Measuring Customer Experience in E-Retail

Abstract

In digital retail, where competition intensifies and customer expectations evolve rapidly, understanding the interplay among customer experience metrics is critical for strategic decision-making. Despite widespread adoption of feedback programmes, practitioners struggle to interpret how improvements in transactional ease, experiential quality, and satisfaction propagate across the customer journey to influence retention and growth. This study addresses this gap by examining the interrelations among Customer Effort Score (CES), Customer Experience (CX), Customer Satisfaction (CSAT), Customer Loyalty Index (CLI), and Net Promoter Score (NPS) within a unified framework. A quantitative, cross-sectional survey of recent online shoppers yielded 359 valid responses; the model was estimated with PLS-SEM. All hypothesised direct paths were positive and significant, evidencing a cascade from effort through experience and satisfaction to repurchase and recommendation intentions. Mediation analyses confirmed that CES and CX affect loyalty and advocacy indirectly via CSAT. Out-of-sample prediction validated predictive relevance. The study integrates transactional and relational indicators, establishes the empirical utility of single-item measures within PLS-SEM, and advances a portfolio view of CX metrics. Managerially, effort reduction and experience enhancement emerge as high-leverage interventions: improvements cascade through satisfaction to elevate loyalty and advocacy in digital retail.

1. Introduction

Customer experience (CX) is a strategic lever in digital retail, as omnichannel journeys, automation, and service customisation heighten competition (Banik & Gao, 2023). CX is intricate and influenced by institutions, environment (Ruiz-Alba et al., 2023), and the experience economy (Steffen & Doppler, 2020). It perceives seamless, well-coordinated encounters as precursors of satisfaction and ensuing behaviours (Suh & Moradi, 2023). Thus, customer experience is a strategic imperative in digital retail.

But despite investment in feedback programmes, the metric landscape is still fragmented, with managers tracking transactional ease, satisfaction and loyalty but struggling to see improvements at one stage affecting retention and growth (Bawack et al., 2021). Recent evidence links customer feedback metrics to company performance, and research recommends combining user experience and CX metrics (Agag et al., 2023). Analytics work shows that advocacy can be predicted from text, highlighting its importance to managers (Holmlund et al., 2020). Customer Loyalty Index (CLI) integrates repurchase and recommendation intentions (often alongside satisfaction) to measure loyalty (Eric Hostler et al., 2012). The link between satisfaction and loyalty is well documented in service and tourism contexts (Chi et al., 2018). Yet, empirical research often examines these indicators in isolation. Practice-oriented scholarship contrasts memorable experiences with frictionless ones, while CX research focuses on reducing effort to increase loyalty (Castell et al., 2023). A coherent account is needed to explain how interactional ease affects experience (Zhou et al., 2024) and how experience leads to satisfaction and loyalty (Cui et al., 2024). Studies that examine individual metrics are unable to provide comprehensive recommendations, and therefore, the adoption of integrated directional models is recommended.

This paper addresses that gap with the following research objective: to examine and quantify the directional relationships among Customer Effort Score (CES), CX, Customer Satisfaction (CSAT), CLI, and Net Promoter Score (NPS) within a unified framework, thereby clarifying how reductions in customer effort cascade through experience and satisfaction to influence loyalty and advocacy in online retail, using a prediction-oriented PLS-SEM approach. In addition, the study will evaluate the mediation of CES and CX from loyalty/advocacy via CSAT. The design draws on prior PLS-SEM applications in e-retail and adheres to established reporting of reliability, validity, and predictive fit (Rodríguez et al., 2020).

The study contributes in three ways. Theoretically, it integrates transactional ease, holistic experience, and evaluative satisfaction within a single nomological network that clarifies their roles in loyalty and advocacy. Methodologically, it demonstrates that a concise instrument, consistent with usability and CX guidance, can yield reliable structural estimates and out-of-sample predictive performance. Managerially, it reframes dashboard design as portfolio management: organisations should prioritise interventions by understanding how reductions in effort cascade through experience and satisfaction to shift CLI and NPS (Agag et al., 2023).

The remainder of the paper proceeds as follows. First, we review the literature on CES, CX, CSAT, CLI, and NPS. Next, we describe the research design, instrument, and estimation strategy. We then report measurement and structural results, followed by a discussion of implications for theory and practice. The paper concludes with limitations and avenues for future research.

2. Theoretical Framework

2.1. Customer Effort Score

According to Agag and Eid (2020), CES evaluates the perceived effort customers expend to resolve a request, typically via a single five-point Likert item that asks, “How much effort did you personally have to put forth to handle your request?”. In fact, the metric is intentionally backwards-looking because it captures appraisals of prior service performance rather than forecasting future behaviour (Williams et al., 2020). CES is commonly used alongside CSAT and NPS, yet it distinctively targets interactional ease rather than overall satisfaction or advocacy, thereby isolating friction as a diagnostic focus (Arce et al., 2022). Orr et al. (2023) underline that, while related to usability, CES directs attention to the perceived exertion required in specific service encounters rather than to global sentiment.

The antecedents of perceived effort primarily concern the resources customers invest to achieve resolution. Kalamas et al. (2002) show that time, money, and cognitive demands shape expectations and evaluation criteria, which subsequently inform CES judgements. In addition, attitudes toward the firm and its personnel influence effort assessments, since perceived provider commitment consolidates knowledge and strengthens relational bonds (Sharma et al., 1999). It is important to note that effort is intertwined with service-quality dimensions, and visible extra effort by providers can lift evaluations across multiple facets of quality (Koerner, 2000). Consequently, the same practices that improve service quality tend, by design, to suppress perceived effort by removing barriers to goal completion (Agag & Eid, 2020). In practical terms, strategies that lower CES focus on making interactions easier, quicker, and more predictable. Wirtz (2020) indicates that process redesign eliminates redundant steps, tiers delivery, and optimises capacity so that customers expend less toil across journeys. Since customers increasingly co-produce outcomes digitally, Osorio-Gómez et al. (2024) emphasise intuitive participation mechanisms so that the effort of navigation and choice is minimised. A shift to genuine customer centricity, rather than a purely cost-cutting logic, ensures that these changes remove friction from the customer’s point of view (Osorio-Gómez et al., 2024).

Furthermore, there is growing evidence that technology acts as an enabler of low-effort journeys. Grewal et al. (2023) show that automation, real-time information, and integrated platforms compress handoffs and shorten waits, thereby reducing perceived exertion. Likewise, decision support through up-to-date information and personalised recommendations lightens cognitive load during search and choice (Gauri et al., 2021). When interfaces are highly usable, adoption rises, and the effort required for routine activities declines (Pizzi et al., 2022). Faster responses, enabled by efficient communication systems, also curtail the time and energy needed to achieve resolution (Pitchay Muthu @ Chelliah et al., 2016) and reducing hassles often secures loyalty more reliably than attempts to delight, because customers chiefly value friction-free solutions (Gattiker, 2012). Töllner et al. (2011) similarly argue that frictions are remembered more vividly than pleasant surprises, which prioritises effort-reduction. Extending these insights to retail operations, Vivaldini and Vivaldini (2025) describe how smart stores and autonomous checkout compress tasks, while personalisation tools help customers feel understood and thereby lower exertion during search (Agyei-Boapeah et al., 2022). In addition, stable, user-friendly platforms stabilise expectations and diminish confusion, which supports repeatably low-effort completion (Ali & Maelah, 2025). Efficient CRM-led communication affords proactive, low-effort support, and sales enablement tools coordinate channels to speed recovery when things go wrong (Buehrer et al., 2005; Ryu & Lee, 2018).

With respect to measurement, a single Likert-type item remains the modal approach to CES (Agag & Eid, 2020). Other options include asking about the ease of handling issues on a five- or seven-point scale, still focused on perceived difficulty (Arce et al., 2022), and these instruments tap interactional friction rather than whole-of-journey experience (Orr et al., 2023). Swinyard (2003) also illustrates multi-item variants, where several seven-point items are summed to represent customer service effort, with higher scores denoting greater exertion. In addition, pre/post comparisons and internal process metrics can be integrated with CES to assess whether design interventions genuinely reduce effort along the journey (Chehade et al., 2023).

In e-commerce, the salience of CES is well established because ease of completion is tightly connected to repurchase intentions and spending (Agag & Eid, 2020). Nair and Manohar (2024) argue that a seamless omnichannel experience reduces confusion and removes friction when customers move between online and offline touchpoints. Likewise, perceived organisational effort, reflected in the calibre of digital services, shapes judgments of service quality in ways that directly influence CES (Lin et al., 2022). Over time, intuitive platforms and low-effort interfaces build familiarity and satisfaction, which can raise engagement and conversion propensity (Boccia & Palmieri, 2025).

Placed within the broader CX paradigm, CES isolates a critical dimension, interactional ease, within a construct that spans cognitive, emotional, behavioural, sensorial, and social responses (Bawack et al., 2021). From a managerial standpoint, focusing on effort yields actionable diagnostics for removing frictions that undermine value (Funk, 2017). As Wieland et al. (2024) note, CX quality influences satisfaction, loyalty, and value creation, to which effort-reduction contributes directly. In measurement terms, CES is often complemented by NPS and the CX to furnish a more comprehensive perspective (Sauro & Lewis, 2012). T. Keiningham et al. (2020) argue that engagement and value co-creation benefit when obstacles are systematically removed, while Filieri et al. (2023) connect ease, usability, and enjoyability to the ease of conducting business ethos that resonates with CES. Albert (2023) accordingly recommends portfolios of metrics rather than single indices, and Sauro (2016) advocates mixing transactional with relational indicators so that journey-level frictions can be traced to organisational causes. On the managerial side, firms are urged to initiate customer engagement and orchestrate experiences that integrate customers’ time, money, and effort with company resources (Ruiz-Alba et al., 2023). In empirical terms, CES can equal or even outperform NPS and satisfaction in predicting repurchase, though its direct impact on firm performance appears more indirect (Agag et al., 2023). Integrating CES into CX management helps identify pain points, strengthen emotional connections, and increase wallet share through repeatable, low-effort journeys (Pekovic & Rolland, 2020). Finally, friction-reduction tends to translate into more frequent visits and deeper relationships over time (Klaus, 2022), which positions effort management as a foundation for loyalty rather than a peripheral optimisation (Srivastava & Kaul, 2016).

