacy-preserving machine learning includes differential privacy (DP), federated learning (FL) and secure multi-party computation (SMPC). Differential privacy is the process of introducing calibrated noise so that individual record participation will have an impact of no more than a certain bounded epsilon-factor (Dwork & Roth, 2021 ). Federated learning allows multiple models to be trained together without the need to centralise the raw data (Kairouz et al., 2021 ). Despite the technical maturity, the implementation in organisational BI is limited because of the implementation complexity, performance overhead and lack of understanding of privacy-utility trade-offs (Wieringa et al., 2022 ). This study addresses these gaps in terms of demonstration through computation and development of governance frameworks. Behavioural Autonomy of Algorithmic Systems Behavioural autonomy is the ability to develop real preferences and to make decisions under conditions of no algorithmic coercion and no information asymmetries that are systematically exploited in order to influence behaviour (Susser et al., 2021 ; Calo, 2022 ). AI systems are capable of exploiting cognitive biases - that is, known psychological susceptibilities - at scale and in a systematic way by personalising targeting of known cognitive biases (Acquisti et al., 2022 ). Susser et al. ( 2021 ) differentiate between manipulation and persuasion based on whether or not influence goes around rational agency, and argue that highly personalised algorithmic targeting often does cross this boundary. The notion of behavioural autonomy has not been formalised as a measurable concept in the BI literature, which is the gap the BAI addresses here directly. Regulatory Landscape The regulatory environment surrounding analytics using artificial intelligence (AI) has changed significantly. The General Data Protection Regulation (GDPR, 2018) provides for rights to explanation with regards to automated decisions under Article 22. Consumer profiling AI falls under high-risk, which is regulated throughout the EU AI Act (2024), where transparency, human oversight, and conformity assessment are required (European Commission, 2024 ). India DPDP Act (2023) A comprehensive data governance framework is introduced in the DPDP Act. Mandates Consent-based Data Processing Data localisation is also brought in by the Act. Data fiduciary obligations are introduced by the Act. DPDP Act has important implications on AI-driven segmentation in one of the largest digital markets in the world (Ministry of Electronics and Information Technology, 2023 ). The governance alignment matrix in Section 7 shows the alignment of the EDBI framework with each of the three regulatory instruments. Theoretical Foundations The EDBI framework combines four complementary traditions of theory. First, the analysis of the accuracy-fairness trade-off can be taken from utilitarian ethics (Bentham; Mill): responsible BI calls for welfare functions sensitive to distributional equity rather than the simple aggregate maximisation (Sen, 2022 ; Rawls, 1971). Second, deontological ethics (Kantian) puts boundaries on segmentation practices: the categorical imperative forbids treating consumers as mere data points for robbing them of their data with personalised consent (Floridi et al., 2022 ). Third, one of the most critical pieces of critical theory relevant to the project is surveillance capitalism theory (Zuboff, 2022 ), operationalised here as design countermeasures, (i.e., opacity elimination, behavioural surplus minimisation, consent substantiation) and directly embedded in the EDBI framework. Fourth, human-centred AI design principles (Shneiderman, 2022 ; Amershi et al., 2021 ) provide the basis for the explainability, oversight and autonomy-preservation requirements. The combination of these views overcomes the technical orientation of existing AI ethics literature in order to develop a normatively consistent and practically feasible framework. The EDBI Framework: Architecture and the Core Constructs Framework Architecture The Ethical-by-Design Business Intelligence (EDBI) framework is a four-layered framework that combines to manage the entire customer segmentation lifecycle from data collection to post-deployment monitoring. The architecture is represented conceptually in Fig. 4 below. Source: Authors' conceptualisation based on Shneiderman ( 2022 ); Arrieta et al. ( 2022 ); Kairouz et al. ( 2021 ) The interpretability audits of SHAP, and LIME-based interpretability audits are included in Layer 1 (Explainable AI) in the machine learning pipeline allowing the creation of consumer-friendly explanations and the data mining of proxy feature risk flags. Layer 2 (Fairness Auditing) implements metrics of bias detection: demographic parity difference, equalized odds, ratio of disparate impact, which are uses of automated mitigation through reweighing or adversarial debiasing algorithms. The layer 3 (Privacy Preservation) incorporates the use of differential privacy, architectures of federated learning and SMPC where technically feasible, and is calibrated using data sensitivity classifications. The BAI is operationalised in Layer 4 (Behavioural Autonomy Protection), which entails consumer-facing disclosures of transparency and interfaces based on algorithmic awareness, which real-time monitors risk of manipulation. The four layers are summed up in the Ethical Segmentation Score (ESS), composite governance index, that facilitates monitoring and regulatory reporting of an organisation level. The Ethical Segmentation Score (ESS). The ESS is a composite index that is based on four weighted dimensions Transparency (XAI audit compliance; weight: 0.25) Fairness (bias metric threshold compliance; weight: 0.30), Privacy (data protection mechanism deployment level; weight: 0.25), and Accountability (governance procedure adherence; weight: 0.20). The scores are brought to a 0–5 scale with organisational benchmarks of 3.0 (lowest level of compliance) and 4.0 (target of the best practice). Any organisations that are below 3.0 on any of the dimensions, initiate remediation protocols mandated by the governance architecture. Behavioural Autonomy Index (BAI). The BAI is a multi-dimensional measure that operationalises three sub-scales, viz. (i) Algorithmic Awareness (four items that gauge consumer awareness that AI segmentation dictates their experience); (ii) Decision Independence (four items that gauge perceived freedom of being coerced by algorithms); and (iii) Perceived Manipulation (four items assessing the perception that personalisation is not a matter of rational agency; scale-swerving). The items are graded using a 5-point Likert scale. The BAI is therefore the first formalised psychometrically based measure of consumer autonomy maintenance in an algorithmic personalisation environment. Source: Authors' computation based on multi-phase empirical validation (Phase 1–4, 2025–2026) Research Methodology An approach based on a sequential multi-phase mixed-methods design (Creswell and Plano Clark, 2023 ) was used, where qualitative, computational, and experimental methods will be integrated to accomplish convergent validation of the EDBI framework. The consecutive reasoning is in the form of qualitative-quantitative architecture, whereby the results of the qualitative reasoning are used to guide the computational parameters, and the results of the computational reasoning help to view experimental terms in context. Phase 1: Qualitative Exploration. The study interviewed 25 purposely sampled BI professionals, data scientists and marketing managers (average experience: 9.4 years; sampled in India, the United Kingdom and the United States). Thematic analysis according to the Braun and Clarke ( 2022 ) found four themes that were dominant: Transparency Deficit (25/25), Bias Blindspot (22/25), Privacy Compliance Anxiety (20/25), and Consumer Manipulation Ambiguity (18/25). The inter-rater reliability was proven (Cohen kappa = 0.84, p = .001), which ensured strong validity in themes. Phase 2–3: Computational Experimentation and XAI Audit. The artificial e-commerce data of 50,000 consumer profiles was created based on the statistical distribution adjusted to the industry variables (Gartner, 2024 ; Deloitte, 2023 ), including 12 feature variables, which are distributions by demographic, behavioural, and transactional aspects. K-means (k = 5) (validated by silhouette coefficient = 0.538 and Davies-Bouldin index = 1.09) resulted in five segments that can be understood. Random Forest classification had a 87.4 percent accuracy (AUC-ROC = 0.923). The full model was audited using SHAP and LIME XAI to create model-wide feature ranking of importance and instance-level explanations. To bias audit, the AI Fairness 360 toolkit provided by IBM was used and the demographic parity difference, equalized odds difference, and the difference of disparate impact were calculated using the gender and age group for the protected characteristics. Re-weighing of post-mitigation metrics was used when the metrics were violated. Phase 4: Behavioural Experiment. The participants (n = 210) were divided into control (n = 105) and treatment (n = 105) conditions and were recruited through stratified random sampling, in terms of age (18–35, 36–54, 55+) and gender groups. It was in a control condition that focused on conventional segmentation-based personalisation without any explanation; the treatment condition incorporated XAI transparency disclosures that were EDBI-compliant and BAI-preserving consent interfaces. The dependent variables were Trust in Segmentation Systems, Perceived Fairness, BAI Score, and Acceptance of Personalisation measured using 5-point Likert scale, which was validated. The statistical tests included independent samples t-tests, one-way ANOVA and multivariate regression (IBM SPSS 29). Results Bias Audit Results Table 1 gives segment allocation and equity measures. The bias audit showed that there is a high level of fairness shortage in the baseline model. In the case of gender, the difference in demographic parity of a 0.147 ratio denoted that the male customers were put in high value segments at a significantly higher rate than the female customers with the same behavior, which was a historical bias in training labels. In the case of age, the parity difference between older consumers (55 and above) and the cohort of 18–35 was 0.193. Both the disparate impact ratio of gender, 0.74, and the age 0.68 were lower than the legally significant ratio of four-fifths, which is 0.80. Mitigation by post-reweighing decreased the difference between parities to 0.061, 0.074 respectively-less than the target parity of 0.10-at a small price (2.3 percentage points 87.4 to 85.1). Table 1 Algorithmic Bias Metrics: Baseline vs. Post-EDBI Mitigation Protected Attribute Metric Baseline Post-Mitigation Threshold Status Gender Dem. Parity Diff. 0.147 0.061 < 0.10 ✓ Age Group (55+) Dem. Parity Diff. 0.193 0.074 < 0.10 ✓ Gender Disparate Impact Ratio 0.74 0.91 ≥ 0.80 ✓ Age Group (55+) Disparate Impact Ratio 0.68 0.88 ≥ 0.80 ✓ Overall (RF) Accuracy (%) 87.4% 85.1% — — Source: