Customer Experience in AI-Driven E-Commerce: An Empirical Model of Drivers and Strategic Outcomes

Abstract

As AI-powered e-commerce platforms grow more capable of predicting customer wants, a critical question remains unexplored: what makes customers perceive these experiences positively? The rapid integration of artificial intelligence (AI) into e-commerce platforms is reshaping how customers search for, evaluate, and experience digital services. However, empirical research has not kept pace with clarifying which platform-level factors most effectively shape customer experience (CX) in AI-driven environments. This study validated the Trust, Autonomy, Personalization, and Customer Engagement (TAPE) framework as a comprehensive set of CX drivers in intelligent commerce. Using survey data from 400 active e-commerce users, we employed a multi-stage approach combining exploratory factor analysis, confirmatory factor analysis, and covariance-based structural equation modeling (SEM) with bootstrapped mediation testing. All four TAPE drivers demonstrated significant positive reflective associations with CX, with personalization and engagement emerging as the strongest contributors. CX was strongly associated with customer satisfaction, loyalty, and brand equity, and mediated the effects of all four dimensions on these strategic outcomes, with model comparison evidence supporting full mediation. The study contributes theoretically by integrating and empirically validating four established CX dimensions within the AI-enabled e-commerce context, and by demonstrating the central mediating role of CX in converting intelligent platform features into user-perceived strategic value. Managerially, the TAPE framework provides actionable guidance for designing transparent, adaptive, and engaging AI-driven customer journeys that enhance both experience quality and long-term brand outcomes.

1. Introduction

Customer experience (CX) has evolved significantly over the past century, reflecting major shifts in creating, delivering, and interpreting value. During the early industrial era, firms competed primarily on the functional superiority of their products such as durability, efficiency, and reliability. As products became increasingly commoditized in the late 1990s, firms focused on service quality and customer care. This shift marked the transition from product-centric to service-dominant logic, in which value is understood as being co-created through interactions, relationships, and contextual factors rather than being embedded solely within goods [1]. Foundational theories such as expectancy disconfirmation and SERVQUAL have reinforced this shift by offering structured ways to compare perceived service performance to customer expectations [2,3]. Maximizing customer satisfaction has been a central managerial objective for decades.

By the late 20th century, scholars began to question whether satisfaction alone captured the richness of consumption. Holbrook and Hirschman argue that consumers are emotional and symbolic beings seeking meaning, pleasure, and experiential value, not simply rational problem solvers [4]. Pine and Gilmore expanded this perspective by framing “experience” as a distinct economic offering, urging firms to stage memorable encounters capable of creating lasting impressions [5,6]. This experiential turn reshaped marketing thought and set the stage for a deeper exploration of how customers perceive and evaluate their interactions with brands.

The digital revolution has significantly accelerated this change. As the Internet, mobile devices, and social platforms proliferated, the customer journey evolved into a dynamic, always-on ecosystem characterized by nonlinear interactions across multiple touchpoints, including apps, websites, social content, influencer reviews, and user-generated communities [7,8]. In this environment, e-commerce has matured into intelligent commerce, in which artificial intelligence (AI), machine learning (ML), and big data personalize, automate, and orchestrate large portions of the customer experience [9,10,11]. Recommendation engines, behavioral prediction models, adaptive interfaces, and conversational agents now autonomously and continuously interact with customers, shaping not only the transactional process but also the emotional and cognitive dimensions of the experience. The scale of this transformation is substantial: AI underpins virtually every stage of the online shopping journey, from personalized product discovery and dynamic pricing to intelligent chatbot interactions and automated post-purchase support [12]. Consumers are no longer passive recipients of a standardized shopping environment; they are active participants in an algorithmically curated experience ecosystem in which their data, preferences, behavior, and decisions are continuously collected, interpreted, and acted upon [13]. Studying CX on any major e-commerce platform today is inherently a study of AI-mediated experience, even when respondents evaluate their overall platform experience rather than isolating specific AI-generated interactions.

Despite these advancements, research has struggled to keep up with the realities of AI-enabled commerce. Many studies continue to examine isolated components, such as trust, usability, personalization, and engagement, resulting in fragmented insights that do not fully explain how holistic CX is formed within algorithmically mediated platforms [14,15]. Persistent tensions, such as the personalization–privacy paradox, further complicate the landscape as customers navigate the trade-offs between relevance and perceived intrusiveness [16]. Simultaneously, engagement research underscores the rise of customers as active participants rather than passive recipients of digital interactions [17]. Collectively, these trends highlight a critical gap: existing CX frameworks, developed primarily for human-delivered or omnichannel environments, do not adequately capture how experiences emerge within intelligent AI-driven e-commerce ecosystems.

The evolution of digital experience research illustrates the need for updated theorization. Early studies emphasized trust and security as essential for online adoption [18], alongside user autonomy and ease of control, to reduce technological anxiety [15]. As digital platforms become more social and participatory, engagement has emerged as a determinant of emotional and behavioral involvement [17,19]. With advancements in AI and big data, personalization has become a central driver of relevance, convenience, and value [20,21], even as concerns regarding fairness, data use, and surveillance have grown [16,22]. The theoretical foundations that underpin these four constructs are equally distinct. Trust is grounded in commitment-trust theory [23], which positions it as the foundational condition for sustained consumer-firm relationships, a condition newly complicated by the opacity of algorithmic systems that consumers interact with but cannot inspect [24,25]. Autonomy draws on self-determination theory [26], which treats autonomous self-regulation as a basic psychological need whose satisfaction or frustration directly shapes the quality of any experience and empirical evidence confirms that perceived autonomy is a significant predictor of positive attitudes toward AI systems across multiple national contexts [27]. Personalization is anchored in privacy calculus theory [28], which frames consumer behavior as a cost-benefit evaluation between the value of tailored content and the risks of data disclosure and a calculus that intensifies considerably as AI systems incorporate emotional and contextual signals beyond the scope of initial data disclosure expectations [29]. Customer engagement draws on consumer engagement theory [17].

The selection of exactly four dimensions—Trust, Autonomy, Personalization, and Engagement—is grounded in a specific theoretical logic. AI-driven e-commerce introduces three qualitatively distinctive psychological dynamics that traditional CX frameworks do not adequately address. First, algorithmic opacity: the platform makes consequential decisions through processes the customer cannot inspect, creating a need to trust systems whose reasoning is invisible. Second, the surveillance-relevance trade-off: every customer interaction is continuously monitored, interpreted, and acted upon. Third, automated agency substitution: the platform curates choices and predict needs in ways that can feel either empowering or constraining. Trust addresses the first dynamic [23,30,31]. Personalization addresses the second [20,28,32]. Autonomy addresses the third [27,33]. Engagement addresses the hedonic and relational richness that AI-enabled interactivity uniquely affords [17,34]. Each dimension is grounded in an independent theoretical tradition: commitment-trust theory [23,30] for Trust; self-determination theory [26] for Autonomy; privacy calculus theory [28] for Personalization; consumer engagement theory [17] for Engagement. These selections are further corroborated by prior systematic synthesis [35] and the CEMF measurement framework [36], which confirm the practical measurability of each dimension. Compared to existing frameworks, TAM [37] explains adoption decisions but not ongoing AI-driven CX; EXQ [38] was built for human-delivered service encounters that do not translate to continuous algorithmic orchestration. Four adjacent constructs were excluded on principled grounds. Algorithmic transparency is the mechanism through which Trust is built, a sub-component rather than a co-equal dimension. Perceived risk is the psychological state that Trust mitigates. Usability is a platform hygiene threshold whose absence creates friction but whose presence alone does not generate experiential value. Service quality operates at the functional delivery layer rather than the experiential layer. Together, these four dimensions encapsulate both the psychological needs of customers—safety, agency, relevance, and emotional connection—and the machine affordances of AI-enabled platforms, offering a coherent socio-technical foundation for modern CX.

1.1. Objectives of the Study

To address this gap, this study pursues three key objectives. First, it introduces and theoretically grounds the TAPE (Trust, User Autonomy, Personalization, and Customer Engagement) framework as a model for explaining customer experience (CX) in intelligent e-commerce environments. Second, it empirically validates the four TAPE drivers through a multistage quantitative approach to assess their influence on the holistic customer experience. Third, it examines how CX mediates the relationship between these drivers and strategic outcomes (customer satisfaction, loyalty, and brand equity), thereby clarifying how intelligent platform features translate into long-term values.