Research reveals that CES offers a focused, backwards-looking perspective on interactional friction, whereas technology, process overhaul, and employee empowerment serve as effective mechanisms for minimising effort across journeys (Grewal et al., 2023). In conclusion, while CES does not encompass the complete scope of CX, it covers a critical aspect that shapes satisfaction, loyalty, and repurchase; thus, it should be integrated into a multi-metric CX dashboard for comprehensive, organisation-wide learning (Bawack et al., 2021). Therefore, we hypothesise that CES is positively associated with CX (H1).

2.2. Customer Experience

CX is a multidimensional set of cognitive, emotional, physical, sensorial, and social responses evoked through interfaces that may be human or non-human (T. Keiningham et al., 2020). In addition, the value of shopping draws on both hedonic and utilitarian facets, which jointly shape evaluations of the journey (Merle et al., 2022). Synthesising the literature, Rajala et al. (2025) underline that CX is personal and holistic because touchpoints and journey stages interact to generate meaning. Steffen and Doppler (2020) add that this interactional nature makes each experience unique, while Srivastava and Kaul (2016) emphasise the role of effort as a determinant of satisfaction and loyalty. In view of these points, the key components that define CX are its multidimensionality, its integrated cognitive-emotional-sensorial-social-physical elements, and the interplay between hedonic and utilitarian value across touchpoints (Radia et al., 2022; Srivastava & Kaul, 2016).

The reliance on traditional surveys and NPS tends to privilege affective responses while under-capturing the multidimensional journey (Bawack et al., 2021). To address this, Godovykh and Tasci (2020) recommend experience sampling, grid technique, netnography, and structured content analysis, since these methods better reflect sequential and contextual nuance. Additionally, multidimensional scales such as EXQ outperform traditional satisfaction measures in predicting loyalty and word-of-mouth, though adoption remains limited in practice (Williams et al., 2020). Organisations should therefore monitor CX continuously through brand tracking, employee climate surveys, and touchpoint audits so that corrective action can be timely (Birnik et al., 2010).

CX is increasingly recognised as a strategic lever in e-retail. Javed and Wu (2020) show that improved service efforts and enabling technologies differentiate retailers by raising convenience and satisfaction. In related terms, product experience, service procedures, the shopping environment, and staff interactions co-determine satisfaction and retention (Mustafa et al., 2022). Technological advances, particularly automation and machine learning, support personalisation and convenience, which build trust and strengthen purchase intentions as experience with online shopping grows (Mofokeng, 2023). Liu et al. (2024) stress that every touchpoint, from browsing to post-purchase, contributes to the holistic impression of the retailer. In addition, antecedents of CX operate through value co-creation, prompting retailers to orchestrate interactions that enhance attitudes and repurchase (Anshu et al., 2022). The omnichannel context amplifies these demands because online platforms are simultaneously sales, communication, and service-delivery channels (Cortinas et al., 2019). Additional insights claim that CX drives market share, retention, and sales, including in cross-border settings, which keeps it central to academic and managerial agendas (Hsu & Tsou, 2011; Rose et al., 2012; Siqueira et al., 2019).

Nevertheless, the literature emphasises the association between CX and CSAT. J. Kim et al. (2009) position satisfaction as cumulative and central to relationship programmes, given its role in loyalty and repurchase intentions. Duarte et al. (2018) show that convenience and ease of use raise perceived value and satisfaction, which increases reuse. Nguyen (2020) adds that seller service, incentives, and privacy and security provisions contribute positively to satisfaction, highlighting CX’s multifaceted nature. Javed and Wu (2020) observe that the total retail experience, including product and support, shapes service satisfaction beyond the product itself. Subramanian et al. (2014) note that CSAT-type models remain widely used, while Lian (2021) underscores the role of after-sales service in overall satisfaction. Therefore, the relationship between CX and CSAT in online retail is multifaceted, where positive customer experiences, characterised by convenience, service quality, ethical alignment, and enjoyment, lead to higher satisfaction. Then, we hypothesise that CX is positively associated with CSAT (H2).

2.3. Customer Satisfaction

In marketing and consumer behaviour theory, CSAT is seen as a subjective evaluation based on the difference between expected and actual performance, emphasising the importance of the customer’s perception in each service encounter (Francesco & Roberta, 2019). In other words, it is the customer’s overall evaluation of a product or service after purchase, reflecting how well the offering meets or exceeds their expectations and needs (Shaikh et al., 2018). Castro et al. (2007) indicate that CSAT can be captured by a single, global item or via attribute-level disconfirmation judgements comparing expectations and outcomes. In operational terms, surveys typically cover cognitive, affective, and behavioural domains, touching on perceived quality, efficiency, and innovation (Bandaru et al., 2015). From a managerial perspective, satisfied customers tend to be less price-sensitive and more inclined to recommend, which supports reputation building (Albahar et al., 2023). Historically, complaint analysis and service-recovery effectiveness have also been used to assess satisfaction and target staff development (Gordon, 2002). In e-commerce, satisfaction reflects order fulfilment, assortment breadth, and site usability, which together express the CX (Cuesta-Valiño et al., 2023). Choi and Chu (2001) highlight that background characteristics and external stimuli shape satisfaction through service availability and quality.

CSAT is closely tied to CX, with organisations focusing on generating positive experiences that cumulatively enhance satisfaction (Cuesta-Valiño et al., 2023). It is based on customer needs, effective coordination, and the ability to gratify customer expectations (Balaji et al., 2021), quality and speed in service delivery, process improvement, teamwork (Aytekin et al., 2023), and is also shaped by both intrinsic and extrinsic variety-seeking motivations (Lee et al., 2020). Indeed, CSAT is multifaceted in nature, being shaped by experience, quality, motivation, feedback, co-creation, and the interplay between core and peripheral service elements (Pires et al., 2024).

The measurement process may consider factors like product quality, price, and service, as these influence whether customers become loyal and repeat buyers (Albahar et al., 2023). It may use several approaches, each capturing different aspects of the CX (Bandaru et al., 2015), or may apply a single, global item (Castro et al., 2007). Many other scales exist (Pires et al., 2024), but the most widely disseminated of these is the American Customer Satisfaction Index (ACSI). In the case of a single-item assessment, the evaluation is carried out using the generic question ‘How would you rate your overall satisfaction with the service you received?’ and assessed on an interval scale (Menon & Dubé, 2000).

CSAT is recognised as a central driver of business profitability, influencing both repeat consumption and word-of-mouth referrals, which can attract new customers and deter potential losses from negative experiences (L. Zhu et al., 2022). It is a cornerstone of success in online retail, serving as both a critical performance outcome and a primary predictor of an internet retailer’s durability and long-term success (Bressolles et al., 2014). Nevertheless, CLI may require a broader understanding of customer relationships, given the complex nature of loyalty (Sundström & Hjelm-Lidholm, 2020). In the online context, CSAT is attitudinal and it is recognised as the most important factor influencing CLI, which in turn drives repeat visits and purchases (Rodríguez et al., 2020), a relationship also supported by J.-N. Wang et al. (2018). Therefore, we hypothesise that there is a positive association between CSAT and CLI (H3).

NPS is widely recognised as a simple yet powerful metric for evaluating CSAT (Jouve et al., 2012). In practice, NPS is used in B2B and B2C contexts to assess CX and satisfaction, demonstrating its versatility across industries (Barnum, 2021), and it is increasingly adopted as a research tool to gauge satisfaction through a single, standardised question (Bitencourt et al., 2023). Additionally, research comparing NPS with ACSI indicated that both are comparable measures of CSAT, reinforcing the validity of NPS (von Janda et al., 2021). In fact, NPS, willingness to recommend, is seen as a proxy for satisfaction (Hasebrook et al., 2023). Hence, we hypothesise that there is a positive association between CSAT and NPS (H4).

2.4. Customer Loyalty Index

As Gadár et al. (2024) explain, CLI integrates behavioural, emotional, and intention-based elements to yield a comprehensive indicator that correlates with performance over time. The two behavioural indicators, specifically the likelihood to repurchase and the likelihood to recommend, are frequently combined with an overall satisfaction element (Chi & Qu, 2008). CLI is grounded in the duality of loyalty: attitudinal commitment and behavioural loyalty (Öztayşi et al., 2011). Positive experiences, perceived value, and product-service satisfaction shape loyalty, which is often used to gauge market share and retention propensity (Mahmoodjanloo et al., 2020). In service sectors, loyalty manifests in intent to revisit or endorse, reinforcing the cumulative character of commitment (C. E. Kim et al., 2017). Over time, loyalty is visible in repeat purchasing and deep ties to the brand or provider (Xu & Gursoy, 2015). CLI can also be interpreted inversely to churn rate, since intention to remain and renewal behaviours reduce attrition (Hwang et al., 2004).

CLI is measured using a combination of repurchase likelihood and recommendation intention, gathered through surveys. This multi-dimensional approach provides a comprehensive and predictive measure of customer loyalty (Auh & Johnson, 2005). T. L. Keiningham et al. (2007) and Curtis et al. (2011) support the idea of three questions: (1) How likely are you to recommend the online shop to family or friends? (2) How likely are you to shop at this online shop again? (3) How likely are you to try other products from this shop?

For managerial purposes, retention gains from loyal customers reduce acquisition costs and stabilise revenue (Murali et al., 2016). Ren et al. (2024) add that comparative evaluations, plans to switch, and renewal intentions capture attitudinal and behavioural loyalty concurrently. Berkowitz (2006) frames loyalty in terms of repeat purchase frequency and breadth of service usage. Srivastava and Rai (2018) argue that emotional attachment interacts with satisfaction to produce long-term commitment. Of even greater significance, Jothi Krishnan (2021) details cognitive and affective ingredients (perceived quality, cost–benefit, belief, involvement, preference, trust, and purchase intention) that anchor the index. Objective history matters as well. Chow et al. (2007) showed that complaints, revenue per order, and reorder cadence provide behavioural texture, while Auh and Johnson (2005) synthesised price tolerance with repurchase likelihood to assess resilience to competitive offers. Xu and Jackson (2019) added perceived value, fairness, and image as determinants of loyalty, while Baumann et al. (2017) caution that modelling must reflect market-specific factors. Hence, CLI serves as a robust, multidimensional tool for measuring, monitoring, and enhancing loyalty in practice (Gadár et al., 2024).

