Relative advantage, as articulated by Rogers (1995), denotes the extent to which an invention is regarded as superior to the concept it supplants, rendering it a pivotal element in technology adoption. Some studies have emphasised the beneficial impact of relative advantage on the adoption of FinTech and associated technologies. Marei et al. (2023) and Owusu et al. (2021) asserted that it is improbable for prospective adopters to accept new technology before a conviction that its advantages surpass those crystallising from current options. Almashhadani et al. (2023) reinforce this assertion, revealing that consumers are more predisposed to embrace technologies that offer improved productivity, less effort and time efficiency. Iluba and Phiri (2021) and Moghavvemi et al. (2021) similarly identified a significant association between relative advantage and FinTech adoption, confirming its status as a crucial predictor.

Further research, including that of Qalati et al. (2021) and Mehra, Rajput and Paul (2022), similarly determined that perceived advantages, such as enhancements in efficiency and productivity, incentivise SMEs and individuals to embrace technology. Nevertheless, divergent results are present in extant literature, as Pan et al. (2022) and Sukumaran, Bee and Wasiuzzaman (2022) observed no substantial impact of relative advantage on the adoption of artificial intelligence and other technologies, respectively. Rani and Kumar (2023) as well as Mangula, Van de Weerd and Brinkkemper (2014) similarly found that relative advantage did not significantly affect behavioural intentions to use technology.

These divergent findings emphasise that although relative advantage frequently serves as a crucial factor in adoption decisions, its relevance may differ across contexts, technologies and sectors. Consequently, results from prior studies cannot be relied upon to accurately predict what the exact situation would be among SMEs in South Africa as it specifically concerns the nexus of relative advantage and FinTech adoption. This study remains duly cognisant of the different positions in the literature but is also mindful of the highly competitive nature of the South African SME sector that heightens the quest of a business to seek advantage over others in order to remain in business. It is against this background that this study elects to hypothesise (H) that in the context of South Africa as a developing economy:

Trialability denotes the degree to which prospective adopters can test an innovation on a restricted scale, as articulated by Ho et al. (2020) and Ullah et al. (2021). The capacity to trial a product or service is generally regarded as a characteristic that enables adoption, though research evidence about its role in technology adoption is inconsistent. Results stemming from studies conducted by Iluba and Phiri (2021) as well as Sun and Chi (2018) indicate that trialability is crucial for fostering adoption, as it enables prospective adopters to gain direct experience, establish confidence and evaluate the functioning and compatibility of an innovation.

Kaur et al. (2023) as well as Johnson et al. (2018) also emphasise that trialability instils confidence in users regarding the validity and security of products, hence facilitating adoption. More specifically, Tiwari et al. (2023) observed that trialability facilitates user acclimatisation to FinTech goods, hence promoting their utilisation. Conversely, some other studies, including those by Bureshaid (2021), Anh (2023), as well as Rani and Kumar (2023), revealed no significant effect of trialability on the adoption of technologies like mobile banking, suggesting that its effects may differ across settings. The divergent findings indicate that although trialability may influence technology adoption, the nature of its role may vary as it is contingent on the characteristics of the innovation, the user demographic and contextual elements such as perceived risk and technological familiarity.

In the quest to hypothesise the nature of the relationship between trialability and adoption of FinTech, this study is mindful of the different positions advanced by scholars but is swayed by the reality that SMEs are wary of the high failure rates that plague the sector and can ill-afford to institutionalise business practices that they have not tested. It is on this basis that it is hypothesised that in the South African context, that:

As articulated by Dearing and Cox (2018), observability denotes the extent to which the results and advantages of employing a technology are apparent to others. Minishi-Majanja and Kiplang’at (2005) contended that observable outcomes of FinTech generate social discourse that promotes adoption, whereas Mehra et al. (2022) observed that younger demographics frequently display their utilisation of new technology to their friends, resulting in increased adoption rates. Indeed, observability is regarded as a significant factor influencing acceptance (Hsu & Lin 2016; Kolodinsky, Hogarth & Hilgert 2004). Corroborating this position, Badi et al. (2021) emphasised the significance of observability in the process of the adoption of smart contracts within the United kingdom (UK) construction sector. Observability diminishes uncertainty by demonstrating beneficial outcomes, hence facilitating the dissemination of technology (Elmustapha et al. 2018; Sun & Chi 2018).