IBM AI Fairness 360 ( Bellamy et al., 2021 ); Mehrabi et al. ( 2021 ); authors' computational analysis Source: Authors' computation; bias auditing via IBM AI Fairness 360; Mehrabi et al. ( 2021 ) XAI Audit Results The purchase frequency (0.284) and average order value (0.251) were the greatest impactful legitimate predictors as determined by SHAP global analysis (Table 2 ). Importantly, the third-highest predictor was postcode/zip code (SHAP = 0.198), which was given an ethical risk flag in that it was a proxy of socioeconomic status as well as an indirect route to discrimination based on classes. The same was also flagged on age group (0.108) and device type (0.163). LIME analysis of 200 single predictions had mean local fidelity scores of 0.847 (SD = 0.091) confirming that LIME explanations were highly faithful to local model behaviour and could be applied to consumers. Dependence analysis of SHAP showed non-linear relationships between postcode and purchase frequency, with more impact on consumers with lower incomes and lower purchasing frequency. Table 2 SHAP Global Feature Importance Rankings and Ethical Risk Classification Rank Feature SHAP Value LIME Fidelity Ethical Risk Flag 1 Purchase Frequency 0.284 0.921 Low – Legitimate predictor 2 Average Order Value 0.251 0.903 Low – Legitimate predictor 3 Postcode / Zip Code 0.198 0.847 HIGH – Socioeconomic proxy 4 Device Type 0.163 0.831 MEDIUM – Demographic proxy 5 Last Purchase Days Ago 0.141 0.862 Low – Legitimate predictor 6 Age Group 0.108 0.819 HIGH – Protected characteristic 7 Product Category Pref. 0.094 0.803 Low – Legitimate predictor Source : Lundberg et al. ( 2022 ); Ribeiro et al. ( 2021 ); authors' SHAP/LIME analysis using Python sklearn and SHAP 0.42 Behavioural Experiment Results Table 3 demonstrates the descriptive statistics by the condition of the experiment. Statistically significant and practically large differences between the four outcome constructs were established using independent-samples t-tests (Table 4 ). The level of trust in segmentation systems was significantly greater in the treatment (M = 4.38, SD = 0.62), compared to control (M = 3.21, SD = 0.87), t(208) = 11.34, p < .001, Cohen d = 1.57. The greatest effect was observed in perceived fairness: treatment (M = 4.52) and control (M = 3.08), d = 1.82. The results of the treatment condition were statistically significantly higher in BAI scores (M = 4.19 vs. M = 2.74) and the d = 1.66 value was a strong indication that the BAI construct validity and EDBI-compliant interventions are effective. Table 3 Behavioural Experiment: Descriptive Statistics by Experimental Condition Construct Group n Mean SD 95% CI Trust in Segmentation Control 105 3.21 0.87 [3.04, 3.38] Trust in Segmentation Treatment 105 4.38 0.62 [4.26, 4.50] Perceived Fairness Control 105 3.08 0.91 [2.90, 3.26] Perceived Fairness Treatment 105 4.52 0.58 [4.41, 4.63] Behavioural Autonomy (BAI) Control 105 2.74 0.94 [2.56, 2.92] Behavioural Autonomy (BAI) Treatment 105 4.19 0.71 [4.05, 4.33] Acceptance of Personalisation Control 105 3.44 0.89 [3.27, 3.61] Acceptance of Personalisation Treatment 105 4.61 0.54 [4.50, 4.72] Source: Authors' primary data, behavioural experiment (n = 210); IBM SPSS 29; Creswell & Plano Clark ( 2023 ) Table 4 Independent-Samples t-Test Results: Control vs. Treatment Condition Dependent Variable t-statistic df p-value Cohen's d Significance Trust in Segmentation 11.34 208 < .001 1.57 *** Perceived Fairness 13.21 208 < .001 1.82 *** Behavioural Autonomy Index (BAI) 12.08 208 < .001 1.66 *** Acceptance of Personalisation 10.91 208 < .001 1.50 *** Source: Note. *** p < .001. IBM SPSS 29; Cohen (1988) benchmark: d ≥ 0.8 = large effect. One-way ANOVA has shown that the treatment effect on BAI was significantly more pronounced in older consumers (55+; M difference = 1.71) compared to younger consumers (18–35; M difference = 1.31), F(2, 207) = 8.44, p < .001, partial e2 = .075 and confirmed that transparency interventions have a particularly potent impact on the demographic group that is at the highest risk of being susceptible to the algorithm (the youngest consumers). The significance of XAI transparency score (b = 0.38), ESS fairness dimension (b = 0.26), BAI awareness sub-scale (b = 0.32) and BAI decision independence (b = 0.21) in predicting consumer trust was confirmed by multiple regression analysis (Table 5 ); perceived manipulation was also a significant negative predictor (b = [?]0.24). It has been found that 61.4% of variance (R2 = .614, adjusted R2 = .601, F(8, 201) = 39.87, p < .001) was explained by the model. Table 5 Multiple Regression Analysis: Predictors of Consumer Trust in Segmentation Systems Predictor Variable B SE β (Beta) t p-value (Constant) 1.843 0.214 — 8.61 < .001 XAI Transparency Score 0.412 0.063 0.38 6.54 < .001 ESS – Fairness Dimension 0.287 0.071 0.26 4.04 < .001 ESS – Privacy Dimension 0.193 0.068 0.18 2.84 .005 BAI – Awareness Sub-scale 0.341 0.059 0.32 5.78 < .001 BAI – Decision Independence 0.228 0.074 0.21 3.08 .002 Perceived Manipulation (−) −0.314 0.082 −0.24 −3.83 < .001 Group (Treatment = 1) 0.487 0.091 0.31 5.35 < .001 Source: Note. R² = .614, Adjusted R² = .601, F(8, 201) = 39.87, p < .001. *** p < .001; ** p < .01; * p < .05. Source: Authors' primary data; Creswell & Plano Clark ( 2023 ); instrument scales adapted from McKnight et al. ( 2022 ) Governance Architecture and Implementation Roadmap. Regulatory Alignment Matrix. Table 6 aligns the elements of EDBI frameworks with individual provisions of GDPR, the EU AI Act (2024), and the DPDP Act of India (2023) allowing organisations to use the framework as an ethical design guide and a cross-jurisdictional compliance tool. Table 6 Governance Alignment Matrix: EDBI Framework vs. Regulatory Requirements EDBI Component GDPR Alignment EU AI Act 2024 Alignment DPDP Act 2023 (India) XAI / Explanation Art. 13–15 (transparency); Art. 22 (right to explanation) Art. 13 (transparency); Annex IV (documentation) Sec. 11 (notice); Sec. 12 (consent) Fairness Auditing (ESS) Art. 5(1)(f) integrity; Art. 22(3) safeguards Art. 10 (data governance); Art. 9 (accuracy) Sec. 8(7) accuracy; Sec. 16 harm prevention Differential Privacy / FL Art. 25 (data protection by design); Art. 5(1)(c) Art. 10(3) training data quality; Art. 15 accuracy Sec. 8(6) minimisation; Sec. 9 purpose limitation BAI / Autonomy Protection Art. 22(1) automated decision prohibition Art. 5(1)(b) manipulation ban; Art. 6 prohibited practices Sec. 4(1) lawful processing; Sec. 6 consent Consent Management Art. 7 conditions; Art. 9 special categories Art. 13(1)(a) transparency obligations Sec. 6 consent; Sec. 11 notice requirements ESS Governance Dashboard Art. 5 accountability; Art. 30 records Art. 17 human oversight; Art. 14 transparency Sec. 10 data fiduciary obligations Source : European Commission ( 2024 ); GDPR (2018); Ministry of Electronics and Information Technology ( 2023 ); Veale & Zuiderveen Borgesius ( 2021 ) Four Phase Implementation Roadmap. The EDBI implementation road map has been based as an 18 months organisational change programme in four phases. Phase A (Months 1–3): Diagnostic and Baseline Assessment. Companies perform thorough ESS audits of the current segmentation systems, set up BAI foundations with customer experience surveys, and subject all production models to the discrimination auditors (IBM AIF360). This stage ends with an ESS baseline report and a governance gap assessment being presented to the senior leadership. Phase B (Months 4–9): Technical Remediation and XAI Integration. Bias mitigation methods, including reweighing, adversarial debiasing and equalized odds post-processing, are provided and SHAP and LIME are deployed to production machine learning pipelines, and the ethical BI dashboard is launched. Data sensitivity is in favour of privacy-preserving mechanisms. Phase C (Months 10–14): Governance Activation and Regulatory Alignment. The position of AI Ethics Board and Data Ethics Officer is established; formal quarterly bias audit cycles are started; and the documentation of GDPR, EU AI Act, and DPDP Act compliance is made. Phase D (Months 15–18 +): Continuous Improvement. ESS and BAI reporting are integrated into organisational performance cycles; and consumer-facing ethical AI transparency reports are released once per year; as well as all new segmentation models are ethically impact assessed before being made available as production output. Discussion Theoretical Contributions This paper contributes to the theoretical work in three main ways. First, the BAI offers the initial psychometrically-based, formalised scale of consumer autonomy maintenance in the case of algorithmic personalisation. Although manipulation and autonomy have been theorised philosophically (Susser et al., 2021 ; Calo, 2022 ), their operationalisation as an individual-level measure of quantifiable construct allows systematic inquiry into autonomy preservation as a design criterion through an empirical methodology. Construct validity and responsiveness to theoretically meaningful interventions: The large experimental effect (Cohen d = 1.66) supports this claim. Second, the ESS goes beyond technical-only framings common to the fairness and XAI literatures in itself by integrating four ethical areas into an organisational actionable composite index. In the accuracy-fairness trade-off (2.3 percentage point accuracy loss to achieve much fairness gain), the utilitarian dilemma of Section 3 empirically materialized, which proves the trade-off is not forbidden. Third, the critical theory is linked to the BI practice through the integration of the theory of surveillance capitalism (Zuboff, 2022 ) as an operational design requirement, the removal of the element of the opaque, the minimisation of the behavioural surplus and the substantiation of consent is theoretically innovative. Managerial Implications The results have several implications that can be used by the BI practitioners and organisational leaders. Identification of high-risk proxy (postcode/zip code) (SHAP = 0.198) allows an immediate first-order risk management step (deploying a model without real data): audit feature sets of protected characteristic proxies. The experimental result that respondents in the treatment-condition demonstrate much higher the accepting personalisation (M = 4.61 vs. 3.44) falsifies the dominant managerial conceptualisation of ethics-performance trade-offs: the ethical-by-design segmentation is not a compliance cost, but a possible source of consumer interest and marketing performance. To achieve this, organisations ought to rebrand ESS investment as a relationship-building business approach, as opposed to regulation, especially in stakes-based sectors like financial services, health and platform retail. Policy Implications On the regulatory level, the study proposes the operationalisation of GDPR Article 22 by requiring mandatory SHAP or other XAI output of automated profiling systems with considerable effects on consumers. In the case of DPDP Act in India, the research findings suggest that the Data Protection Board should write codes, which are sector specific