Unpacking Customer Experience in Online Shopping: Effects on Satisfaction and Loyalty

Abstract

Drawing on experience–satisfaction–loyalty, this study models how specific digital retail interface attributes translate into behavioural outcomes. Survey data from Portuguese online shoppers were analysed with PLS-SEM to test a formative–reflective framework linking Interactivity and Technologies, Trust–Security–Privacy, Fulfilment and Service Quality, Usability and Web Design, Personalisation and Customisation and Omnichannel Integration to customer experience (CX), customer satisfaction (CS), customer loyalty (CL) and electronic word of mouth (eWOM). The model explains 62.6% of CX, 70.1% of CS and 66.7% of CL. CX is strongly associated with CS and CS, in turn, with CL; associations with eWOM are non-significant, revealing a theoretical blind spot around advocacy. Interactivity and Technologies, Trust–Security–Privacy and Fulfilment and Service Quality emerge as the most significant antecedents of CX, whereas Omnichannel Integration is inert. The findings advance digital commerce theory by decoupling advocacy from evaluative satisfaction and by reconceptualising integration as multidimensional. Practically, they prioritise investment in interactive, secure and fulfilment capabilities while signalling that loyalty is not associated with advocacy. This study concludes by outlining measurement refinements and longitudinal avenues to capture social–motivational drivers of eWOM.

1. Introduction

Over the past two decades, scholars have progressively repositioned customer experience (CX) from a set of isolated service encounters to a holistic, journey-oriented construct that spans multiple touchpoints and unfolds over time [1,2]. CX is a multidimensional phenomenon involving cognitive, emotional, sensory and behavioural responses to interactions with firms, employees, partners and the environment [1].

Digital retail has heightened the salience of CX: consumers’ overall perception of online stores is a decisive competitive lever [3]. In intensely contested markets, memorable experiences are required to enhance satisfaction, loyalty, and favourable eWOM [4], consistent with creating a positive experience–satisfaction–loyalty (ESL) pathway along the customer journey (CJ). Online CX is shaped at every digital and physical touchpoint, underscoring customer journey mapping (CJM) imperatives [5].

Despite notable advances, significant theoretical and empirical gaps continue to hinder progress. To begin with, there is a dearth of empirically validated, integrative models that transcend sector boundaries. Although the influential framework proposed by Lemon and Verhoef [1] links touchpoints, internal capabilities and external factors, most contributions remain predominantly conceptual; using cross-sector validation is still the exception rather than the rule [1,2,6]. Consequently, robust longitudinal and comparative designs are urgently required. Outcome coverage is also narrow: scholarship privileges satisfaction and behavioural loyalty while downplaying emotional, cognitive, relational, and financial outcomes, and although Gupta and Zeithaml [7] link CX to financial performance, few studies operationalise comprehensive outcome sets within a coherent framework [1,8]. Thus, multidimensional instruments and continuous monitoring are required to track experience–value linkages over time, in accordance with ESL and CJM.

Fundamental theoretical debates persist. One such approach involves the integration of universal CX principles with generalist models [1]. Debate continues over the relative weight of internal versus external drivers of CX: some foreground organisational resources and capabilities [9], whereas others emphasise customer-centric factors and market actors [6,8]. SDL offers a reconciliation by viewing CX as multi-level resource integration across actors, while ESL and CJM align expectation formation and confirmation along mapped touchpoints. Against this lens, the field still lacks integrative models, multidimensional measurement, and dynamic (longitudinal) validation, gaps whose closure would deepen theory and equip practitioners in complex, digitalised markets. Recent work reconceptualises e-commerce CX through consolidated consumer behaviour theories [3], defining it as holistic, subjective reactions, simultaneously functional and emotional, unfolding across the online journey. Although the literature portrays CX as multimodal and integrated, resisting reduction to isolated touchpoints [10], decomposition remains common: Yin and Xu [10] distinguish functional, emotional, and behavioural dimensions; Urdea and Constantin [4] apply CX and CL; and Yi et al. [11] concentrate on repeated purchases.

Building upon these theoretical foundations, the present research conducts an updated review to isolate the principal determinants of online experience. Subsequently, a conceptual model is advanced and empirical hypotheses are derived. Therefore, the research question can be formulated as follows: “What are the main determinants of online customer experience, and how can these be integrated into a conceptual model that captures the multidimensional, dynamic, and multi-level nature of CX across the entire customer journey?” The following structure has been adopted for the presentation of this research. Section 2 provides a review of the pertinent literature, while Section 3 outlines the model and hypotheses. Section 4 details the methodology, followed by Section 5, which reports the results. Section 6 offers a discussion of the results within a theoretical context, and finally, Section 7 summarises the conclusions drawn from the study.

2. Literature Review

2.1. The Precursors of the Experience of Customers

Lemon and Verhoef [1] demonstrate that a constellation of determinants underpins customer experience in digital retail contexts. Building on their insight, recent scholarship up to 2025 has isolated seven pivotal dimensions that merit systematic analysis. The literature depicts a coherent framework within which Usability and Website design, Trust–Security–Privacy, Fulfilment and Service Quality, Interactivity and Technology, and Omnichannel Integration blend to define the contemporary digital retail experience. The five antecedents selected are not an arbitrary enumeration but reflect core mechanisms highlighted in ESL [3] and Service-Dominant Logic (SDL) [12,13]. Each addresses a different stage of the online journey: usability and design shape pre-purchase expectations of ease and accessibility [14]; trust and privacy provide relational assurances during transaction [15,16]; Fulfilment and Service Quality validate post-purchase experiences [17,18]; Interactivity and Technology enable dynamic engagement and co-creation [19,20]; and Omnichannel Integration ensures systemic coherence across touchpoints [21,22]. In fact, each antecedent corresponds to a key phase (search, transaction, fulfilment, and post-purchase) and they exhaust the touchpoint set in digital retail. Jointly, they cover functional, relational, experiential, and systemic dimensions of digital commerce, thereby constituting a comprehensive and theory-driven set of precursors to customer experience. Whereas prior research often isolates individual drivers, this study advances the field by integrating them into a unified formative–reflective model that captures the multidimensional and longitudinal character of online CX.

2.1.1. Usability and Website Design

Website usability is a pivotal determinant of satisfaction, enabling efficient, effective, and enjoyable goal attainment [23]. U-WD is defined as pertaining to the foundational ease and efficiency of achieving basic goals and aligns with classic usability goals (effectiveness and efficiency). It also has robust links to overall satisfaction and affective appraisal [14,24] and shapes perceived service quality, supported by sound information architecture, seamless navigation, and intuitive ordering [25,26]. Usability and functionality are reciprocally reinforcing [27], and clear descriptions with logical navigation foster trust and satisfaction, warranting systematic assessment within overall website quality [14,28,29]. Economically, usability lowers search costs and errors and, combined with accurate search and interactivity, elevates CX and loyalty [24,30]. Usability also bears on eWOM: information quality, reliability, and service foster advocacy and responsive, fast, mobile-optimised design and rich design attributes shape experience and purchase [3,31,32]. Yet trade-offs persist as content richness can depress immediate satisfaction via overload, requiring calibrated density and navigational simplicity [3,33]. Finally, navigation speed and responsiveness reduce friction [34,35,36], though loyalty may stem from delight as much as from the avoidance of dissatisfaction [37]. Therefore, we hypothesise that website usability and design are positively associated with CX (H1).

2.1.2. Omnichannel Integration and Consistency

Omnichannel Integration (OI) is critical because consumers expect frictionless movement across digital–physical touchpoints: coherent coordination reduces perceived privacy risk and reinforces loyalty [38], while transparency and information consistency lower cognitive effort during transitions [39]. Advancing this view, OI aligns technical configuration with experiential uniformity, bolstering utilitarian and hedonic value and, hence, satisfaction and future use [22]; yet many firms struggle to operationalise integration amid organisational silos [40,41]. Consistency heightens perceived fluency and service-use propensity [42], harmonised content curtails risk and search costs [43], and uniform data presentation builds trust in stewardship [44], although messaging inconsistencies can erode credibility [45]. Seamless switching enhances deeper engagement [38], increases share of wallet [45], and reduces anxiety to support eWOM [46]; integration can raise revenue but also privacy concerns without transparent governance [47]. Accordingly, managers prioritise synchronising inventory, marketing, and service, supporting real-time stock reliability [48] and loyalty growth [49]. Historically, rapid channel proliferation risked fragmentation [50], prompting the online CX concept to emphasise consistent assortments, pricing, and identity [21]; discrepancies such as uneven pricing undermine satisfaction and repurchase [51]. Personalisation amplifies integration effects via portable profiles [52], data flows from store to online recommendations [53], and buy-online-pick-up-in-store synergies that lift involvement and loyalty [54]. Despite progress, research remains fragmented [55], though longitudinal evidence links cross-channel integration to purchase frequency [56], and consistency accelerates decisions by reducing uncertainty [57]. Hence, we hypothesise that Omnichannel Integration and Consistency across channels are positively associated with CX (H2).

2.1.3. Personalisation and Customisation

Personalisation positions business as architects [58,59]. By contrast, customisation grants users control over attributes across design, fabrication, assembly, delivery, and use, reducing uncertainty and fostering mastery even without full uniqueness [60,61,62]. Accordingly, the distinction rests on the initiator (firms for personalisation and customers for customisation), yet both pursue preference fit; personalisation can still yield unique experiences, while in virtual contexts, customisation confers a relative advantage and engagement [63,64,65]. Despite challenges, such as privacy regulation, algorithmic opacity, and implementation complexity, personalisation remains a strong lever for satisfaction [63,66]. Recent work has sharpened its scope and control: customisation spans consumption, marketing, and lifecycle stages, whereas firms silently refine offerings via behavioural data [63,67,68]. Moreover, personalisation can precede deeper customisation, with the latter shaping “what” is delivered and the former “how”; surfacing latent preferences elevates valuation and empowers customers [69,70,71]. Sectoral evidence is consistent: customisable menus enhance profitability [72], personalised recommendations heighten psychological ownership, and bespoke room options support loyalty, while multichannel personalisation strengthens trust and consistency [73,74,75,76]. Post pandemic, consumers expect deep personalisation embedded within omnichannel online CX, with AI recommenders increasing repurchase intent [21,55]. However, intrusive or opaque implementations erode confidence; privacy concerns diminish efficacy, underscoring the need for transparent, consent-based practices [77,78,79]. Accordingly, we hypothesise that Personalisation and Customisation are positively associated with CX (H3).