Chow et al. (2007) indicate that CLI helps identify loyal customers who are more likely to repeat purchases and sustain profitability. In the same vein, J. Kim et al. (2009) describe CLI as a favourable commitment that elevates willingness to pay and amplifies referrals. In addition, accurate and consistent input from CLI supports targeted campaigns that convert buyers into loyalists (Z. Wang et al., 2022). Facing fierce competition, retailers must prioritise retention because only a small minority remain loyal to one outlet (Kamran-Disfani et al., 2017). With e-commerce growth, building and maintaining loyalty in a digital environment is crucial (Pereira et al., 2016). Retailers are therefore encouraged to analyse drivers of loyalty within quality online contexts (Llach et al., 2013). As Levenburg (2005) noted, perceived value is strategically important, while trust, reputation, and switching costs reinforce loyalty and make retention improvements highly profitable (X. Zhu et al., 2024).

2.5. Net Promoter Score

NPS is designed to measure satisfaction and loyalty through a single recommendation question scored 0–10 (Nanath & Olney, 2023). Respondents are classified as Promoters (9–10), Passives (7–8), and Detractors (0–6), and the score is computed as Promoters minus Detractors, ranging from −100 to +100 (Aguinis & Burgi-Tian, 2021). At its core, NPS is derived from responses to a single, standardised question that asks customers how likely they are to recommend a company, product, or service to others, typically rated on a scale from 0 (not at all likely) to 10 (extremely likely) (Nanath & Olney, 2023). This question is central to the NPS methodology and is designed to capture the likelihood to recommend, which is a direct indicator of customer advocacy (Baumgartner et al., 2023).

Shen et al. (2018) showed that the same computation applies across sectors, which supports benchmarking and comparability. In the same vein, Barnum (2021) documents adoption in both B2B and B2C settings, and, to complement, Stepanovic et al. (2025) reported that NPS is often used as a key performance indicator to track change over time.

According to Aguinis and Burgi-Tian (2021), the metric’s simplicity yields highly actionable insight into loyalty risk and advocacy potential. Nanath and Olney (2023) argue that the standardised question allows for rapid implementation across touchpoints, from post-purchase pulses to ongoing engagement surveys. In competitive contexts, NPS is widely recognised as a key indicator of loyalty and satisfaction, which are central to growth (Ungar et al., 2017). Seufert (2014) notes that focusing on extremes helps identify champions and churn risks that directly influence profitability through repeat and organic growth. Because low NPS often heralds churn and negative word-of-mouth, targeted interventions can be prioritised where they will matter most (Shi & Wei, 2024). In addition, integrating NPS with attributes such as browsing, purchase, and support behaviour enables granular analysis and targeted personalisation (Jouve et al., 2012). Its validity and ubiquity in satisfaction studies have made NPS a trusted tool for building long-term relationships in online retail (Munoz et al., 2020). NPS is indispensable in online retail for its ability to provide a clear, actionable measure of customer loyalty, drive continuous improvement, and support strategic decision-making aimed at enhancing CSAT and business growth (Bitencourt et al., 2023).

NPS is so widely used because it is simple, universally applicable, provides actionable insights for business growth, and is adaptable to various contexts, all while being supported by evidence of its validity and predictive power (Aguinis & Burgi-Tian, 2021; Bitencourt et al., 2023; Head et al., 2020; Koladycz et al., 2018; Mecredy et al., 2018; Nanath & Olney, 2023; Nelson et al., 2023; Shi & Wei, 2024; Siering et al., 2018).

2.6. Synthesis and Tensions

The preceding review reveals a landscape of customer metrics characterised not by consensus, but by strategic tension. A central debate pivots on whether a frictionless, low-effort interaction (Dixon et al., 2010) or a memorable, emotionally resonant experience (Godovykh & Tasci, 2020) is the primary driver of loyalty. Proponents of the CES argue that reducing friction is a more reliable and cost-effective strategy for securing loyalty than exceeding expectations, as customers vividly remember negative hassles (Töllner et al., 2011). In contrast, multidimensional CX frameworks posit that holistic, sensorial, and emotional engagements are crucial for differentiation and fostering deep emotional connections that transcend mere satisfaction (Klaus, 2022). A further tension exists regarding the predictive power and sufficiency of individual metrics. The NPS is lauded for its simplicity and claimed correlation with growth (Reichheld, 2003), yet this is contested by studies finding that its predictive validity can be context-dependent and sometimes inferior to traditional satisfaction metrics (T. L. Keiningham et al., 2007). Similarly, while CSAT is a well-established cornerstone, it has been criticised as a passive, backwards-looking measure that may not fully capture future behavioural intentions, particularly in dynamic digital environments where switching costs are low (Sundström & Hjelm-Lidholm, 2020). These conflicting findings and theoretical positions do not, however, invalidate the metrics. Instead, they highlight their complementary, rather than competing, nature. Each metric illuminates a different facet of the customer journey: CES targets transactional efficiency, CX captures the holistic quality of the journey, CSAT provides an overarching evaluative judgement, and CLI/NPS gauge future-oriented behavioural intentions. Relying on any single metric provides a fragmented and potentially misleading diagnosis. Consequently, a portfolio approach is warranted (Albert, 2023; Sauro, 2016), as it allows managers to identify specific friction points (via CES), assess the emotional impact of the journey (via CX), understand cumulative satisfaction (via CSAT), and predict business outcomes (via CLI/NPS). The present study addresses this synthesis by integrating these distinct yet interconnected metrics into a single nomological network, thereby clarifying their specific roles within a coherent causal sequence.

3. Development of Hypotheses and Conceptual Framework

This section sets out the development of the hypotheses and the conceptual framework that underpins the empirical analysis. Building on prior arguments that effort reduction enhances experience quality and shapes downstream outcomes, we articulate directional links among CES, CX, CSAT, CLI, and NPS. Specifically, we posit that CES improves CX (H1), CX elevates CSAT (H2), and CSAT, in turn, drives both CLI and NPS (H3–H4). In summary, the framework provides a principled basis for model estimation and prepares the ground for the operationalisation reported subsequently.

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

Research Hypothesis	Authors
H1: CES is positively associated with CX	(Bawack et al., 2021; Pekovic & Rolland, 2020; Srivastava & Kaul, 2016)
H2: CX is positively associated with CSAT	(Duarte et al., 2018; Javed & Wu, 2020; J. Kim et al., 2009; Lian, 2021; Nguyen, 2020)
H3: CSAT is positively associated with CLI	(Rodríguez et al., 2020; J.-N. Wang et al., 2018; L. Zhu et al., 2022)
H4: CSAT is positively associated with NPS	(Barnum, 2021; Bitencourt et al., 2023; Hasebrook et al., 2023; Jouve et al., 2012; von Janda et al., 2021)
H5: CSAT mediates the relationship between CX and CLI	(Duarte et al., 2018; J. Kim et al., 2009)
H6: CSAT mediates the relationship between CX and NPS	(Barnum, 2021; Bitencourt et al., 2023; Gadár et al., 2024; Hasebrook et al., 2023; von Janda et al., 2021)
H7: CX mediates the relationship between CES and CSAT	(Pekovic & Rolland, 2020; Srivastava & Kaul, 2016)
H8a: The effect of CES on CLI is serially mediated by CX and CSAT	(Pekovic & Rolland, 2020; Rodríguez et al., 2020; Srivastava & Kaul, 2016)
H8b: The effect of CES on NPS is serially mediated by CX and CSAT	(Pekovic & Rolland, 2020; Srivastava & Kaul, 2016; Williams et al., 2020)

presents the research hypotheses, their associated constructs, and the theoretical bases that motivate each directional path.

The table consolidates the rationale for four positive associations (CES → CX, CX → CSAT, CSAT → CLI, and CSAT → NPS) by drawing together evidence reviewed earlier in the manuscript. In fact, this synopsis clarifies expectations before estimation and guides interpretation of the structural results reported later. Beyond these direct paths, H5–H8b formalise the mediation logic: H5 and H6 posit that CSAT transmits the influence of CX to CLI and NPS, respectively; H7 asserts that CX conveys the effect of CES to CSAT; and H8a–H8b specify serial mediation from CES to CLI/NPS via CX and CSAT. Accordingly, we expect positive and statistically significant specific and serial indirect effects (CX → CSAT → CLI/NPS; CES → CX → CSAT; CES → CX → CSAT → CLI/NPS), with any residual direct paths attenuating once mediators are included. Comparing models with and without direct links from CES/CX to CLI/NPS will determine whether mediation is partial or full and sharpen the theoretical reading of the cascade.

Figure 1 shows the proposed customer-experience metrics framework for online shopping, providing a visual summary of the hypothesised relations among the five constructs. As shown, CES precedes CX (H1), CX precedes CSAT (H2), and CSAT has dual positive links to CLI and NPS (H3–H4). Therefore, the diagram offers a concise map linking interactional ease to satisfaction and loyalty intentions.

4. Materials and Methods

4.1. Nature of the Study and Methodological Approach

The present study adopts a quantitative, cross-sectional research design to investigate the proposed CX metrics model. It uses a survey-based approach to collect primary data from online retail consumers, focusing on individuals with recent e-commerce purchase experiences. This design is appropriate for testing the hypothesised relationships in the conceptual model (Figure 1) under real-world conditions. The cross-sectional design, while restricting causal inference, is appropriate for this exploratory study, which aims to map relationships among constructs within a single time frame. Despite the possibility of evolving consumer perceptions over time or across different product categories, the present study focuses on identifying general patterns applicable to online retail contexts, as opposed to category-specific dynamics. The utilisation of longitudinal and experimental designs in future studies would serve to strengthen causal claims.

The decision to use a Partial Least Squares Structural Equation Modelling (PLS-SEM) technique was based on the exploratory nature of the model and the study’s predictive aims. PLS-SEM is particularly useful for analysing complex models with both single-item and multi-item constructs, and it emphasises variance explanation and prediction. We pre-registered a prediction-oriented PLS-SEM workflow, so we treat out-of-sample performance (PLSpredict/Q²_predict) as primary evidence, reporting residual-based fit transparently but not as the decisive criterion. Accordingly, we complement exact-fit indices with PLSpredict results and compare against a linear benchmark.