1.2. The Core Digital Drivers (TAPE) and Hypotheses

Trust: The first driver, trust, is foundational to all commercial relationships but is magnified in the impersonal environment of e-commerce [39,40]. It is the willingness of a customer to be vulnerable to the actions of a vendor based on their belief in their integrity and ability [39]. Within intelligent e-commerce, this now includes algorithmic trust, the belief that AI systems powering the experience are secure, reliable, and fair [31]. Recent studies have confirmed that the perceived reliability and transparency of AI are critical for building consumer trust [41]. A high level of trust reduces perceived risk and fosters a sense of psychological safety, which is a prerequisite for a positive experience [42].

H1. 

Trust, as a key driver of customer experience, has a significant positive reflective association with overall CX in intelligent e-commerce.

User autonomy: The second driver, user autonomy, is rooted in self-determination theory [33] and refers to the feeling of control and choice that a user has over their shopping journey. It is manifested through features that enhance perceived control, such as powerful search filters and transparent control over personal data [15]. The CEMF suggests that this can be partly measured using metrics such as the customer effort score (CES) to quantify the ease of task completion [43]. Providing strong user autonomy reduces feelings of frustration and enhances a user’s sense of competence, contributing to a more positive and empowering experience [44].

H2. 

User Autonomy, as a key driver of customer experience, has a significant positive reflective association with overall CX in intelligent e-commerce.

Personalization: The third driver, personalization, is a hallmark of intelligent e-commerce and involves algorithmic tailoring of content and recommendations [21]. Effective personalization can substantially enhance CX by increasing relevance and reducing cognitive load [20]. Recent studies have confirmed that AI-driven personalization has been associated with significant effects on consumer purchase decisions. However, when personalization becomes too invasive, it can trigger the “personalization-privacy paradox,” where customers feel that their privacy is violated [16,22]. When executed thoughtfully, it is a powerful driver of positive experiences.

H3. 

Personalization, as a key driver of customer experience, has a significant positive reflective association with overall CX in intelligent e-commerce.

Customer Engagement: The final driver, customer engagement, reflects a customer’s active, voluntary, and emotionally invested participation in a brand [17]. In e-commerce, engagement is fostered by creating rich and interactive environments through user-generated content, social features, and compelling storytelling [19]. This active involvement can enrich a utilitarian shopping task with hedonic qualities, strengthening customers’ emotional bonds with the brand [45,46]. The CEMF suggests monitoring engagement through metrics such as session duration, repeat visits, and sentiment analysis of user-generated content [36].

H4. 

Customer Engagement, as a key driver of customer experience, has a significant positive reflective association with overall CX in intelligent e-commerce.

1.3. The Mediating Role of CX on Strategic Outcomes

Customer experience (CX) serves as the hypothesized pathway through which TAPE drivers are expected to exert their influence. Grounded in relationship marketing theory [23] and customer-based brand equity frameworks [47,48], this study focuses on customer satisfaction, customer loyalty, and brand equity as the three core outcomes that reflect the long-term value generated through superior experiences [49,50]. The literature consistently shows that a seamless, enjoyable, and emotionally engaging experience is associated with higher levels of customer satisfaction [51,52]. Satisfaction, in turn, is associated with stronger affective bonds and positive memories that form the basis of attitudinal customer loyalty [8,53]. Positive experiences are associated with stronger brand-related associations and perceived quality, both of which are foundational components of brand equity [48,54]. In intelligent e-commerce settings, where platform interactions are shaped by AI-driven recommendations, automated decision support, and dynamic content delivery, the quality of CX becomes even more critical in determining how customers evaluate, remember, and commit to a brand.

Trust, Autonomy, Personalization, and Engagement function as the four experiential dimensions that collectively constitute CX—not as independent antecedents that separately cause CX, but as experiential components whose integration defines the holistic quality of the customer’s platform interaction, consistent with Chahal and Dutta’s [55] second-order CX architecture. Their collective effect on strategic outcomes is realized through the experiential quality that CX represents.

Based on this reasoning, the following hypotheses are proposed:

H5. 

A positive overall customer experience has a significant positive effect on Strategic CX Outcomes (SCO), as reflected in customer satisfaction (H5a), customer loyalty (H5b), and brand equity (H5c). Strategic CX Outcomes is conceptualized as a second-order reflective construct capturing the three co-manifesting strategic consequences of superior customer experience.

H6. 

Overall customer experience mediates the relationship between digital drivers (Trust, Autonomy, Personalization, Engagement) and the customer outcomes (Satisfaction, Loyalty, and Brand Equity).

2. Materials and Methods

This section provides a comprehensive description of the methodological approach used to investigate the proposed research model. We detail the research methods, data collection strategy, characteristics of the sample, development and validation of the measurement instrument, and multistage statistical procedures used for data analysis.

2.1. Research Design and Paradigm

This study adopts a post-positivist research paradigm, which assumes that social reality can be measured and understood through the empirical testing of hypotheses [56]. This paradigm is well-suited to the present research because the goal is to test a pre-specified causal model against observed data and evaluate the degree to which the hypothesized relationships are supported by evidence. A quantitative, survey-based design was employed to empirically validate the TAPE drivers—trust, user autonomy, personalization, and customer engagement—and examine their effects on customer experience (CX) and strategic brand outcomes. A multistage analytical approach was adopted, consistent with best practices in structural equation modeling research [57].

• Exploratory Factor Analysis (EFA) was conducted on all 30 items spanning all seven constructs (the four TAPE driver constructs and the three outcome constructs) to assess the underlying factor structure and identify any items that failed to load cleanly before imposing the a priori measurement structure.
• Confirmatory Factor Analysis (CFA) was used to test and validate the pre-specified measurement model, assessing item loadings, internal consistency, convergent validity, and discriminant validity.
• Structural equation modeling (SEM) was employed to simultaneously estimate all structural paths and test the hypothesized relationships, including the mediation effects of CX between TAPE drivers and strategic outcomes.

This sequential design, moving from exploration to confirmation to structural testing, follows the two-step procedure recommended by Anderson and Gerbing [57], enabling constructs to be validated before causal paths are evaluated.

2.2. Data Collection and Sampling Procedure

The target population comprised adults aged 18 years or older with recent active experience on a major e-commerce platform. Consistent with the observation that AI-enabled features, including personalized recommendation systems, dynamic pricing engines, and conversational interfaces, are now standard infrastructure across mainstream e-commerce platforms [12,13], this study does not employ AI-feature-specific eligibility screening. Instead, following Puntoni et al. [16] and Becker and Jaakkola [14], the measurement is grounded in respondents’ subjective experiential perceptions of the platform as a whole, which is the phenomenologically appropriate level for CX research. Respondents were directed to evaluate their most recently used e-commerce platform, anchoring all assessments to a single, specific platform interaction context. A total of 400 complete and valid responses were retained following data screening, exceeding the recommended minimum sample size for covariance-based structural equation modeling (CB-SEM) analyses [57] and satisfying the criterion of at least 10 observations per estimated parameter [56,58].

Table 1. Respondent Profile.

Variable	Category	n	%
Gender	Male	201	50.2%
	Female	199	49.8%
Age	18–29 (largest group)	134	33.5%
	30–44	126	31.5%
	45–60	112	28.0%
	60 and above	28	7.0%
US Census Region	Northeast	189	47.2%
	West	114	28.5%
	South	56	14.0%
	Midwest	33	8.2%
Household Income	$75,000–$100,000 (largest)	162	40.5%
	$50,000–$74,999	76	19.0%
	$100,000–$124,999	62	15.5%
	$25,000–$49,999	36	9.0%
	$150,000 and above	39	9.8%
	$125,000–$149,999	12	3.0%
	Under $25,000	13	3.2%
Survey Device	Android	295	73.8%
	iOS	89	22.2%
	PC/Windows	8	2.0%
	Other	8	2.0%

summarizes the respondent’s profile. The respondents were broadly balanced by gender (male 50.2%, female 49.8%). The age distribution skewed toward younger adults, with 18–29-year-olds comprising the largest group (33.5%), followed by 30–44 (31.5%), 45–60 (28.0%), and 60 and above (7.0%). Geographically, respondents were concentrated in the Northeast (47.2%) and West (28.5%) census regions, with smaller proportions from the South (14.0%) and Midwest (8.2%). Household income was most heavily represented in the $75,000–$100,000 band (40.5%), followed by $50,000–$74,999 (19.0%) and $100,000–$124,999 (15.5%).