A slightly different opinion was canvassed by Ahmad et al. (2023), whose study revealed that observability, albeit influential, was the least important determinant of home technology adoption in Pakistan. Contrarily, reliant on findings from different studies, Kapoor, Dwivedi and Williams (2013) as well as Iluba and Phiri (2021) concluded that observability did not significantly affect the adoption of mobile payments and technologies in India and Zambia, respectively. Similarly, in a study undertaken by Lin et al. (2020a), it was determined that observability did not demonstrate a statistically significant relationship with technology adoption by SMEs. The conflicting findings indicate that although observability might facilitate adoption by demonstrating advantages and diminishing ambiguity, the definitive role that it plays may differ based on different contexts. Some of the positions advocated in prior research, the fact that positive social perceptions often dictate business behaviour of SMEs and the reality that observable benefits would encourage particular business practices, influence the study to hypothesise that:

Compatibility, characterised as the congruence of a technology with the end-user’s beliefs, expectations and needs (Ullah et al. 2021), may be pivotal in the adoption of technologies. This notwithstanding, some studies, such as that conducted by AlBar and Hoque (2019) found no significant correlation between compatibility and technology adoption in rural Saudi Arabia. Similarly, the results of the study undertaken by Maroufkhani et al. (2020) revealed that compatibility did not influence the adoption of big data analytics by SMEs.

In contrast, Okoli and Tewari (2021) contend that technologies that align with established standards and previous experiences are more likely to be embraced. Wang, Zhao and Cheng (2022) emphasised that compatibility affects technology adoption. Compatibility is recognised as a significant factor in FinTech acceptance in Tanzania (Iluba & Phiri 2021; Koloseni & Mandari 2024). This position is given credence by the finding that compatibility has been demonstrated to facilitate the adoption of mobile banking (Ho et al. 2020) and digital banking (Tiong 2020).

In addition to its direct effects, compatibility indirectly facilitates FinTech adoption by improving perceived usefulness and perceived ease of use (Purwantini, Athief & Waharini 2020; Sunardi et al. 2022). Also, compatibility promotes financial inclusion, as emphasised by Hayat and Ul Hameed (2024), by facilitating the use of FinTech services. Though there are contrasting thoughts as it concerns how compatibility – a complex and context-sensitive phenomenon – relates to FinTech adoption, a recurrent opinion in the literature seems to be that they have a positive association. Additionally, from a business practice perspective, it would seem plausible that SMEs would be driven to employ systems that are compatible with their businesses. It is in light of these that the current study projects that:

Complexity, defined as the level of difficulty consumers encounter in comprehending or utilising a technology (Ching, Teoh & Amran 2020), is commonly regarded as an impediment to the adoption of FinTech. Technologies regarded as intricate or challenging to utilise often dissuade prospective adopters, as consumers exhibit reluctance to dedicate time and effort to mastering their use (Shanmugam & Shanmugam 2021; Tiwari et al. 2023). Research indicates that reduced complexity facilitates the acceptance of technology solutions, since customers are more inclined to accept simpler systems (Riahi & Garrouch 2023). In studies conducted by Matsepe and Van der Lingen (2022) as well as Moghavvemi et al. (2021), it was revealed that users promptly discarded technology perceived as complex. The notion of complexity deters adoption by inducing user reluctance to engage with the technology (Nasution et al. 2023).

Nonetheless, studies by AlBar and Hoque (2019) as well as Bureshaid (2021) identified no substantial correlation between complexity and technology adoption. Research on mobile banking by Anh (2023) also revealed that complexity did not significantly influence its adoption. The varied findings indicate that although complexity is typically viewed as an obstacle, its effect on adoption may differ from context to context. Though there is some extent of discordance in various scholarly perspectives presented in existing literature, rational thinking suggests that complex systems may not necessarily hold sufficient appeal to SMEs that would lead to their adoption, as they may complicate business processes. Based on this conviction, the current study posits that:

Leveraging the hypothesised relationships deductively derived by considering different positions in extant literature concerning DOI components and technology adoption, a research model was developed for the study, and it is presented in Figure 1.

