1. Introduction
The accelerating digitalization of services has fundamentally transformed the way consumers interact with technology. Among various innovations, self-service technologies (SSTs) such as kiosks have become a central driver of modern service delivery, enabling transactions, information retrieval, and other processes without human intermediaries [
1,
2,
3]. The COVID-19 pandemic further accelerated this transformation, making contactless solutions essential for ensuring service continuity and public safety [
4,
5]. Even after the pandemic, the demand for kiosks has not declined; rather, they have become a permanent service infrastructure that reshapes consumer expectations and operational standards across industries [
6,
7].
Against this backdrop, kiosk adoption is no longer a question of “whether” but of “how.” The challenge for both academia and industry lies not only in implementing the technology but in ensuring that it is used continuously and effectively. Identifying the drivers of adoption is therefore crucial, not only from the perspective of functional performance but also in terms of the psychological and social mechanisms that influence user acceptance and sustained use [
8].
The Technology Acceptance Model (TAM) has provided a theoretical foundation for technology adoption research, identifying perceived usefulness (PU) and perceived ease of use (PEU) as key determinants [
9,
10,
11,
12]. However, in unattended service environments, socio-psychological factors such as digital readiness (DR), trust in technology (TT), and perceived barriers (PB) may play a more critical role [
13,
14]. Social Trust Theory posits that trust reduces uncertainty and perceived risk, thereby facilitating adoption [
15,
16], while the concept of barriers provides an extended pathway in TAM to capture negative inhibitory factors beyond usefulness and ease of use [
17]. Nevertheless, empirical studies that integrate TT with DR, PU, and PB in kiosk settings remain scarce, and studies that explicitly compare different user groups are still limited in scope [
3,
8].
Existing kiosk research has also tended to focus on specific industries (e.g., quick-service restaurants, transportation hubs) or restricted samples, limiting generalizability [
18,
19]. For example, defined as individuals born after the early 1980s who have grown up with digital technologies as part of their everyday lives [
20], tend to have high technology familiarity and faster learning curves, and may therefore be more sensitive to barrier reduction than to trust formation [
21]. In contrast, non-digital natives, with limited experience using such technologies, may view trust building as a central factor in adoption [
22]. Similarly, high-frequency users may rely more on functional efficiency and obstacle removal, while low-frequency users may depend more heavily on lowering initial entry barriers and fostering psychological comfort. Understanding these differences is essential for advancing theory and refining practical strategies.
In response, this study extends TAM by integrating DR, TT, PU, and PB into a unified socio-cognitive analytical framework and empirically examines their effects on intention to use kiosks (IU) in the context of China, a country with advanced digital infrastructure. It further explores segmental differences by generation and usage frequency, thereby reinterpreting kiosk adoption through both a trust-centered and barrier-sensitive lens. From a practical perspective, the study highlights the need for differentiated strategies, balancing trust-building measures for less experienced users with barrier-reduction strategies for digitally proficient users.
The remainder of this paper is organized as follows.
Section 2 reviews relevant literature and hypotheses.
Section 3 outlines the research design and methodology.
Section 4 reports the empirical results.
Section 5 discusses the findings and robustness checks, and
Section 6 concludes with implications and future research directions.
3. Methodology
The proposed hypotheses were empirically examined using a structured survey instrument based on previously validated scales. The items were carefully refined to reflect the specific context of kiosk usage, ensuring both content validity and theoretical alignment. Structural equation modeling (SEM) was employed to assess the direct and indirect relationships among digital readiness, trust in technology (TT), perceived usefulness, perceived barriers, and intention to use kiosks. In addition, multi-group moderation analyses were conducted based on generational cohort and kiosk usage frequency, and a robustness check using the Maximum Likelihood Robust (MLR) estimation method was performed to validate the stability of the results.
3.1. Data Collection
To investigate the psychological and behavioral determinants of kiosk adoption, this study targeted only individuals with prior kiosk usage experience. Participants were recruited through a combination of snowball and convenience sampling via WeChat, the most widely used digital communication platform in China. This approach enabled broad participation across diverse generational and regional backgrounds.
A screening question was positioned at the beginning of the survey to confirm prior usage experience; only those who responded affirmatively were allowed to proceed. IP tracking was employed to prevent duplicate entries, and all responses were collected anonymously in compliance with data privacy standards. After excluding incomplete or invalid submissions, a total of 750 valid responses were retained for analysis.