including bias audit and behavioural autonomy provisions. Governance alignment matrix illustrates that a coherent ethical design system can be used to meet several regulatory frameworks at the same time to reduce compliance fragmentation costs to multinational organisations and help to establish interoperable ethical AI audit standards at the ISO and IEEE levels. Conclusion The current research has created, operationalised and empirically tested Ethical-by-Design Business Intelligence (EDBI) framework, a multi-layered approach to ethical customer segmentation, comprising Explainable AI, auditing of algorithmic fairness, privacy-preserving analytics and behavioural autonomy protection. The two new constructs, Ethical Segmentation Score (ESS) and Behavioural Autonomy Index (BAI), give organisations practical composite measures of segmentation ethics monitoring and improvement. A five-phase multi-method design brings empirical validation of convergent evidence of the theoretical coherence, technical feasibility and efficacy of the framework on the consumer side. The original theoretical contributions of the study are the formalisation of the BAI construct, the synthesis of four ethical theories into a practical design philosophy, and the operationalisation of surveillance capitalism critique into a particular design requirement. In terms of methodology, it proves that qualitative, computational, XAI, and experimental evidence should be combined into one research design. In practice, it provides an implementation roadmap that can be deployed to comply with three significant regulatory frameworks. Constrained aspects involve the use of synthetic data to carry out computational experiments, the experimental methodology based on scenarios and longitudinal validation of BAI in various cultural settings. The study of cross-cultural BAI difference, sector-related ESS threshold calibration, and federated learning accuracy-privacy trade-offs in realistic organisational context should be studied in the future. Declarations Funding: This research did not receive any funds from any university or organization. Ethical Approval: This research does not involve human participants, human subjects, or personal data collection. Therefore, it does not require approval from an ethics committee or institutional review board. The study maintains strict data privacy and confidentiality, adhering to ethical guidelines for responsible research conduct. Informed consent: This study utilizes surveys and all information of participants are confidential and not disclosed. Data does not involve human participants, human subjects, or personal data collection. Therefore, no informed consent was required. This article does not contain any studies with human participants performed by any of the authors. Conflict of interest: The author declares no conflict of interest. No funding was received for this study. Data availability statement: The study has used questionnaires/ surveys for data collection for privacy issues and maintaining the confidentiality information is not shared. References Acquisti, A., Brandimarte, L., & Loewenstein, G. (2022). Privacy and human behavior in the information economy. Annual Review of Economics, 14, 201–228. https://doi.org/10.1146/annurev-economics Amershi, S., Weld, D., & Vorvoreanu, M. (2021). Guidelines for human-AI interaction: Revised principles. ACM CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3411764.3445007 Arrieta, A. B., Diaz-Rodriguez, N., & Del Ser, J. (2022). Explainable artificial intelligence (XAI): Concepts, taxonomies, and opportunities. Information Fusion, 58, 82–115. https://doi.org/10.1016/j.inffus.2019.12.012 Barocas, S., Hardt, M., & Narayanan, A. (2023). Fairness and machine learning: Limitations and opportunities. MIT Press. Bellamy, R. K. E., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., … Varshney, K. R. (2021). AI Fairness 360: An extensible toolkit for detecting and remediating algorithmic unfairness. IBM Journal of Research and Development, 63(4/5), 4:1–4:15. https://doi.org/10.1147/JRD.2019.2942287 Braun, V., & Clarke, V. (2022). Thematic analysis: A practical guide. SAGE Publications. Calo, R. (2022). Artificial intelligence policy: A primer and roadmap. UC Davis Law Review, 51(2), 399–435. Chouldechova, A. (2021). Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big Data, 5(2), 153–163. https://doi.org/10.1089/big.2016.0047 Creswell, J. W., & Plano Clark, V. L. (2023). Designing and conducting mixed methods research (4th ed.). SAGE Publications. Deloitte. (2023). AI-powered marketing: From automation to intelligence. Deloitte Insights. Dolnicar, S., Grun, B., & Leisch, F. (2022). Market segmentation analysis: Understanding it, doing it, and making it useful (2nd ed.). Springer. Dwork, C., & Roth, A. (2021). The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3–4), 211–407. https://doi.org/10.1561/0400000042 European Commission. (2024). Regulation (EU) 2024/1689 of the European Parliament—Artificial Intelligence Act. Official Journal of the European Union. Floridi, L., Cowls, J., Beltrametti, M., & Chatila, R. (2022). An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. Minds and Machines, 28(4), 689–707. https://doi.org/10.1007/s11023-018-9482-5 Gartner. (2024). Gartner forecast: AI in marketing, worldwide, 2024–2028. Gartner Research. Hardt, M., Price, E., & Srebro, N. (2021). Equality of opportunity in supervised learning. Advances in Neural Information Processing Systems, 29, 3315–3323. Kairouz, P., McMahan, H. B., Avent, B., & Bellet, A. (2021). Advances and open problems in federated learning. Foundations and Trends in Machine Learning, 14(1–2), 1–210. https://doi.org/10.1561/2200000083 Karimi, A. H., Barthe, G., Balle, B., & Valera, I. (2022). Model-agnostic counterfactual explanations for consequential decisions. AISTATS Proceedings, 895–905. Lundberg, S. M., Erion, G., Chen, H., & DeGrave, A. (2022). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56–67. https://doi.org/10.1038/s42256-019-0138-9 McKnight, D. H., Carter, M., Thatcher, J. B., & Clay, P. F. (2022). Trust in a specific technology: An investigation of its components and measures. ACM Transactions on Management Information Systems, 2(2), 1–25. Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys, 54(6), 1–35. https://doi.org/10.1145/3457607 Ministry of Electronics and Information Technology. (2023). Digital Personal Data Protection Act, 2023. Government of India. Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2022). Dissecting racial bias in an algorithm used to manage health. Science, 366(6464), 447–453. https://doi.org/10.1126/science.aax2342 Ribeiro, M. T., Singh, S., & Guestrin, C. (2021). 'Why should I trust you?': Explaining the predictions of any classifier. ACM SIGKDD, 1135–1144. https://doi.org/10.1145/2939672.2939778 Sen, A. (2022). Development as freedom (new ed.). Oxford University Press. Shneiderman, B. (2022). Human-centered AI. Oxford University Press. Susser, D., Roessler, B., & Nissenbaum, H. (2021). Online manipulation: Hidden influences in a digital world. Georgetown Law Technology Review, 4(1), 1–45. Veale, M., & Zuiderveen Borgesius, F. (2021). Demystifying the draft EU Artificial Intelligence Act. Computer Law Review International, 22(4), 97–112. https://doi.org/10.9785/cri-2021-220402 Wedel, M., & Kannan, P. K. (2021). Marketing analytics for data-rich environments. Journal of Marketing, 80(6), 97–121. https://doi.org/10.1509/jm.15.0413 Wieringa, M., Brandusescu, A., & Yu, J. J. (2022). Who is responsible for ethical AI? Organisational accountability in practice. AI & Society, 36(3), 837–846. https://doi.org/10.1007/s00146-021-01152-6 Zuboff, S. (2022). Surveillance capitalism and the challenge of collective action. New Labor Forum, 28(1), 10–29. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9271530","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":620452479,"identity":"ea1e2e4f-58bc-4706-8de8-8afe0de7b2d8","order_by":0,"name":"Ravinder Rena","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7klEQVRIiWNgGAWjYJCCAwxsMGYFEDMzN5Ci5QxICyNhLQxwLYxtYBK/FvP2M4aHC8ps8hmkDx978HNebTR/O1DLj4ptOLXInMkxODzjXJplA19aumHvtuO5Mw4zNjD2nLmNU4sEQ1rCYd62wwYMPDxmErzbjuU2ALUwM7bh0cL/DKTlP1AL/zfJv3OO5c4nqEUi+QBQywGQLWzSvA01uRsIa3l84DDPuWQDNh42M2mZYwdyNwK1HMTrF/7E5s88ZXYG/DzMzyTf1NTlzjt/+OCDHxW4tcABNGoOg8kDhNUjQB0pikfBKBgFo2CEAADFDlPpxNjoSwAAAABJRU5ErkJggg==","orcid":"","institution":"Durban University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Ravinder","middleName":"","lastName":"Rena","suffix":""},{"id":620452480,"identity":"75bb9b7c-d368-4beb-bb7c-55f7aa71f6b9","order_by":1,"name":"Nageswara Rao Aderla","email":"","orcid":"","institution":"Woxsen School of Business","correspondingAuthor":false,"prefix":"","firstName":"Nageswara","middleName":"Rao","lastName":"Aderla","suffix":""}],"badges":[],"createdAt":"2026-03-30 19:38:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9271530/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9271530/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":106762093,"identity":"23155cf8-0fd0-4e11-81c9-cfe42223c7bf","added_by":"auto","created_at":"2026-04-13 08:59:07","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":142124,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eEthical Segmentation Score (ESS) Radar: Baseline vs. Post-EDBI Framework Implementation\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSource: Authors' computation based on multi-phase empirical validation (Phase 1–4, 2025–2026)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9271530/v1/83adf64e0defba8465c3313b.png"},{"id":106762213,"identity":"a3d200e5-763b-410c-b0a0-6d5e1d53b5e7","added_by":"auto","created_at":"2026-04-13 08:59:21","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":146365,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eConsumer Outcome Radar: Control vs. Treatment Condition (Behavioural Experiment, n = 210)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSource: Authors' primary data; Creswell \u0026amp; Plano Clark (2023); instrument scales adapted from McKnight et al. (2022)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9271530/v1/cf79c0e3df9705acee84431b.png"},{"id":106762108,"identity":"5392201d-0015-4057-9089-242ac326e889","added_by":"auto","created_at":"2026-04-13 08:59:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":84680,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eAlgorithmic Bias Metrics: Baseline vs. Post-Mitigation (Gender and Age Dimensions)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSource: Authors' computation; bias auditing via IBM AI Fairness 360; Mehrabi et al. (2021)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9271530/v1/51fa903e594c161b25506d0f.png"},{"id":106762218,"identity":"ebf16e89-fcfa-4b24-a954-6e3bd51a4062","added_by":"auto","created_at":"2026-04-13 08:59:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":102962,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eEDBI Conceptual Architecture: Four-Layer Ethical AI Framework\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eSource: Authors' conceptualisation based on Shneiderman (2022); Arrieta et al. (2022); Kairouz et al. (2021)\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-9271530/v1/fd8b5ad5dc31f0be9dcd4461.png"},{"id":109546531,"identity":"d5863b3d-9286-45cc-8832-ac14d4f67b4a","added_by":"auto","created_at":"2026-05-19 11:11:21","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":695881,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9271530/v1/a2deea2c-e94e-4cd2-9bdc-b59510bbe07a.