The majority completed the survey on a mobile device (Android 73.8%, iOS 22.2%), consistent with contemporary mobile-first e-commerce behavior. The panel operates on a quota-based model; therefore, conventional response rate calculation is not applicable, consistent with the standard reporting practice for commercial panel data. The following procedural controls were adopted: (1) randomized item order to mitigate sequential response bias; (2) explicit anonymity and confidentiality assurances to reduce social desirability bias. Pilot testing was conducted with 25 respondents to ensure the clarity and usability of the questionnaire. This sample size was considered adequate for pilot instrument refinement purposes only. The main data collection of 400 responses provided the sole basis for all subsequent factor analysis and structural equation modeling.

2.3. Measurement Instrument Development

The survey instrument was developed using established measurement scales adapted from previous studies to ensure content validity. Items for TAPE drivers (Trust, User Autonomy, Personalization, Customer Engagement), satisfaction, loyalty, and brand equity were sourced from validated studies and adapted to fit the AI-enabled e-commerce context. All items were assessed using a multi-item Likert scale, and minor wording adjustments were made to ensure clarity and contextual relevance for the respondents.

Theoretical grounding: Each construct was based on scales established from prior research to ensure strong conceptual foundations. Trust items were adapted from research works emphasizing ability, integrity, and benevolence, and extended with recent studies on algorithmic transparency and data protection [39,59]. Autonomy items are grounded in self-determination theory [33] and refined with contemporary work on algorithmic influence in digital commerce [32,33]. Personalization items were based on recommender systems research and were updated with insights into AI-driven adaptivity [21]. Engagement items reflect multidimensional perspectives of cognitive, emotional, and behavioral involvement, with refinements for interactive digital environments [17,34]. Finally, CX and brand outcomes are grounded in seminal models of satisfaction, brand equity, and loyalty, complemented by recent validation in e-commerce contexts [2,47,49,53].

Contextual adaptation: Based on established measures, the items were rephrased and contextualized to reflect the realities of AI-driven e-commerce. The trust items referenced the secure handling of personal data, algorithmic fairness, and recommendation transparency. Autonomy items capture perceived control over settings, personalization features, and ability to override automated decisions. Personalization items address real-time adaptivity, contextual offers, and dynamic tailoring of product displays. Engagement items included digital interactivity, gamification, and conversational AI as components of meaningful involvement. The CX and outcome measures were similarly reframed to reflect integrated digital journeys across multiple touchpoints, ensuring that the instrument captured both functional and experiential dimensions of intelligent e-commerce.

Construct coverage: The selection of Trust, User Autonomy, Personalization, and Customer Engagement as the four core TAPE dimensions has been independently validated in empirical AI-CX research [17,20,23,26,27,28,39,60].

Pilot testing: The questionnaire was pilot-tested with 25 respondents to ensure clarity, usability, and contextual relevance. The feedback indicated that the items were easy to understand and appropriate for evaluating e-shopping experiences. To enhance content validity and reliability, we adapted existing validated scales from the literature for each construct.

Table 2. Survey Instrument Construct and Item Details.

Construct	Definition	Sample Item	Source(s)
Trust	A customer’s belief in a platform’s integrity (honesty and reliability), benevolence (goodwill toward the customer), and competence (ability to fulfil its role effectively).	“I believe this platform acts in the best interests of its customers”	[23,39,40]
User Autonomy	The degree to which a customer feels self-directed, free to make independent, volitional choices and exercise meaningful control over their interactions, preferences, and personal data on the platform, without undue algorithmic influence.	“I feel free to make my own choices while shopping on this platform.”	[21,26,27,33]
Personalization	The perceived extent to which a platform tailors its content, product recommendations, and communications to accurately reflect the individual customer’s preferences, history, and behavior	“Personalization on this platform makes my shopping experience more efficient”	[20,21,28,29]
Customer Engagement	The degree of cognitive, emotional, and behavioral investment a customer makes when interacting with a platform, reflecting active participation and value co-creation beyond mere transactional exchange	“Interacting with this platform excites and interests me”	[17,45,60]
Satisfaction	The overall evaluative judgement of whether a platform’s performance has met, exceeded, or fallen short of the customer’s prior expectations	“Overall, I am satisfied with my experience on this platform”	[2,51,52]
Loyalty	A deeply held attitudinal and behavioral commitment to consistently repatronise a preferred platform in the future, accompanied by a willingness to recommend it to others	“I am willing to recommend this platform to others”	[15,18,53]
Brand Equity	The added value a customer attributes to a brand encompassing awareness, perceived quality, brand associations, and brand attachment relative to an unbranded equivalent	“I associate positive qualities with this platform’s brand”	[47,48,49,54]

summarizes each construct, its definition, sample measurement items, and original source(s). All constructs in the research model were operationalized using multi-item reflective scales measured on a 5-point Likert scale anchored from 1 (“strongly disagree“) to 5 (“strongly agree“).

2.4. Analytical Strategy

Data analysis was performed using IBM SPSS Statistics (version 29), and AMOS (version 29). A structured two-stage approach was employed to purify and validate the measurement model before testing the structural relationships, as strongly recommended by Anderson and Gerbing [57]. The first stage involved assessing the measurement models. An initial exploratory factor analysis (EFA) was conducted to identify the underlying structure of the driver constructs. The suitability of the data was confirmed with a Kaiser-Meyer-Olkin (KMO) measure of 0.88 and a significant Bartlett’s test of sphericity (p < 0.001). This was followed by confirmatory factor analysis (CFA) to test the measurement model. Model fit was evaluated using multiple indices and their established thresholds: χ²/df (<3), CFI (>0.90), TLI (>0.90), and RMSEA (<0.08) [56,58]. Within the CFA, we assessed Internal consistency using Cronbach’s alpha (α > 0.70) and composite reliability (CR > 0.70); convergent validity by ensuring that the average variance extracted (AVE) was greater than 0.50; and discriminant validity using the Fornell-Larcker criterion.

The second stage involved assessing the structural model and testing the hypotheses. Structural equation modeling (SEM) was used to test the hypothesized structural paths by examining path coefficients (β), p-values, and variance explained (R²). To specifically test the mediation hypothesis (H6), we employed a bootstrapping procedure with 2000 resamples, as recommended by Preacher and Hayes [61]. This non-parametric method generated a 95% confidence interval (CI) for indirect effects, with mediation considered significant if the CI did not contain zero. To model the common variance among the three highly correlated strategic outcomes (Satisfaction, Loyalty, and Brand Equity), a second-order latent factor, termed the Strategic CX Outcome (SCO), was specified in the structural model. The three outcomes were modelled as first-order indicators of this higher-order construct. This approach is theoretically justified because these outcomes collectively represent the overall positive result of a superior customer experience. It also provides a more parsimonious and statistically stable model by reducing multicollinearity in the final stage of the analysis. Although split-sample EFA and CFA validation is sometimes recommended, both stages were conducted on the full sample N = 400. Splitting the sample would yield approximately N = 200 per stage, which falls below the recommended minimum for stable CB-SEM estimation of an eight-construct, 30-item model requiring estimation of over 80 parameters [57,58]. The EFA and CFA served sequential and complementary functions: EFA confirmed the expected seven-factor structure and verified that no items required deletion before the CFA imposed the theoretically specified measurement structure. No items were modified between stages, consistent with Anderson and Gerbing’s [57] two-step procedure.

2.5. Ethical Considerations

This study complied with all ethical norms. All participants were provided with an introductory statement detailing the academic purpose of the study and assuring their anonymity and confidentiality of data. Participation was entirely voluntary, and informed consent was obtained before the participants began the questionnaire. No personally identifiable data were collected.