Our use of single-item measures for CES, CSAT, and NPS warrants careful justification given the methodological guidance of Fuchs and Diamantopoulos (2009), who emphasise that constructs at the heart of a study require multi-item scales to generate detailed insights into their nature. We acknowledge this principle fully. Our decision reflects a pragmatic, prediction-oriented objective rather than an attempt to validate or deeply characterise these constructs psychometrically. Specifically, we aim to demonstrate whether industry-standard operationalisations, as deployed in managerial dashboards, can yield actionable predictive insights when integrated into structural models, despite their known psychometric constraints. This exploratory aim aligns with recent calls to balance measurement rigour with practical relevance in applied CX research (Agag & Eid, 2020; Dixon et al., 2010; Nanath & Olney, 2023). Critically, we do not claim that single items adequately capture the full conceptual domain of these constructs, nor do we assert that our findings substitute for confirmatory studies employing validated multi-item scales. Rather, we provide preliminary evidence that parsimonious measurement, while insufficient for theory testing, may suffice for prediction and practice when constructs are concrete and unambiguous, as argued by Bergkvist and Rossiter (2007). The limitations of this choice—such as the inability to assess reliability, potential bias in effect sizes, and restricted construct validity—are comprehensively addressed in the relevant sections. We explicitly acknowledge that this choice, whilst managerially relevant and theoretically justified for concrete constructs (Bergkvist & Rossiter, 2007; Fuchs & Diamantopoulos, 2009), carries significant psychometric limitations, most notably the inability to assess internal consistency reliability or measurement error, which are addressed in the limitations section. To partially mitigate these concerns, we selected items with established face validity and extensive prior use, conducted cognitive pretesting to ensure uniform comprehension, reported out-of-sample predictive performance (PLSpredict/Q²_predict) alongside in-sample fit, and explicitly framed findings as exploratory rather than confirmatory.

In this study, PLS-SEM enabled simultaneous assessment of the measurement model (the reliability and validity of the survey constructs) and the structural model (the paths corresponding to our hypotheses). The analysis followed the recommended two-step approach: first, evaluating the reflective measurement model, and subsequently examining the structural model for hypothesis testing and model fit.

In accordance with the objectives of theory-building, an initial investigation is conducted using a parsimonious model devoid of covariates to establish baseline relationships among core CX metrics. While the incorporation of demographic controls (e.g., age, education, purchase frequency) and product-category moderators could refine estimates, their inclusion would complicate interpretation at this exploratory stage. Subsequent confirmatory studies should incorporate such variables to test boundary conditions and alternative specifications.

4.2. Sample Target

The target population comprised adults (≥18 years) with prior experience purchasing in online environments. Data were collected via non-probability convenience sampling. Specifically, an anonymous online questionnaire was disseminated through official social media channels (LinkedIn and Instagram) of a Portuguese university. Posts invited participation in research on online shopping experiences and included a link to the survey hosted in Google Forms. No incentives were offered. The survey remained open for seven weeks, and additional reach was achieved through digital word-of-mouth (participants sharing the link within their networks).

Acknowledgement is extended that the use of social media for recruitment via university channels introduces selection bias, resulting in a sample that is skewed towards younger, digitally native, and highly educated respondents. While this demographic profile is characteristic of heavy online shoppers and thus relevant to e-retail contexts, the homogeneity of the sample limits the generalisability of the findings to broader consumer populations, particularly older adults and individuals with lower digital literacy.

Following the PLS-SEM 10-times rule (Hair et al.), the minimum sample size equals ten times the larger of: (i) the maximum number of structural paths pointing to any endogenous construct, or (ii) the largest number of indicators in any measurement block. In our model, each endogenous construct is predicted by a single antecedent (CES → CX, CX → CSAT, CSAT → CLI/NPS), so the maximum number of incoming paths is one, and the largest indicator block is CX with three reflective items; hence, the rule implies N_min = 10 × max(1, 3) = 30. Eligibility criteria were enforced through mandatory screening items at the survey outset: (i) age ≥ 18 years; (ii) at least one online purchase within the previous 6 months; (iii) voluntary informed consent to participate. Respondents failing any criterion were automatically redirected to a thank-you screen and excluded from the dataset. No other exclusion criteria were applied; the survey was open to all qualifying individuals regardless of demographics, purchase category, or retailer.

4.3. Data Collection Instrument

4.3.1. Data Protection and Security

Data collection followed GDPR. All participants knew their responses would be anonymous and confidential. They were assured that the information would be used only for academic research and analysed statistically, not individually. This protected respondents’ identities and followed ethical and legal data processing principles.

4.3.2. Item Development and Validation

All survey items were adapted from established sources in the literature to ensure content validity (see

Table 2. Research model constructs.

Construct	Items	Adapted from (Authors)
CES	How easy was it for you to complete your purchase?	(Dixon et al., 2010)
CX	My overall experience with this online shop was positive I felt comfortable throughout the purchase process I find this online shop pleasant to use	(Pires et al., 2024)
CSAT	How would you rate your overall satisfaction with the online shop?	(Oliver, 1980)
CLI *	How likely are you to recommend the online shop to family or friends? How likely are you to shop at this online shop again? How likely are you to try other products from this shop?	(Cossío-Silva et al., 2019)
NPS	How likely are you to recommend the online shop to a friend or colleague?	(Reichheld, 2003)

* The three items in the construct are combined by definition into one item, calculated as an average of the three original items. Our operationalisation of CLI as the arithmetic mean of three intentions rests on established theory, precedent, and empirical performance in our data. We treat the three intentions as interchangeable indicators of a single underlying continuity and expansion with the focal firm disposition; under this reflective conception, equal weighting is customary when items are on the same scale and designed to sample the same domain. This stance is aligned with widely used loyalty summaries that combine repurchase and recommendation intentions, often with a cross-buy element.

). However, adaptation involved minor wording modifications to suit the online retail context and Portuguese language administration (the survey was conducted in Portuguese and subsequently back translated for reporting). The development and validation process comprised three stages.

• Stage 1: Item Selection and Adaptation—For CX, we selected three items (Pires et al., 2024) that capture holistic experiential evaluation (overall positivity, emotional comfort, and hedonic pleasure). These items were chosen for their brevity and alignment with the cognitive-emotional-hedonic dimensions central to CX theory. For CLI, we adopted the three-item structure from Cossío-Silva et al. (2019), integrating recommendation intent, repurchase intent, and cross-buy intent. Single-item measures for CES, CSAT, and NPS were taken verbatim from Dixon et al. (2010), Oliver (1980), and Reichheld (2003), respectively, as these represent canonical operationalisations.
• Stage 2: Expert Review and Cognitive Pretesting—The initial item pool was reviewed by three academic experts in consumer behaviour and marketing, who assessed face validity, clarity, and conceptual alignment. Minor wording adjustments were made to ensure Portuguese linguistic fluency while preserving construct meaning. Subsequently, cognitive interviews were conducted with 8 individuals representative of the target population (online shoppers aged 22–58). Participants completed the survey and were then asked to paraphrase each item, explain their interpretation, and identify any ambiguities. This process confirmed that all items were uniformly understood, with no systematic misinterpretations detected.
• Stage 3: Pilot Testing—A pilot survey (10 persons) was administered to assess item performance, survey flow, and completion time. Based on pilot feedback, minor formatting adjustments were made to improve visual clarity before full data collection commenced.

CES was captured with a single item asking respondents how easy it was to complete their purchase. CX was operationalised as a reflective construct using three statements about the overall positivity, comfort, and pleasantness of the online shopping experience. CSAT was measured by a single global satisfaction question (“How would you rate your overall satisfaction with the online shop?”). Similarly, NPS was measured with the standard likelihood-to-recommend question on a 0–10 scale, following its conventional definition. The CLI was initially represented by three items addressing future recommendation and repurchase intentions; following prior practice, these three loyalty indicators were combined into a single index value (by averaging) to serve as one composite measure of loyalty. The three intentions jointly index a single loyalty disposition that encompasses retention (repurchase), advocacy (recommendation), and relationship deepening (cross-buy). This content-valid, precedent-backed, and empirically well-behaved composite justifies our use of the averaged CLI score in the structural model. All items (except the 11-point NPS scale) were scored on a Likert-type agreement scale. Before deployment, the survey instrument was reviewed for clarity and relevance, ensuring that questions were interpreted consistently.

5. Results

5.1. Characterisation and Description of the Sample

The study comprised a total of 359 individuals. The sociodemographic characteristics and digital consumption patterns of these individuals are presented in

Table 3. Sample characterisation.

Age			Gender			Education Level
18–34 Years Old	35–54 Years Old	55 Years Old or Older	Female	Male	I Prefer Not to Answer	Primary Education	Secondary Education	Higher Education
267	62	30	206	151	2	4	114	241
74.37%	17.27%	8.36%	57.38%	42.06%	0.56%	1.11%	31.75%	67.13%

.

A total of 359 valid responses were obtained and analysed. As shown in Table 3, the sample was relatively young and well-educated, with a slight majority of female respondents. About 74.37% of participants were between 18 and 34 years old, 17.27% were 35–54, and 8.36% were 55 or older. In terms of gender, 57.38% of respondents identified as female and 42.06% as male, with only 0.56% choosing not to disclose gender. The educational profile of the sample was high: 67.1% of respondents reported having higher education (university degree or above), 31.75% had completed secondary education, and only 1.11% had attained primary education as their highest level. This demographic breakdown suggests that the typical respondent was a young adult with advanced education, which is characteristic of digitally active online shoppers. Such a profile may reflect a segment of customers who are experienced with e-commerce and likely to have well-formed opinions on the online shopping experience.

This demographic concentration is indicative of both the sampling method and the target population of active online shoppers. However, it is imperative to acknowledge that findings should be interpreted with the understanding that younger, educated consumers may exhibit distinct effort sensitivities, experience expectations, and loyalty patterns compared to older or less tech-savvy segments. The enhancement of external validity would be served by replication studies employing probability sampling or quota sampling across age and education strata.