Because the sampling strategy was non-probabilistic, the findings should be interpreted with caution regarding their generalizability. In particular, the exclusion of non-users and digitally excluded groups inherently limits the extent to which the results can be generalized beyond active kiosk users. Nevertheless, the inclusion of respondents across multiple demographic and behavioral strata—such as gender, generational cohort, city tier, occupational group, and kiosk usage frequency—enhances the external validity of the study within the context of kiosk users in China.
3.2. Measures and Instruments
All measurement items were adapted from previously validated scales and refined to suit the context of kiosk usage. Excluding the first six demographic and screening questions, the remaining 25 items were mapped onto five core latent constructs—Digital Readiness (DR), Trust in Technology (TT), Perceived Usefulness (PU), Perceived Barriers (PB), and Intention to Use (IU). Each construct was measured using five items, and all responses were recorded on a seven-point Likert scale ranging from 1 (Strongly Disagree) to 7 (Strongly Agree).
Statistical analyses were conducted using R (version 4.5.1). Confirmatory factor analysis (CFA), structural equation modeling (SEM), and tests of reliability and validity (including HTMT) were performed using the lavaan and semTools packages. Multicollinearity was assessed via VIF, and descriptive statistics were computed to ensure analytical precision and replicability.
Digital Readiness (DR) was measured using items adapted from Compeau and Higgins [
52], Bandura [
53], Davis [
9], and Venkatesh and Davis [
24], capturing respondents’ confidence in using digital technologies. Trust in Technology (TT) was assessed using modified items from Mcknight, Carter, Thatcher and Clay [
47], Gefen, Karahanna and Straub [
50], and Pavlou [
34], reflecting users’ perceptions of system reliability, integrity, and security. Perceived Usefulness (PU) was based on scales developed by Davis [
9] and Venkatesh, Morris, Davis and Davis [
25] measuring the degree to which kiosk services are perceived to enhance task performance and autonomy. Perceived Barriers (PB) drew from Venkatesh, Morris, Davis and Davis [
25] and Pavlou [
34], assessing psychological and practical concerns associated with kiosk usage—such as technological difficulty, breakdown risk, or discomfort with non-human interaction. Intention to Use (IU) was measured using items derived from Venkatesh, Morris, Davis and Davis [
25] and Fishbein and Ajzen [
54], representing respondents’ behavioral intentions to continue using kiosks and recommend them to others.
All items were originally written in English, translated into simplified Chinese, and back-translated by bilingual experts to ensure semantic equivalence and cultural appropriateness.
3.3. Sample Characteristics
In this study, a “kiosk user” is defined as an individual who has used a self-service terminal for ordering, payment, or information retrieval in commercial or public settings such as restaurants, hospitals, or transportation hubs. To ensure behavioral context relevance, kiosk usage experience was first confirmed through a screening question at the beginning of the survey, and respondents without prior experience were not allowed to proceed. This design choice was intentional, as the purpose of this study was to examine the psychological and cognitive mechanisms that emerge after direct usage experience.
The updated dataset comprises 750 valid responses collected through a combination of snowball and convenience sampling via WeChat in mainland China. This context reflects a high degree of digital infrastructure maturity and widespread kiosk penetration, providing an appropriate setting to explore psychological and behavioral mechanisms of kiosk adoption.
The sample exhibits considerable demographic diversity. Females accounted for 54.7%, males for 45.1%, and 0.2% identified as other or non-disclosed. In terms of generational cohort, 39.3% were Gen Z (born 1997–2010), 36.8% Gen Y (1980–1996), and 23.9% Gen X or older (born 1979 or earlier). City tier distribution included 24.0% from Tier 1 (highly developed cities), 36.7% from Tier 2, 27.7% from Tier 3, and 11.6% from rural or less-developed areas. Occupations were distributed across customer-facing roles (25.2%), office/administrative (23.2%), education/public/professional (26.0%), and self-employed/freelancer/others (25.6%).
Regarding kiosk usage frequency, 32.9% reported very frequent use (weekly or more), 43.9% reported frequent use (2–3 times/month), while 12.0% used kiosks occasionally (about once/month), and 11.2% rarely (less than once/month). This variation allows for comparative analysis across high- and low-experience user segments.