2.1.4. Trust, Security and Privacy

CX is inseparable from trust, security, and privacy: consumers weigh providers’ assurances in adoption decisions, and such assurances are persuasive only when evidenced by reliable conduct [80]; yet firms often overrate protection while customers remain wary [81]. Privacy concerns therefore impair CX via the mediating erosion of trust, with emotional instability exacerbating this effect [82]. Building trust requires tangible safeguards: perceived security and privacy raise willingness to transact and, with experience, reduce generalised concerns [83,84]. Consistently, stronger information security perceptions increase trust [15,85]; trust combines belief in a party’s honesty with willingness to rely on them despite limited control [85], and reputational trust can even relax privacy demands in e-health [86]. Trust remains central to digital CX and tightly coupled to security [87,88]; it entails perceived honesty, reliability, and competence over quality, delivery, and data protection [16,89]. Because digital transactions are risky, security and privacy guarantees are indispensable [90,91]. Accordingly, trust, perceived risk, and information security are fundamental to CX, to satisfaction and loyalty, in integrative models [3]. Lapses can negate otherwise positive experiences, whereas credibility can fortify confidence. In omnichannel journeys, uniform data protection standards are expected; security cues amplify other experiential elements or nullify them when absent [21]. Thus, we hypothesise a positive T–S–P → CX association (H4).

2.1.5. Fulfilment and Service Quality

Fulfilment, operational commitment to explicit and implicit promises, augments trust and perceived expertise and often translates into loyalty, while employee responsiveness and personalised communication elevate perceived service quality [17,92,93]. Experiential trust further mitigates risk and improves evaluations, making CX management a route to gaining advantages; customer orientation amplifies satisfaction [94,95,96]. Strategically, effective fulfilment resolves basic needs before higher-order desires, is embedded in corporate operations, and, in livestreaming commerce, requires intelligent, real-time logistics [97,98,99]. Moreover, fulfilment regarding customisation suits sophisticated buyers while standardisation competes on cost and reliability; however, process integration yields seamless experiences and higher perceived value without prohibitive expenses [100,101,102]. Supplier integration improves reliability and delivery, strengthening market and financial outcomes; close, interactive customer relationships reinforce co-created processes; and well-designed fulfilment catalyses CX and loyalty [103,104,105]. Post-purchase fulfilment execution and service define the ultimate experience: accurate processing, timely despatch, intact arrival, and efficient returns are essential and should be assessed alongside service quality [18,106]. Empirically, punctual, dependable logistics and out-of-home options raise satisfaction; swift, convenient returns boost repurchase; and demand for same-/next-day delivery with hassle-free returns is now integral [51,54,107]. Consistency across web, mobile, and physical stores enhances perceived service quality; responsive support mitigates incidents; personable, competent agents encourage revisits and eWOM; rapid responses sustain usage; perceived availability itself builds security; and cohesive cross-channel service consolidates allegiance [5,52,53,108,109,110]. Finally, reliable delivery, precise billing, and effective after-sales reinforce trust and satisfaction and, consistent with ESL, post-purchase experiences validate expectations and shape loyalty [4,111]. Hence, we hypothesise that Fulfilment and Service Quality are positively associated with CX (H5).

2.1.6. Interactivity and Technological Features

Interactivity is central to contemporary customer experience because it enables bi-directional dialogue and heightens consumer empowerment [112]. Unlike unidirectional promotions, digital platforms create immediate feedback loops that deepen engagement and perceived value; accordingly, highly interactive social media adverts are viewed as more useful and entertaining, strengthening purchase intentions [112,113]. Technological affordances can streamline journeys or, if poorly executed, generate frustration [114]. As an example, artificial reality elevates engagement and satisfaction, boosting willingness to buy and aligning convenience with firms’ value propositions [65]. I-T is framed as encompassing advanced, dynamic functionalities that enable bidirectional communication and engagement beyond basic navigation. This distinction is well-supported in the literature [27,115]. Moreover, interactivity powers personalisation and co-creation: real-time preference capture and customer involvement foster tailored offerings, loyalty, profitability, and innovation [116,117,118]. On e-commerce platforms, live chat, 360-degree customer visualisation, and user reviews elicit enthusiasm, involvement, and pleasure; fashion retailers therefore embed Instagram feeds, style quizzes, and virtual try-ons to sustain satisfaction [119,120,121,122]. Social-sharing and multi-device compatibility further unify journeys, while evidence from Romania links advanced features, AI personalisation, and immersive design to engagement, trust, retention, and repurchase [5,123,124]. In parallel, interactive chatbots and community mechanisms reduce perceived risk, nurture social support, and strengthen affinity and loyalty [125,126,127,128]. Yet complexity can erode these gains; thus, simplicity, judicious gamification, and continuous, user-centred testing are essential [129,130,131,132,133]. Overall, combining interactivity with sophisticated technologies renders interactions engaging, personalised, and convenient, whereas misalignment breeds frustration; ongoing optimisation is therefore vital [134]. Therefore, we hypothesise that interactivity and technological functionalities are positively associated with customer experience (H6).

2.2. The Downstream Constructs of CX

2.2.1. Customer Satisfaction

Understanding customer satisfaction constitutes a cornerstone of e-commerce scholarship, since it delineates the extent to which actual outcomes match prior expectations. Previous research showed that customer experience represents the sole direct antecedent of satisfaction within their model; this finding corroborates earlier empirical work according to [3]. Likewise, Singh and Söderlund [135] highlighted online experience as a robust predictor of subsequent satisfaction, which in turn stimulates positive word-of-mouth communication. Researchers also distinguish between transactional satisfaction, linked to a single encounter, and overall satisfaction, which accumulates across multiple touchpoints [3]. Yin and Xu [10] demonstrate that satisfaction garnered over several experiential dimensions significantly affects long-term allegiance. Nevertheless, despite these nuances, the literature consistently recognises customer experience as a decisive determinant of satisfaction, which not only promotes repurchase intentions but also converts patrons into brand advocates, a theme explored in the ensuing sections. Therefore, we theorise that customer experience has a positive association with customer satisfaction (H7).

2.2.2. Customer Loyalty

Customer loyalty, defined as sustained repeat purchasing accompanied by attitudinal commitment, remains pivotal in digital markets where switching costs are minimal [136]. Harris and Goode [137] conceptualise loyalty as comprising affective, cognitive, and conative strata, a taxonomy still widely employed. Recent investigations underscore the mediating role of satisfaction: when Pires, Prisco, Delgado and Santos [3] controlled for satisfaction, the direct path from experience to loyalty lost significance, thereby implying full mediation in certain contexts. Such evidence aligns with earlier theoretical positions that privilege satisfaction as a prerequisite for loyalty [35].

Conversely, a corpus of studies identifies circumstances in which experience exerts a direct influence on loyalty. Williams, Buoye, Keiningham and Aksoy [37] argue that memorable episodes nurture emotional bonds that transcend mere contentment. In hyper-competitive arenas, distinctive experiential elements, such as swift problem resolution, can differentiate one platform from another despite comparable satisfaction levels [10].

The literature also separates attitudinal loyalty from behavioural loyalty. While survey-based metrics capture intentions to recommend a product or service, transactional data reveal genuine repurchase patterns [138]. More comprehensive models now integrate affective, cognitive, and conative components, thereby mirroring loyalty’s multidimensionality in the digital milieu [3,139]. In summary, cultivating an exceptional customer experience remains a strategic conduit to loyalty, whether its effect is mediated by satisfaction or partially direct [140]. Consequently, we hypothesise that customer experience has a positive association with customer loyalty (H8).

2.2.3. Electronic Word of Mouth

eWOM refers to consumers’ propensity to voice opinions via digital channels, such as reviews, social platforms, or thematic forums [141,142]. Positive eWOM operates as a cost-effective acquisition mechanism, amplifying a retailer’s reach. Nhi et al. [143] observe that extraordinary experiences precipitate spontaneous online endorsements, thereby corroborating Anderson’s [144] foundational service research. Singh and Söderlund [135] further determine that experience shapes eWOM indirectly through satisfaction; the higher the delight, the stronger the advocacy.

Negative experiences conversely incite detrimental eWOM, underscoring the necessity for proactive experience management. Taheri et al. [145] show that frictionless service design mitigates complaint volumes. Overall, scholarly consensus since 2018 affirms that enriching customer experience elevates satisfaction, which in turn fuels loyalty and propels eWOM, positioning consumers as authentic brand ambassadors [146]. Subsequently, we theorise that customer experience has a positive association with electronic word of mouth (H9).

2.2.4. Integrative Synthesis

Collectively, the reviewed studies converge on a sequential linkage: superior customer experience enhances satisfaction, which subsequently fortifies loyalty and stimulates positive eWOM [147]. Memorable episodes further magnify this trajectory, given their propensity to be shared widely online [148]. Consequently, firms that invest in holistic experiential design not only secure repeat business but also harness voluntary promotional activity, thereby achieving sustainable competitive advantage [149].

Moreover, drawing upon the preceding literature, it is further posited that customer satisfaction has a positive association with customer loyalty (H10) and customer satisfaction has a positive association with eWOM (H11).

3. Development of Hypotheses and Conceptual Framework

A comprehensive review of the preceding literature yielded a total of eleven research hypotheses, which sought to delineate the relationships between the antecedents of customer experience and its downstream constructs.

Table 1. Research hypotheses, respective constructs and theoretical basis.

Research Hypothesis	Authors
H1: Website usability and design are positively associated with customer experience (U-WD → CX)	[3,31,33]
H2: Omnichannel integration and consistency across channels are positively associated with customer experience (OI → CX)	[21,52,53,54]
H3: Personalisation and customisation are positively associated with customer experience (P-C → CX)	[21,55,77]
H4: The perception of trust, security and privacy are positively associated with customer experience (T-S-P → CX)	[3,16,89,150]
H5: Fulfilment and service quality are positively associated with customer experience (F-SQ → CX)	[5,18,51,107]
H6: Interactivity and technological functionalities are positively associated with the customer experience (I-T → CX)	[5,119,123]
H7: Customer experience is positively associated with customer satisfaction (CX → CS)	[3,4,135]
H8: Customer experience is positively associated with customer loyalty (CX → CL)	[3,10,17]
H9: Customer experience is positively associated with electronic word of mouth (CX → eWOM)	[31,141,143,151]
H10: Customer satisfaction is positively associated with customer loyalty (CS → CL)	[3,10,17]
H11: Customer satisfaction is positively associated with eWOM (CS → eWOM)	[3,135,141]

provides a concise overview of each hypothesis, including the associated construct and the authors who provide theoretical support.