5.2. Results of the PLS-SEM Analysis

PLS-SEM was applied to test the research model, and the results are presented in two parts: the measurement model assessment and the structural model evaluation. Before examining the hypotheses, the reliability and validity of the reflective constructs were verified. Then, the structural relationships were analysed to determine the support for each hypothesis. The analysis was conducted using SmartPLS v. 4.1.1.4 software. Bootstrapping (10,000 resamples) was employed to gauge the statistical significance of path coefficients. The following sections detail the findings of the PLS-SEM analysis, including measurement model metrics, model fit indices, and the outcomes of hypothesis testing.

5.3. Measurement Model Evaluation—Reflective Measurement Model

Given the reliance on single-item measurement for CES, CSAT, CLI, and NPS, we acknowledge that conventional reliability and convergent validity metrics cannot be computed for these constructs. The implicit assumption of perfect reliability (λ = 1.0, error = 0) means that path coefficients involving single-item constructs may be positively biassed due to unmodeled measurement error, R² values may overestimate explained variance, and discriminant validity assessments (Fornell-Larcker, HTMT) are less stringent when comparing single-item with multi-item constructs. These constraints do not invalidate the model but require cautious interpretation; specifically, the reported relationships should be understood as preliminary evidence subject to confirmation via multi-item operationalisation. Furthermore, the Q²_predict results provide some reassurance of predictive relevance despite measurement simplicity.

The reflective measurement model was evaluated for internal consistency, indicator reliability, convergent validity, and discriminant validity.

Table 4. Outer loadings, composite reliability and convergent validity of the constructs.

Constructs	Indicators	Outer Loadings	Cronbach’s Alpha	rho_A	Composite Reliability	AVE
CES	1	1	1	1	1	1
CX	1	0.948	0.934	0.934	0.958	0.883
	2	0.944
	3	0.927
CSAT	1	1	1	1	1	1
CLI	1	1	1	1	1	1
NPS	1	1	1	1	1	1

presents the outer loadings of the indicators for each construct, alongside the corresponding values for internal consistency and convergent validity. The table includes results for outer loadings, Cronbach’s alpha, rho_A, composite reliability, and Average Variance Extracted (AVE) for all constructs.

Each multi-item measure demonstrated strong indicator loadings well above the recommended threshold of 0.70. In particular, the three items measuring CX had outer loadings of 0.948, 0.944, and 0.927, indicating that these indicators were highly correlated with the underlying CX construct. Single-item constructs (CES, CSAT, CLI, NPS) have an implicit loading of 1.0 by definition. Cronbach’s alpha and composite reliability (CR) values confirmed excellent internal consistency for the one multi-item construct. For CX, Cronbach’s α = 0.934 and CR = 0.958, both exceeding the common cutoff of 0.70. The average variance extracted (AVE) for CX was 0.883, well above the 0.50 benchmark, demonstrating convergent validity (i.e., the construct explains the majority of variance in its indicators). The single-item constructs each have an AVE of 1.0 since the item itself captures all variance for those constructs.

The limitations of the psychometric approach under investigation are recognised. Firstly, the use of single items precludes the assessment of internal consistency reliability (Cronbach’s α, composite reliability), meaning that the quantification of measurement error or the distinction between true score variance and random error is not possible. Secondly, they prevent evaluation of convergent validity via AVE, thereby limiting our ability to confirm that items fully capture their intended constructs.

Table 5. The Fornell–Larcker matrix.

	CES	CLI	CSAT	CX	NPS
CES	1
CLI	0.486	1
CSAT	0.543	0.600	1
CX	0.570	0.595	0.645	0.940
NPS	0.446	0.654	0.565	0.477	1

reports the Fornell–Larcker discriminant validity matrix. The diagonal entries present the square root of AVE for each construct, while the off-diagonals show inter-construct correlations.

The model also satisfied discriminant validity criteria. The Fornell–Larcker criterion was met, as each construct’s square root of AVE was greater than its correlations with any other construct (Table 5). For example, the square root of AVE for CX is 0.94, which is higher than the correlations between CX and any other latent variable (the highest being 0.645 with CSAT). Likewise, the single-item constructs CES, CSAT, CLI, and NPS each have an AVE of 1.0, exceeding their intercorrelations (all of which are below 0.66). In addition, the Heterotrait–Monotrait (HTMT) ratios were calculated for all pairs of constructs (

Table 6. HTMT (Heterotrait–Monotrait) ratio.

	CES	CLI	CSAT	CX	NPS
CES
CLI	0.486
CSAT	0.543	0.600
CX	0.590	0.616	0.667
NPS	0.446	0.654	0.565	0.494

).

The HTMT values ranged from 0.45 to 0.67, which is well beneath the conservative threshold of 0.85, providing further evidence of discriminant validity. Lastly, an inspection of cross-loadings (

Table 7. Cross-loading matrix.

	CES	CLI	CSAT	CX	NPS
CES	1	0.486	0.543	0.570	0.446
CLI	0.486	1	0.60	0.595	0.654
CSAT	0.543	0.600	1	0.645	0.565
CX1	0.526	0.585	0.615	0.948	0.465
CX2	0.554	0.536	0.577	0.944	0.405
CX3	0.528	0.557	0.625	0.927	0.475
NPS	0.446	0.654	0.565	0.477	1

) confirmed that each indicator was most strongly associated with its intended construct.

The CX indicators, for instance, loaded between 0.927 and 0.948 on the CX construct, whereas their correlations with other constructs were substantially lower (all cross-loadings involving CX items were ≤0.625). Overall, these results indicate that the measurement model is sound: the survey measures are reliable, and each construct is empirically distinct from the others.

5.4. Structural Model Evaluation

Having established measurement validity, we next assessed the structural model. This involved checking for multicollinearity, evaluating the path coefficients for significance and relevance, and examining the model’s explanatory power, predictive capability, and overall fit.

Table 8. Inner VIF values.

	CES	CLI	CSAT	CX	NPS
				1
CES
CLI		1			1
CSAT			1
CX

presents the collinearity diagnostics for the inner (structural) model. It reports the Variance Inflation Factor (VIF) values for each predictor associated with the respective endogenous constructs.

Collinearity among predictor constructs was not a concern, since each endogenous construct in our model is predicted by only a single antecedent, and all inner VIF values are 1.0. This indicates an absence of multicollinearity issues in the structural model.

The estimated path coefficients and their bootstrapped significance levels are presented in

Table 9. Path coefficients.

	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values
CES ⟶ CX	0.570	0.571	0.039	14.621	0
CSAT ⟶ CLI	0.600	0.598	0.051	11.841	0
CSAT ⟶ NPS	0.565	0.565	0.055	10.256	0
CX ⟶ CSAT	0.645	0.645	0.043	14.990	0

.

All four hypothesised relationships were positive and statistically significant. Specifically, CES had a significant positive effect on CX (β = 0.570, t = 14.621, p < 0.001). Likewise, CX exerted a strong positive influence on CSAT (β = 0.645, t = 14.99, p < 0.001). In turn, CSAT was a significant predictor of both CLI (β = 0.600, t = 11.841, p < 0.001) and NPS (β = 0.565, t = 10.256, p < 0.001). These results confirm support for H1 through H4, indicating that the hypothesised links between the CX metrics are robust. In terms of effect magnitude, the standardised coefficients suggest that CX → CSAT was the strongest direct relationship in the model, followed closely by CSAT → CLI. By comparison, CSAT → NPS and CES → CX showed slightly smaller (though still substantial) effects.

To formally test the mediation hypotheses (H5–H8b), we employed bootstrapping procedures (10,000 resamples) to estimate indirect effects and construct bias-corrected 95% confidence intervals.

Table 10. Specific indirect effects for each individual mediation path.

	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values
CX ⟶ CSAT ⟶ CLI	0.387	0.387	0.052	7.383	0
CX ⟶ CSAT ⟶ NPS	0.364	0.364	0.040	9.064	0
CES ⟶ CX ⟶ CSAT	0.368	0.369	0.041	9.078	0
CES ⟶ CX ⟶ CSAT ⟶ CLI	0.221	0.222	0.038	5.796	0
CES ⟶ CX ⟶ CSAT ⟶ NPS	0.208	0.208	0.030	6.919	0

presents the specific indirect effects for each hypothesised mediation path.

The results provide strong support for all mediation hypotheses. CSAT significantly mediates the CX → CLI relationship (H5: β_indirect = 0.387, t = 7.383, p < 0.001) and the CX → NPS relationship (H6: β_indirect = 0.364, t = 9.064, p < 0.001), confirming that satisfaction serves as the proximal mechanism through which experiential quality translates into behavioural intentions. Additionally, CX significantly mediates the CES → CSAT linkage (H7: β_indirect = 0.368, t = 9.078, p < 0.001), demonstrating that effort reductions influence satisfaction primarily by enhancing holistic experience quality rather than merely reducing transaction costs. Lastly, the serial mediation paths from CES to CLI (H8a: β_indirect = 0.221, t = 5.796, p < 0.001) and from CES to NPS (H8b: β_indirect = 0.208, t = 6.919, p < 0.001) via CX and CSAT are both significant, confirming the complete cascade hypothesised in our model.

Importantly, because our model specification excludes direct paths from CES to CLI/NPS and from CX to CLI/NPS, these indirect effects constitute the total effects of distal antecedents on outcomes (see

Table 11. Total effects for every antecedent–outcome pair.

	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values
CES ⟶ CLI	0.221	0.222	0.038	5.796	0
CES ⟶ CSAT	0.368	0.369	0.041	9.078	0
CES ⟶ NPS	0.208	0.208	0.030	6.919	0
CX ⟶ CLI	0.387	0.387	0.052	7.383	0
CX ⟶ NPS	0.364	0.364	0.040	9.064	0

). This full mediation structure implies that managerial interventions targeting effort reduction will influence loyalty and advocacy exclusively through their impact on experience and satisfaction. The absence of direct paths is theoretically justified: effort is a transaction-specific attribute unlikely to directly shape relationship-level commitments (CLI) or global advocacy (NPS) without first being integrated into broader experiential and evaluative judgments.

Table 11 (total effects) confirms that CES contributes meaningfully to variations in CSAT, CLI, and NPS entirely via the mediating variables, and similarly, CX contributes to CLI and NPS through CSAT. This pattern of full mediation via intermediate constructs suggests a chain-like mechanism. Low customer effort improves CX, which in turn elevates satisfaction, which subsequently drives loyalty intentions and recommendation likelihood.