Although the sampling strategy is non-probabilistic, the inclusion of respondents across diverse demographic and behavioral strata demonstrates that the sample reflects a broad cross-section of actual kiosk users in China. This diversity strengthens the external validity of the study by capturing multiple usage contexts and user groups, thereby enhancing confidence that the findings are representative of the active kiosk user population. However, it is acknowledged that the study does not include kiosk non-users or digital avoiders, which may limit interpretability for digitally excluded populations.
Table 2 presents a comprehensive demographic and behavioral profile of the final sample.
4. Empirical Results
4.1. Measurement Model Evaluation
Framed by the Technology Acceptance Model (TAM) and informed by the Unified Theory of Acceptance and Use of Technology (UTAUT) for measurement practice, the measurement model was first evaluated to ensure that it met the psychometric standards required for subsequent structural analysis. The assessment covered four key aspects: internal consistency, convergent validity, discriminant validity, and overall model fit.
As shown in
Table 3, all five latent constructs—Digital Readiness (DR), Trust in Technology (TT), Perceived Usefulness (PU), Perceived Barriers (PB), and Intention to Use (IU)—exhibited exceptionally high internal consistency. Cronbach’s alpha and composite reliability (CR) values ranged from 0.975 to 0.980, well above the conventional 0.70 threshold, while the average variance extracted (AVE) values ranged from 0.885 to 0.906, exceeding the 0.50 criterion. These results confirm that the items within each construct consistently measure the same underlying concept, indicating strong convergent validity.
Confirmatory factor analysis (CFA) results further supported the robustness of the measurement model. As reported in
Table 4, incremental fit indices and residual-based indices indicated excellent model fit (CFI = 0.999, TLI = 0.999, RMSEA = 0.014, SRMR = 0.011), notwithstanding a marginally significant χ
2 (χ
2 = 305.792, df = 265,
p = 0.043), a common outcome in large-sample CFA applications. Additionally, all standardized factor loadings exceeded 0.94, underscoring the strong association between each item and its corresponding latent construct. While such high loadings can sometimes suggest redundancy, variance inflation factor (VIF) checks revealed no multicollinearity issues, indicating that the items capture distinct facets of the construct while maintaining conceptual cohesion.
Discriminant validity was confirmed using the Heterotrait–Monotrait (HTMT) ratio of correlations (
Table 5). All HTMT values ranged from 0.48 to 0.87, remaining below the conservative 0.90 threshold. Notably, PU and PB—both evaluative constructs—were clearly distinguished, supporting the notion from dual-process theory [
55] that positive and negative evaluations are processed through separate cognitive pathways.
Overall, the results in
Table 3,
Table 4 and
Table 5 collectively indicate that the measurement model demonstrates outstanding reliability and validity, providing a strong empirical foundation for testing the proposed structural relationships.
4.2. Descriptive Statistics and Multicollinearity Assessment
Before estimating the structural model, descriptive statistics were analyzed to understand the general perception patterns across the five constructs, and multicollinearity diagnostics were performed to ensure the independence of predictors.
As shown in
Table 6, mean scores ranged from 3.72 (PB) to 4.41 (IU) on a seven-point Likert scale. The highest mean for IU suggests that respondents generally hold favorable intentions toward kiosk usage. In contrast, PB recorded both the lowest mean and the highest standard deviation (SD = 2.00), indicating not only a relatively low perception of barriers overall but also substantial variation across individuals—implying that certain user segments perceive kiosk usage as significantly more challenging than others.
Internal consistency remained exceptionally high for all constructs (Cronbach’s α = 0.975–0.980), reaffirming the reliability of the measurement scales. VIF values ranged from 1.78 (DR) to 3.52 (PB), far below the common cutoff of 5.00, confirming the absence of problematic multicollinearity. While PU and PB exhibited relatively higher VIFs compared to other constructs, they still remained well within acceptable bounds, suggesting that the constructs capture related but statistically distinguishable aspects of the user experience.
Taken together, the descriptive statistics highlight a generally positive orientation toward kiosks, with variation in perceived barriers serving as a meaningful source of heterogeneity in adoption behavior. The multicollinearity results further confirm that the constructs are sufficiently independent for inclusion in the structural model.
4.3. Structural Model and Hypothesis Testing
The structural equation modeling (SEM) results indicated an excellent overall model fit (CFI = 0.989, TLI = 0.988, RMSEA = 0.041, SRMR = 0.068). All structural paths were statistically significant (p < 0.001), and the standardized estimates aligned closely with theoretical expectations, with one notable exception.