The conceptual model illustrated in Figure 1 was methodically developed, drawing upon the constructs and research hypotheses presented in the preceding table and substantiated by the preceding literature.

4. Materials and Methods

4.1. Nature of the Study and Methodological Approach

The present investigation adopts a quantitative, deductive approach and seeks to validate empirically a conceptual model grounded in the scientific literature. The associations among constructs that shape the e-commerce experience are delineated in the model, and their subsequent association with behavioural variables, namely satisfaction, loyalty, and electronic word of mouth, is designed. In addition, this research takes a descriptive–explanatory orientation because it not only documents the observed phenomena but also elucidates the underlying mechanisms in line with pre-established hypotheses. Data were gathered through a structured questionnaire designed in Google Forms, thereby facilitating broad distribution, easy access, and efficient retrieval of responses.

4.2. Sample Target

The survey directed adults who had completed at least one purchase in an online environment. By eschewing sectoral boundaries, this study sought a panoramic understanding of how customers take their digital shopping experiences, thereby enhancing external validity. This study employed an online questionnaire, and participants were recruited via a university social media community (LinkedIn and Instagram), yielding a non-probability convenience sample. Digital word of mouth further extended the reach and attracted a heterogeneous yet technologically literate audience, albeit without achieving statistical randomness. It is important to note that this expedient strategy precludes probabilistic generalisation to the wider population. To ensure analytical robustness, we adopted Hair et al.’s [152] rule that the minimum sample should equal ten times the greatest number of direct paths to any endogenous construct. Because the construct CX is predicted by six antecedents (H1–H6), at least sixty valid responses were required. Nevertheless, to obtain more stable estimates and stronger statistical power, we sought a minimum of one hundred and fifty complete responses. Only individuals aged at least eighteen who had previously purchased online and who consented voluntarily to complete the web-based instrument were eligible. These conditions were set out at the survey’s outset and enforced through mandatory screening questions in its opening section.

4.3. Data Collection Instrument

The data collection instrument employed in this research study was a structured questionnaire, comprising closed-ended questions and subdivided into three sections. The development of questionnaire was supported in the literature review and the adaptation of scales validated in previous studies, ensuring methodological rigour and the theoretical relevance of the items included.

Data collection was conducted in accordance with the General Data Protection Regulation (GDPR). All participants were informed in advance about the conditions of participation, namely the anonymous and confidential nature of their responses. They were assured that the information provided would be used exclusively for academic research purposes and would be analysed statistically, in aggregate form, and never individually. This ensured the protection of the respondents’ identity and compliance with the ethical and legal principles inherent in the processing of personal data.

Table 2. Research model constructs.

Construct (Type)	Items	Adapted from [Authors]
Usability and Website Design (U-WD) (Formative)	The online shop where I made my purchase is easy to navigate The layout of the elements in the online shop where I made my purchase is clear and intuitive I can easily find the information I’m looking for in the online shop	[3,31,33]
Omnichannel Integration and Consistency (OI) (Reflective)	The online shop offers a consistent experience across all its channels (physical shop, website, social media) Information and offers are consistent across all online shop channels	[21,52,53,54]
Personalisation and Customisation (P-C) (Formative)	When I’m browsing the online shop, I get recommendations for products that are relevant to me The online shop adapts to my browsing preferences I feel that my experience is personalised	[21,55,77]
Trust, Security and Privacy (T-S-P) (Formative)	I feel safe making transactions in this online shop I trust the privacy policies of this shop I have confidence in the company associated with the online shop	[3,16,89,150]
Fulfilment and Service Quality (F-SQ) (Formative)	The products were delivered within the expected time I received exactly what was described in the online shop The post-payment service (e.g., customer support, delivery tracking, exchanges or returns) works efficiently	[5,18,51,107]
Interactivity and Technologies (I-T) (Formative)	The online shop offers useful interactive features (e.g., search filters, zooming in on images, reviews from other customers, customer support chat, among others) Navigation in the online shop is pleasant, fluid and visually appealing	[5,119,123]
Customer Experience (CX) (Reflective)	My overall experience with this online shop was positive I felt comfortable throughout the purchase process I find this online shop pleasant to use	[3,4,135]
Customer Satisfaction (CS) (Reflective)	I am satisfied with my shopping experience at this shop This online shop met my expectations I was satisfied with the service provided	[3,4,135]
Customer Loyalty (CL) (Reflective)	I intend to buy from this online shop again I would recommend this shop to others I consider this shop to be one of my favourites	[3,10,17]
eWOM (Reflective)	I have shared my experiences with this online shop on social media I have already spoken positively about this shop to friends or family I often leave online reviews when satisfied with a purchase	[31,141,143,151]

presents description of each latent construct. It should be noted that applying conventional scales would create an impossible questionnaire with hundreds of items. As a result, it would be impossible to obtain enough responses for construct validation and evaluate the relationship between them. Therefore, this research uses select scales adapted from other authors.

As shown by Rigdon et al. [153] and Hair Jr. et al. [154], reducing the number of indicators per construct increases measurement uncertainty, but it also increases model fit. This is because fewer indicators may not fully capture the construct’s variance, leading to lower reliability and validity [155]. The use of scales with a reduced number of indicators was only due to the reasons mentioned previously and was not intended to obtain an adjusted model. Furthermore, uncertainty can be addressed by repeating the study, as stated by Sarstedt et al. [156].

The fundamental yet often unresolved choice between specifying a construct’s measurement model as reflective or formative necessitates rigorous theoretical justification to prevent mis-specification and ensure the validity of research findings. The revision of each latent construct was performed against the four classic decision rules proposed by Jarvis et al. [157] and subsequently endorsed in PLS-SEM handbooks (e.g., Hair et al. [158]). What follows is a discursive examination of the optimal construct type for the constructs that have been identified as being questionable in nature: U-WB; OI;P-C;T-S-P;F-SQ; and I-T.

Within the domain of U-WD, the construct has been conceptualised either reflectively or formatively. It is important to note that the Human–Computer Interaction community regards usability as a desirable system quality [159]. Nevertheless, representing it reflectively seems inappropriate, and adopting a formative stance remains technically challenging. From a design perspective, usability encompasses evaluates user’ support in achieving online goals efficiently and effectively, as Chang et al. [160] show. These dimensions are frequently operationalised through multidimensional measurement models capturing effectiveness, efficiency and satisfaction [27]. Under formative measurement, items capture heterogeneous facets, remain non-interchangeable, and need not covary, as illustrated by Pee et al. [161].

OI has been treated as a reflective construct in existing studies. However, there is an ongoing discussion in the academic community about whether it should be considered formative, suggesting that more research is needed to definitively categorise it. Therefore, while the current literature leans towards a reflective construct, the debate is not entirely settled [162,163,164].

It is important to note that empirical evidence increasingly indicates that personalisation is better conceptualised within a formative measurement model [165,166]. According to Sarmento, Simões and Lages [165], the formative model is particularly appropriate because the various indicators of personalisation contribute to shaping the construct, rather than merely reflecting it. In fact, this conceptualisation aligns with earlier insights offered by Huang, Goo, Nam and Yoo [166], who similarly underscored the formative nature of personalisation indicators. In addition, research has further substantiated that the indicators of personalisation are formative, given that they actively contribute to the development of the overarching construct, rather than being its outcome [167]. This finding is reinforced by Lee et al. [168], who argued that the formative model is inherently characterised by such causal contributions from indicators to the latent construct. Moreover, Marimuthu [169] elaborated on this relationship, highlighting that the formative approach allows for a more nuanced understanding of how specific personalisation elements coalesce to define the construct itself. As for customization, scholarship to date leans towards a formative first-order representation because indicators often capture heterogeneous elements of preparation, data handling, and interface design, which collectively form the construct [67,170]. Nonetheless, when customisation operates as a sub-dimension of a reflective higher-order factor, or when its indicators genuinely reflect a single underlying attitude, a reflective specification may be appropriate [171,172]. Consequently, model choice should remain contingent upon theoretical justification and empirical verification rather than methodological convention.

Trust is frequently conceptualised as a reflective construct, primarily because its constituent dimensions—such as integrity, ability, and benevolence—are assumed to interact and exhibit mutual covariance, thereby reflecting a shared underlying theme [173,174]. According to Alsaad et al. [175], this reflective approach has become prevalent in numerous empirical studies. Nevertheless, it is important to note that in some contexts, trust has been approached as a second-order formative construct, composed of first-order reflective sub-constructs [176]. This less common approach highlights the flexibility inherent in modelling trust, as it allows researchers to adapt their models to specific research objectives. The constructs of privacy and security exhibit a duality in modelling approaches, as they can be conceptualised either reflectively or formatively. For example, Gao et al. [177] modelled privacy risks using a reflective–formative framework, whereby first-order constructs such as information privacy and relationship privacy collectively form the higher-order construct. In a related vein, privacy inhibitors—comprising dimensions such as privacy concerns and perceived risk—are often modelled formatively, given that these dimensions are not necessarily interchangeable but together define the overarching construct [178]. This observation suggests that the choice between reflective and formative models is contingent upon the nature of the dimensions and their interactions. Therefore, trust is predominantly conceptualised using a reflective approach, whereas privacy and security constructs lend themselves to either reflective or formative modelling. This flexibility underscores the necessity of carefully considering the specific dimensions and their interrelations when determining the most appropriate modelling framework.

Recent work has described service quality as a third-order formative construct [179]. In a related, albeit less complex, design, service quality has been modelled as formatively [180]. More broadly, the literature tends to converge on the view that overall service quality is most appropriately treated as formative; however, research on e-service quality often employs purely reflective indicators [181]. Although some investigations employ reflective indicators for individual facets, the composite construct is still frequently regarded as formative. For instance, the well-known SERVQUAL instrument has been characterised as a type-II reflective–formative second-order model [182]. Therefore, mounting evidence indicates that service quality is conceptually formative at higher levels of abstraction. Additionally, the fulfilment construct can be modelled as either reflective or formative depending on the theoretical framework and research objectives. It is essential to consider the nature of the indicators and the construct’s role in the model to determine the appropriate specification [183,184,185].

As for the Interactivity and Technologies construct, the consensus is that interactivity is best understood and measured as a formative construct, where the indicators define and influence the construct rather than the other way around [20,115]. The

Table 3. Construct specification (formative vs. reflective) and justification using Jarvis, MacKenzie and Podsakoff [157] decision rules.