The purposive non-probability sampling technique was employed to determine the pool of respondents for the study. Purposive sampling was considered suitable for this investigation primarily because of the absence of a comprehensive sampling frame. Additionally, randomisation was deemed largely impractical, given the extensive and heterogeneous population of SMEs in South Africa (Small Enterprise Development Agency [SEDA] 2022), which complicates the establishment of a reliable sampling frame (Etikan, Musa & Alkassim 2016). The study is cross-sectional and utilised an online survey methodology.

Online surveys offer researchers several benefits, such as cost-effectiveness, the ability to pause and continue the survey, and the swift completion of the research instrument by participants (Nayak & Narayan 2019). Of the 1533 online responses gathered, 1065 were classified as valid, while 468 were regarded as invalid as a result of incompleteness, resulting in a usable rate of 69.47%. Based on prior research regarding the determinants of FinTech adoption, which utilised approximately 362 responses (Sharma et al. 2024), 386 responses (Wu & Peng 2024) and 399 responses (Amnas et al. 2023), the 1065 responses employed in this study are deemed sufficient.

A structured questionnaire was employed to acquire data for the study. The constructs relative advantage (RA), trialability (TRI), observability (OBS), compatibility (COMPT) and complexity (CPLEX) were regarded as singular constructs and assessed with measurement instruments adapted from the studies of Moore and Benbasat (1991), Al-Jabri and Sohail (2012), as well as Gholami, Abdekhoda and Gavgani (2018). The adaptation mainly involved minor word changes from ‘mobile banking’ to ‘FinTech adoption’ and the integration of scale items drawn from the studies of the identified authors. This culminated in having five items for the measurement of each of the DOI component constructs. Additionally, adoption of FinTech (ADO) was assessed using an instrument derived from the study of Dulcic, Pavlic and Silic (2012) along with that of Shahbaz et al. (2020). The RA, TRI, OBS, COMPT, CPLEX and ADO measuring scales were evaluated on a five-point Likert scale with answer options ranging from strongly disagree (1) to strongly agree (5).

A pilot study involving 30 respondents (comprising owners and managers) was conducted prior to the main study, primarily to help determine if any questions were ambiguous or lacked clarity in a bid to determine the content validity of the instrument. This act aligns with the thoughts of Creswell and Creswell (2017), who contend that a pilot study enables the researcher to establish content validity. The results of the pilot study in relation to content validity and internal consistency of scale items were deemed satisfactory.

Instructively, the study recognised that utilising the same questionnaire to gather data for both independent and dependent variables can lead to common method bias (CMB). Consequently, the Harman single-factor test was employed to analyse CMB, to establish the existence a single-factor variance percentage below 50%, as per the precedence of Abbasi et al. (2021). The Harman single-factor test results indicated that one factor accounted for 35.51% of the total variance, suggesting that CMB is not an issue in this study.

The hypothesised relationships were examined using Covariance-based Structural Equation Modelling (CB-SEM). AMOS version 29 (IBM Corp., Armonk, NY, US) was used to perform the CB-SEM analysis in this study. Covariance-based structural equation modelling is a statistical analysis method designed to investigate the interrelationships among many variables within a model (Mia, Majri & Rahman 2019). Covariance-based structural equation modelling is suitable for validating theoretically hypothesised relationships (Reinartz, Haenlein & Henseler 2009), as demonstrated in this study. The CB-SEM consists of two main phases: (1) it evaluates the measurement model (i.e. reliability and validity) and (2) it analyses the structural model by investigating the linear correlations between the suggested latent variables (De Carvalho & Chima 2014). The measuring model was evaluated by assessing the reliability of the scales using Cronbach’s alpha, while convergent validity was examined through the average variance extracted (AVE), and discriminant validity was investigated through the heterotrait-monotrait ratio of correlations (HTMT). The second stage in CB-SEM concentrates on calibrating the structural model by evaluating the significance of the interactions among latent variables (Iacobucci 2010).