The model’s R² values indicate moderate to substantial explanatory power for the key endogenous constructs (

Table 12. Explanatory power for the conceptual model.

		CLI	CSAT	CX	NPS
R Square	Original Sample	0.36	0.416	0.325	0.319
	Sample Mean	0.36	0.418	0.327	0.322
	p Values	0	0	0	0
R Square Adjusted	Original Sample	0.358	0.414	0.323	0.317
	Sample Mean	0.358	0.416	0.325	0.32
	p Values	0	0	0	0

).

The model explains about 41.6% of the variance in CSAT (R² = 0.416) and 36.0% of the variance in CLI. It also accounts for 32.5% of the variance in CX and 31.9% in NPS. These values suggest that while additional factors beyond those in the model contribute to CSAT, loyalty, and advocacy, the included constructs (CES, CX, CSAT) capture a significant portion of the variance in these outcomes. The adjusted R² values were nearly identical, reflecting no overfitting despite the complexity of the model.

Nevertheless, we acknowledge limitations due to the psychometric approach. Unmodeled measurement error may inflate path coefficients, as the structural model treats single items as perfect indicators (loading = 1.0, error = 0). Consequently, the reported R² values (0.32–0.42) represent upper-bound estimates that may overstate true relationships.

We calculated Cohen’s f² to assess the relative impact of each exogenous construct on its endogenous variable (

Table 13. Effect size (f²) in the structural model relationships.

	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values
CES ⟶ CX	0.482	0.493	0.100	4.807	0
CSAT ⟶ CLI	0.562	0.577	0.152	3.695	0
CSAT ⟶ NPS	0.469	0.488	0.140	3.358	0.001
CX ⟶ CSAT	0.713	0.734	0.171	4.160	0

).

The results show large effect sizes for all structural paths. In particular, the effect of CX on CSAT was very strong (f² = 0.713), underscoring that CX is a critical driver of satisfaction in this context. Similarly, CSAT’s effect on CLI (f² = 0.562) and on NPS (f² = 0.469) was large, indicating that satisfaction has a considerable influence on both loyalty intentions and likelihood to recommend. The influence of CES on CX was also large (f² = 0.482). These f² values corroborate the earlier interpretation of the path coefficients. They show that the relationships are not only statistically significant but also practically significant in terms of explanatory power.

According to Cohen’s guidelines, f² values of 0.02, 0.15, and 0.35 represent small, medium, and large effect sizes, respectively. All four structural paths in our model exhibit large effects (f² > 0.35), indicating that each antecedent explains substantial unique variance in its dependent construct even after accounting for other predictors. The particularly strong effect of CX on CSAT (f² = 0.713) suggests that experience quality alone accounts for over 41% of satisfaction variance—a finding with clear managerial implications, as improving CX emerges as the single most impactful lever for elevating satisfaction. The similarly large effects of CSAT on CLI (f² = 0.562) and NPS (f² = 0.469) confirm satisfaction’s status as a high-leverage driver of both retention and advocacy. Finally, the large effect of CES on CX (f² = 0.482) validates effort reduction as a foundational CX intervention. Collectively, these effect sizes demonstrate that the hypothesised relationships are not merely statistically significant but practically meaningful, justifying resource allocation toward effort-reduction and experience-enhancement initiatives.

In addition to explanatory power, the model’s predictive performance was evaluated using the PLSpredict procedure (

Table 14. Predictions from PLS path model estimations.

	MV Predict				LV Predict
	PLS		LM		PLS
	RMSE	Q2_Predict	RMSE	Q2_Predict	RMSE	MAE	Q2_Predict
CLI	15.993	0.161	15.335	0.228	0.931	0.654	0.161
CSAT	0.582	0.260	0.571	0.287	0.871	0.714	0.261
CX1	0.579	0.270	0.580	0.270	0.832	0.646	0.32
CX2	0.576	0.302	0.576	0.302
CX3	0.625	0.273	0.625	0.272
NPS	1.635	0.138	1.585	0.190	0.939	0.699	0.139

).

The Stone–Geisser Q²_predict values for all endogenous constructs were well above zero, demonstrating out-of-sample predictive relevance. For instance, Q²_predict was 0.161 for CLI and 0.138 for NPS, indicating that the model has meaningful ability to predict loyalty and recommendation measures for new cases. The Q²_predict for CSAT was even higher at 0.260, and the three reflective indicators of CX each had Q²_predict values around 0.27–0.30. We also compared the prediction errors with a linear benchmark model (LM). The LM’s RMSE values were marginally lower than the PLS-SEM’s for certain measures, suggesting that the linear model performed as well as or slightly better than PLS in those cases. Because all Q²_predict values are >0 and substantively sized, the model exhibits out-of-sample predictive relevance, even as residual fit flagged constraints to revisit.

Finally, global goodness-of-fit measures were inspected (

Table 15. Global model fit.

	Saturated Model	Estimated Model
SRMR	0.025	0.123
d_ULS	0.018	0.424
d_G	0.041	0.142
Chi-Square	90.081	263.082
NFI	0.948	0.849
rms Theta	0.357

).

The standardised root mean square residual (SRMR) was 0.025 for the saturated model (unconstrained relationships) and 0.123 for the estimated model. The saturated model SRMR well below 0.08 indicates an excellent fit of the variance–covariance structure of the data. The estimated model’s SRMR of 0.123 is slightly above common cut-off values (0.08–0.10), which may signal some model misspecification. The saturated model showed excellent residual fit (SRMR = 0.025), whereas the estimated model SRMR = 0.123 and NFI = 0.849 indicate misspecification under conventional cut-offs. We therefore treat these indices as a signal to (i) probe potentially omitted but theoretically plausible paths and (ii) report prediction-oriented diagnostics alongside exact-fit results. In line with our a priori focus on explanation-to-prediction, we retain full transparency about both families of evidence. Furthermore, the use of single-item CES/CSAT/NPS (and a single-score CLI) reduces degrees of freedom for exact-fit tests, which can make residual indices more sensitive to any small misspecification.

Other indices, such as the normed fit index (NFI), yielded 0.849 for the estimated model, below the desired 0.90 threshold, reflecting room for improving model fit. Importantly, all model constructs and paths were theoretically justified. The strong significance of every pathway suggests that any minor lack of fit is not due to irrelevant links. It is more likely due to the omission of minor cross-linkages or other factors beyond this study’s scope. Overall, the structural model evaluation confirms that the hypothesised model has substantial explanatory and predictive power.

Table 16. Results of the research hypotheses.

H	Constructs	Original Sample	Sample Mean	Standard Deviation	T Statistics	p Values	Results
H1	CES ⟶ CX	0.570	0.571	0.039	14.621	0	Supported
H2	CX ⟶ CSAT	0.645	0.645	0.043	14.99	0	Supported
H3	CSAT ⟶ CLI	0.600	0.598	0.051	11.841	0	Supported
H4	CSAT ⟶ NPS	0.565	0.565	0.055	10.256	0	Supported
H5	CX ⟶ CSAT ⟶ CLI	0.387	0.387	0.052	7.383	0	Supported
H6	CX ⟶ CSAT ⟶ NPS	0.364	0.364	0.040	9.064	0	Supported
H7	CES ⟶ CX ⟶ CSAT	0.368	0.369	0.041	9.078	0	Supported
H8a	CES ⟶ CX ⟶ CSAT ⟶ CLI	0.221	0.222	0.038	5.796	0	Supported
H8b	CES ⟶ CX ⟶ CSAT ⟶ NPS	0.208	0.208	0.030	6.919	0	Supported

synthesises the hypothesis testing results: each of the NINE hypotheses (H1–H8b) is supported by the data at a high level of statistical confidence.

It is important to note the psychometric limitations of this approach, which may result in unmodeled measurement error. This, in turn, has the potential to inflate path coefficients, consequently leading to the reported path coefficients (β = 0.57–0.65) representing upper-bound estimates that may overstate true relationships. The survey data can be found in the Supplementary Materials.

6. Discussion

This study set out to examine how key CX metrics, namely CES, CX, CSAT, CLI, and NPS, interrelate in the online retail context. The findings provide empirical support for all hypothesised links in our model, underscoring a coherent narrative from customer effort through experience and satisfaction to loyalty and advocacy outcomes. In this section, we discuss these results in light of prior research and highlight their implications.

Firstly, the significant positive effect of customer effort on overall CX (H1) confirms that reducing friction in the shopping process enhances the perceived quality of the experience. This outcome is consistent with the observations of prior studies that emphasise ease of service as a critical component of CX. When customers have to expend minimal effort to complete their purchase, they tend to rate the experience more favourably. Our result aligns with Agag and Eid (2020)’s findings that lower effort correlates with better service appraisals and echoes Srivastava and Kaul (2016)’s view that effortless experiences contribute directly to loyalty formation. It extends the literature on CES by quantitatively demonstrating its impact on holistic CX in an e-commerce setting. In practical terms, this suggests that online retailers can improve customers’ experience evaluations by streamlining processes, improving website usability, and otherwise minimising the work customers must do to have their purchase completed. This finding reinforces theoretical propositions that frictionless experiences are foundational to positive customer impressions.

Secondly, the positive relationship between CX and CSAT (H2) corroborates the widely held notion that delivering a superior experience is pivotal for cultivating satisfaction. This finding agrees with earlier research in online retail, which has found that enjoyable, convenient, and well-orchestrated customer experiences tend to elevate overall satisfaction levels. Our results add empirical weight to the argument by J. Kim et al. (2009) that satisfaction is essentially an outcome of accumulated positive experiences in both the digital and physical facets of service. The strong effect size of CX on CSAT (nearly 42% of variance explained) highlights that CX is not an abstract concept but a concrete driver of evaluative satisfaction. It suggests that retailers need to manage the entire customer journey (pre-purchase, during purchase, and post-purchase) to ensure consistency and delight, as any single weak touchpoint could undermine the overall satisfaction. This integrated perspective resonates with contemporary CX frameworks that view satisfaction as emerging from the sum of all interactions and touchpoints (Duarte et al., 2018; Javed & Wu, 2020). Our evidence aligns with the recommendations of Godovykh and Tasci (2020), who advocated multidimensional experience monitoring. It affirms that beyond product and price, experiential elements are decisive in shaping CSAT. Factors such as website aesthetics, responsiveness, and personalisation strongly influence how satisfied customers feel.