Table 7 summarizes the numerical results of hypothesis testing, and
Figure 2 illustrates the structural model with standardized path coefficients and hypothesis testing outcomes. Together, they provide both statistical and visual confirmation of the hypothesized relationships.
For H1, digital readiness (DR) exerted a strong positive effect on trust in technology (TT; β = 0.512,
p < 0.001), confirming that greater digital competence fosters higher levels of trust. This aligns with the self-efficacy perspective of Compeau and Higgins [
52] and Bandura [
53], which views familiarity and capability as foundational to trust formation. H2 predicted a positive relationship between DR and perceived usefulness (PU), which was supported (β = 0.481,
p < 0.001), indicating that digitally proficient users are more likely to view kiosks as enhancing efficiency and performance. Consistent with H3, DR had a significant negative impact on perceived barriers (PB; β = −0.266,
p < 0.001), suggesting that digital familiarity reduces cognitive and psychological resistance to technology use.
H4 hypothesized a positive direct effect of DR on intention to use (IU). However, the observed coefficient was significant but negative (β = −0.131, p < 0.001), opposite to the hypothesized direction. Accordingly, H4 was not supported in its direct form. This pattern is consistent with a suppression effect, wherein DR influences IU primarily through indirect pathways—via TT, PU, and PB—which are positive, while the residual direct path captures other factors. One possible explanation is that highly digitally proficient users may hold elevated performance expectations, making them more sensitive to even minor usability flaws or service delays. They may also prefer alternative channels, such as mobile applications, that they already trust and perceive as offering greater autonomy. As a result, while DR fosters adoption indirectly through trust and perceived usefulness, its direct effect can become negative when heightened expectations and alternative preferences outweigh its facilitating role.
H5 and H6 examined the effects of TT on PU and PB, respectively. TT significantly increased PU (β = 0.355,
p < 0.001) and reduced PB (β = −0.604,
p < 0.001), underscoring its dual role in enhancing positive evaluations and lowering perceived friction. For H7, TT emerged as the strongest direct predictor of IU (β = 0.541,
p < 0.001). This finding reaffirms prior studies by Pavlou [
34] and Gefen, Karahanna and Straub [
50] that highlight the centrality of trust in self-service technology adoption decisions.
Finally, H8 and H9 assessed the influence of motivational appraisals. PU exerted a small but significant positive effect on IU (β = 0.135, p < 0.001), whereas PB had a much stronger negative effect (β = −0.443, p < 0.001). This suggests that reducing barriers is more critical for ensuring sustained usage than enhancing usefulness.
Overall, eight of the nine hypotheses (H1–H3, H5–H9) were supported; H4 was not supported in its direct form due to the opposite sign, although indirect pathways via TT, PU, and PB were positive.
4.4. Multi-Group Structural Equation Modeling (SEM) Analysis
Multi-group SEM analyses were conducted to examine whether the structural relationships varied according to generational cohort and kiosk usage frequency.
Generational Cohort Comparison (Gen Z vs. Non-Gen Z) (
Table 8) revealed clear differences in the role of digital readiness (DR; H1–H4). Among Non-Gen Z users, DR exerted significant positive effects on trust in technology (TT; β = 0.338) and perceived usefulness (PU; β = 0.667), and a significant negative effect on perceived barriers (PB; β = −0.411). This indicates that for less digitally native users, digital competence plays a pivotal role in fostering trust, enhancing utility perceptions, and reducing friction. In contrast, for Gen Z, DR had negligible effects on TT (β = 0.044) and PU (β = 0.033), suggesting that baseline trust and utility perceptions are already high, leaving little room for improvement through readiness.
Across cohorts, the direct path from DR to IU remained small and negative (Non-Gen Z: β = −0.037; Gen Z: β = −0.038), indicating no support for H4 at the subgroup level.
The effects of TT (H5–H7) also varied. While TT increased PU (Non-Gen Z: β = 0.127; Gen Z: β = 0.635) and reduced PB (Non-Gen Z: β = −0.509; Gen Z: β = −0.706) in both cohorts, the magnitudes were larger for Gen Z, implying that once trust is established, younger users translate it more strongly into positive evaluations and barrier reduction. However, TT had a much stronger direct effect on intention to use (IU) among Non-Gen Z (β = 0.670) than among Gen Z (β = 0.154), indicating that trust serves as a primary motivational driver for older cohorts, whereas for Gen Z, barrier presence or absence is more decisive.