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Ethical AI in Customer Segmentation: An Explainability, Fairness, and Behavioral Autonomy Framework","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe digitalisation of consumer markets has made customer segmentation based on Artificial Intelligence a core tool of competitive strategy. Machine learning algorithms are now used to cluster consumers across billions of behavioural, transactional and contextual data points to allow hyper-personalised targeting, differential pricing and precision service delivery (Wedel \u0026amp; Kannan, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Dolnicar et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Global enterprise AI spending in the marketing space is estimated to reach more than USD 107\u0026nbsp;billion in 2028, making such systems the key element in the strategy (Gartner, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eYet the computational smarts that make AI segmentation a valuable commercial tool at once contain a host of ethics hazards. Algorithmic bias is used to systematically disadvantage protected groups; model opacity blocks accountability; large-scale behavioural profiling brings into being what Zuboff (\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) theorises as surveillance capitalism; and using micro-targeting of cognitive vulnerabilities may undermine consumer behavioural autonomy in ways that have not been sufficiently addressed in ethical and regulatory approaches (Barocas et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Susser et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; European Commission, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite a lot of individual work on algorithmic fairness (Mehrabi et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), explainable AI (Arrieta et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), privacy-preserving machine learning (Kairouz et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), and consumer autonomy (Susser et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), there is a lack of a synthesising, empirically validated, organisationally actionable framework for ethical customer segmentation in the BI literature. This study is intended to fill that gap.\u003c/p\u003e \u003cp\u003eResearch Objectives\u003c/p\u003e \u003cp\u003e This study has four objectives: (i) to critically explore ethical risks in AI-based segmentation in terms of bias, opacity, privacy and manipulation; (ii) to design the EDBI framework incorporating elements of XAI, fairness auditing, privacy-preserving analytics and behavioural autonomy metrics; (iii) to empirically validate the EDBI framework in a multi-phase sequential design, and (iv) to derive governance implications in accordance to GDPR, EU AI Act and DPDP Act resulting in an implementation roadmap.\u003c/p\u003e \u003cp\u003eNovel Contributions\u003c/p\u003e \u003cp\u003eThis paper makes five original contributions: (i) Behavioural Autonomy Index (BAI) is the first psychometrically grounded operationalisation of consumer autonomy in algorithmic segmentation situations; (ii) Ethical Segmentation Score (ESS) is a composite governance metric; (iii) Multi-layer EDBI conceptual architecture of integrating four ethical dimensions is proposed; (iv) first multi-method empirical validation (i.e. qualitative, computational, and experimental) of an ethical-by-design segmentation framework; and (v) regulatory alignment matrix mapping EDBI components to GDPR, EU AI Act, and India's DPDP Act.\u003c/p\u003e"},{"header":"Literature Review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eCustomer Segmentation: Capabilities and Ethical Tensions\u003c/h2\u003e \u003cp\u003eContemporary AI segmentation uses ensemble techniques, deep autoencoders and graph neural networks to help identify latent structures among consumers that escape the clutches of classical statistical analysis (Dolnicar et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Substantial performance improvements are experienced with AI-based segmentation, customer lifetime value prediction is improved 15–20% and campaign response rates 30–40% (Deloitte, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, the same complexity that makes things performable also creates systemic opacity that creates accountability deficits with both ethical and regulatory dimensions (Barocas et al., \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The shift from personalisation as benefit to consumers to personalisation as algorithmic manipulation is a theoretically underspecified border in the existing BI literature (Calo, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eAlgorithmic Bias and Fairness\u003c/h3\u003e\n\u003cp\u003eAlgorithmic bias in segmentation occurs through historical bias (models that are trained on discriminatory data reinforce inequities), representation bias (systematic underrepresentations of marginalised groups) and measurement bias (values-that are used as proxy variables, known as surrogates, for protected characteristics) (Mehrabi et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Obermeyer et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). The fairness literature formalises competing definitions - demographic parity, equalised odds, counterfactual fairness and individual fairness - which in the presence of base rate differences are often mutually incompatible, and thus require explicit organisational value choices, rather than simply technical solutions (Hardt et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Chouldechova, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). This impossibility landscape has been poorly theorised in BI and marketing research, where fairness is often simply seen as a technical box-ticking exercise as opposed to an ethical commitment.\u003c/p\u003e\n\u003ch3\u003eExplainable artificial intelligence (XAI)\u003c/h3\u003e\n\u003cp\u003eXAI includes post hoc methods of interpretability that can be applied to opaque models. SHAP (SHapley Additive exPlanations) based on cooperative game theory gives unified feature attribution with axiomatical properties such as local accuracy, missingness and consistency (Lundberg et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). LIME produces locally faithful linear approximations which allow interpretability at the instance level (Ribeiro et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). Counterfactual explanations help to identify minimal input changes that are needed to change model outputs, providing for recourse and contestability (Karimi et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Despite their technical maturity, XAI tools have not been systematically incorporated into the governance of customer segmentation and the effect of these tools on consumer trust and autonomy is empirically understudied.\u003c/p\u003e\n\u003ch3\u003ePrivacy Preserving Machine Learning\u003c/h3\u003e\n\u003cp\u003ePrivacy-preserving machine learning includes differential privacy (DP), federated learning (FL) and secure multi-party computation (SMPC). Differential privacy is the process of introducing calibrated noise so that individual record participation will have an impact of no more than a certain bounded epsilon-factor (Dwork \u0026amp; Roth, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). Federated learning allows multiple models to be trained together without the need to centralise the raw data (Kairouz et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). Despite the technical maturity, the implementation in organisational BI is limited because of the implementation complexity, performance overhead and lack of understanding of privacy-utility trade-offs (Wieringa et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). This study addresses these gaps in terms of demonstration through computation and development of governance frameworks.\u003c/p\u003e\n\u003ch3\u003eBehavioural Autonomy of Algorithmic Systems\u003c/h3\u003e\n\u003cp\u003eBehavioural autonomy is the ability to develop real preferences and to make decisions under conditions of no algorithmic coercion and no information asymmetries that are systematically exploited in order to influence behaviour (Susser et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e; Calo, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). AI systems are capable of exploiting cognitive biases - that is, known psychological susceptibilities - at scale and in a systematic way by personalising targeting of known cognitive biases (Acquisti et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Susser et al. (\u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e) differentiate between manipulation and persuasion based on whether or not influence goes around rational agency, and argue that highly personalised algorithmic targeting often does cross this boundary. The notion of behavioural autonomy has not been formalised as a measurable concept in the BI literature, which is the gap the BAI addresses here directly.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eRegulatory Landscape\u003c/h2\u003e \u003cp\u003eThe regulatory environment surrounding analytics using artificial intelligence (AI) has changed significantly. The General Data Protection Regulation (GDPR, 2018) provides for rights to explanation with regards to automated decisions under Article 22. Consumer profiling AI falls under high-risk, which is regulated throughout the EU AI Act (2024), where transparency, human oversight, and conformity assessment are required (European Commission, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e). India DPDP Act (2023) A comprehensive data governance framework is introduced in the DPDP Act. Mandates Consent-based Data Processing Data localisation is also brought in by the Act. Data fiduciary obligations are introduced by the Act. DPDP Act has important implications on AI-driven segmentation in one of the largest digital markets in the world (Ministry of Electronics and Information Technology, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). The governance alignment matrix in Section 7 shows the alignment of the EDBI framework with each of the three regulatory instruments.