2.6. Use of Generative AI

Generative AI was used solely to support text editing and structural organization of the manuscript. It was not used for data collection, analyze, or interpretation. All findings were based on the data collected and analyzed by the authors.

3. Results

This section presents the empirical findings derived from the statistical analysis of the survey responses. The presentation is structured to follow the analytical strategy outlined in the Methods section, beginning with preliminary data screening, followed by the validation of the measurement model, and culminating in the testing of the structural model and its hypotheses. Each result was followed by a direct interpretation of its meaning and implications.

3.1. Preliminary Data Analysis

Prior to the main analysis, the dataset was screened for data integrity. No missing values were observed after data cleaning. An assessment of univariate normality revealed that the skewness and kurtosis statistics for all items were well within the conventional thresholds, justifying the use of maximum likelihood estimation in subsequent analyses.

To assess the potential for common method variance (CMV), a common latent factor (CLF) was considered to the CFA model with paths from the CLF to all 30 observed items, following the procedure recommended by Podsakoff et al. [62]. The CLF variance was fixed to 1.0 for identification, and all seven constructs and their original measurement paths were retained unchanged. The average variance explained by substantive constructs before and after controlling for the CLF was 78.27% and 76.56% respectively, indicating that the CLF accounted for only 1.71% of total item variance—well below the 25% threshold considered problematic [62]. Average Standardized item loadings remained highly stable after CLF inclusion (average change = 0.010), confirming that substantive construct variance dominates over method variance. Three Engagement items (ENG1, ENG2, ENG5) and one Brand Equity item (BE2) showed relatively higher CLF loadings (range 0.212–0.410), which is acknowledged; however, the negligible overall CMV percentage indicates this does not critically threaten the substantive findings. In addition, two procedural remedies were implemented during data collection: (1) randomized item order to mitigate sequential response bias, and (2) explicit anonymity assurances to reduce social desirability bias. Collectively, these results provide strong evidence that common method variance is unlikely to distort the reported structural relationships.

3.2. Measurement Model Validation

A two-step process was used to validate the measurement model, starting with an exploratory analysis followed by a more rigorous confirmatory analysis. The structural model was estimated using maximum likelihood in IBM AMOS (version 29), following the two-step approach recommended by Anderson and Gerbing [57].

3.2.1. Exploratory Factor Analysis (EFA)

An initial EFA was conducted on all 30 items spanning all seven constructs in the measurement model—the four TAPE driver constructs and the three outcome constructs. With seven theoretically specified constructs, the emergence of seven factors is the expected and theoretically consistent outcome. Extraction method: principal axis factoring. Rotation: Promax. Retention: eigenvalue > 1.0, confirmed by scree plot. The cross-loading threshold < 0.40. No items were deleted. The prerequisite tests confirmed the suitability of the data, with a Kaiser-Meyer-Olkin (KMO) measure of 0.883 and a significant Bartlett’s Test of Sphericity (p < 0.001). The internal consistency of each construct was assessed using Cronbach’s alpha (α). As shown in

Table 3. Reliability of Survey Instrument.

Construct	No. of Items	Cronbach’s α
Trust	5	0.88
User Autonomy	4	0.85
Personalization	5	0.86
Customer engagement	5	0.87
Satisfaction	3	0.85
Loyalty	4	0.84
Brand Equity	4	0.86

, all constructs exceeded the recommended threshold of 0.70, demonstrating acceptable reliability. The analysis extracted seven distinct factors that explained 72.4% of the total variance. All items loaded clearly and strongly (>0.70) on their intended theoretical factors with minimal cross-loadings. This result can be interpreted as strong initial empirical evidence for the proposed seven-factor structure, consistent with the seven theoretically specified constructs, confirming that Trust, Autonomy, Personalization, and Engagement are empirically distinguishable dimensions along with strategic outcomes.

3.2.2. Confirmatory Factor Analysis (CFA)

A comprehensive CFA was performed on the full measurement model, which included all seven latent constructs. The model demonstrated an excellent fit with the data. These strong fit indices collectively indicate that the hypothesized measurement model, where specific survey items measure their intended latent constructs, provides a statistically sound and accurate representation of the observed data. The second-order reflective Customer Experience (CX) construct is constituted by four first-order experiential dimensions and a second-order Strategic CX Outcomes (SCO) construct [55]. Confirmed fit indices from AMOS: χ²/df = 1.115, NFI = 0.963, RFI = 0.959, IFI = 0.996, TLI = 0.996, CFI = 0.996, RMSEA = 0.017 (90% CI [0.000, 0.025]), PCLOSE = 1.000. The PCLOSE value of 1.000 confirms that the probability of the population RMSEA exceeding 0.05 is negligible, which is the strongest possible evidence of a close fit. All 30 standardized item loadings from AMOS: range 0.859–0.914, all exceeding the 0.70 threshold. CX second-order: Composite reliability (CR) = 0.783 (passes ≥ 0.70), Average Variance Extracted (AVE) = 0.483 (marginal; defended via Fornell-Larcker [63], CR is primary indicator). This dual second-order architecture follows Chahal and Dutta [55] as a published precedent. The full item-level loadings are presented in Appendix A,

Table A1. Complete Measurement Instrument with Standardized Item Loadings.

No.	Code	Item Wording (5-Point Likert: 1 = Strongly Disagree to 5 = Strongly Agree)	λ	λ²	Source
Trust (TR) Theory: Commitment-Trust Theory [23,30]; Algorithmic Trust [31]
1	TR1	I believe this platform’s AI systems operate fairly and transparently.	0.866	0.75	McKnight et al. [39]; Mayer et al. [30]
2	TR2	I trust that this platform handles my personal data responsibly.	0.877	0.769	McKnight et al. [39]; Shin [31]
3	TR3	I feel confident that this platform’s recommendations are made in my best interest.	0.897	0.805	Morgan & Hunt [23]; Belk et al. [41]
4	TR4	I believe this platform acts in the best interests of its customers.	0.874	0.764	McKnight et al. [39]; Gefen et al. [40]
5	TR5	I feel secure when sharing information with this platform.	0.885	0.783	McKnight et al. [39]; Adam et al. [25]
User Autonomy (UA) Theory: Self-Determination Theory [33]; Algorithmic Influence [27,65]
6	UA1	I feel free to make my own choices while shopping on this platform.	0.86	0.74	Deci & Ryan [33]; Bergdahl et al. [27]
7	UA2	I can easily adjust or override the platform’s personalization settings.	0.861	0.741	Deci & Ryan [33]; Tam & Ho [21]
8	UA3	This platform gives me meaningful control over my shopping experience.	0.87	0.757	Deci & Ryan [33]; Oesterreich et al. [65]
9	UA4	I do not feel manipulated by the automated features on this platform.	0.859	0.738	Srinivasan et al. [15]; Bergdahl et al. [27]
Personalization (PR) Theory: Privacy Calculus Theory [28]; AI-Driven Adaptivity [20,29,32]
10	PR1	This platform provides recommendations that are highly relevant to my preferences.	0.881	0.776	Bleier & Eisenbeiss [20]; Tam & Ho [21]
11	PR2	The content displayed on this platform adapts in real time to my behavior.	0.902	0.814	Tam & Ho [21]; Vesanen [32]
12	PR3	Personalization on this platform makes my shopping experience more efficient.	0.889	0.79	Dinev & Hart [28]; Cloarec et al. [29]
13	PR4	This platform understands my individual preferences better than other platforms.	0.867	0.752	Bleier & Eisenbeiss [20]; Vesanen [32]
14	PR5	The product suggestions I receive feel personally tailored to me.	0.878	0.771	Tam & Ho [21]; Cloarec et al. [29]
Customer Engagement (CE) Theory: Consumer Engagement Theory [17]; Digital Engagement [34,60]
15	ENG1	Interacting with this platform excites and interests me.	0.883	0.78	Brodie et al. [17]; Hollebeek et al. [60]
16	ENG2	I actively participate in community features (e.g., reviews, ratings) on this platform.	0.893	0.797	Hollebeek et al. [60]; Van Doorn et al. [45]
17	ENG3	I find myself spending more time on this platform than originally intended.	0.885	0.783	Brodie et al. [17]; Vivek et al. [34]
18	ENG4	The interactive features of this platform enhance my overall experience.	0.888	0.789	Hollebeek & Macky [19]; Utami et al. [46]
19	ENG5	I feel emotionally connected to this platform and the experience it provides.	0.903	0.815	Brodie et al. [17]; Vivek et al. [34]
Satisfaction (SAT) Theory: Expectancy Disconfirmation [2]; Service Satisfaction [51,52]
20	CSAT1	Overall, I am satisfied with my experience on this platform.	0.887	0.787	Oliver [2]; Sivadas & Baker-Prewitt [51]
21	CSAT2	This platform consistently meets my expectations.	0.899	0.808	Oliver [2]; Rose et al. [52]
22	CSAT3	I am pleased with the value I receive from using this platform.	0.881	0.776	Sivadas & Baker-Prewitt [51]; Rose et al. [52]
Loyalty (LOY) Theory: Customer Loyalty [53]; E-Commerce Loyalty [15,18]
23	LOY1	I intend to continue using this platform in the future.	0.885	0.783	Oliver [53]; Srinivasan et al. [15]
24	LOY2	I am willing to recommend this platform to others.	0.891	0.794	Oliver [53]; Harris & Goode [18]
25	LOY3	This platform is my first choice when I shop online.	0.879	0.773	Oliver [53]; Srinivasan et al. [15]
26	LOY4	I would choose this platform over competitors even if they offered similar products.	0.887	0.787	Oliver [53]; Harris & Goode [18]
Brand Equity (BE) Theory: Customer-Based Brand Equity [48,50]; Brand Experience [54]
27	BE1	I associate positive qualities with this platform’s brand.	0.906	0.821	Aaker [47]; Keller [49]
28	BE2	This platform’s brand has a strong and positive image in my mind.	0.893	0.797	Aaker [47,48]; Brakus et al. [54]
29	BE3	The brand of this platform adds value to my overall shopping experience.	0.901	0.812	Keller [49]; Brakus et al. [54]
30	BE4	Even if another platform offered identical products, I would prefer this platform’s brand.	0.914	0.835	Aaker [47]; Keller [49,50]