Consistent with H8–H9, PU positively influenced IU in both groups (Non-Gen Z: β = 0.094; Gen Z: β = 0.245), while PB exerted strong negative effects, with a substantially larger impact among Gen Z (Non-Gen Z: β = −0.306; Gen Z: β = −0.684). This highlights younger users’ heightened sensitivity to friction in service interactions.
Usage Frequency Comparison (High vs. Low) (
Table 9) produced a similar pattern. Among low-frequency users, DR significantly increased TT (β = 0.335) and PU (β = 0.689), and reduced PB (β = −0.423), indicating that digital competence is crucial for fostering positive perceptions among less experienced users. In contrast, for high-frequency users, DR had no meaningful effect on TT (β = −0.195), PU (β = −0.010), or PB (β = 0.003), suggesting that accumulated experience overrides initial competence.
Similarly, the direct DR → IU path was near zero and negative for both groups (Low: β = −0.033; High: β = −0.020), further indicating non-support for H4 across usage segments.
TT’s effects (H5–H7) were positive in both groups, enhancing PU (Low: β = 0.127; High: β = 0.625) and reducing PB (Low: β = −0.500; High: β = −0.710), with larger magnitudes for high-frequency users. This suggests that experienced users channel trust more directly into efficiency appraisals and friction reduction. However, TT’s direct influence on IU was much stronger for low-frequency users (β = 0.658) than for high-frequency users (β = 0.140), implying that trust formation is essential for novices, while minimizing barriers is more critical for habitual users.
H8–H9 analysis further confirmed that PU positively and PB negatively influenced IU in both groups, with the deterrent effect of PB particularly pronounced among high-frequency users (β = −0.758). This finding suggests that experienced users are more likely to reduce their intention sharply when encountering usability obstacles.
Taken together, both generational cohort and usage frequency function as important moderators in the pathways linking DR, TT, and motivational appraisals to IU. For digital natives and highly experienced users, removing barriers is the primary strategy for fostering adoption, whereas for Non-Gen Z and low-frequency users, building trust is the critical bridge between capability and intention.
Consistent with these patterns, the absence of a positive direct DR → IU effect suggests that DR operates primarily through TT, PU, and PB rather than directly influencing intention.
5. Robustness Check
To verify that the structural equation modeling (SEM) results were not biased by violations of distributional assumptions such as non-normality and heteroskedasticity, the hypothesized model was re-estimated using the robust maximum likelihood estimator (MLR). This method adjusts standard errors and chi-square statistics, enabling more reliable inference under less-than-ideal data conditions.
As shown in the robustness check results (
Table 10), the fit indices from the MLR estimation were virtually identical to those from the ML estimation, confirming the stability of overall model fit (CFI = 0.989, TLI = 0.988, RMSEA = 0.041, SRMR = 0.068). All values comfortably exceeded conventional thresholds (CFI/TLI ≥ 0.95, RMSEA ≤ 0.05, SRMR ≤ 0.08), indicating that the model maintained excellent fit even under robust estimation.
Path coefficients also remained stable in both magnitude and significance. Notably, the direct DR → IU path (H4) remained negative and significant under MLR (β = −0.131, p < 0.001), thus failing to support the hypothesized positive direction. In contrast, digital readiness (DR) continued to positively influence trust in technology (TT; H1, β = 0.512), while TT enhanced perceived usefulness (PU; H5, β = 0.355) and reduced perceived barriers (PB; H6, β = −0.604). TT also remained the strongest direct predictor of intention to use (IU; H7, β = 0.541). The negative effect of PB on IU (H9, β = −0.443) again outweighed the positive effect of PU (H8, β = 0.135).
Furthermore, the bootstrap results in
Table 11 reaffirmed the robustness of these relationships. All 95% confidence intervals excluded zero, and the effect sizes were consistent with the MLR results. In particular, the DR → IU path was again negative and significant (β = −0.116, 95% CI [−0.16, −0.07]), confirming non-support for H4 due to the opposite sign. The DR → TT path (β = 0.475, 95% CI [0.40, 0.54]) and TT → IU path (β = 0.517, 95% CI [0.43, 0.60]) both exhibited high significance and stability, reinforcing TT as the most influential direct predictor of IU.
In summary, as evidenced in
Table 10 and
Table 11, the proposed structural relationships are robust to both estimation method and sampling variability. Trust in technology (TT) functions not only as a strong direct determinant of IU but also as a consistent mediator in the DR → TT → IU pathway. Although indirect effects are not reported here, including bootstrapped indirect effects in future drafts would further quantify the mediation mechanism.