Construct (Result)	Direction of Causality	Interchangeability	Covariance	Nomological Net
U-WB (Formative)	Indicators define what “good usability” means.	Not interchangeable. “Ease of navigation” is distinct from “clear layout”.	Not required. A site can be beautiful but hard to navigate.	Antecedents/consequences for each dimension can differ.
P-C (Formative)	Indicators (recommendations, adaptation, feeling) form the overall perception of a tailored experience.	Not interchangeable. “Receiving recommendations” is different from “feeling personalised”.	Not required. A user may receive recommendations without feeling the experience is for them.	Drivers of good algorithms are different from drivers of a personal feel.
T-S-P (Formative)	The dimensions of trust, security, and privacy combine to form the overall belief in the retailer’s integrity.	Not interchangeable. “Trust in policies” is conceptually distinct from “feeling safe to transact”.	Not required. A user might trust a brand but doubt its payment security.	Antecedents differ (e.g., reputation builds trust, encryption signals security).
F-SQ (Formative)	Indicators (on-time, as-described, service works) are the definition of good fulfilment.	Not interchangeable. “Timely delivery” does not imply “efficient returns service”.	Not required. A product can arrive on time but be wrong, or vice versa.	Causes of delivery delays are different from causes of poor customer support.
I-T (Formative)	The features create the perception of an interactive and technological experience.	Not interchangeable. “Useful features” is a different concept from “pleasant navigation”.	Not required. A site can have great tools but be clunky and unpleasant to use.	The decision to implement a feature is separate from designing a fluid interface.
OI (Reflective)	The latent trait of “seamless integration” causes the perception of consistency across channels.	Interchangeable. Both items reflect the same core idea of cross-channel harmony.	Required. Perceived consistency and seamless experience should highly correlate.	Both items should share the same antecedents and consequences.

contains the type for each construct, based on Jarvis, MacKenzie and Podsakoff [157]’s four criteria and the indications obtained from the bibliography.

The data were collected online via the Google Forms platform during the month of June 2025. The first part of the questionnaire is designed to characterise the sociodemographic profile of the participants, encompassing variables such as age, gender, educational attainment, professional status, frequency of online purchases, categories of products typically procured, and the time of the most recent online purchase. The second part comprises a series of statements pertaining to the various constructs of the conceptual model. These constructs are operationalised using 5-point Likert scales, ranging from “1—Strongly disagree” to “5—Strongly agree”. This second part comprises sections that have been organised into thematic blocks, corresponding to the following constructs: usability and design of the online shop, Omnichannel Integration and consistency between channels, Personalisation and Customisation of the experience, Trust, Security and Privacy, Fulfilment and Quality of Service, Interactivity and Technologies, customer experience, satisfaction, loyalty and eWOM. The inclusion of open-ended questions aimed to identify the name of the shop and the rationale behind the purchase, thereby facilitating a more nuanced articulation of responses.

The statements were formulated in a clear and neutral manner and organised by construct to guarantee the internal coherence of each dimension and facilitate statistical analysis. The questionnaire was implemented on the Google Forms platform, as it allows for rapid and accessible dissemination in a digital environment, as well as automatic data collection.

5. Results

5.1. Characterisation and Description of the Sample

A total of 360 individuals took part in the study, whose sociodemographic characteristics and digital consumption patterns are presented below (

Table 4. Sample characterisation.

Variables	Categories	Frequency	Percentage
Gender	Male | Female | Prefer not to answer	151 | 207 | 2	41.9% | 57.5% | 0.6%
Age	18–24 | 25–34 | 35–44 | 45–54 | 55+	179 | 89 | 30 | 32 | 30	49.7% | 24.7% | 8.3% | 8.9% | 8.3%
Education	Primary | Secondary | Bachelor’s | Master’s or PhD	4 | 114 | 158 | 84	1.1% | 31.7% | 43.9% | 23.3%

).

The majority of the sample was female (57.5%), followed by male (41.9%), and only 0.6% chose not to indicate. In terms of age, most participants were between 18 and 24 years old (49.7%), followed by the 25 to 34 age group (24.7%). The distribution by education level reveals a majority with higher education: 43.9% have a bachelor’s degree and 23.3% have a master’s degree or doctorate. Regarding professional status, the group of employees stands out (51.4%), followed by students (22.2%) and student–workers (12.2%).

5.2. Results of the PLS-SEM Analysis

To evaluate the proposed conceptual model, the Partial Least Squares Structural Equation Modelling (PLS-SEM) technique was used, using the SmartPLS v. 4.1.1.4 software. This approach was chosen for its robustness in exploratory models with multiple constructs and complex relationships, as advised by Hair et al. [158].

No multivariate outliers (respondent-level anomalies) were identified using factor scores and then, applying robust Mahalanobis distance, grounded in polychoric correlation matrix, as specified by Hair et al. [186]. No missing values were detected. The PLS-SEM approach does not impose specific distribution assumptions; however, it is imperative to ascertain that the data do not exhibit substantial deviations from the normal distribution, as such non-normal data can potentially compromise the determination of parameter significance [158]. Consequently, the variables were subjected to a thorough assessment for skewness and kurtosis. This analysis revealed that none of the variables exhibited asymmetric distributions or highly peaked distributions. The systematic assessment procedure for the outcomes of PLS-SEM is composed of two phases and adheres to the guidance set out by Hair et al. [158]: (i) measurement model evaluation and (ii) structural model evaluation. In phase (i), reflective and formative constructs must be assessed. Reflective constructs include an evaluation of the following: indicator reliability (outer loadings); internal consistency (composite reliability and Cronbach’s α); convergent validity (AVE); and discriminant validity (the HTMT ratio, the Fornell–Larcker criterion and cross-loading). As for the formative constructs, the following assessments must be included: multicollinearity (indicator VIF); convergent validity (redundancy analysis); indicator relevance (weight significance with bootstrapping); and directionally congruency with theory. Phase (ii), structural model evaluation, includes collinearity among latent variables; path significance and relevance; explanatory power (R² values for endogenous constructs and effect-size statistics); predictive relevance (Stone–Geisser Q² values); model fit (SRMR); and direct and indirect effects (mediations) proposed in the research hypotheses.

5.3. Measurement Model Evaluation—Reflective Measurement Model

Table 5. Outer loadings, composite reliability and convergent validity of the constructs.

Constructs	Indicators	Outer Loadings	Cronbach’s Alpha	rho_A	Composite Reliability	AVE
Omnichannel Integration	1	0.89	0.812	0.867	0.912	0.839
	2	0.942
Customer Experience	1	0.953	0.938	0.939	0.96	0.89
	2	0.947
	3	0.93
Customer Satisfaction	1	0.947	0.944	0.944	0.964	0.899
	2	0.944
	3	0.953
Customer Loyalty	1	0.932	0.894	0.911	0.934	0.826
	2	0.948
	3	0.843
WOM	1	0.917	0.761	0.779	0.892	0.806
	3	0.877

presents the factor loadings of the indicators, in addition to the internal reliability and convergent validity coefficients for each construct in the model.

Table 5 contains the psychometric properties of the five reflective constructs retained in the study. All outer loadings are well above the 0.70 benchmark [158], ranging from 0.843 (CL-3) to 0.953 (CX-1), indicating that each indicator shares more than 50% of its variance with the underlying latent variable. Internal consistency metrics likewise surpass accepted thresholds: Cronbach’s α varies between 0.761 (eWOM) and 0.944 (customer satisfaction), while composite reliability (CR), the preferred index in PLS-SEM, lies between 0.892 and 0.964, comfortably exceeding the recommended minimum of 0.70.

Convergent validity is confirmed through the average variance extracted (AVE). All constructs display AVE values above the 0.50 cut-off [187], ranging from 0.806 for eWOM to 0.899 for customer satisfaction, thus demonstrating that, on average, each latent variable explains more than half of the variance in its indicators. The only purification required concerned the electronic word-of-mouth (eWOM) scale: the removal of item 2 (loading = 0.982) increased its CR from 0.694 to 0.892 and AVE from 0.466 to 0.806, while Cronbach’s α stabilised at 0.761. Such two-indicator constructs are acceptable when reliability and validity criteria are met [188].

The evidence pertaining to the reliability of the indicators, the internal consistency of the indicators, and the convergent validity of the indicators is indicative of the satisfactory psychometric quality of the reflective measurement model. This permits the subsequent evaluation of the structural relationships hypothesised in the study.

The discriminant validity of the model is assessed using the Fornell–Larcker criterion and the HTMT (Heterotrait–Monotrait) ratio. The values pertaining to the Fornell–Larcker are displayed in

Table 6. The Fornell–Larcker matrix.

	CL	CS	CX	OI	eWOM
CL	0.909
CS	0.813	0.948
CX	0.722	0.837	0.943
OI	0.408	0.455	0.472	0.916
eWOM	0.091	0.049	0.084	0.072	0.898

.

The matrix supplied lists the square root of the AVE for the five reflective constructs (CL, CS, CX, OI, and eWOM) on the diagonal and shows the construct–construct correlations in the off-diagonal cells. An interpretation of the reflective block is provided in

Table 7. Interpretation for the Fornell–Larcker matrix.

Highest Correlation	Correlation	Evaluation	Comment
CS–CX	0.837	0.837 < 0.948 (CS) and 0.943 (CX)	Acceptable but high; flag for HTMT check
CS–CL	0.813	0.813 < 0.948 (CS) and 0.909 (CL)	Acceptable
CX–CL	0.722	0.722 < 0.943 and 0.909	Acceptable
Other correlations	≤0.625	All < relevant √AVE	Acceptable

.

All reflective constructs pass the Fornell–Larcker criterion. The closest call is the correlation between customer satisfaction and customer experience (0.837), which is still comfortably below either construct’s square root of the AVE but merits a confirmatory HTMT test. Therefore, what follows is a display of the HTMT ratio, as depicted in

Table 8. HTMT (Heterotrait–Monotrait) ratio.

	CL	CS	CX	OI	eWOM
CL
CS	0.88
CX	0.782	0.89
OI	0.473	0.51	0.529
eWOM	0.153	0.062	0.101	0.088

.

The HTMT ratio is the preferred metric for assessing discriminant validity. All pair-wise HTMT values fall below the conservative threshold of 0.90 recommended for conceptually related constructs and well below the stricter 0.85 benchmark for conceptually distinct constructs [189]. The highest ratio, observed between customer satisfaction (CS) and customer experience (CX), is 0.89; although comparatively elevated, it remains safely under the threshold, indicating these constructs are empirically distinct. The application of the Fornell–Larcker matrix in conjunction with the HTMT ratio provided substantial evidence that discriminant validity has been successfully established across all latent variables. This finding serves to effectively mitigate any concerns pertaining to the occurrence of construct overlap within the measurement model. Cross-loadings are also presented and discussed in Appendix A.