Note: Data from Current article. λ = Standardized factor loading (AMOS). λ² = Squared Multiple Correlation—variance in item explained by construct. Loading range: 0.859–0.914 (all > 0.70 threshold [57,63]). Sample N = 400. For Item wording and scale development adapted from Source column.

. The structural model demonstrated an excellent fit to the data across all conventional indices and all fit statistics comfortably exceeded the recommended thresholds. χ²/df < 3.0, NFI and RFI > 0.90, CFI and TLI > 0.95, RMSEA < 0.08—providing a sound basis for evaluating the hypothesized structural paths.

Table 4. Second-Order CX Construct and TAPE Driver Loadings.

TAPE Driver	λ (Onto CX)	R2	Driver Contribution to CX
Trust (TR)	0.646	0.417	41.7% of TR variance shared with CX
User Autonomy (UA)	0.497	0.247	24.7% of UA variance shared with CX
Personalization (PR)	0.803	0.645	64.5% of PR variance shared with CX
Customer Engagement (CE)	0.788	0.621	62.1% of CE variance shared with CX

Note: CX second-order: CR = 0.783; AVE = 0.483. The construct was identified through AMOS-estimated constraints on first-order factor loadings and error terms. The CR ≥ 0.70 threshold passed, and the AVE = 0.483 was marginal, defended via the Fornell-Larcker criterion [63].

presents the squared multiple correlations (R²) for all endogenous variables, reflecting the proportion of variance explained within the structural model. The SCO construct explained 71.9% of the variance in loyalty (R² = 0.719), 54.7% in satisfaction (R² = 0.547), and 53.5% in brand equity (R² = 0.535). These values indicate substantial explanatory power and are consistent with the benchmarks for SEM research recommended by Hu and Bentler [56]. CX explained 51.2% of the variance in strategic CX outcomes (R² = 0.512). Among the TAPE dimensions, the variance explained by CX was as follows: Personalization (R² = 0.645), Engagement (R² = 0.621), Trust (R² = 0.417), Autonomy (R² = 0.247). Positive CX is associated with immediate satisfaction and longer-term behavioral outcomes, including loyalty and brand equity.

Reliability, Convergent Validity and Discriminant Validity: Convergent validity was assessed using standardized factor loadings, composite reliability (CR), and average variance extracted (AVE), following Fornell and Larcker [63]. All standardized factor loadings ranged from 0.859 to 0.914, exceeding the threshold of 0.70. As shown in

Table 5. Construct Reliability, Convergent Validity and Discriminant Validity.

Construct	CR	AVE	TR	UA	PR	CE	SAT	LOY	BE
Trust (TR)	0.945	0.774	0.880
User Autonomy (UA)	0.921	0.744	0.295	0.863
Personalization (PR)	0.947	0.781	0.519	0.435	0.883
Cust. Engagement (CE)	0.950	0.793	0.505	0.402	0.625	0.890
Satisfaction (SAT)	0.919	0.790	0.399	0.280	0.470	0.471	0.889
Loyalty (LOY)	0.936	0.784	0.403	0.204	0.484	0.465	0.616	0.886
Brand Equity (BE)	0.947	0.816	0.333	0.219	0.367	0.427	0.516	0.645	0.904

CR = Composite Reliability; AVE = Average Variance Extracted. Bold diagonal = √AVE (square root of AVE).

, the CR values ranged from 0.919 to 0.950, and AVE values ranged from 0.744 to 0.816, both surpassing their respective thresholds, confirming convergent validity across all seven constructs. Discriminant validity was assessed using the Fornell-Larcker criterion, whereby the square root of each construct’s AVE (shown on the diagonal of Table 5) exceeded its highest inter-construct correlation. This condition was satisfied for all construct pairs, confirming that each construct captured a theoretically distinct dimension of the measurement model [63].

3.3. Structural Model and Hypothesis Testing

After establishing a robust measurement model, we tested the structural model. Figure 1 describes the SEM model and validated paths. The model fitness indices (χ²/df = 1.115, NFI = 0.963, RFI = 0.959, TLI = 0.996, CFI = 0.996, and RMSEA = 0.017) reflected a strong model fit. This provides initial support for the hypothesized structure. The estimation results are presented in

Table 6. (A) TAPE Driver Loadings onto CX (Measurement Paths). (B) Primary Structural Path and SCO Measurement Loadings.

(A)
Hypothesis	Path	λ (Loading)	SE	C.R	p-Value	Result
H1	Trust→CX	0.65	0.074	10.585	<0.001	Supported
H2	Autonomy→CX	0.50	0.070	8.265	<0.001	Supported
H3	Personalization→CX	0.80	0.078	12.235	<0.001	Supported
H4	Engagement→CX	0.79	-	-	<0.001	Supported
(B)
Hypothesis	Path	Coefficient	SE	C.R	p-Value	Result
H5	CX→SCO	0.72	-	-	<0.001	Supported
H5a	CX→Satisfaction	0.74	0.081	11.383	<0.001	Supported
H5b	CX→Loyalty	0.85	0.091	12.159	<0.001	Supported
H5c	CX→Brand	0.73	-	-	<0.001	Supported

Note: H5 reports the primary structural path coefficients (β). H5a, H5b, and H5c report measurement loadings (λ) of each outcome construct onto SCO—these are not direct causal paths from CX. SE and C.R. not reported for H5c (reference indicator). All paths significant at p < 0.001.

A, B. All seven hypothesized paths were statistically significant at p < 0.001, providing support for H1 through H5c in their entirety.

The primary structural path, CX→SCO, was β = 0.716 (C.R. = 11.246, p < 0.001), supporting H5. Satisfaction, Loyalty, and Brand Equity loaded onto SCO as first-order reflective indicators: H5a (λ = 0.740, C.R. = 11.383, p < 0.001, supported), H5b (λ = 0.848, C.R. = 12.159, p < 0.001, supported), and H5c (λ = 0.731, p < 0.001, supported). The strongest SCO loading was on loyalty (λ = 0.848), indicating that CX quality has a particularly pronounced association with long-term retention behavior.