6. Conclusions
This study empirically examined how digital readiness (DR), trust in technology (TT), perceived usefulness (PU), and perceived barriers (PB) shape the intention to use kiosks (IU), incorporating multi-group analyses by generational cohort and usage frequency. Beyond simple hypothesis testing, the results provide deeper insights into the mechanisms of kiosk adoption.
First, the analysis confirms that adoption in self-service technology contexts is not driven solely by usefulness, as suggested by the traditional Technology Acceptance Model (TAM) [
9,
25], but rather by a more complex interplay in which trust and barrier reduction emerge as the most decisive factors. TT exerted both direct and indirect positive effects on IU, while PB consistently demonstrated a stronger negative influence than the positive effect of PU. This finding underscores the importance of reducing friction as a precondition for sustained adoption.
Second, the results reveal the dual role of digital readiness. While DR indirectly fosters adoption by strengthening TT and PU and lowering PB, its direct path to IU was significantly negative. This paradox suggests that high digital proficiency may raise performance expectations, heighten sensitivity to usability issues, and foster preferences for alternative channels or greater autonomy. Accordingly, DR should be conceptualized not as a simple enabler but as a contingent factor with both facilitative and constraining effects.
Third, the multi-group analysis shows that adoption pathways differ across user segments. For non-digital natives and low-frequency users, trust formation is the primary driver, whereas for digital natives and high-frequency users, barrier minimization is more decisive. This highlights the need for tailored adoption strategies that account for generational and experiential diversity rather than relying on uniform approaches.
Taken together, these findings extend TAM by demonstrating the centrality of trust and barriers, validating the context-dependent role of DR, and empirically confirming the segmented nature of kiosk adoption mechanisms. These insights provide both theoretical refinement and practical guidance for developing differentiated strategies.
6.1. Theoretical Implications
Unlike UTAUT2, which extends TAM mainly by adding motivational and social influence factors, this study takes a different direction. It emphasizes negative determinants (PB), redefines digital readiness (DR) as an ambivalent construct, and incorporates segmentation logic. In this way, the model is not just an additional extension but a theoretical redirection that explains SST adoption through friction minimization and contextual variability.
First, by adding trust in technology (TT) and perceived barriers (PB) to TAM, the study moves the focus away from the traditional usefulness-centered view. Instead, it shows that building trust and reducing barriers are critical drivers of behavioral intention. This helps to correct the imbalance in earlier TAM studies that paid too much attention to perceived usefulness (PU) and perceived ease of use (PEU).
Second, the findings redefine digital readiness (DR). Rather than being a simple positive enabler, DR shows a dual role. It strengthens adoption indirectly through TT and PU but can also suppress adoption when expectations or autonomy concerns become too strong. This helps explain why user competence can sometimes encourage adoption but in other cases create resistance.
Third, the study introduces segmentation into TAM by showing that adoption pathways differ across generations and usage frequencies. Non-digital natives and low-frequency users rely more on trust, while digital natives and high-frequency users are influenced more by barrier reduction. This highlights why a one-size-fits-all model is not sufficient and why TAM needs to be adaptive to user segments.
Finally, the results show that PB has a stronger and more stable effect than PU. This asymmetry suggests that reducing friction and barriers may be more effective than increasing usefulness. It calls for rethinking SST adoption from a barrier-sensitive perspective, where inhibitors are seen as central, not secondary.
In summary, this study differentiates itself from earlier TAM extensions by focusing on inhibitors, redefining readiness, and embedding segmentation logic. These contributions refine TAM into a trust-centered, barrier-sensitive, and context-aware framework that better explains SST adoption in today’s diverse digital environments.
6.2. Business Implications
The findings of this study suggest that kiosk adoption strategies must go beyond generic implementation and directly address the asymmetries revealed in our model.
First, the stronger impact of perceived barriers compared to perceived usefulness indicates that firms should shift their managerial focus from promoting benefits to eliminating frictions. Traditional marketing emphasizes “why kiosks are useful,” but our evidence shows that adoption depends more on removing doubts, inconveniences, and risks. This insight calls for a strategic reorientation: reducing friction is not supplementary, it is the core value proposition of self-service technologies, consistent with prior SST research emphasizing the salience of barriers [
23].