5.4. Measurement Model Evaluation—Formative Measurement Model

Severe multicollinearity among formative indicators inflates standard errors, destabilises weight estimates, and can lead to spurious conclusions about indicator relevance [158,190]. Consequently, routine collinearity diagnostics based on the variance inflation factor (VIF) are indispensable when evaluating formative blocks [191,192]. Contemporary PLS-SEM guidelines advise that VIFs should preferably remain below 3; values above this mark warrant closer scrutiny, while values exceeding 5 signal problematic collinearity that calls for remedial action such as merging or removing indicators [158,190].

Table 9. Collinearity statistics for the indicators of the formative constructs.

Item	U-WD	P-C	T-S-P	F-SQ	I-T
1	2.656	2.526	2.586	1.886	1.599
2	3.393	2.722	3.131	1.837	1.599
3	3.025	2.517	3.855	1.431

contains the outer VIF values for the indicators of the formative constructs.

The outer VIF diagnostics show that none of the formative indicators exceeds the critical cut-off of 5 that signals harmful multicollinearity. Most VIFs fall below the often-recommended cautionary ceiling of 3.0. The only exceptions are U-WD-2 (3.393), U-WD-3 (3.025), T-S-P-2 (3.131), T-S-P-3 (3.855). Although these four values are slightly above the conservative threshold, they remain well under 5. Hence, the formative indicators do not exhibit collinearity severe enough to inflate standard errors or destabilise weight estimates, and no remedial action (e.g., indicator removal, merging, or residual-centring) is required. All remaining indicators post VIFs comfortably under 3, further confirming that multicollinearity is not a substantive concern for any of the formative blocks, so the analysis can proceed to interpret indicator weights and structural paths with confidence.

Bootstrapping was performed (10,000 subsamples, bias-corrected and accelerated bootstrap confidence intervals) and it reveals that most formative indicators contribute significantly to their respective composites (|t| > 1.96, p < 0.05), but a few do not (

Table 10. Outer-weight results for the formative blocks.

Formative Construct	Significant Indicators (p < 0.05)	Non-Significant Indicators (p ≥ 0.05)	Outer Loadings *		Comment Retain/Drop?
			Lowest	Highest
Trust–Security–Privacy (T-S-P)	1 (β = 0.509, t = 3.73) 2 (β = 0.501, t = 3.83)	3 (β = 0.076, t = 0.47)	0.878	0.924	Even 3 (weight not significant earlier) loads very strongly, so retain.
Fulfilment and Service Quality (F-SQ)	1–3 all significant (t ≥ 3.16)	—	0.786	0.879	Retain all.
Interactivity and Technology (I-T)	1 (β = 0.200, t = 2.68) 2 (β = 0.865, t = 15.32)	—	0.730	0.987	Retain all.
Personalisation and Customisation (P-C)	2 (β = 0.872, t = 4.48)	1 (β = 0.091, t = 0.40) 3 (β = 0.077, t = 0.35)	0.783	0.996	1 and 3 load strongly, justifying their inclusion despite non-significant weights. Retain all
Usability and Web Design (U-WD)	1 (β = 0.632, t = 7.04) 3 (β = 0.359, t = 2.65)	2 (β = 0.087, t = 0.63)	0.855	0.960	All three items are robust. Retain all.

* Retain non-significant weights when their outer loading is ≥0.50 or when strong substantive arguments exist (the item is theoretically essential); otherwise, remove to avoid diluting the composite [158,192].

).

For formative constructs such as T-S-P, P-C and U-WD, indicator weights are the primary criterion but given the loadings above, even indicators with marginal weights possess strong absolute contributions and should remain in the model to preserve content validity. Therefore, the outer loading results corroborate the excellent psychometric quality of both reflective and formative measurement models. No item shows problematic reliability; hence, all indicators can be retained for the subsequent structural model assessment.

We acknowledge as a limitation that we could not conduct redundancy analyses for the formative indices because our instrument lacks independent global single-item criteria.

5.5. Structural Model Evaluation

Table 11. Inner VIF values.

	CL	CS	CX	F-SQ	I-T	OI	P-C	T-S-P	U-WD	eWOM
CL
CS	3.348									3.348
CX	3.348	1								3.348
F-SQ			1.821
I-T			1.954
OI			1.566
P-C			1.241
T-S-P			1.645
U-WD			1.877
eWOM

summarises the values of the inner-model VIFs to flag possible multicollinearity among the latent variables that jointly predict an endogenous construct.

The following conclusions can be drawn from the values presented in Table 10:

• Customer Loyalty: Both CX and customer satisfaction are moderately correlated as drivers of loyalty (3.348—CX and CS). A VIF above the conservative caution level of 3.0 but below the critical 5.0 indicates no harmful collinearity; but results should be interpreted with an awareness of the overlap.
• Customer Satisfaction: CX is the sole predictor, so collinearity is irrelevant (VIF = 1).
• CX: In the current model CX is endogenous only to the five formative blocks, all with VIFs < 2 (see below).
• CX Antecedents (Predicting CX): Usability and Web design (U-WD), Fulfilment and Service Quality (F-SQ), Interactivity and Technology (I-T), Personalisation and Customisation (P-C) and Trust–Security–Privacy (T-S-P) all post VIFs well below 3 (1.241–1.954), indicating negligible collinearity.

Therefore, all inner VIFs are <5.0, the generally accepted ceiling for PLS-SEM. Additionally, only the loyalty equation shows VIFs slightly above the 3.0 caution guideline. This is expected because satisfaction and experience are conceptually close. The values do not reach problematic territory, so no remedial action (e.g., deleting a predictor or orthogonalising) is required. All other paths exhibit comfortable VIF margins, confirming that multicollinearity will not inflate standard errors or distort the significance tests. Hence, the structural estimates can be interpreted confidently with respect to the unique contribution of each latent predictor.

Table 11 summarises the structural model results. Following current PLS-SEM guidelines, we used the path-weighting scheme and ran a 10,000-subsample bootstrap with bias-corrected and accelerated confidence intervals to obtain standard errors, t-values, and two-tailed p-values for every hypothesised path. The table reports the standardised coefficients (β), their bootstrap statistics, and an indication of whether each hypothesis is supported at the 5% significance level. Because all inner VIFs fall below the critical threshold of 5, collinearity does not distort the estimates; thus, the coefficients in

Table 12. Path coefficients.

Constructs	Original Sample	Sample Mean	Standard Deviation	T Statistics (|O/STDEV|)	p Values	Comment
CS → CL	0.7	0.702	0.091	7.724	0	Supported
CS → eWOM	−0.07	−0.067	0.147	0.475	0.635	Not supported
CX → CL	0.135	0.133	0.091	1.481	0.139	Not supported
CX → CS	0.837	0.836	0.032	26.489	0	Supported
CX → eWOM	0.143	0.142	0.132	1.082	0.279	Not supported
F-SQ → CX	0.218	0.219	0.057	3.854	0	Supported
I-T → CX	0.28	0.274	0.055	5.077	0	Supported
OI → CX	0.01	0.008	0.04	0.238	0.812	Not supported
P-C → CX	0.092	0.095	0.033	2.804	0.005	Supported
T-S-P → CX	0.223	0.225	0.053	4.229	0	Supported
U-WD → CX	0.215	0.218	0.058	3.719	0	Supported

can be interpreted as the unique effects of each antecedent on its respective outcome.

The values obtained from bootstrapping show that seven of the eleven hypothesised links attain statistical significance at p < 0.05, thereby being supported, whereas four paths remain non-significant and are not supported.

Starting from the down-stream outcomes, customer satisfaction exerts a large positive effect on customer loyalty (β = 0.700; t = 7.724), explaining much of the variance in loyalty by itself. By contrast, satisfaction has no discernible influence on electronic word of mouth (β = −0.070; n.s.). Customer experience strongly predicts satisfaction (β = 0.837; t = 26.489) but does not directly influence loyalty (β = 0.135; n.s.) or eWOM (β = 0.143; n.s.), suggesting that the relationship between CX and behavioural outcomes is statistically accounted for by satisfaction.

As for the antecedents of customer experience, six formative composites jointly explain CX. The most potent drivers are Interactivity and Technology (β = 0.280; t = 5.077) and Trust–Security–Privacy (β = 0.223; t = 4.229), both exhibiting medium-sized effects. Fulfilment and Service Quality (β = 0.218) and Usability and Web Design (β = 0.215) also make significant contributions, while Personalisation and Customisation has a smaller yet still significant impact (β = 0.092; t = 2.804). Conversely, Omnichannel Integration shows no meaningful effect on CX (β = 0.010; n.s.). The pattern indicates that technological richness, transactional reliability, and interface quality are the principal levers for enhancing online customer experience, which in turn elevates satisfaction and, indirectly, loyalty. The absence of direct CX → CL and CX → eWOM paths underscores the mediating role of satisfaction in driving post-purchase behaviours.

To complement the direct-path results, we next report the specific indirect effects derived from the bias-corrected and accelerated bootstrap procedure. Presenting these coefficients is essential for evaluating each hypothesised mediation chain individually and for determining whether the intervening construct fully or partially transmits the influence of the antecedent to the outcome.

Table 13. Specific indirect effects for each individual mediation path.