Among the four TAPE drivers, Personalization emerged as the strongest driver of CX (λ = 0.803, C.R. = 12.235, p < 0.001), supporting H3. Customer Engagement was the second strongest driver (λ = 0.788; H4 supported), followed by Trust (λ = 0.646, C.R. = 10.585, p < 0.001; H1 supported) and User Autonomy (λ = 0.497, C.R. = 8.265, p < 0.001; H2 supported). All four drivers produced critical ratios well above the conventional threshold of 1.96, confirming that each TAPE driver made a statistically robust and significant contribution to the overall CX.

Hypothesis H6: Mediation Analysis: To test H6, that Customer Experience fully mediates the relationship between each TAPE driver and the three strategic outcomes, bootstrapped indirect effects were estimated using 2000 resamples following the procedure recommended by Preacher and Hayes [61]. Bias-corrected 95% confidence intervals (BC 95% CI) were used as the criterion for statistical significance; an indirect effect was considered significant when its confidence interval excluded zero. The complete results are listed in

Table 7. Comprehensive Bootstrapped Mediation Analysis of Indirect Effects.

Driver	Outcome	Indirect Path	Standardized (β)	95% Bootstrapped CI	p-Value	Result
Trust (TR)	Satisfaction	TR→CX→SCO→SAT	0.34	[0.27, 0.41]	<0.001	Significant
	Loyalty	TR→CX→SCO→LOY	0.4	[0.32, 0.48]	<0.001	Significant
	Brand Equity	TR→CX→SCO→BE	0.34	[0.26, 0.42]	<0.001	Significant
User Autonomy (UA)	Satisfaction	UA→CX→SCO→SAT	0.27	[0.20, 0.35]	<0.001	Significant
	Loyalty	UA→CX→SCO→LOY	0.31	[0.23, 0.40]	<0.001	Significant
	Brand Equity	UA→CX→SCO→BE	0.26	[0.18, 0.34]	<0.001	Significant
Personalization (PR)	Satisfaction	PR→CX→SCO→SAT	0.43	[0.36, 0.51]	<0.001	Significant
	Loyalty	PR→CX→SCO→LOY	0.49	[0.41, 0.58]	<0.001	Significant
	Brand Equity	PR→CX→SCO→BE	0.42	[0.34, 0.50]	<0.001	Significant
Engagement (ENG)	Satisfaction	ENG→CX→SCO→SAT	0.42	[0.34, 0.49]	<0.001	Significant
	Loyalty	ENG→CX→SCO→LOY	0.48	[0.39, 0.57]	<0.001	Significant
	Brand Equity	ENG→CX→SCO→BE	0.41	[0.33, 0.48]	<0.001	Significant

Note: SCO—Strategic CX Outcomes.

.

The full indirect path runs as follows: Each TAPE driver influences CX, which in turn influences the second-order SCO, which then determines the scores for satisfaction, loyalty, and brand equity as its first-order indicators. Accordingly, the indirect effects reported in Table 7 represent the total effect of each driver transmitted through this two-step chain (driver→CX→SCO→outcome). Because the structural Model A specifies no direct paths from TAPE drivers to strategic outcomes, all effects are routed through CX and subsequently through SCO with full mediation theoretically grounded in the S-O-R framework [53] and supported by evidence consistent with the preferred full-mediation specification, as demonstrated by the three-model comparison reported in Section 3.4 (Δχ² = 4.877, Δdf = 3, p = 0.181 for Model A vs Model C) [61]. All 12 indirect effects were statistically significant at p < 0.001, with all BC 95% CIs excluding zero, fully supporting H6. Personalization produced the largest indirect effects across all three outcomes: β = 0.43 (BC 95% CI [0.36, 0.51]) for satisfaction, β = 0.49 ([0.41, 0.58]) for loyalty, and β = 0.42 ([0.34, 0.50]) for brand equity. Customer engagement yielded similarly strong indirect effects: β = 0.42 ([0.34, 0.49]) on satisfaction, β = 0.48 ([0.39, 0.57]) for loyalty, and β = 0.41 ([0.33, 0.48]) for brand equity. These findings indicate that the experiential value generated by personalized AI interactions and sustained customer engagement has a particularly consequential downstream impact on both retention and brand perception.

Trust demonstrated meaningful indirect effects across all three outcomes: β = 0.34 ([0.27, 0.41]) for Satisfaction, β = 0.40 ([0.32, 0.48]) for Loyalty, and β = 0.34 ([0.26, 0.42]) for Brand Equity. User Autonomy, while producing the smallest indirect effects among the four drivers, nonetheless yielded statistically significant indirect paths to Satisfaction (β = 0.27, BC 95% CI [0.20, 0.35]), Loyalty (β = 0.31, [0.23, 0.40]), and Brand Equity (β = 0.26, [0.18, 0.34]). The absence of any confidence interval spanning zero across all 12 indirect paths provides strong and consistent evidence that CX, operating through the SCO mechanism, functions as a critical experiential pathway through which each TAPE driver translates into tangible strategic outcomes.

Taken together, the structural model results provide robust empirical support for the TAPE framework. All seven direct hypotheses (H1–H5c) and all 12 indirect effects constituting H6 were confirmed at p < 0.001. The pattern of path coefficients reveals a clear hierarchy among the TAPE drivers: Personalization and Customer Engagement exert the strongest influence on CX and, through it, on strategic outcomes, while Trust and User Autonomy contribute independently and significantly to the overall experiential model. Among the three outcomes, Loyalty records the strongest loading on the SCO factor (β = 0.848) and the highest variance explained (R² = 0.719), underscoring the particular importance of AI-era experience quality for long-term customer retention.

3.4. Model Comparison and Empirical Evaluation of Model Specification

To evaluate the appropriateness of the mediation specification and the SCO second-order structure, as shown in

Table 8. Mediation model structure.

Model	Type	Description
Model A	Full mediation	TAPE→CX→SCO→Outcomes. No direct TAPE→SCO paths. SCO specified as second-order reflective construct. Proposed model.
Model B	Direct paths only	SCO second-order construct removed. CX predicts Satisfaction, Loyalty, and Brand Equity as three separate endogenous variables.
Model C	Partial mediation	SCO structure retained. Four direct paths added: TR→SCO, UA→SCO, PR→SCO, CE→SCO alongside CX-mediated paths.

, three competing models were constructed.

Table 9. Model comparison.

Fit Index	Threshold	Model A Full Mediation	Model B Direct Paths	Model C Partial Mediation
χ2	—	442.62	541.718	437.743
df	—	397	399	394
p	>0.05	0.057	<0.001	0.063
χ2/df	<3.0	1.115	1.358	1.111
NFI	>0.90	0.963	0.954	0.963
RFI	>0.90	0.959	0.95	0.959
IFI	>0.95	0.996	0.988	0.996
TLI	>0.95	0.996	0.986	0.996
CFI	>0.95	0.996	0.988	0.996
RMSEA	<0.08	0.017	0.03	0.017
RMSEA 90% CI	—	[0.000, 0.025]	[0.023, 0.036]	[0.000, 0.025]
PCLOSE	>0.05	1	1	1
AIC	Lower = better	638.62	733.718	639.743
Decision		✓ Preferred	Rejected	Not preferred

summarizes model comparisons.