Second, the dual role of digital readiness challenges a common managerial assumption. Highly skilled digital users are not necessarily easy adopters. Instead, they are the most demanding customers, quick to abandon kiosks when performance is suboptimal. For managers, this implies that sustaining adoption among digital natives requires investments in reliability, autonomy-respecting design, and multi-channel flexibility. This aligns with prior findings that technologically ready users often hold elevated expectations [
38].
Third, industry contexts reveal distinct managerial priorities. In hospitality, adoption hinges on reducing psychological barriers through multilingual support and visible assistance, particularly for international tourists. In healthcare, kiosk use will fail unless trust is actively cultivated through strict privacy safeguards and transparent data practices. In transportation, speed and error-free performance are not operational add-ons but the decisive conditions for adoption in high-volume environments. In retail and food service contexts, streamlined ordering processes and seamless payment integration can further strengthen adoption.
Finally, our results emphasize that adoption strategies must be temporally staged. Initial adoption requires trust-building and onboarding; sustained usage requires continuous reliability; and long-term loyalty requires personalization and predictive maintenance. Misallocating resources across these stages—for example, overspending on long-term personalization before establishing trust—may undermine adoption.
By reframing adoption as a process where friction minimization outweighs utility maximization and by identifying the dual role of user readiness, this study provides actionable guidance that challenges prevailing managerial assumptions. These insights equip service providers to design not just functional kiosks, but adoption ecosystems aligned with diverse user segments and industry contexts [
20].
6.3. Limitations and Future Research
This study has several limitations, and each of them suggests avenues for further research.
Research design limitations. Because the study used a cross-sectional design, it cannot establish causal relationships. In addition, the use of a non-probability sampling method restricts generalizability. These methodological choices were reasonable for capturing a snapshot of kiosk adoption among active users, but they do not allow for the observation of long-term changes or resistance at the early stages of adoption. Future studies should therefore employ longitudinal or experimental designs to examine how trust, perceived barriers, and adoption behaviors evolve over time.
Sample-specific limitations. The analysis was based exclusively on current kiosk users. This design made it possible to investigate adoption mechanisms grounded in actual experience, yet it also meant that non-users and digitally excluded populations were not represented. As a result, important early-stage barriers such as technophobia, distrust of automation, or digital inequality were not fully captured. Because the study relied on active users and a non-probabilistic method, the findings should be generalized with caution. Even so, by focusing on experienced users, this study provides a benchmark for understanding the mechanisms that operate after direct interaction, which can complement future research on first-time adoption.
Segmentation limitations. Respondents were segmented by generational cohort and usage frequency, both of which are meaningful dimensions for digital proficiency and exposure. Other demographic factors, such as gender, occupation, or city tier, may also influence adoption pathways. Future research could explore these additional dimensions through latent class analysis, machine learning approaches, or mixed-method designs in order to build more refined user profiles. Such methods would allow for a more integrated view of psychological traits, trust orientations, and technology experience. Acknowledging this limitation points to concrete directions for advancing both methodology and theory.
Contextual limitations. The research was conducted in China, which is characterized by advanced digital infrastructure and widespread kiosk penetration. This makes China a highly suitable setting for studying kiosk adoption, but it also restricts the ability to generalize the findings to markets with lower levels of digital maturity. Nevertheless, prior Western SST adoption studies have confirmed universal mechanisms such as trust, barriers, and technology-related anxiety [
23,
27,
56]. These parallels suggest that the processes identified in this study are not uniquely Chinese but share common ground with global patterns, thereby enhancing the cross-cultural relevance of our findings. At the same time, this study contributes beyond earlier work by revealing the dual role of digital readiness and by demonstrating that adoption pathways differ across generational and usage-frequency segments—insights that open new avenues for theory building and comparative research.
Theoretical opportunities. The findings challenge the traditional assumption in TAM that effects are linear and consistently positive. The suppression effect of digital readiness and the stronger influence of perceived barriers relative to perceived usefulness suggest that future extensions of TAM should take nonlinear and context-dependent dynamics into account. Future research should also examine affective responses such as frustration, anxiety, or confidence as possible mediating or moderating factors [
57]. These dynamics are likely to become even more important in AI-augmented or autonomous SST environments.
By organizing the limitations thematically and reflecting on their implications, this study not only clarifies its boundaries but also positions itself as both an empirical and theoretical benchmark for future research across different user groups and market contexts.