Constructs	Original Sample	Sample Mean	Standard Deviation	T Statistics (|O/STDEV|)	p Values	Comment
U-WD → CX → eWOM	0.031	0.031	0.031	0.977	0.329	Supported
U-WD → CX → CS → eWOM	−0.013	−0.012	0.029	0.44	0.66	Not supported
U-WD → CX → CS → CL	0.126	0.128	0.04	3.171	0.002	Supported
U-WD → CX → CS	0.18	0.182	0.049	3.638	0	Supported
U-WD → CX → CL	0.029	0.029	0.022	1.332	0.183	Supported
T-S-P → CX → eWOM	0.032	0.032	0.031	1.015	0.31	Supported
T-S-P → CX → CS → eWOM	−0.013	−0.013	0.029	0.448	0.654	Not supported
T-S-P → CX → CS → CL	0.131	0.132	0.037	3.529	0	Supported
T-S-P → CX → CS	0.187	0.188	0.045	4.118	0	Supported
T-S-P → CX → CL	0.03	0.03	0.022	1.355	0.176	Supported
P-C → CX → eWOM	0.013	0.014	0.014	0.938	0.348	Supported
P-C → CX → CS → eWOM	−0.005	−0.005	0.013	0.433	0.665	Not supported
P-C → CX → CS → CL	0.054	0.055	0.02	2.704	0.007	Supported
P-C → CX → CS	0.077	0.079	0.027	2.831	0.005	Supported
P-C → CX → CL	0.012	0.013	0.011	1.175	0.24	Supported
OI → CX → eWOM	0.001	0.001	0.008	0.171	0.864	Not supported
OI → CX → CS → eWOM	−0.001	0	0.006	0.097	0.923	Not supported
OI → CX → CS → CL	0.006	0.006	0.024	0.235	0.814	Not supported
OI → CX → CS	0.008	0.007	0.034	0.238	0.812	Not supported
OI → CX → CL	0.001	0	0.006	0.2	0.842	Not supported
I-T → CX → eWOM	0.04	0.039	0.037	1.075	0.282	Supported
I-T → CX → CS → eWOM	−0.016	−0.015	0.034	0.478	0.632	Not supported
I-T → CX → CS → CL	0.164	0.162	0.041	3.955	0	Supported
I-T → CX → CS	0.234	0.229	0.046	5.104	0	Supported
I-T → CX → CL	0.038	0.035	0.025	1.522	0.128	Supported
F-SQ → CX → eWOM	0.031	0.031	0.031	1.003	0.316	Supported
F-SQ → CX → CS → eWOM	−0.013	−0.012	0.028	0.453	0.651	Not supported
F-SQ → CX → CS → CL	0.128	0.129	0.038	3.349	0.001	Supported
F-SQ → CX → CS	0.183	0.184	0.049	3.752	0	Supported
F-SQ → CX → CL	0.029	0.029	0.022	1.315	0.189	Supported
CX → CS → eWOM	−0.059	−0.056	0.124	0.471	0.637	Not supported
CX → CS → CL	0.586	0.587	0.083	7.076	0	Supported

lists, for every predictor–mediator–criterion combination, the standardised indirect effect (β), its bootstrap-based standard error, t-value, and two-tailed p-value, together with an indication of support for the corresponding mediation hypothesis.

Bootstrapping reveals consistency with a two-step indirect path. All service–interface antecedents—Usability and Web Design (U-WD), Trust–Security–Privacy (T-S-P), Personalisation and Customisation (P-C), Interactivity and Technology (I-T), and Fulfilment and Service Quality (F-SQ)—show statistically significant indirect associations with customer loyalty via the sequential chain CX → CS → CL (β = 0.126–0.164, p ≤ 0.007). Each antecedent also displays a more proximate, single-step mediation through CX → CS (β = 0.077–0.234, p ≤ 0.005), underscoring customer satisfaction as the immediate conduit through which customer experience improvements take hold.

In contrast, Omnichannel Integration (OI) does not produce any significant indirect influence along either the single- or double-mediation routes (all p > 0.80), mirroring its non-significant direct effect on CX reported earlier.

Regarding electronic word of mouth, none of the double-mediation paths (CX → CS → eWOM) reach significance (p > 0.63). Likewise, the single-step paths from each antecedent to eWOM via CX remain non-significant (p = 0.17–0.35). These findings suggest that although interface and service quality enhancements cascade through CX and CS to foster loyalty, they do not translate into a higher propensity to share positive eWOM within the present sample. Considering all the previously outlined assertions, the results confirm customer satisfaction as the pivotal mediator that converts experiential gains into behavioural loyalty, while eWOM appears to be driven by other factors beyond the scope of the current model.

We provide a holistic view of each construct’s overall influence, that is, the sum of all direct and indirect pathways linking every predictor to its respective outcomes, in the next table. Presenting these aggregate coefficients is recommended in current PLS-SEM guidelines because they reveal a construct’s net impact and facilitate importance ranking for theory testing and managerial prioritisation [158].

Table 14. Total effects for every antecedent–outcome pair.

Constructs	Original Sample	Sample Mean	Standard Deviation	T Statistics (|O/STDEV|)	p Values	Comment
U-WD → eWOM	0.018	0.019	0.014	1.289	0.197	Supported
U-WD → CX	0.215	0.218	0.058	3.719	0	Supported
U-WD → CS	0.18	0.182	0.049	3.638	0	Supported
U-WD → CL	0.155	0.157	0.044	3.552	0	Supported
T-S-P → eWOM	0.019	0.019	0.014	1.355	0.176	Supported
T-S-P → CX	0.223	0.225	0.053	4.229	0	Supported
T-S-P → CS	0.187	0.188	0.045	4.118	0	Supported
T-S-P → CL	0.161	0.162	0.04	3.993	0	Supported
P-C → eWOM	0.008	0.008	0.007	1.19	0.234	Supported
P-C → CX	0.092	0.095	0.033	2.804	0.005	Supported
P-C → CS	0.077	0.079	0.027	2.831	0.005	Supported
P-C → CL	0.067	0.068	0.024	2.764	0.006	Supported
OI → eWOM	0.001	0.001	0.004	0.194	0.846	Not Supported
OI → CX	0.01	0.008	0.04	0.238	0.812	Not Supported
OI → CS	0.008	0.007	0.034	0.238	0.812	Not Supported
OI → CL	0.007	0.006	0.029	0.239	0.811	Not Supported
I-T → eWOM	0.023	0.024	0.017	1.384	0.166	Supported
I-T → CX	0.28	0.274	0.055	5.077	0	Supported
I-T → CS	0.234	0.229	0.046	5.104	0	Supported
I-T → CL	0.202	0.197	0.038	5.33	0	Supported
F-SQ → eWOM	0.018	0.019	0.015	1.262	0.207	Supported
F-SQ → CX	0.218	0.219	0.057	3.854	0	Supported
F-SQ → CS	0.183	0.184	0.049	3.752	0	Supported
F-SQ → CL	0.157	0.158	0.042	3.706	0	Supported
CX → eWOM	0.084	0.087	0.058	1.446	0.148	Supported
CX → CS	0.837	0.836	0.032	26.489	0	Supported
CX → CL	0.722	0.72	0.04	18.265	0	Supported
CS → eWOM	−0.07	−0.067	0.147	0.475	0.635	Not Supported
CS → CL	0.7	0.702	0.091	7.724	0	Supported

lists, for every antecedent–outcome pair, the standardised total effect (β), its bootstrap standard error, t-value, and two-tailed p-value.

An inspection of the aggregate coefficients, each combining all direct and mediated influences, corroborates the pivotal role of customer experience (CX) and customer satisfaction (CS) in shaping downstream outcomes. CX exerts the single largest total effect on satisfaction (β = 0.837, t = 26.49, p < 0.001) and, together with CS (β = 0.700, t = 7.72, p < 0.001), accounts for the bulk of variance in customer loyalty (CL), with CX’s overall impact (β = 0.722, t = 18.27) marginally exceeding that of CS.

Among the service–interface antecedents, Interactivity and Technology delivers the strongest total influence on CX (β = 0.280, p < 0.001) and, by extension, on both satisfaction (β = 0.234) and loyalty (β = 0.202). It is followed closely by Trust–Security–Privacy (βs = 0.223 → CX; 0.187 → CS; 0.161 → CL) and Fulfilment and Service Quality (βs ≈ 0.218 → CX; 0.183 → CS; 0.157 → CL). Usability and Web Design contributes comparably (β = 0.215 → CX; 0.155 → CL), whereas Personalisation and Customisation shows a more modest but still significant chain of effects (β = 0.092 → CX; 0.067 → CL). In contrast, Omnichannel Integration remains non-significant across all outcomes, echoing its earlier null direct and indirect paths.

With respect to electronic word of mouth (eWOM), none of the antecedent composites nor CX itself reach conventional significance thresholds (all p > 0.10), and CS even exhibits a trivial, non-significant negative total effect (β = −0.070, p = 0.64). These findings confirm that while the experiential and satisfaction pathways are decisive for fostering loyalty, they do not automatically translate into customers’ propensity to spread positive eWOM within the present context.

Table 15. Explanatory power for the conceptual model.

Explanatory Power	Metrics	CL	CS	CX	eWOM
R Square	Original Sample (O)	0.667	0.701	0.626	0.009
	Sample Mean (M)	0.669	0.7	0.638	0.019
	p Values	0	0	0	0.421
R Square Adjusted	Original Sample (O)	0.665	0.7	0.62	0.003
	Sample Mean (M)	0.667	0.699	0.631	0.013
	p Values	0	0	0	0.78

documents how well the structural model accounts for variance in each key outcome.

The coefficients of determination (R²) reveal that the model explains around two-thirds of the variance in customer loyalty (R² = 0.667) and customer satisfaction (R² = 0.701) and just over three-fifths of the variance in customer experience itself (R² = 0.626). After adjusting for model complexity, the explanatory power remains virtually unchanged, underscoring the model’s parsimony (adjusted R² between 0.620 and 0.700). By contrast, electronic word of mouth is barely explained (R² = 0.009; p > 0.40), signalling that additional drivers, beyond those captured here, govern customers’ propensity to share their experiences online. These results position the model’s explanatory ability as moderate to substantial for loyalty, satisfaction and experience, while highlighting an important gap with respect to eWOM that future research could address.

Table 16. Effect size (f²) in the structural model relationships.

	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values	Comment
CS → CL	0.440	0.463	0.157	2.810	0.005	Large
CS → eWOM	0.001	0.008	0.009	0.163	0.87	-
CX → CL	0.016	0.023	0.026	0.622	0.534	-
CX → CS	2.348	2.438	0.619	3.793	0	Large
CX → eWOM	0.006	0.011	0.010	0.625	0.532	-
F-SQ → CX	0.070	0.078	0.041	1.709	0.087	Small
I-T → CX	0.107	0.110	0.047	2.267	0.023	Small
OI → CX	0.000	0.003	0.004	0.037	0.97	-
P-C → CX	0.018	0.022	0.015	1.267	0.205	-
T-S-P → CX	0.081	0.089	0.041	1.990	0.047	Small
U-WD → CX	0.066	0.073	0.037	1.792	0.073	Small

gauges the practical importance of every structural path by reporting Cohen’s f² effect sizes.