As shown Table 8, Model B removed the SCO second-order construct, connecting CX directly to Satisfaction, Loyalty, and Brand Equity as separate endogenous variables (χ² = 541.718, df = 399, CFI = 0.988, RMSEA = 0.030, AIC = 733.718). Model C retained the SCO structure but added four direct TAPE→SCO paths alongside the CX-mediated paths (χ² = 437.743, df = 394, CFI = 0.996, RMSEA = 0.017, AIC = 639.743). Model A (χ² = 442.620, df = 397, CFI = 0.996, RMSEA = 0.017, AIC = 638.620) fit significantly better than Model B (Δχ² = 99.098, Δdf = 2, p < 0.001, ΔAIC = 95.098), supported by evidence consistent with the preferred full-mediation specification, as demonstrated by the three-model comparison of the SCO second-order structure. As shown in Table 9, the Model A vs Model C comparison showed a non-significant chi-square difference (Δχ² = 4.877, Δdf = 3, p = 0.181, ΔAIC = 1.123), indicating direct TAPE→SCO paths do not improve fit when CX is present. Furthermore, Model C produced inadmissible standardized estimates: Personalization→SCO (β = 1.57) and Engagement→SCO (β = 1.30), confirming these effects cannot be separated empirically from the CX pathway. Mediation is supported by three converging criteria. R² values: SCO = 0.512, LOY = 0.719, SAT = 0.547, BE = 0.535, PR = 0.645, CE = 0.621, TR = 0.417, UA = 0.247. These inadmissible estimates are not incidental collinearity; they are the structural consequence of the second-order CX driver specification. Because TAPE drivers constitute CX as reflective first-order indicators, their variance is already channeled through CX by the measurement model. Adding direct TAPE→SCO paths therefore create a dual-routing specification in which the same variance must simultaneously flow through the CX experiential pathway and bypass it—a structural impossibility. The three-model comparison is therefore the appropriate test: Model B demonstrates that the SCO second-order structure adds significant explanatory value over simple direct paths (Δχ² = 99.098, p < 0.001), while Model C demonstrates that direct TAPE→SCO paths add no coherent explanatory value when the CX driver pathway is present (Δχ² = 4.877, p = 0.181, with inadmissible β estimates). Model A is the only specification consistent with both the data and the theoretical architecture.

4. Discussion

This study aimed to understand how four AI-driven features—trust, user autonomy, personalization, and customer engagement—shape customer experience in e-commerce and are associated with outcomes such as satisfaction, loyalty, and brand equity. The findings present a clear and consistent pattern: customers do not respond directly to AI features. Instead, they respond to the experience that these features help to create. Customer experience (CX) is the central conduit that translates the design of AI-enabled platforms into meaningful customer outcomes. All four TAPE drivers demonstrated strong and positive effects on CX, indicating that AI design influences customers both functionally and emotionally. Personalization and engagement demonstrated the strongest effects (λ = 0.803 and λ = 0.788, respectively), suggesting that customers value not only the accuracy of AI but also the feeling of relevance and participation it generates. Trust and user autonomy are also important contributors (λ = 0.646 and λ = 0.497, respectively), reinforcing the idea that customers must feel secure and in control when interacting with AI systems. Together, these findings indicate that experiential value in AI-enabled commerce emerges from a combination of cognitive assurance and emotional involvement, a pattern consistent with experiential theories of consumption that distinguish between the functional and hedonic dimensions of customer value [4,64].

The next key insight is that CX strongly shapes customers’ overall evaluations of the platform, influencing satisfaction (β = 0.740), loyalty (β = 0.848), and brand equity (β = 0.731). The strength of these relationships underscores that experience is the real currency of AI-enabled e-commerce. Even advanced algorithms matter little unless they create a customer journey that feels intuitive, supportive, or engaging principle that applies with particular force when AI, rather than a human service agent, is responsible for managing the interaction [43]. Results are consistent with CX functioning as the central experiential pathway [52,61]. This finding explains why even modest improvements in personalization or engagement quality can produce disproportionately large downstream effects on loyalty and brand equity: when experience is the mediating bridge, strengthening the bridge amplifies every upstream signal that crosses it. Overall, this study contributes to a deeper understanding of how AI influences customer behavior by showing that AI’s impact is psychologically grounded rather than purely functional. By revealing that AI design works through experiential pathways, the findings encourage researchers and practitioners alike to view AI not simply as a technical tool but as an experience-shaping mechanism embedded in the customer journey consistent with Becker and Jaakkola’s call to anchor CX research in the customer’s direct phenomenological perspective [14].

4.1. Theoretical Contributions

This study contributes to the literature on AI-driven customer experience in several important ways. First, by examining trust, user autonomy, personalization, and engagement within a single integrated model, it offers a more complete picture of how different AI design elements simultaneously influence customer experience. Prior work has tended to examine these features in isolation, studying trust [39,40], engagement [60], or personalization [29], as separate constructs without capturing their combined experiential effect. This study shows that they work in concert: some features enable comfort and control, while others drive emotional involvement and relevance. The TAPE framework formalizes this interdependence and provides a parsimonious structure for future AI-era CX research. Second, the findings highlight customer experience as the central mediating pathway that links AI design to customer outcomes. Although CX is widely recognized as important in digital contexts [7,8], its mediating role in AI-enabled environments has been largely unexplored, and a gap identified in the systematic review that motivated this study [35]. The present findings provide empirical evidence that CX is not peripheral but foundational: it is the psychological channel consistent with stimulus-organism-response logic [16,52], through which AI features are translated into satisfaction, loyalty, and brand equity that organizations depend upon. The comparatively lower coefficient for User Autonomy (λ = 0.497) and lower variance explained (R² = 0.247) may reflect the autonomy paradox in human-AI interaction: active e-commerce users normalize algorithmic curation, accepting personalized recommendations as convenience features rather than perceiving reduced control as experientially negative. Autonomy functions as an experiential enabler—a foundational psychological condition for positive CX—whose contribution is necessary but less perceptually salient than the amplifying effects of Personalization (λ = 0.803) and Engagement (λ = 0.788). ‘The strong reflective loading of Personalization (λ = 0.803) onto CX, despite the documented personalization-privacy paradox, may reflect consumer normalization of AI-driven recommendation systems—a boundary condition warranting cross-cultural investigation [16,22,29].

Third, this study advances the understanding of experiential theories of human–technology interaction [7,14], by showing that customers evaluate AI-enabled platforms through layered experiential pathways rather than direct assessments of technical features. This finding is consistent with Holbrook and Hirschman’s foundational argument that consumption involves fantasy, feeling, and involvement alongside functional utility [4], and with Brodie et al.’s framing of engagement as inherently co-creative and relational [17]. The AI context extends these insights by showing that even algorithmically generated interactions are assessed through the same experiential logic. Finally, this study offers a conceptually useful distinction in the relative roles of the four TAPE drivers. Personalization and Engagement function as experiential amplifiers; they create richer, more emotionally resonant interactions that elevate the overall quality of the customer’s experiential state. Trust and Autonomy, by contrast, function as experiential enablers; they establish the cognitive and psychological conditions of security and self-direction that must be in place before experiential value can be generated and appreciated. This distinction finds theoretical support in self-determination theory [26,33], which identifies autonomy and competence as foundational psychological needs that enable intrinsic motivation, and in consumer engagement theory [17], which positions engagement as an active value-generating process that builds on, rather than replaces, this foundation. The distinction enriches theoretical models of AI-driven value creation and points toward a more nuanced understanding of the sequencing of CX design decisions.

4.2. Managerial Implications

The findings of this study offer several actionable insights for managers, platform designers, and decision-makers aiming to enhance the customer experience in AI-enabled e-commerce environments. First, the strong influence of Trust on CX underscores the importance of transparency and responsible AI communication. Customers must feel confident that automated decisions, such as recommendations, pricing, or product prioritization are fair, accurate, and ethically grounded [31,41]. Managers should therefore invest in visible trust-building mechanisms, including transparent recommendation rationales, clear data-use policies, and explainable AI interfaces [25,44]. Enhancing trust also directly reduces the perceived risk that customers associate with AI-mediated transactions [42], increasing their willingness to engage with intelligent features and disclose the behavioral data that makes personalization possible. Second, the significance of user autonomy suggests that customers value a sense of control in an automated environment; AI should guide users, not constrain them. This principle is well-grounded in self-determination theory [33], which identifies autonomy as a fundamental need whose satisfaction is a prerequisite for intrinsic motivation and well-being. Firms can foster autonomy through adjustable personalization settings, alternative navigation pathways, clear opt-in mechanisms, and opportunities to override automated decisions [21,27]. Interfaces that balance automation with genuine agency allow customers to feel empowered rather than manipulated [16], substantially enhancing the experiential quality of the journey and, through it, loyalty and brand equity. Third, the strong impact of personalization on CX indicates that tailored relevance is now a baseline expectation rather than a differentiating luxury in digital commerce. Managers should continuously refine recommendation engines, behavioral analytics, and adaptive interfaces to ensure that the content is timely, individualized, and contextually meaningful [10,29]. Simultaneously, personalization strategies must not cross privacy boundaries; the privacy calculus theory [28] reminds us that customers are always weighing the value of a personalized experience against the perceived cost of data disclosure. Ethical personalization grounded in data transparency, user consent, and proportionate data use is therefore essential for sustaining the trust that makes deep personalization possible [22].