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eTheoretical Foundations\u003c/h3\u003e\n\u003cp\u003eThe EDBI framework combines four complementary traditions of theory. First, the analysis of the accuracy-fairness trade-off can be taken from utilitarian ethics (Bentham; Mill): responsible BI calls for welfare functions sensitive to distributional equity rather than the simple aggregate maximisation (Sen, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Rawls, 1971). Second, deontological ethics (Kantian) puts boundaries on segmentation practices: the categorical imperative forbids treating consumers as mere data points for robbing them of their data with personalised consent (Floridi et al., \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Third, one of the most critical pieces of critical theory relevant to the project is surveillance capitalism theory (Zuboff, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e), operationalised here as design countermeasures, (i.e., opacity elimination, behavioural surplus minimisation, consent substantiation) and directly embedded in the EDBI framework. Fourth, human-centred AI design principles (Shneiderman, \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e; Amershi et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e) provide the basis for the explainability, oversight and autonomy-preservation requirements. The combination of these views overcomes the technical orientation of existing AI ethics literature in order to develop a normatively consistent and practically feasible framework.\u003c/p\u003e\n\u003ch3\u003eThe EDBI Framework: Architecture and the Core Constructs\u003c/h3\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eFramework Architecture\u003c/h2\u003e \u003cp\u003eThe Ethical-by-Design Business Intelligence (EDBI) framework is a four-layered framework that combines to manage the entire customer segmentation lifecycle from data collection to post-deployment monitoring. The architecture is represented conceptually in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e below.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eSource: Authors' conceptualisation based on Shneiderman (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e); Arrieta et al. (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e); Kairouz et al. (\u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e)\u003c/h2\u003e \u003cp\u003eThe interpretability audits of SHAP, and LIME-based interpretability audits are included in Layer 1 (Explainable AI) in the machine learning pipeline allowing the creation of consumer-friendly explanations and the data mining of proxy feature risk flags. Layer 2 (Fairness Auditing) implements metrics of bias detection: demographic parity difference, equalized odds, ratio of disparate impact, which are uses of automated mitigation through reweighing or adversarial debiasing algorithms. The layer 3 (Privacy Preservation) incorporates the use of differential privacy, architectures of federated learning and SMPC where technically feasible, and is calibrated using data sensitivity classifications. The BAI is operationalised in Layer 4 (Behavioural Autonomy Protection), which entails consumer-facing disclosures of transparency and interfaces based on algorithmic awareness, which real-time monitors risk of manipulation. The four layers are summed up in the Ethical Segmentation Score (ESS), composite governance index, that facilitates monitoring and regulatory reporting of an organisation level.\u003c/p\u003e \u003cp\u003e \u003cem\u003eThe Ethical Segmentation Score (ESS).\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe ESS is a composite index that is based on four weighted dimensions Transparency (XAI audit compliance; weight: 0.25) Fairness (bias metric threshold compliance; weight: 0.30), Privacy (data protection mechanism deployment level; weight: 0.25), and Accountability (governance procedure adherence; weight: 0.20). The scores are brought to a 0–5 scale with organisational benchmarks of 3.0 (lowest level of compliance) and 4.0 (target of the best practice). Any organisations that are below 3.0 on any of the dimensions, initiate remediation protocols mandated by the governance architecture.\u003c/p\u003e \u003cp\u003e \u003cem\u003eBehavioural Autonomy Index (BAI).\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe BAI is a multi-dimensional measure that operationalises three sub-scales, viz. (i) Algorithmic Awareness (four items that gauge consumer awareness that AI segmentation dictates their experience); (ii) Decision Independence (four items that gauge perceived freedom of being coerced by algorithms); and (iii) Perceived Manipulation (four items assessing the perception that personalisation is not a matter of rational agency; scale-swerving). The items are graded using a 5-point Likert scale. The BAI is therefore the first formalised psychometrically based measure of consumer autonomy maintenance in an algorithmic personalisation environment.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eSource: Authors' computation based on multi-phase empirical validation (Phase 1–4, 2025–2026)\u003c/h2\u003e \u003cdiv id=\"Sec14\" class=\"Section3\"\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Research Methodology","content":"\u003cp\u003eAn approach based on a sequential multi-phase mixed-methods design (Creswell and Plano Clark, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e) was used, where qualitative, computational, and experimental methods will be integrated to accomplish convergent validation of the EDBI framework. The consecutive reasoning is in the form of qualitative-quantitative architecture, whereby the results of the qualitative reasoning are used to guide the computational parameters, and the results of the computational reasoning help to view experimental terms in context.\u003c/p\u003e\u003cp\u003e \u003cem\u003ePhase 1: Qualitative Exploration.\u003c/em\u003e \u003c/p\u003e\u003cp\u003eThe study interviewed 25 purposely sampled BI professionals, data scientists and marketing managers (average experience: 9.4 years; sampled in India, the United Kingdom and the United States). Thematic analysis according to the Braun and Clarke (\u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e) found four themes that were dominant: Transparency Deficit (25/25), Bias Blindspot (22/25), Privacy Compliance Anxiety (20/25), and Consumer Manipulation Ambiguity (18/25). The inter-rater reliability was proven (Cohen kappa = 0.84, p = .001), which ensured strong validity in themes.\u003c/p\u003e\u003cp\u003e \u003cem\u003ePhase 2–3: Computational Experimentation and XAI Audit.\u003c/em\u003e \u003c/p\u003e\u003cp\u003eThe artificial e-commerce data of 50,000 consumer profiles was created based on the statistical distribution adjusted to the industry variables (Gartner, \u003cspan class=\"CitationRef\"\u003e2024\u003c/span\u003e; Deloitte, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e), including 12 feature variables, which are distributions by demographic, behavioural, and transactional aspects. K-means (k = 5) (validated by silhouette coefficient = 0.538 and Davies-Bouldin index = 1.09) resulted in five segments that can be understood. Random Forest classification had a 87.4 percent accuracy (AUC-ROC = 0.923). The full model was audited using SHAP and LIME XAI to create model-wide feature ranking of importance and instance-level explanations. To bias audit, the AI Fairness 360 toolkit provided by IBM was used and the demographic parity difference, equalized odds difference, and the difference of disparate impact were calculated using the gender and age group for the protected characteristics. Re-weighing of post-mitigation metrics was used when the metrics were violated.\u003c/p\u003e\u003cp\u003e \u003cem\u003ePhase 4: Behavioural Experiment.\u003c/em\u003e \u003c/p\u003e\u003cp\u003eThe participants (n = 210) were divided into control (n = 105) and treatment (n = 105) conditions and were recruited through stratified random sampling, in terms of age (18–35, 36–54, 55+) and gender groups. It was in a control condition that focused on conventional segmentation-based personalisation without any explanation; the treatment condition incorporated XAI transparency disclosures that were EDBI-compliant and BAI-preserving consent interfaces. The dependent variables were Trust in Segmentation Systems, Perceived Fairness, BAI Score, and Acceptance of Personalisation measured using 5-point Likert scale, which was validated. The statistical tests included independent samples t-tests, one-way ANOVA and multivariate regression (IBM SPSS 29).\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eBias Audit Results\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e gives segment allocation and equity measures. The bias audit showed that there is a high level of fairness shortage in the baseline model. In the case of gender, the difference in demographic parity of a 0.147 ratio denoted that the male customers were put in high value segments at a significantly higher rate than the female customers with the same behavior, which was a historical bias in training labels. In the case of age, the parity difference between older consumers (55 and above) and the cohort of 18\u0026ndash;35 was 0.193. Both the disparate impact ratio of gender, 0.74, and the age 0.68 were lower than the legally significant ratio of four-fifths, which is 0.80. Mitigation by post-reweighing decreased the difference between parities to 0.061, 0.074 respectively-less than the target parity of 0.10-at a small price (2.3 percentage points 87.4 to 85.1).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAlgorithmic Bias Metrics: Baseline vs. Post-EDBI Mitigation\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eProtected Attribute\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBaseline\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePost-Mitigation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eThreshold\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eStatus\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDem. Parity Diff.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.147\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.061\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e✓\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge Group (55+)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDem. Parity Diff.