Ethical clearance to conduct this study was obtained from the University of Pretoria, Faculty of Economic and Management Sciences Research Ethics Committee (No. EMS129/23).

The study employed Confirmatory Factor Analysis (CFA) to assess the measurement model utilised in the study. Table 1 presents the results for internal consistency reliability, convergent validity and discriminant validity for RA, TRI, OBS, COMPT, CPLEX and ADO.

The factor loadings were assessed using a cut-off point of 0.600, following Palmieri et al. (2023:5), where indicators below this threshold are considered to demonstrate weaker item reliability. Following the identification of lower-loading indicators, a conservative approach was adopted in line with Howard (2020:916). Items below the threshold were carefully reviewed, and indicators including TRI1 (0.551), TRI5 (0.448), OBS1 (0.504), OBS5 (0.101), COMPT1 (0.506), COMPT5 (0.536) and CPLEX2 (0.476) were removed. However, CPLEX1 (0.585) and ADO1 (0.335) were retained because removing these indicators negatively affected construct reliability, convergent validity and overall model fit. Their retention was also supported by theoretical relevance, ensuring that the conceptual meaning of the constructs was preserved. Therefore, a conservative decision was made to retain these items rather than rely solely on statistical thresholds.

The retained indicators demonstrated acceptable measurement quality, with Cronbach’s alpha values ranging from 0.744 to 0.860 and composite reliability values ranging from 0.745 to 0.862, exceeding the recommended threshold of 0.700. Convergent validity was supported for most constructs; however, the AVE values for TRI (AVE = 0.494), OBS (AVE = 0.501), and ADO (AVE = 0.610) indicate marginal convergence for some constructs. In particular, TRI recorded an AVE slightly below the recommended 0.500 threshold. Following Fornell and Larcker (1981) as well as Wach, Stephan and Gorgievski (2016), TRI was retained because AVE values between 0.400 and 0.500 may be acceptable when composite reliability is satisfactory and theoretical validity is established. Nevertheless, the marginal AVE value of TRI, alongside the relatively modest AVE values of OBS and ADO, is acknowledged as a limitation, and the findings involving these constructs should be interpreted with appropriate caution.

An HTMT criterion of 0.850 was adopted to assess discriminant validity, following Henseler, Ringle and Sarstedt (2015) and Rashid, Noranee and Bahtar (2023). Although HTMT was initially developed within the partial least square-structural equation modelling (PLS-SEM) context, Voorhees et al. (2016) extended its application to CB-SEM and provided empirical evidence supporting its effectiveness as a reliable assessment of discriminant validity. Therefore, HTMT was adopted in this study as a complementary and more sensitive measure compared with traditional approaches such as the Fornell–Larcker criterion, which may fail to identify discriminant validity issues in certain circumstances. The HTMT ratios for the study constructs ranged from 0.003 to 0.649, which are below the recommended threshold of 0.850. These findings confirm adequate discriminant validity, indicating that the constructs are empirically distinct and that the measurement items capture different theoretical concepts. Consequently, the measurement model demonstrates satisfactory construct differentiation and provides support for subsequent structural model assessment.

A structural model was developed to examine the proposed interactions, as illustrated in Figure 2. The model analysed the correlation among RA, TRI, OBS, COMPT, CPLEX and ADO. The outcomes of the model fit for the path analysis model depicted in Figure 2 indicate that the model achieved a satisfactory fit, as evidenced by Chi-square (minimum discrepancy) divided by degrees of freedom (CMIN/DF) = 2.864; comparative fit index (CFI) = 0.967; Tucker-Lewis index (TLI) = 0.960; normed fit index (NFI) = 0.951; goodness of fit index (GFI) = 0.953; adjusted goodness of fit index (AGFI) = 0.936; root mean square error of approximation (RMSEA) = 0.042; and standardised root mean square residual (SRMR) = 0.050.

The structural path analysis model results show that the path coefficient for the association between RA and ADO is −0.28, whereas the path coefficients for the independent variables TRI, OBS, COMPT, CPLEX and the dependent variable ADO are 0.50, −0.01, 0.64 and 0.09, respectively.