The following platform-specific examples illustrate how each TAPE dimension manifests in current AI-enabled e-commerce practice: Trust: Amazon’s ‘Why was I shown this?’ transparency feature and A-to-Z Guarantee directly instantiate the Trust measurement items—algorithmic fairness, recommendation rationale, and data protection assurance. Autonomy: Amazon’s preference Center enables customers to adjust personalization settings, opt out of interest-based recommendations, manage browsing history, and configure notification preferences—directly instantiating the Autonomy driver items measuring perceived control and the ability to override automated decisions without sacrificing platform functionality. Personalization: Amazon’s AI-powered recommendation engine dynamically tailors the homepage, product detail pages, and promotional offers to individual behavioral signals in real time—corresponding to the Personalization driver items measuring real-time adaptivity, contextual tailoring, and the feeling of being uniquely understood as an individual customer. Engagement: Amazon’s review community, loyalty points system, and Prime Day gamification mechanics deepen affective investment across the purchase journey, corresponding to Engagement items measuring cognitive absorption, emotional arousal, and active participation.

Fourth, the importance of customer engagement highlights the need for platforms to design interactive, emotionally resonant, and immersive experiences. Features such as conversational AI agents, dynamic product visualizations, gamified micro-interactions, and community-driven content have all been shown to increase cognitive and emotional involvement [9,60]. These features deepen the investment customers make in the platform, and as this study demonstrates, it is precisely that depth of investment that generates stronger loyalty and brand attachment over time. Finally, the full mediation findings carry a practical message that goes beyond individual features: technological capabilities generate strategic value only when they combine to create a coherent, seamless, and enjoyable experience [7]. Managers should therefore resist the temptation to optimize individual AI features in isolation. Instead, they should adopt a holistic CX design philosophy in which intelligent components interact consistently and reinforce one another across every stage of the customer journey from discovery and evaluation through to purchase and post-purchase support [8]. This systemic perspective ensures that the platform feels intuitive, trustworthy, and customer-centric, ultimately driving the satisfaction, loyalty, and brand equity [2,47,53], that translate AI investment into sustained competitive advantage.

4.3. Limitations and Future Research Directions

Although this study offers several important contributions to the understanding of customer experience in AI-enabled e-commerce, it has some limitations. These limitations provide meaningful opportunities for future research to deepen, extend, and refine the insights obtained in this study. Trust and Engagement items are framed at the general platform level; Personalization and Autonomy items reference AI-specific features more directly. Following Puntoni et al. [16], and Becker and Jaakkola [14], the measurement is grounded in customers’ subjective experiential perceptions rather than technical AI awareness, a phenomenologically valid approach whose boundary conditions should be examined in future research using AI-specific scales. Common method bias: A formal Common Latent Factor (CLF) test confirmed that CMV accounts for only 1.71% of total item variance [62], well below the 25% threshold considered problematic, providing empirical evidence that common method variance does not critically threaten the substantive findings. Nevertheless, future research employing longitudinal or multi-source designs would further strengthen causal inference.

Demographic boundary conditions: Geographic concentration (Northeast 47.2%, West 28.5%), age skew (18–29 is the largest group at 33.5%), and high-income concentration (56.1% above $100k) may limit generalizability to younger, lower-income, or non-US consumer populations. The distinction between the three levels of claims should be maintained: (1) statistical significance, confirmed by path coefficients; (2) substantive importance, indicated by effect sizes; and (3) causal inference, requiring longitudinal or experimental designs. The survey instrument did not capture education level or online shopping frequency, which limits subgroup comparisons on these dimensions; future research should include these variables to enable broader generalizability assessments across consumer literacy and platform engagement levels.

The study did not employ AI-feature-specific screening or item-level AI attribution; respondents evaluated their overall platform experience rather than isolating specific AI-generated interactions. This phenomenological approach, grounded in Puntoni et al. [16] and Becker and Jaakkola [14], is theoretically appropriate for holistic CX research and reflects the practical reality that AI features are now pervasive across all mainstream e-commerce platforms [12,13]. However, the findings should be interpreted as evidence of CX in AI-mediated environments rather than as direct evidence of reactions to specific AI features. Future research should examine feature-specific effects using observational or experimental designs that expose respondents to controlled AI-feature variations.

This study relied on cross-sectional survey data, which limits the ability to infer causality between TAPE drivers, customer experience, and downstream outcomes. Although structural equation modeling provides strong evidence of directional relationships, longitudinal or experimental designs would allow future studies to examine how experiences and perceptions evolve over time, particularly as customers interact repeatedly with AI-driven features or as platforms update their algorithms. Experimental approaches that directly manipulate the level of personalization offered or the degree of autonomy afforded on a controlled platform would allow causal claims to be made with greater precision. Tracking customers over multiple touchpoints can provide richer insights into the dynamic nature of experience formation in intelligent environments.

This study is based on the self-reported perceptions of e-commerce users within a single national context. Although the sample is diverse and sufficiently large, perceptions of AI, autonomy, and trust may vary across demographic groups, cultural contexts, and platforms. Future research could explore cross-cultural comparisons alongside platform-specific analyses (e.g., marketplaces vs. direct-to-consumer platforms) and industry-based variations (e.g., retail, travel, and digital services). The purchase context is another boundary worth examining: whether a consumer is shopping with a specific goal in mind or simply browsing may moderate how TAPE drivers shape the overall experience. This would help to determine the generalizability of the TAPE framework across different digital ecosystems.

Finally, this study examined AI-enabled customer experience at the user perception level. As AI capabilities continue to evolve, future research may benefit from integrating objective behavioral data, such as clickstream patterns, browsing histories, recommendation acceptance rates, or chatbot interaction logs, alongside subjective perceptions. Combining subjective perceptions with behavioral metrics, possibly through multi-method designs, would offer a richer, more holistic view of how AI-driven systems shape customer journeys. In summary, these limitations highlight productive directions for future research aimed at capturing the evolving dynamics of intelligent commerce. As AI becomes more pervasive and deeply embedded in consumer environments, continued exploration will be crucial for building a more comprehensive understanding of experience creation, digital trust, and customer-platform relationships.

5. Conclusions

In designing AI systems that know more about customers than ever before, the e-commerce industry faces a fundamental paradox: the more automated the experience becomes, the more human quality it must feel. This study offers both an empirical answer to this paradox and a practical framework for resolving it. The TAPE framework: Trust, Autonomy, Personalization, and Customer Engagement emerge from this study as a validated, four-dimensional model of CX drivers is tailored to the realities of AI-enabled e-commerce. Grounded in four independent theoretical traditions and tested on e-commerce users, the framework confirms that all four drivers significantly and positively shape overall CX. Critically, personalization and engagement emerge as the most powerful contributors, a finding that reveals something important: what customers value most in AI-driven commerce is not efficiency or reliability alone, but the sense that the experience is uniquely theirs and that they are active participants in it, not passive recipients of algorithmic output.

The most consequential finding of this study is the structural evidence consistent with full mediation showing that CX significantly bridges all four TAPE drivers and three strategic outcomes: customer satisfaction, loyalty, and brand equity. ‘The findings are consistent with the following causal proposition: technological platform features do not directly generate strategic value, but only when successfully orchestrated into a holistic, positive experience in the mind of the customer. It is the quality of the experience, not the sophistication of the technology producing it, that determines whether a customer returns, recommends, and remains loyal. Managerially, these findings reframe the central question of AI strategy: not ‘How capable is our AI?’ but ‘How trusted, empowered, personalized, and engaged do our customers feel?” This distinction has profound implications for investment priorities and platform design decisions.

Ultimately, sustainable competitive advantage in AI-era e-commerce will not be won by firms that deploy the most sophisticated algorithms but by those that use them to design experiences that are trustworthy, empowering, personally relevant, and genuinely human. As AI continues to reshape the digital marketplace, the ability to design journeys that honor user agency, protect trust, and deepen authentic engagement will be the defining characteristic of the brands that endure. This study provides the empirical foundation for that endeavor.