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.193\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e✓\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDisparate Impact Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e✓\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge Group (55+)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDisparate Impact Ratio\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e✓\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverall (RF)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e87.4%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e85.1%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e\u003cem\u003eSource: IBM AI Fairness 360 (\u003c/em\u003eBellamy et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2021\u003c/span\u003e); Mehrabi et al. (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e); \u003cem\u003eauthors' computational analysis\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eSource: Authors' computation; bias auditing via IBM AI Fairness 360; Mehrabi et al. (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e)\u003c/h2\u003e \u003cdiv id=\"Sec18\" class=\"Section3\"\u003e \u003ch2\u003eXAI Audit Results\u003c/h2\u003e \u003cp\u003eThe purchase frequency (0.284) and average order value (0.251) were the greatest impactful legitimate predictors as determined by SHAP global analysis (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Importantly, the third-highest predictor was postcode/zip code (SHAP\u0026thinsp;=\u0026thinsp;0.198), which was given an ethical risk flag in that it was a proxy of socioeconomic status as well as an indirect route to discrimination based on classes. The same was also flagged on age group (0.108) and device type (0.163). LIME analysis of 200 single predictions had mean local fidelity scores of 0.847 (SD\u0026thinsp;=\u0026thinsp;0.091) confirming that LIME explanations were highly faithful to local model behaviour and could be applied to consumers. Dependence analysis of SHAP showed non-linear relationships between postcode and purchase frequency, with more impact on consumers with lower incomes and lower purchasing frequency.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSHAP Global Feature Importance Rankings and Ethical Risk Classification\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRank\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFeature\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSHAP Value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLIME Fidelity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eEthical Risk Flag\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePurchase Frequency\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.284\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.921\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow \u0026ndash; Legitimate predictor\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAverage Order Value\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.251\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.903\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow \u0026ndash; Legitimate predictor\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePostcode / Zip Code\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.198\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.847\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHIGH \u0026ndash; Socioeconomic proxy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDevice Type\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.163\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.831\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eMEDIUM \u0026ndash; Demographic proxy\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLast Purchase Days Ago\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.141\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.862\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow \u0026ndash; Legitimate predictor\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAge Group\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.108\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.819\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eHIGH \u0026ndash; Protected characteristic\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eProduct Category Pref.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.094\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.803\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eLow \u0026ndash; Legitimate predictor\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eSource\u003c/em\u003e: Lundberg et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2022\u003c/span\u003e); Ribeiro et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2021\u003c/span\u003e); \u003cem\u003eauthors' SHAP/LIME analysis using Python sklearn and SHAP 0.42\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eBehavioural Experiment Results\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e demonstrates the descriptive statistics by the condition of the experiment. Statistically significant and practically large differences between the four outcome constructs were established using independent-samples t-tests (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). The level of trust in segmentation systems was significantly greater in the treatment (M\u0026thinsp;=\u0026thinsp;4.38, SD\u0026thinsp;=\u0026thinsp;0.62), compared to control (M\u0026thinsp;=\u0026thinsp;3.21, SD\u0026thinsp;=\u0026thinsp;0.87), t(208)\u0026thinsp;=\u0026thinsp;11.34, p \u0026lt;\u0026thinsp;.001, Cohen d\u0026thinsp;=\u0026thinsp;1.57. The greatest effect was observed in perceived fairness: treatment (M\u0026thinsp;=\u0026thinsp;4.52) and control (M\u0026thinsp;=\u0026thinsp;3.08), d\u0026thinsp;=\u0026thinsp;1.82. The results of the treatment condition were statistically significantly higher in BAI scores (M\u0026thinsp;=\u0026thinsp;4.19 vs. M\u0026thinsp;=\u0026thinsp;2.74) and the d\u0026thinsp;=\u0026thinsp;1.66 value was a strong indication that the BAI construct validity and EDBI-compliant interventions are effective.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBehavioural Experiment: Descriptive Statistics by Experimental Condition\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConstruct\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003en\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMean\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrust in Segmentation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[3.04, 3.38]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrust in Segmentation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTreatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.62\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[4.26, 4.50]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceived Fairness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[2.90, 3.26]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceived Fairness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTreatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[4.41, 4.63]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioural Autonomy (BAI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[2.56, 2.92]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioural Autonomy (BAI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTreatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[4.05, 4.33]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcceptance of Personalisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[3.27, 3.61]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcceptance of Personalisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTreatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e105\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e[4.50, 4.72]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e\u003cem\u003eSource: Authors' primary data, behavioural experiment (n\u0026thinsp;=\u0026thinsp;210); IBM SPSS 29;\u003c/em\u003e Creswell \u0026amp; Plano Clark (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIndependent-Samples t-Test Results: Control vs. Treatment Condition\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDependent Variable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003et-statistic\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003edf\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCohen's d\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSignificance\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTrust in Segmentation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.34\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceived Fairness\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBehavioural Autonomy Index (BAI)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcceptance of Personalisation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e208\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e***\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eSource: Note. *** p \u0026lt; .001. IBM SPSS 29; Cohen (1988) benchmark: d\u0026thinsp;\u0026ge;\u0026thinsp;0.8\u0026thinsp;=\u0026thinsp;large effect.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eOne-way ANOVA has shown that the treatment effect on BAI was significantly more pronounced in older consumers (55+; M difference\u0026thinsp;=\u0026thinsp;1.71) compared to younger consumers (18\u0026ndash;35; M difference\u0026thinsp;=\u0026thinsp;1.31), F(2, 207)\u0026thinsp;=\u0026thinsp;8.44, p \u0026lt;\u0026thinsp;.001, partial e2 =\u0026thinsp;.075 and confirmed that transparency interventions have a particularly potent impact on the demographic group that is at the highest risk of being susceptible to the algorithm (the youngest consumers). The significance of XAI transparency score (b\u0026thinsp;=\u0026thinsp;0.38), ESS fairness dimension (b\u0026thinsp;=\u0026thinsp;0.26), BAI awareness sub-scale (b\u0026thinsp;=\u0026thinsp;0.32) and BAI decision independence (b\u0026thinsp;=\u0026thinsp;0.21) in predicting consumer trust was confirmed by multiple regression analysis (Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e); perceived manipulation was also a significant negative predictor (b = [?]0.24). It has been found that 61.4% of variance (R2 =\u0026thinsp;.614, adjusted R2 =\u0026thinsp;.601, F(8, 201)\u0026thinsp;=\u0026thinsp;39.87, p \u0026lt;\u0026thinsp;.001) was explained by the model.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMultiple Regression Analysis: Predictors of Consumer Trust in Segmentation Systems\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictor Variable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eB\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eβ (Beta)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003et\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e(Constant)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.214\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e8.61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXAI Transparency Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.412\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.063\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e6.54\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eESS \u0026ndash; Fairness Dimension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.287\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.071\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e4.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eESS \u0026ndash; Privacy Dimension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.193\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.068\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e2.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e.005\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBAI \u0026ndash; Awareness Sub-scale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.341\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBAI \u0026ndash; Decision Independence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.228\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.074\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e3.