Table 16’s Cohen f² results make clear which links in the structural model really matter. CX is the powerhouse: removing it from the CX → customer satisfaction (CS) path would slash the model’s explanatory power for CS by more than two full R-square units (f² = 2.348, p < 0.001), a massive effect. Satisfaction, in turn, is the only driver that has a large practical impact on customer loyalty (CL) (f² = 0.440, p = 0.005). Every other relationship shows, at best, a small incremental value. Three service–interface blocks—Interactivity and Technology (I-T, f² = 0.107, p = 0.023), Trust–Security–Privacy (T-S-P, f² = 0.081, p = 0.047) and Fulfilment and Service Quality (F-SQ, f² = 0.070, p = 0.087)—make modest yet non-trivial contributions to shaping CX. Usability and Web Design (U-WD) just reaches the small-effect threshold (f² = 0.066, p = 0.073), while Personalisation and Customisation (P-C) stays below it (f² = 0.018, n.s.), and Omnichannel Integration (OI) adds virtually nothing (f² ≈ 0). Finally, none of the modelled antecedents, neither CX nor CS, provide meaningful incremental explanation for electronic word of mouth (all f² ≤ 0.006, p > 0.50). Therefore, the data spotlight a tight two-step chain (CX → CS → CL) where CX is strongly shaped by a handful of interface factors, but eWOM remains unexplained by the current set of predictors.

Table 17. Stone–Geisser Q² via blindfolding.

	PLS		LM
Construct	RMSE	Q2_Predict	RMSE	Q2_Predict
CL 1	0.57	0.393	0.557	0.42
CL 2	0.579	0.423	0.584	0.413
CL 3	0.86	0.3	0.872	0.28
CS 1	0.495	0.509	0.494	0.511
CS 2	0.516	0.501	0.52	0.494
CS 3	0.454	0.518	0.454	0.519
CX1	0.477	0.536	0.485	0.52
CX2	0.485	0.535	0.488	0.527
CX3	0.52	0.522	0.528	0.507
eWOM3	1.397	0.008	1.364	0.055
eWOM1	1.291	0.002	1.244	0.073

documents the predictive-relevance step of the PLS-SEM evaluation. Specifically, it lists the Stone–Geisser Q² statistics and associated RMSE values obtained through the blindfolding procedure for each indicator, juxtaposed with the same metrics from a simple linear model (LM) benchmark.

For customer loyalty (CL), all three CL indicators show positive Q²_predict values (0.300–0.423), comfortably above the 0.25 “medium” relevance threshold. For two of the three indicators (CL-2 and CL-3), the PLS RMSE is lower than the LM RMSE, while the reverse is true for CL-1. Overall, the loyalty block achieves medium predictive power. As for customer satisfaction (CS), Q²_predict values exceed 0.50 on every item (0.501–0.518), signalling large predictive relevance. PLS errors are on par with, or very slightly better than, the LM baseline except for CS-1, where LM has a negligible edge. The block therefore combines strong relevance with adequate accuracy. Moving to the customer experience (CX), each CX indicator posts a Q²_predict above 0.52, again in the large range. The PLS RMSE is consistently lower than the LM (e.g., CX-1: 0.477 vs. 0.485), confirming that the structural model adds tangible predictive value for experience outcomes. In the other extreme stands electronic word of mouth (eWOM). Both eWOM indicators yield Q²_predict values that hover around zero (0.002–0.008), well below the threshold for even small relevance, and PLS errors are higher than the LM. The current drivers therefore fail to predict customers’ propensity to engage in eWOM. Therefore, the model-level verdict, and supported by the counting indicators where PLS outperforms the LM (6 of 11), places the model in the medium predictive power category according to recent PLSpredict guidelines. Predictive strength is concentrated in the CX → CS → CL chain; the eWOM outcome remains practically unpredictable, highlighting a gap for future model refinement.

Table 18. Global model fit.

	Saturated Model	Estimated Model
SRMR	0.042	0.064
d_ULS	0.653	1.559
d_G	0.429	0.489
Chi-Square	931.238	1029.472
NFI	0.886	0.874
rms Theta	0.179

provides the final step in the PLS-SEM reporting sequence: the overall verification of the adequacy of the research model. This allows us to assess whether the specified measurement and structural models, when considered together, provide an acceptable reproduction of the empirical covariance matrix. This is important before interpreting the substantive results.

Table 18 summarises the global fit check for the PLS-SEM model, juxtaposing the unrestricted (saturated) solution with the fully specified (estimated) model. For the standardised root-mean-square residual (SRMR), both the saturated (0.042) and estimated (0.064) values sit comfortably below the 0.08 guideline often used in PLS-SEM, signalling that, even after the structural constraints are imposed, the reproduced correlations deviate only modestly from the observed ones.

However, the Discrepancy based on Unweighted Least Squares (d_ULS) value of 1.559 indicates some level of discrepancy between the sample and model-implied covariance matrices, suggesting room for improvement in the model fit. The Geodesic Discrepancy (d_G) value of 0.489 further supports the notion that there is some divergence between the observed data and the proposed model, though it remains within a reasonable range.

As for the Normed Fit Index (NFI), the index drops slightly from 0.886 (saturated) to 0.874 (estimated). Although it misses the conventional 0.90 heuristic borrowed from CB-SEM, values greater than 0.80 → are often deemed acceptable in composite-based models, especially when predictive accuracy, as opposed to exact covariance reproduction, is the primary goal. The Chi-square exact-fit test was conducted. The χ² statistic naturally inflates as constraints are added (931.238 → 1029.472). Given PLS-SEM’s prediction orientation and the large sample, the chi-square significance test is not decisive evidence of misfit.

Lastly, and for RMStheta, the value of 0.179 indicates a moderate average difference between the observed and predicted correlations, again suggesting that while the model fits the data reasonably well, some improvements could enhance its accuracy.

As an overall assessment, the most important SRMR statistic affirms an acceptable global fit, and the two discrepancy measures do not raise red flags. The NFI is close to, but not quite at, the typical 0.90 benchmark, while RMStheta advises some caution. Therefore, the model demonstrates adequate but not perfect covariance reproduction, well within the range generally considered satisfactory for variance-oriented PLS-SEM studies, so subsequent interpretation can proceed while noting the minor residual misfit.

Table 19. Results of the research hypotheses.

H	Constructs	Original Sample (B)	Sample Mean (B)	Standard Deviation	T Statistics	p Values	Results
H1	U-WD → CX	0.215	0.218	0.057	3.742	0	Supported
H2	OI → CX	0.010	0.008	0.04	0.239	0.812	Not supported
H3	P-C → CX	0.092	0.095	0.033	2.791	0.005	Supported
H4	T-S-P → CX	0.223	0.225	0.053	4.215	0	Supported
H5	F-SQ → CX	0.218	0.218	0.057	3.803	0	Supported
H6	I-T → CX	0.280	0.274	0.055	5.116	0	Supported
H7	CX → CS	0.837	0.836	0.031	27.122	0	Supported
H8	CX → CL	0.135	0.134	0.09	1.493	0.139	Not supported
H9	CX → eWOM	0.143	0.142	0.131	1.085	0.279	Not supported
H10	CS → CL	0.700	0.701	0.09	7.789	0	Supported
H11	CS → eWOM	−0.070	−0.067	0.146	0.478	0.635	Not supported

consolidates the structural model hypothesis testing stage of this research. It lists every theorised relationship (H1–H11) together with the estimated path coefficient, its confidence interval, and associated t- and p-values. The final column provides a concise summary of whether each hypothesis is supported or not based on the observed data.

Several key insights emerge from the table. First, the relationships between usability and website design (U-WD), personalization and customization (P-C), Trust–Security–Privacy (T-S-P), Fulfilment and Service Quality (F-SQ), and Interactivity and Technology (I-T) with customer experience (CX) are all statistically significant, as evidenced by their p-values being less than 0.05. This indicates that these factors have a meaningful impact on CX. However, the relationship between Omnichannel Integration (OI) and CX is not supported by the data, indicating that OI may not significantly influence CX in this particular context.

The most significant observation of this study is that experience converts almost one to one into satisfaction. The huge coefficient (0.837) shows that if one can lift perceived CX by one standard deviation, satisfaction jumps by roughly the same amount, underscoring CX’s centrality. Therefore, this relationship is strongly supported.

Another relevant insight is related to satisfaction: Satisfaction, not raw experience, locks in loyalty. Once CS is in the model, CX’s direct link to loyalty dies (H8), confirming full mediation: the emotional appraisal (satisfaction) is what cements repeat-purchase intentions. As for eWOM, it remains an enigma. Neither CX nor CS explains customers’ willingness to recommend online products or services. The non-significant, even slightly negative, CS → eWOM path (β = −0.070) suggests other motives, such as social capital seeking, brand enthusiasm, and surprise, among others, may drive digital word of mouth. This mirrors the minuscule R² for eWOM reported earlier and flags a research gap.

6. Discussion

U-WD has a significant yet modest effect on CX. Its influence on loyalty is indirect, channelled through CX and then satisfaction, rather than a direct behavioural effect. The lack of any link to eWOM shows that better U-WD alone does not stimulate advocacy. This pattern aligns with the hygiene factor view, necessary to avoid dissatisfaction but insufficient to trigger enthusiastic promotion, and with evidence linking usability to satisfaction and loyalty, as the literature supports [14,24]. Notwithstanding recent evidence linking usability to eWOM [3,31], the null effect likely reflects (i) omitted social–motivational drivers (not modelled here); (ii) information density that dampens satisfaction and offsets usability gains [33]; and (iii) UX ceiling effects, whereby widespread baseline competence yields only marginal CX improvements.

OI was non-significant across outcomes in our model, despite the extensive literature support. A plausible explanation is a ceiling (hygiene) effect: for a digitally native sample, baseline integration may be sufficiently high that variance in perceived integration is limited, muting its explanatory power relative to more proximal levers. OI is systemic and often invisible when functioning well. We therefore interpret OI as a context-setting capability, necessary for service fluency but insufficient to shift experiential evaluations unless it translates into perceived experiential coherence that customers can notice and value. The null paths also suggest measurement mis-specification: many integration scales privilege technical connectivity over experiential alignment. The non-significant finding for the Omnichannel Integration path suggests that our operationalization of OI as a holistic, first-order perception may have been too simplistic. Future research should reconceptualize OI as a formative higher-order construct, potentially comprising distinct dimensions such as technical, informational, and experiential alignment. This would allow researchers to test which specific facet of integration truly influences customer experience. As a summary, the following explanations are postulated: firstly, the absence of variation in omnichannel usage among respondents; secondly, measurement mis-specification; and thirdly, ceiling effects.