Thirdly, our analysis confirms that CSAT has a strong influence on loyalty intentions (H3). Satisfied customers in our study were substantially more likely to exhibit loyalty, as captured by the combined CLI measure (which incorporated repeat purchase and recommendation intent). This is consistent with the classic satisfaction–loyalty paradigm established in marketing literature. It also mirrors recent findings in e-commerce contexts that tie higher satisfaction to greater repurchase rates and customer retention (Rodríguez et al., 2020; J.-N. Wang et al., 2018; L. Zhu et al., 2022). The fact that CSAT explained 36% of the variance in the loyalty index is noteworthy for an online retail setting where consumers often face abundant alternatives. It underscores that even in a digital market with low switching costs, building CSAT can yield tangible increases in loyalty behaviours. Our results echo those of previous research (Murali et al., 2016) and others who have stressed satisfaction’s role as a cornerstone of long-term customer relationships. In addition, by employing a multi-faceted loyalty metric (CLI) that goes beyond just repeat purchase, this study contributes to loyalty research by showing that satisfaction’s effect is broad-based. In our results, higher satisfaction boosted both the willingness to continue patronage and the willingness to recommend. This dual impact reinforces the idea that loyalty is a composite of attitudinal and behavioural commitment (Öztayşi et al., 2011). Both components are strengthened by positive customer experiences and high satisfaction.

A comparison of path coefficients and effect sizes offers managerial insight into the relative leverage of satisfaction for driving different behavioural outcomes. While CSAT significantly predicts both CLI (β = 0.600, f² = 0.562) and NPS (β = 0.565, f² = 0.469), the slightly stronger coefficient and effect size for loyalty suggest that satisfaction has a marginally greater influence on repurchase and relationship-deepening intentions than on advocacy. This asymmetry may reflect the distinct psychological mechanisms underlying these outcomes: loyalty (CLI) is primarily calculative and habitual, directly contingent on cumulative satisfaction, whereas advocacy (NPS) involves social risk and requires satisfaction to exceed a higher threshold before customers willingly endorse the brand publicly. Managerially, this implies that satisfaction improvements will reliably boost retention (CLI), but converting satisfied customers into active promoters (high NPS) may require additional interventions, such as referral incentives, social proof mechanisms, or identity-affirming brand positioning, that amplify satisfaction’s effect on word-of-mouth. Organisations seeking to optimise metric-specific outcomes should therefore calibrate investments accordingly: satisfaction-focused strategies effectively drive loyalty, but advocacy may demand satisfaction plus supplementary engagement tactics.

Finally, the strong positive effect of satisfaction on NPS (H4) validates the premise that satisfied customers are far more likely to become promoters of the brand. This result was expected, given that willingness to recommend is inherently linked to one’s satisfaction with the product or service (Jouve et al., 2012). Nevertheless, demonstrating this link empirically is valuable, as it confirms that in our dataset, a one-unit improvement in CSAT leads to a significant rise in the likelihood-to-recommend score. The finding aligns with studies that have found NPS to be a reasonable proxy for CSAT and loyalty in many cases (Barnum, 2021; Bitencourt et al., 2023). It also supports the argument that fostering high satisfaction will naturally increase a company’s NPS, because delighted customers tend to share their positive experiences. By integrating NPS into our model, we provide evidence of how a single-item advocacy metric fits within the nomological network of CX constructs. It sits at the very end of the chain, driven largely by satisfaction (which itself is influenced by prior CX quality). Our data suggest that improvements upstream (e.g., making the experience easier and more pleasant) propagate through satisfaction to lift NPS. This emphasises that companies looking to raise their NPS should invest in fundamental CX enhancements rather than focusing narrowly on the score itself.

The mediation analyses provide critical insight into the mechanisms underlying CX’s influence on behavioural outcomes. The significant indirect effects (H5–H8b) confirm that satisfaction operates as a cognitive gateway through which experiential impressions and effort appraisals are transformed into actionable loyalty and advocacy. This finding aligns with the expectancy-disconfirmation paradigm (Oliver, 1980), which positions satisfaction as a summary judgement that mediates stimulus-response linkages in consumer behaviour. The serial mediation from CES to CLI/NPS via CX and CSAT (H8a, H8b) is particularly noteworthy, as it validates the complete cascade posited by our framework: operational improvements in effort reduction trigger a chain reaction, enhancing perceived experience quality, which elevates satisfaction, which ultimately drives retention and referral intentions. From a theoretical standpoint, this cascade structure suggests that CX metrics are hierarchically organised, with transactional attributes (effort) feeding into holistic experiential judgments, which in turn shape global evaluations (satisfaction), which finally determine behavioural outcomes. Managerially, these findings imply that efforts to improve CLI or NPS by directly targeting those metrics (e.g., incentivising recommendations) may be less effective than addressing root causes (effort reduction and experience enhancement), whose benefits propagate forward through the satisfaction mechanism.

The empirical support for full mediation has important theoretical implications. It suggests that the influence of operational efficiency (effort) and experiential quality on customer relationship outcomes is not automatic but rather cognitively mediated: customers first integrate transactional and experiential cues into holistic satisfaction judgments, which then drive behavioural commitments. This finding challenges managerial assumptions that delighting customers through exceptional experiences or eliminating friction will immediately secure loyalty; rather, these interventions must translate into demonstrable satisfaction gains to affect retention and advocacy. The absence of significant direct effects from CES and CX to CLI/NPS (when CSAT is controlled) also clarifies construct boundaries: effort and experience are antecedent inputs into the satisfaction formation process, not parallel or substitute drivers of loyalty.

Nevertheless, the proposed model’s generalisability depends on boundary conditions. It is possible that older consumers who value customer service and trust more than efficiency may not value effort reduction (CES) as much as younger, digitally native consumers. The model’s parameters may vary per product category. Also, the CES → CX → CSAT cascade may be less direct for high involvement purchases due to brand reputation and post-purchase service. It may be likewise most effective for transactional retail. Therefore, managers must separate the findings due to heterogeneity. A homogeneous CX investment strategy is inferior. Instead, customer and category details should guide resource allocation. Prioritising frictionless journeys provides high leverage for digitally savvy customers and standardised products. Building trust and offering enhanced support should be the strategic priority for other sectors or complex categories to ensure nuanced CX portfolio management.

It is also important to note that a methodological caveat is in place with regard to our measurement approach. The use of single-item measures, while pragmatically aligned with industry practice and respondent burden considerations, limits psychometric scrutiny. We cannot rule out that unmodeled measurement error inflates the observed path coefficients or that content underrepresentation (particularly for CLI) biases structural estimates. However, the consistency of our findings with prior multi-item studies and the demonstrated out-of-sample predictive validity (Q²_predict > 0) suggest that substantive relationships exist beyond measurement artefacts.

We may therefore conclude that this research integrates multiple facets of CX measurement into one model, and it provides clear evidence of their interplay. This approach offers a more unified understanding of how customer perceptions and attitudes form in online retail. The results of the study demonstrate a correlation between the variables under investigation. It is evident that metrics are interdependent, with effort and experience exerting an influence on satisfaction, which in turn exerts an influence on loyalty and advocacy. This integrated perspective is of significance for researchers and practitioners alike, as it underscores the notion that enhancements made at any given stage (for instance, the reduction in customer effort) have the potential to engender a series of advantageous outcomes throughout the customer relationship lifecycle. In the subsequent section, we offer a conclusion that comprises an outline of the theoretical contributions of this study, practical implications for managers, and limitations that suggest avenues for future research.

7. Conclusions

This study aimed to clarify the relationships among key customer experience metrics, specifically effort, experience, satisfaction, loyalty, and advocacy, in the realm of online retail. The amalgamation of these constructs into a cohesive, directional framework, followed by rigorous testing through structural equation modelling, yields empirical evidence of a sequential cascade. The hypothesis posits that transactional ease influences the overall quality of experience, which enhances evaluative satisfaction, subsequently leading to behavioural commitment and word-of-mouth promotion. This study’s findings are significant in three domains: (i) advances customer experience scholarship by illustrating that fragmented metrics are interconnected indicators within a causal chain, where upstream interventions affect downstream outcomes; (ii) demonstrates that parsimonious measurement approaches, when adequately justified and rigorously tested, can provide predictive insights applicable to both academic research and managerial practice; (iii) offers digital retailers a framework for resource allocation grounded in empirical evidence. This roadmap delineates the prioritisation of effort reduction and experience enhancement as essential investments and gains are expected to lead to satisfaction, thereby ensuring loyalty and advocacy. The subsequent sections elaborate on these contributions, discuss managerial implications, and recognise the limitations of the chart directions, thereby laying a foundation for future research.

7.1. Theoretical Contributions

This research makes several contributions to the literature on CX and loyalty metrics. To the best of our knowledge, it is one of the first studies to empirically integrate CES, CX, CSAT, CLI, and NPS into a single comprehensive model within the context of online retail. Therefore, the study extends existing theory in multiple ways. First, it bridges the gap between transaction-specific feedback metrics (such as CES and NPS) and broader relationship measures (satisfaction and loyalty) by demonstrating how these indicators relate to each other. Prior studies often examined these metrics in isolation or in partial combinations; our holistic approach provides a more systemic understanding of how improvements in one domain (e.g., reducing customer effort) can translate into gains in another (e.g., increased loyalty). Second, the findings reinforce and refine the theoretical understanding of the role of CX. The significant link from CX to satisfaction validates multidimensional CX models that posit experience quality as a precursor to evaluative judgments like CSAT. We contribute to this body of work by quantifying the strength of that link and confirming it in an e-commerce scenario. Our approach contrasts with traditional satisfaction-centric models that treat CSAT as the sole or primary metric for CX evaluation. Classical paradigms, such as ACSI or SERVQUAL, position satisfaction as both an outcome of service delivery and a predictor of loyalty, collapsing transactional, experiential, and relational constructs into a unidimensional satisfaction judgement (Balaji et al., 2021). While parsimony is appealing, this conflation obscures critical distinctions: (i) transactional ease (CES) and holistic experience (CX) are distinct antecedents with different managerial levers; (ii) loyalty intentions (CLI) and advocacy behaviours (NPS) are separable outcomes requiring differentiated strategies; (iii) satisfaction itself is an intermediate evaluative state, not a direct manipulation target. Our model decomposes this black box, clarifying that improvements in customer outcomes require sequential investments: reducing effort (operational excellence) → enhancing experience (service design) → elevating satisfaction (quality delivery) → securing loyalty and advocacy (relationship management). This granularity enables diagnostic precision: managers can identify bottlenecks, such as high effort despite positive experiences or high satisfaction but low advocacy, and allocate resources to the specific construct constraining performance. In contrast, satisfaction-only models yield undifferentiated prescriptions, specifically improve satisfaction, without specifying whether the problem lies in operational friction, experiential deficits, or advocacy barriers. Our findings thus advocate for measurement portfolios that respect construct boundaries while mapping their nomological relationships.