Table 2a and Table 2b presents the coefficients, t-values, p-values and statistical findings about the direct relationships between the independent variables RA, TRI, OBS, COMPT, CPLEX and the dependent variable ADO.

The structural equation modelling (SEM) results indicated a significant negative correlation between RA and ADO (β = −0.277, t = −2.750, p = 0.006); thus, H₁ was not supported because of the inverse direction of the relationship as demonstrated by the results. Additionally, a statistically significant positive correlation was shown between TRI and ADO (β = 0.504, t = 5.187, p = 0.000), indicating that H₂ is supported. The study revealed that OBS and ADO exhibited a negative and statistically insignificant correlation (β = −0.014, t = −0.123, p = 0.902), which implies that H₃ was not supported. The findings of the study also indicate that the correlation between COMPT and ADO was positive and statistically significant (β = 0.644, t = 3.390, p = 0.000), which means that H₄ is supported by the study’s results. The research findings reveal a positive and statistically significant association between CPLEX and ADO (β = 0.091, t = 2.560, p = 0.010). This is contrary to the hypothesised negative relationship between the two constructs, and therefore the results signal that H₅ is not supported.

In summary, the study findings indicate that TRI and COMPT promote ADO, while RA, OBS and CPLEX exhibit no statistically significant correlation with ADO among the cohort of studied SMEs in South Africa. Notably, the results of the study also show that the predictive model for RA, TRI, OBS, COMPT, CPLEX and ADO has a coefficient of determination of 0.766 (R² = 0.766). This means that 76.6% of the variance in ADO is explained by RA, TRI, OBS, COMPT and CPLEX.

This research provides theoretical and practical insights on FinTech adoption by SMEs, especially within the South African setting. The study theoretically enhances understanding of the diffusion of innovation theory by offering empirical evidence regarding the varying effects of determinants, including relative advantage, trialability, observability, compatibility and complexity, on FinTech adoption among a sample of SMEs. The theory suggests that these factors often promote technology adoption; however, the study indicates that not all factors exert equal influence in the context of the SMEs in South Africa that participated in this study, as only trialability and compatibility emerged as key drivers relative to others.

These findings enhance the theoretical framework by illustrating the intricate effect of these factors in influencing FinTech adoption in an emerging market context. The study’s emphasis on SMEs, a vital yet frequently overlooked sector, addresses a substantial gap in the literature and enhances the theoretical comprehension of technology adoption in smaller firms. The study elucidates the distinctive elements affecting FinTech adoption among the examined cohort of SMEs, offering critical insights that can enhance theoretical models and frameworks specific to this sector.

This emphasis on SMEs enhances the overall discussion on technology adoption by integrating the viewpoints and experiences of smaller enterprises. The study provides practical insights for FinTech providers and governments seeking to enhance adoption rates among SMEs. By highlighting trialability and compatibility as crucial antecedents of FinTech adoption, the study emphasises the need for creating tech-solutions that are readily testable and can effortlessly integrate with current SME operations. The findings also challenge the premise that perceptions of relative advantage, observability and reduced complexity always enhance adoption, amplifying the need for customised approaches that cater to the specific concerns and operational constraints of SMEs.

This study employed a quantitative technique reliant on self-reported data from participants, potentially introducing biases and inaccuracies in the gathered replies. Future research should implement a mixed-methods approach, as this would engender the generation of more profound insights as it pertains to the antecedents of FinTech adoption in SMEs. The present study is a cross-sectional analysis that does not account for potential changes in SME behaviour over time concerning the adoption of FinTech specifically, and technology broadly. Subsequent studies could utilise a longitudinal methodology to elucidate the fluctuations in SMEs’ behaviour regarding technology adoption. Future researchers should also consider undertaking a comparative analysis of the determinants influencing FinTech adoption by SMEs by using the technology acceptance model (TAM) and the technology-organisation-environment (TOE) as theoretical prisms to elucidate the dynamics associated with the geographical advantages and economies of scale of SMEs. Finally, the findings of this study are ultimately constrained by reliance on a non-random sample for responses, which invariably limits the generalisability of the findings. Consequently, it is recommended that future studies rely on random sampling techniques to create a pool of respondents for their studies.