P-C exerts a statistically significant yet modest effect on CX; its influence on loyalty is indirect, travelling via the CX → CS → CL chain, while links to eWOM remain non-significant. In ESL terms, P-C shapes customers’ expectations pre-purchase and their confirmation across the journey; under SDL, personalisation is business-led, data-driven tailoring, whereas customisation transfers control to the customer. When executed transparently and with robust privacy safeguards, both mechanisms enhance satisfaction and loyalty [58,59,60,63], and personalisation may seed deeper customisation and longer-term commitment [69]. Given heterogeneity, the construct is best modelled formatively, with indicators that shape rather than reflect the latent variable [165,166,167]. Within customer journey mapping, P-C operates most visibly at discovery/choice and usage stages. Compared with post-pandemic reports of stronger P-C effects [21,55], our weaker coefficients likely reflect the following: (i) under-specification—a formative block with few indicators, attenuating its impact; (ii) privacy and algorithmic transparency concerns that mute perceived benefits among Portuguese respondents; (iii) ceiling effects where baseline personalisation is ubiquitous, compressing variance; and (iv) null eWOM paths because advocacy depends on social–motivational levers absent from the model. Accordingly, while prior work positions P-C as a cornerstone of superior CX, its observable impact is contingent on measurement breadth, user-salient execution along the mapped journey, and trust-enabling conditions.

T-S-P is foundational because consumers scrutinise assurances and react strongly to perceived vulnerability. From an ESL perspective, T-S-P shapes customers’ safety expectations pre-purchase and their confirmation at checkout and in post-purchase data handling; under SDL, it operates as an institutional arrangement that lowers perceived risk in resource integration. Empirical work shows privacy concerns depress trust and CX, whereas visible safeguards and reliable conduct foster willingness and credibility [15,83,85]. Our results concur: T-S-P positively predicts CX, but the effect is modest. This attenuation likely reflects (i) lean formative operationalisation, (ii) ceiling effects in a technologically literate sample where baseline security is assumed, and (iii) overshadowing by more proximal levers (interactivity and fulfilment) when modelled jointly. In customer journey terms and on the journey map, T-S-P is most salient at evaluation or checkout and in post-purchase governance, necessary for fluency, yet not, on its own, dominant.

F-SQ shows a clear, statistically significant, though small, incremental effect on CX; its impact propagates along the CX → CS → CL spine to yield a significant total effect on CL, while effects on eWOM are negligible. When it comes to ESL, F-SQ validates post-purchase expectations at delivery, support, and complaint-handling touchpoints on the customer journey map; under SDL, coordinated service processes integrate resources across channels to consolidate brand allegiance. The literature likewise portrays Fulfilment and Service Quality as indispensable post-purchase mechanisms that cement loyalty [5,18,51,107] and shows that cohesive processes, logistical excellence, and after-sales assistance reinforce trust and satisfaction [4,54,111]. Our results accord with this stream (positive F-SQ → CX and mediated effects on CS and CL) yet the empirical magnitude is modest relative to studies reporting stronger direct links to loyalty or advocacy. Plausible explanations include a lean formative operationalisation, ceiling effects, and the omission of social–motivational levers.

I-T is the strongest antecedent of CX, with effects propagating along CX → CS → CL to produce significant indirect impacts on CS and CL, while eWOM remains unaffected. In a model explaining 62.6% of CX and 66.7% of CL, I-T contributes materially to explained variance, unlike OI. From the standpoint of ESL, I-T recalibrates customers’ expectations and their confirmation through interactive encounters; under SDL, it enables collaboration via bidirectional engagement across journey touchpoints mapped at discovery, evaluation, purchase, and use. The literature likewise positions technological interactivity as a critical experiential lever linked to value perceptions and engagement [5,82,119,123] and embeds interactive functionality within broader omnichannel logics [193,194]. Our results confirm a salient I-T → CX path and mediated downstream effects, reflecting value derived from dynamic rather than merely static interface quality, yet the practical magnitude is modest. Prior work often finds larger direct links to loyalty and advocacy when interactivity co-occurs with affective arousal or community identification; our model omits such social–emotional levers, which may explain the null eWOM. Measurement breadth may also attenuate coefficients, which may under-represent its multidimensionality.

Customer satisfaction is the pivotal conduit: CX is strongly associated with CS, and CS is the only substantial driver of CL, while its path to eWOM is trivial and negative. CX is not directly associated with CL or eWOM; thus, the CX → CS → CL chain is a experiential pathway to behavioural intention. The model explains about two-thirds of the variance in CS and CL but virtually none in eWOM, and effect-size diagnostics confirm large CX → CS and CS → CL links with all eWOM paths negligible. In ESL terms, confirmation-driven satisfaction translates into loyalty; on the customer journey map, this sits at post-purchase evaluation and retention touchpoints. Under SDL, value is realised through the experience–satisfaction pathway, whereas public advocacy likely requires social co-creation mechanisms not modelled here. The literature largely anticipates this sequential pattern—experience enhances satisfaction, which fortifies loyalty and stimulates positive eWOM [4,160,174]—and sometimes reports partial direct CX → CL effects [10,17] or eWOM driven directly by delight or indirectly via satisfaction [141,143,151]. Our results concur on mediation towards loyalty but diverge on advocacy: neither CX nor CS predicts eWOM here, contrary to the virtuous cycle described in several studies [31,146], implying omitted drivers of advocacy.

The non-significant paths from CX to eWOM and CS to eWOM, alongside the minimal explained variance in eWOM, indicate that advocacy is not a simple evaluative spillover from experience or satisfaction within this context. ESL primarily accounts for evaluative states (experience → satisfaction) and loyalty intentions. By contrast, eWOM typically hinges on social–motivational mechanisms (community identification, self-presentation motives, emotional contagion, and normative influence) that sit outside the evaluative pathway. In short, people may feel satisfied yet remain silent unless social identity, audience visibility, and platform norms make advocacy instrumentally or symbolically worthwhile. Therefore, the discrepancy can be explained by three factors. Firstly, the number of indicators for eWOM is limited, so differences may not be measured properly and this may reduce effects. Secondly, the model does not consider social and motivational factors (such as community identification, self-presentation motives and what a platform can do) that are often mentioned in eWOM research; without these, satisfaction alone is not enough to encourage people to communicate. Thirdly, advocacy may have lower thresholds than loyalty in a Portuguese sample that can use digital technology: when customer experience and customer service are already high, small increases do not lead to public endorsement, which can result in very high numbers for the effects and lower coefficients.

7. Conclusions

The results support a model of a sequential relationship where customer experience is a strong antecedent of satisfaction, which in turn is a strong antecedent of loyalty, and shows no significant empirical link between satisfaction/loyalty and eWOM. Given the cross-sectional design, these results are associative rather than causal. Among the formative antecedents of experience, Interactivity and Technologies is the strongest driver, followed by Trust–Security–Privacy and Fulfilment and Service Quality, with Usability and Website Design and Personalisation and Customisation playing smaller roles; Omnichannel Integration is empirically inert. The model attains substantial explanatory power for experience, satisfaction and loyalty, but not for communicative advocacy.

7.1. Theoretical Contributions

The uniqueness of this study resides in empirically demonstrating, with unusually large effect sizes, the centrality of the CX → CS → CL chain. By evidencing that customer experience almost translates one to one into satisfaction, which in turn decisively associated with loyalty, this research offers rare quantitative confirmation of experience–satisfaction–loyalty perspectives. This distinctive result positions experiential stimuli as powerful precursors to evaluative states and subsequent loyalty responses, thereby strengthening theoretical foundations while distinguishing this study from prior work that has typically reported weaker or more fragmented effects.

Theoretically, it is also relevant because it decouples advocacy (eWOM) from satisfaction and loyalty, which challenges prior assumptions. The null findings for eWOM challenge prevalent assumptions that satisfaction naturally spills over into advocacy, suggesting instead that eWOM belongs to a distinct nomological network governed by social–motivational drivers absent here. The non-significance of Omnichannel Integration invites a reconceptualization of this construct as multidimensional and possibly higher-order formative, differentiating technical backbone, informational consistency and experiential coherence. Similarly, the modest weights of Personalisation and Customisation and the mixed significance of formative indicators highlight the need to refine measurement breadth and to model privacy transparency and user control as moderators rather than peripheral attributes. Then, this research advances theory by prioritising interactive and trust-building facets of digital interfaces while decoupling advocacy from evaluative satisfaction, thereby redirecting attention to social identity, impression management and platform affordances as complementary lenses.

7.2. Managerial Implications

Managers should recognise that enhancing interactive technologies and visible security/privacy safeguards yields the most leverage on experience, which in turn increases satisfaction and builds loyalty; however, these gains will not automatically generate word of mouth, so advocacy requires explicit social incentives and community features. Post-purchase excellence in fulfilment and service remains essential to confirm expectations and reinforce satisfaction trajectories, yet its incremental returns may be moderate where baseline logistics are already competent. Investments in omnichannel platforms should be audited against actual customer journeys: if most shoppers remain within a single digital stream, heavy integration spending will not be perceived as valuable. Personalisation strategies must balance relevance with transparency, as privacy concerns can blunt perceived benefits; nuanced, opt-in customisation that grants user control is more likely to sustain experiential value.

7.3. Limitations and Future Work

The cross-sectional, self-report design constrains causal inference and raises the spectre of common method variance, even if large coefficients suggest substantive rather than artefactual relationships. Measurement parsimony is another constraint: Omnichannel Integration and eWOM rely on two purified indicators, and several formative blocks include non-significant weights retained for content validity, all of which risk attenuation bias. Also relevant, our Portuguese convenience social media convenience sample is demographically skewed (57.5% female; ~75% under 34; high tertiary education share), limiting external validity and precluding probabilistic generalisation. This composition may overweight experience–satisfaction–loyalty pathways typical of younger, digitally literate consumers; extrapolations should therefore be cautiously conducted.

The cross-sectional nature of the data means that while the results are consistent with the proposed mediational pathways, they cannot definitively establish causality or the direction of the relationships. Longitudinal or experimental designs are required to robustly test the causal claims implied by the model.

Future research should adopt longitudinal or multi-method designs, link perceptual data to behavioural traces (repeat purchases and review postings), and expand the nomological net around advocacy to include social identity, self-presentation motives and platform affordances. Refining and possibly creating a hierarchy of constructs, such as integration, personalisation and interactivity, and testing segment-level moderators like channel-switching propensity or privacy sensitivity would further calibrate theory to practice. Additionally, future research should use probability-based or stratified/quota sampling, consider post-stratification weighting, and undertake cross-cultural replications to secure more balanced age–gender–education profiles and test effect stability across contexts. Furthermore, future studies should include one global item per formative construct to permit redundancy testing and provide an extra layer of convergent validation. This study also did not control demographic variables such as age, gender, or purchase frequency within the structural model. Future research should incorporate these variables as controls or moderators to confirm the stability of the presented model across diverse populations.

Ultimately, this model crystallises a high-yield experiential–satisfaction–loyalty pathway, giving scholars and managers a rigorously validated, decision-ready lens through which to prioritise what truly moves customers.