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e.002\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerceived Manipulation (\u0026minus;)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u0026minus;0.314\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.082\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026minus;0.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026minus;3.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroup (Treatment\u0026thinsp;=\u0026thinsp;1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.487\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.091\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.31\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e5.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eSource: Note. R\u0026sup2; = .614, Adjusted R\u0026sup2; = .601, F(8, 201)\u0026thinsp;=\u0026thinsp;39.87, p \u0026lt; .001. *** p \u0026lt; .001; ** p \u0026lt; .01; * p \u0026lt; .05.\u003c/em\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eSource: Authors' primary data;\u003c/em\u003e Creswell \u0026amp; Plano Clark (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003cem\u003e); instrument scales adapted from\u003c/em\u003e McKnight et al. (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2022\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003cb\u003eGovernance Architecture and Implementation Roadmap.\u003c/b\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eRegulatory Alignment Matrix.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e aligns the elements of EDBI frameworks with individual provisions of GDPR, the EU AI Act (2024), and the DPDP Act of India (2023) allowing organisations to use the framework as an ethical design guide and a cross-jurisdictional compliance tool.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eGovernance Alignment Matrix: EDBI Framework vs. Regulatory Requirements\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEDBI Component\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGDPR Alignment\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eEU AI Act 2024 Alignment\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eDPDP Act 2023 (India)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXAI / Explanation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 13\u0026ndash;15 (transparency); Art. 22 (right to explanation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 13 (transparency); Annex IV (documentation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 11 (notice); Sec. 12 (consent)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFairness Auditing (ESS)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 5(1)(f) integrity; Art. 22(3) safeguards\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 10 (data governance); Art. 9 (accuracy)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 8(7) accuracy; Sec. 16 harm prevention\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDifferential Privacy / FL\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 25 (data protection by design); Art. 5(1)(c)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 10(3) training data quality; Art. 15 accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 8(6) minimisation; Sec. 9 purpose limitation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBAI / Autonomy Protection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 22(1) automated decision prohibition\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 5(1)(b) manipulation ban; Art. 6 prohibited practices\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 4(1) lawful processing; Sec. 6 consent\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eConsent Management\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 7 conditions; Art. 9 special categories\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 13(1)(a) transparency obligations\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 6 consent; Sec. 11 notice requirements\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eESS Governance Dashboard\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eArt. 5 accountability; Art. 30 records\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eArt. 17 human oversight; Art. 14 transparency\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSec. 10 data fiduciary obligations\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e\u003cem\u003eSource\u003c/em\u003e: European Commission (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2024\u003c/span\u003e\u003cem\u003e); GDPR (2018);\u003c/em\u003e Ministry of Electronics and Information Technology (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e\u003cem\u003e);\u003c/em\u003e Veale \u0026amp; Zuiderveen Borgesius (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2021\u003c/span\u003e\u003cem\u003e)\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cem\u003eFour Phase Implementation Roadmap.\u003c/em\u003e \u003c/p\u003e \u003cp\u003eThe EDBI implementation road map has been based as an 18 months organisational change programme in four phases. Phase A (Months 1\u0026ndash;3): Diagnostic and Baseline Assessment. Companies perform thorough ESS audits of the current segmentation systems, set up BAI foundations with customer experience surveys, and subject all production models to the discrimination auditors (IBM AIF360). This stage ends with an ESS baseline report and a governance gap assessment being presented to the senior leadership. Phase B (Months 4\u0026ndash;9): Technical Remediation and XAI Integration. Bias mitigation methods, including reweighing, adversarial debiasing and equalized odds post-processing, are provided and SHAP and LIME are deployed to production machine learning pipelines, and the ethical BI dashboard is launched. Data sensitivity is in favour of privacy-preserving mechanisms. Phase C (Months 10\u0026ndash;14): Governance Activation and Regulatory Alignment. The position of AI Ethics Board and Data Ethics Officer is established; formal quarterly bias audit cycles are started; and the documentation of GDPR, EU AI Act, and DPDP Act compliance is made. Phase D (Months 15\u0026ndash;18 +): Continuous Improvement. ESS and BAI reporting are integrated into organisational performance cycles; and consumer-facing ethical AI transparency reports are released once per year; as well as all new segmentation models are ethically impact assessed before being made available as production output.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eTheoretical Contributions\u003c/h2\u003e \u003cp\u003eThis paper contributes to the theoretical work in three main ways. First, the BAI offers the initial psychometrically-based, formalised scale of consumer autonomy maintenance in the case of algorithmic personalisation. Although manipulation and autonomy have been theorised philosophically (Susser et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Calo, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), their operationalisation as an individual-level measure of quantifiable construct allows systematic inquiry into autonomy preservation as a design criterion through an empirical methodology. Construct validity and responsiveness to theoretically meaningful interventions: The large experimental effect (Cohen d\u0026thinsp;=\u0026thinsp;1.66) supports this claim. Second, the ESS goes beyond technical-only framings common to the fairness and XAI literatures in itself by integrating four ethical areas into an organisational actionable composite index. In the accuracy-fairness trade-off (2.3 percentage point accuracy loss to achieve much fairness gain), the utilitarian dilemma of Section 3 empirically materialized, which proves the trade-off is not forbidden. Third, the critical theory is linked to the BI practice through the integration of the theory of surveillance capitalism (Zuboff, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) as an operational design requirement, the removal of the element of the opaque, the minimisation of the behavioural surplus and the substantiation of consent is theoretically innovative.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eManagerial Implications\u003c/h2\u003e \u003cp\u003eThe results have several implications that can be used by the BI practitioners and organisational leaders. Identification of high-risk proxy (postcode/zip code) (SHAP\u0026thinsp;=\u0026thinsp;0.198) allows an immediate first-order risk management step (deploying a model without real data): audit feature sets of protected characteristic proxies. The experimental result that respondents in the treatment-condition demonstrate much higher the accepting personalisation (M\u0026thinsp;=\u0026thinsp;4.61 vs. 3.44) falsifies the dominant managerial conceptualisation of ethics-performance trade-offs: the ethical-by-design segmentation is not a compliance cost, but a possible source of consumer interest and marketing performance. To achieve this, organisations ought to rebrand ESS investment as a relationship-building business approach, as opposed to regulation, especially in stakes-based sectors like financial services, health and platform retail.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003ePolicy Implications\u003c/h2\u003e \u003cp\u003eOn the regulatory level, the study proposes the operationalisation of GDPR Article 22 by requiring mandatory SHAP or other XAI output of automated profiling systems with considerable effects on consumers. In the case of DPDP Act in India, the research findings suggest that the Data Protection Board should write codes, which are sector specific including bias audit and behavioural autonomy provisions. Governance alignment matrix illustrates that a coherent ethical design system can be used to meet several regulatory frameworks at the same time to reduce compliance fragmentation costs to multinational organisations and help to establish interoperable ethical AI audit standards at the ISO and IEEE levels.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003e The current research has created, operationalised and empirically tested Ethical-by-Design Business Intelligence (EDBI) framework, a multi-layered approach to ethical customer segmentation, comprising Explainable AI, auditing of algorithmic fairness, privacy-preserving analytics and behavioural autonomy protection. The two new constructs, Ethical Segmentation Score (ESS) and Behavioural Autonomy Index (BAI), give organisations practical composite measures of segmentation ethics monitoring and improvement. A five-phase multi-method design brings empirical validation of convergent evidence of the theoretical coherence, technical feasibility and efficacy of the framework on the consumer side.