Third, our study highlights the pivotal mediating role of CSAT in converting positive experiences into tangible outcomes such as loyalty and advocacy. While the satisfaction–loyalty relationship is well documented, we add evidence using the CLI construct, supporting the notion that loyalty is a cumulative outcome influenced heavily by satisfaction. We also empirically substantiate the proposition that NPS is effectively an expression of satisfaction. In our model, satisfaction accounted for a large portion of the variance in NPS, bolstering the conceptual claim that NPS can serve as a proxy for overall satisfaction.

Finally, from a methodological standpoint, this study provides preliminary, exploratory evidence for the practical utility of single-item CX metrics (CES, CSAT, NPS) within integrated models, demonstrating that managerially relevant constructs can yield predictive insights (Q²_predict > 0) despite measurement simplicity. However, we emphasise that this finding should be interpreted strictly as proof-of-concept rather than endorsement: while single items may suffice for exploratory, prediction-oriented research and operational dashboards, confirmatory theory testing demands multi-item measurement to rigorously assess reliability, validity, and unbiased effect sizes. Our study thus contributes to ongoing methodological dialogue by showing what can be learned from parsimonious measurement (predictive patterns, nomological relationships) while acknowledging its fundamental psychometric constraints (construct underrepresentation, inflated coefficients, unmodeled error). We strongly recommend that future research employ multi-item scales to validate our structural findings and establish true effect magnitudes.

7.2. Managerial Implications

The conclusions drawn from this study have practical implications for professionals in four distinct domains.

1.

Prioritise Effort Reduction as a Foundational Intervention. The significant CES → CX → CSAT → CLI/NPS cascade demonstrates that:

• Reducing customer effort is not merely a cost-cutting exercise but a strategic CX investment with downstream revenue implications.
• Specific tactics include: streamlining checkout flows, minimising form fields, providing intelligent search and filtering, enabling one-click reorder, and proactively resolving issues before customers must exert effort.
• Metrics: Track CES longitudinally and set reduction targets (e.g., <2.5 on 5-point scale).

2.

Manage CX Holistically Across All Touchpoints. The strong CX → CSAT link (β = 0.645) underscores that:

• Satisfaction is an emergent property of accumulated experiences; no single touchpoint can compensate for deficiencies elsewhere.
• Organisations must adopt journey-mapping and continuous experience monitoring (via brand tracking, touchpoint audits, and employee feedback).
• Cross-functional coordination (IT, logistics, customer service, marketing) is essential to ensure experiential consistency.

3.

Leverage Satisfaction as the Proximal Driver of Behavioural Outcomes. Given CSAT’s strong influence on CLI (β = 0.600) and NPS (β = 0.565):

• Post-purchase satisfaction surveys should trigger immediate recovery protocols when scores fall below thresholds.
• Satisfaction improvements reliably boost retention and advocacy; even marginal CSAT gains (e.g., +0.5 on a 5-point scale) yield measurable loyalty increases.
• Proactive satisfaction management (anticipating dissatisfaction via predictive analytics) is more effective than reactive complaint handling.

4.

Adopt Integrated Measurement Rather Than Single-Metric Optimisation. The integrated model demonstrates metric interdependence:

• Tracking only NPS or only CSAT provides incomplete diagnostics; portfolio dashboards (CES-CX-CSAT-CLI-NPS) enable root-cause analysis.
• Resource allocation should reflect causal priorities: invest upstream (effort, experience) to drive downstream outcomes (loyalty, advocacy) rather than directly incentivising NPS.
• Segment-level analysis (e.g., comparing high-CES vs. low-CES cohorts on downstream metrics) can identify high-leverage improvement opportunities.

It is therefore concluded that a customer-centric strategy is not a single intervention, but rather a sequential, multi-level optimisation process: the first stage is to make it easy (CES); the second stage is to make it positive (CX); the third stage is to make them satisfied (CSAT); the fourth stage is to make them loyal (CLI) and vocal (NPS).

7.3. Limitations and Future Work

While this study offers valuable insights, it is not without limitations. These limitations also suggest avenues for future research. First, the study employed a cross-sectional survey design with a social-media-based convenience sample, capturing respondent perceptions at a single point in time. Consequently, we cannot definitively infer causality or observe how improvements in one metric might lead to changes in others over time. Longitudinal studies or field experiments could be conducted to track such dynamics and confirm the causal direction of the effects observed. Furthermore, the present study does not differentiate by product category (e.g., hedonic vs. utilitarian; high vs. low involvement), which may moderate the strength of observed relationships. It is hypothesised that categories which exhibit divergent purchase frequencies or emotional significance may manifest distinct CX dynamics. Subsequent research should employ longitudinal panel designs to track metric evolution over time and test for category-specific effects.

Second, all data were self-reported by customers using a single questionnaire, raising the possibility of common method bias. Although our results exhibited strong differentiation (which mitigates this concern to an extent), future research could use procedural remedies (such as temporal separation of measurements) or incorporate data from multiple sources (e.g., combining survey responses with actual purchase or loyalty records) to further reduce common method bias.

Third, the sample, while sizeable, consisted mainly of younger, well-educated consumers who are active online shoppers. This demographic skew may limit the generalisability of the findings to other consumer segments, such as older or less tech-savvy individuals. Future research should validate the model in different demographic and cultural contexts to see if the inter-metric relationships hold universally or if they vary by segment.

Fourth, our study focused on online retail in general without differentiating by product category or shopping context. The importance of ease, experience, and satisfaction may differ between, for example, low-touch commodity purchases and high-involvement luxury purchases. Subsequent studies could examine the model within specific industries (e.g., online grocery vs. fashion retail) or compare results across categories to identify any nuanced differences.

Fifth, our reliance on single-item measures for CES, CSAT, CLI, and NPS may constitute a methodological limitation. The absence of multi-item scales prevents assessment of internal consistency reliability (α, CR, rho_A), meaning we cannot quantify random measurement error or ensure that indicators reliably capture their constructs. Also, convergent validity (AVE) cannot be evaluated for single-item constructs, raising concerns about content underrepresentation, and discriminant validity tests are less stringent when applied to single-item measures, potentially under-detecting empirical overlap among constructs. Moreover, path coefficients and R² values are likely upwardly biassed because single items are treated as perfect indicators. This means the true relationships among CES, CX, CSAT, CLI, and NPS may be weaker than our model suggests, and our R² values (0.32–0.42) may overstate predictive accuracy.

The fifth and most significant limitation concerns our measurement approach. Fuchs and Diamantopoulos (2009) explicitly caution that single-item measures are inappropriate when constructs are central to a study’s theoretical aims, as they preclude detailed examination of construct facets. We acknowledge that our constructs (CES, CX, CSAT, CLI, NPS) are indeed central to our nomological framework, and therefore, our single-item operationalisation of CES, CSAT, and NPS, while aligned with industry practice, falls short of the psychometric standards required for confirmatory construct validation. This choice introduces three critical constraints: (i) Inability to assess internal consistency reliability, meaning we cannot quantify or partition measurement error; (ii) Potential upward bias in path coefficients and R² values, as single items are treated as perfect indicators (loading = 1.0, error = 0); (iii) Content underrepresentation risk, whereby our measures may not fully capture the conceptual breadth of satisfaction, effort, or advocacy. Consequently, our structural estimates (β = 0.57–0.65; R² = 0.32–0.42) should be interpreted as upper-bound approximations subject to attenuation in studies employing error-corrected, multi-item measurement. Replication using validated scales (e.g., EXQ for experience, ACSI for satisfaction) is essential to confirm our findings and establish true effect sizes. Our contribution is therefore best understood as exploratory rather than definitive; we demonstrate that parsimonious metrics can reveal nomological patterns and predict outcomes, but rigorous theory testing requires multi-item operationalisation.

Finally, there are potential extensions to the model that future studies could explore. Incorporating additional antecedent variables, such as trust, perceived value, or service quality, could enrich the explanation of satisfaction and loyalty outcomes. Likewise, examining moderating factors (for instance, whether the effort → experience effect is amplified for certain types of customers or under certain conditions) would provide deeper insight. Another promising avenue would be to investigate feedback loops: for example, do loyal customers subsequently perceive their experiences more positively, creating a reinforcing cycle? Addressing such questions would further illuminate the dynamics among CX metrics. A further limitation stems from the acknowledged demographic and contextual homogeneity of the sample, which precludes the examination of heterogeneity in the model’s pathways. As discussed, the strength of the relationships within the CES–CX–CSAT–CLI/NPS chain is likely contingent on factors such as age, gender, education, purchase frequency, product category, and prior brand relationship. Therefore, a key agenda for future research is to employ multi-group analysis (MGA) in PLS-SEM to test the model’s invariance across these distinct segments. Such an analysis would formally establish the boundary conditions of the framework by determining whether the posited relationships are universally stable or significantly differ for, say, older versus younger consumers or across high- versus low-involvement product categories, thereby yielding more nuanced and segment-specific managerial guidance.

In conclusion, despite these limitations, this study provides a robust empirical foundation for understanding the interplay among customer effort, experience, satisfaction, loyalty, and advocacy in digital retail. It confirms several theoretical expectations and offers a cohesive model that future research can build upon. By integrating multiple CX metrics, we hope to encourage both academics and practitioners to adopt a holistic view of CX management, one that recognises how improvements in one aspect can influence outcomes across the customer lifecycle. Future research that expands, refines, or longitudinally tracks this model will be invaluable for advancing knowledge in CX and for guiding businesses in creating better customer relationships.