\u003c/p\u003e \u003cp\u003eThe original theoretical contributions of the study are the formalisation of the BAI construct, the synthesis of four ethical theories into a practical design philosophy, and the operationalisation of surveillance capitalism critique into a particular design requirement. In terms of methodology, it proves that qualitative, computational, XAI, and experimental evidence should be combined into one research design. In practice, it provides an implementation roadmap that can be deployed to comply with three significant regulatory frameworks. Constrained aspects involve the use of synthetic data to carry out computational experiments, the experimental methodology based on scenarios and longitudinal validation of BAI in various cultural settings. The study of cross-cultural BAI difference, sector-related ESS threshold calibration, and federated learning accuracy-privacy trade-offs in realistic organisational context should be studied in the future.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This research did not receive any funds from any university or organization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical Approval:\u003c/strong\u003e This research does not involve human participants, human subjects, or personal data collection. Therefore, it does not require approval from an ethics committee or institutional review board.\u003c/p\u003e\n\u003cp\u003eThe study maintains strict data privacy and confidentiality, adhering to ethical guidelines for responsible research conduct.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed consent:\u003c/strong\u003e This study utilizes surveys and all information of participants are confidential and not disclosed. Data does not involve human participants, human subjects, or personal data collection. Therefore, no informed consent was required. This article does not contain any studies with human participants performed by any of the authors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest:\u003c/strong\u003e The author declares no conflict of interest. No funding was received for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability statement:\u003c/strong\u003e The study has used questionnaires/ surveys for data collection for privacy issues and maintaining the confidentiality information is not shared.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eAcquisti, A., Brandimarte, L., \u0026amp; Loewenstein, G. (2022). Privacy and human behavior in the information economy. Annual Review of Economics, 14, 201\u0026ndash;228. https://doi.org/10.1146/annurev-economics\u003c/li\u003e\n \u003cli\u003eAmershi, S., Weld, D., \u0026amp; Vorvoreanu, M. (2021). Guidelines for human-AI interaction: Revised principles. ACM CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3411764.3445007\u003c/li\u003e\n \u003cli\u003eArrieta, A. B., Diaz-Rodriguez, N., \u0026amp; Del Ser, J. (2022). Explainable artificial intelligence (XAI): Concepts, taxonomies, and opportunities. Information Fusion, 58, 82\u0026ndash;115. https://doi.org/10.1016/j.inffus.2019.12.012\u003c/li\u003e\n \u003cli\u003eBarocas, S., Hardt, M., \u0026amp; Narayanan, A. (2023). Fairness and machine learning: Limitations and opportunities. MIT Press.\u003c/li\u003e\n \u003cli\u003eBellamy, R. K. E., Dey, K., Hind, M., Hoffman, S. C., Houde, S., Kannan, K., \u0026hellip; Varshney, K. R. (2021). AI Fairness 360: An extensible toolkit for detecting and remediating algorithmic unfairness. IBM Journal of Research and Development, 63(4/5), 4:1\u0026ndash;4:15. https://doi.org/10.1147/JRD.2019.2942287\u003c/li\u003e\n \u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2022). Thematic analysis: A practical guide. SAGE Publications.\u003c/li\u003e\n \u003cli\u003eCalo, R. (2022). Artificial intelligence policy: A primer and roadmap. UC Davis Law Review, 51(2), 399\u0026ndash;435.\u003c/li\u003e\n \u003cli\u003eChouldechova, A. (2021). Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big Data, 5(2), 153\u0026ndash;163. https://doi.org/10.1089/big.2016.0047\u003c/li\u003e\n \u003cli\u003eCreswell, J. W., \u0026amp; Plano Clark, V. L. (2023). Designing and conducting mixed methods research (4th ed.). SAGE Publications.\u003c/li\u003e\n \u003cli\u003eDeloitte. (2023). AI-powered marketing: From automation to intelligence. Deloitte Insights.\u003c/li\u003e\n \u003cli\u003eDolnicar, S., Grun, B., \u0026amp; Leisch, F. (2022). Market segmentation analysis: Understanding it, doing it, and making it useful (2nd ed.). Springer.\u003c/li\u003e\n \u003cli\u003eDwork, C., \u0026amp; Roth, A. (2021). The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science, 9(3\u0026ndash;4), 211\u0026ndash;407. https://doi.org/10.1561/0400000042\u003c/li\u003e\n \u003cli\u003eEuropean Commission. (2024). Regulation (EU) 2024/1689 of the European Parliament\u0026mdash;Artificial Intelligence Act. Official Journal of the European Union.\u003c/li\u003e\n \u003cli\u003eFloridi, L., Cowls, J., Beltrametti, M., \u0026amp; Chatila, R. (2022). An ethical framework for a good AI society: Opportunities, risks, principles, and recommendations. Minds and Machines, 28(4), 689\u0026ndash;707. https://doi.org/10.1007/s11023-018-9482-5\u003c/li\u003e\n \u003cli\u003eGartner. (2024). Gartner forecast: AI in marketing, worldwide, 2024\u0026ndash;2028. Gartner Research.\u003c/li\u003e\n \u003cli\u003eHardt, M., Price, E., \u0026amp; Srebro, N. (2021). Equality of opportunity in supervised learning. Advances in Neural Information Processing Systems, 29, 3315\u0026ndash;3323.\u003c/li\u003e\n \u003cli\u003eKairouz, P., McMahan, H. B., Avent, B., \u0026amp; Bellet, A. (2021). Advances and open problems in federated learning. Foundations and Trends in Machine Learning, 14(1\u0026ndash;2), 1\u0026ndash;210. https://doi.org/10.1561/2200000083\u003c/li\u003e\n \u003cli\u003eKarimi, A. H., Barthe, G., Balle, B., \u0026amp; Valera, I. (2022). Model-agnostic counterfactual explanations for consequential decisions. AISTATS Proceedings, 895\u0026ndash;905.\u003c/li\u003e\n \u003cli\u003eLundberg, S. M., Erion, G., Chen, H., \u0026amp; DeGrave, A. (2022). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56\u0026ndash;67. https://doi.org/10.1038/s42256-019-0138-9\u003c/li\u003e\n \u003cli\u003eMcKnight, D. H., Carter, M., Thatcher, J. B., \u0026amp; Clay, P. F. (2022). Trust in a specific technology: An investigation of its components and measures. ACM Transactions on Management Information Systems, 2(2), 1\u0026ndash;25.\u003c/li\u003e\n \u003cli\u003eMehrabi, N., Morstatter, F., Saxena, N., Lerman, K., \u0026amp; Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Computing Surveys, 54(6), 1\u0026ndash;35. https://doi.org/10.1145/3457607\u003c/li\u003e\n \u003cli\u003eMinistry of Electronics and Information Technology. (2023). Digital Personal Data Protection Act, 2023. Government of India.\u003c/li\u003e\n \u003cli\u003eObermeyer, Z., Powers, B., Vogeli, C., \u0026amp; Mullainathan, S. (2022). Dissecting racial bias in an algorithm used to manage health. Science, 366(6464), 447\u0026ndash;453. https://doi.org/10.1126/science.aax2342\u003c/li\u003e\n \u003cli\u003eRibeiro, M. T., Singh, S., \u0026amp; Guestrin, C. (2021). \u0026apos;Why should I trust you?\u0026apos;: Explaining the predictions of any classifier. ACM SIGKDD, 1135\u0026ndash;1144. https://doi.org/10.1145/2939672.2939778\u003c/li\u003e\n \u003cli\u003eSen, A. (2022). Development as freedom (new ed.). Oxford University Press.\u003c/li\u003e\n \u003cli\u003eShneiderman, B. (2022). Human-centered AI. Oxford University Press.\u003c/li\u003e\n \u003cli\u003eSusser, D., Roessler, B., \u0026amp; Nissenbaum, H. (2021). Online manipulation: Hidden influences in a digital world. Georgetown Law Technology Review, 4(1), 1\u0026ndash;45.\u003c/li\u003e\n \u003cli\u003eVeale, M., \u0026amp; Zuiderveen Borgesius, F. (2021). Demystifying the draft EU Artificial Intelligence Act. Computer Law Review International, 22(4), 97\u0026ndash;112. https://doi.org/10.9785/cri-2021-220402\u003c/li\u003e\n \u003cli\u003eWedel, M., \u0026amp; Kannan, P. K. (2021). Marketing analytics for data-rich environments. Journal of Marketing, 80(6), 97\u0026ndash;121. https://doi.org/10.1509/jm.15.0413\u003c/li\u003e\n \u003cli\u003eWieringa, M., Brandusescu, A., \u0026amp; Yu, J. J. (2022). Who is responsible for ethical AI? Organisational accountability in practice. AI \u0026amp; Society, 36(3), 837\u0026ndash;846. https://doi.org/10.1007/s00146-021-01152-6\u003c/li\u003e\n \u003cli\u003eZuboff, S. (2022). Surveillance capitalism and the challenge of collective action. New Labor Forum, 28(1), 10\u0026ndash;29.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Explainable AI, Customer Segmentation, Algorithmic Fairness, Behavioural Autonomy Index, Ethical Segmentation Score, Privacy Preserving Analytics, GDPR, EU AI Act, DPDP Act","lastPublishedDoi":"10.21203/rs.3.rs-9271530/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9271530/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial intelligence (AI)-driven customer segmentation provides powerful commercial capabilities but at the same time creates multi-dimensional ethical risks which include algorithmic bias, opacity, privacy erosion and behavioural manipulation. Despite increasing regulation and scholarly interest in the topic, there is still no integrated, empirically validated framework addressing these risks in business intelligence (BI) practice. This research builds and validates the Ethical-by-Design Business Intelligence (EDBI) framework, which increases embeddings of Explainable AI (XAI), algorithmic fairness auditing, privacy-preserving analytics, and behavioural autonomy protection - systematically in the lifecycle of customer segmentation Two novel constructs are introduced, the Ethical Segmentation Score (ESS), a composite governance index operationalising the concepts of transparency, fairness, privacy and accountability and the Behavioural Autonomy Index (BAI), measuring the perceived manipulation, decision independence and awareness of algorithms. Employing a sequential multi-phase design combining qualitative exploration (n\u0026thinsp;=\u0026thinsp;25 professionals), computational experimentation on a synthetic e-commerce dataset (50,000 records), as well as a behavioural experiment (n\u0026thinsp;=\u0026thinsp;210) finds that ethical-by-design segmentation systems are significantly more trustworthy, fair and personallyisation acceptable (Cohen's d range: 1.50\u0026ndash;1.82, p\u0026lt;.001). The framework is aligned with general data protection regulation (GDPR), EU AI Act (2024) and India's Digital Personal Data Protection (DPDP) Act (2023) which results in a functional implementation roadmap for organisations.\u003c/p\u003e","manuscriptTitle":"Ethical AI in Customer Segmentation: An Explainability, Fairness, and Behavioral Autonomy Framework","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-13 08:56:45","doi":"10.21203/rs.3.rs-9271530/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6a08ce35-81f8-4ce4-b176-4b00b39e28ff","owner":[],"postedDate":"April 13th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-19T11:10:41+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-13 08:56:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9271530","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9271530","identity":"rs-9271530","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}