Consumer resistance to AI chatbots: barriers and impacts on negative word-of-mouth

AI-powered conversational agents (AICAs) or chatbots are AI-based interfaces that imitate human communication through written or oral means using natural language processing and machine learning (Hentzen et al., 2022; Jan et al., 2023; Yang et al., 2023). The projected growth of the chatbot in the global Banking, Financial Services and Insurance sector, expected to increase from US$586m in 2019 to US$6.83bn by 2023 (Vailshery, 2022). However, 54% of consumers avoid chatbots for sensitive financial matters (McNamee, 2022), highlighting the importance of understanding the factors driving resistance to AICAs.

The extant literature focuses on chatbot adoption in financial services, neglecting resistance, especially in emerging markets such as Egypt (Chaouali et al., 2024; Jisham et al., 2024). On the other hand, findings from developed contexts often fail to account for the cultural and infrastructural complexities of developing markets, where consumer resistance is influenced by fragile banking systems, concerns over e-banking quality and fear of electronic fraud (Elsotouhy et al., 2023). This study addresses this gap by examining consumer resistance in this context.

In Egypt, financial service usage is growing, with internet penetration reaching 72.2% (Go-Globe, 2024), and financial service accounts increasing by 181% between 2016 and 2024 (CBE, 2024). However, cultural and psychological barriers shape consumer resistance, causing a decrease in the adoption rate of financial technologies such as mobile payments, mobile banking and e-wallets (Bakr et al., 2023; El Din et al., 2023). Despite increasing financial inclusion, there is a limited understanding of consumer resistance to AICA in the Egyptian context (El Din et al., 2023; El-Shihy et al., 2024), highlighting a critical gap.

This study adopts Innovation Resistance Theory (IRT) (Ram and Sheth, 1989) to explore consumer resistance to AICAs in financial services. In contrast to the novelty-seeking paradigm (e.g. UTAUT and TAM) that focuses on adoption drivers, IRT focuses on functional and psychological barriers, such as perceived risks, complexity and traditional preferences (Jan et al., 2023; Yang et al., 2023). This framework is particularly relevant for Egypt’s banking sector, where cultural, infrastructural and psychological factors significantly influence resistance to digital transformation (Elsotouhy et al., 2023).

However, despite the advantages of IRT, previous studies have revealed inconsistencies in understanding the barriers that shape consumer innovation resistance (CIR) (Table 1). Such contradictions highlight the need for further exploration of the specific barriers that influence the resistance to AICAs. This leads to the first research question:

Additionally, previous studies examined the impact of IRT barriers on outcomes such as intention to use (Migliore et al., 2022), actual usage (Sivathanu, 2019), non-adoption intention (Behera et al., 2023), hedonist, social and actualized innovativeness (Chu, 2023). However, limited attention has been paid to negative word-of-mouth (NWOM) (Jana, 2022), which is a critical consequence of resistance (Hentzen et al., 2022; Huang et al., 2021). Previous studies on financial service chatbots indicate that individuals who resist chatbots tend to disseminate unfavorable details regarding this innovation (Jana, 2022), which can discourage adoption and increase resistance (Mani and Chouk, 2018). This is particularly relevant in Egypt, where trust and human interaction are pivotal (Elsotouhy et al., 2023; Esawe, 2022). This prompts the following question:

Previous studies on IRT barriers’ influence on recommendation behavior have inconclusive results (Kaur et al., 2020, 2021; Talwar et al., 2021), indicating an overlooked mediating mechanism (Jana, 2022). NWOM significantly impacts businesses because responses to both functional and psychological barriers can lead to consumer resistance (Chu, 2023; Leong et al., 2021), negative emotions (Chang and Hsiao, 2024) and the propagation of NWOM (Talwar et al., 2021), affecting profitability and customer engagement. This leads to the third research question:

Guided by prior research, the first objective of this study is to investigate the impact of functional and psychological barriers on consumer resistance to AICA adoption. Second, it identifies the impact of CIR on NWOM. Finally, it examines the mediating role of CIR in the relationship between these barriers and NWOM.

This research has both theoretical and practical contributions, as it adds to the existing scant knowledge regarding chatbots in financial services by identifying the impact of usage, risk, tradition, value and image barriers (IBs) on CIR; extending the theoretical understanding of CIR and NWOM in the context of financial services; exploring the mediating effects of CIR on the relationship between functional and psychological barriers and NWOM; and providing actionable insights for financial service providers to address consumer resistance and provide guidance on integrating chatbot tools into customer support.

AI systems have become game changers in financial services, offering opportunities, such as reducing costs, enhancing customer experience, providing better services and increasing efficiency (Akyüz and Mavnacıoğlu, 2023). Chatbots are widely adopted across sectors including banking (Abdel Wahab, 2023) and insurance (Dekkal et al., 2023; Patil et al., 2024). Banks use chatbots to offer immediate customer services or redirect customer inquiries to appropriate service employees, thus enhancing the customer experience (CFPB, 2023). However, the implementation of chatbots remains challenging. For instance, although anthropomorphic chatbots may build psychological attachments, they can also raise privacy concerns, discomfort and perceived intrusiveness in critical financial decisions (El Din et al., 2023; Patil et al., 2024; Zhu, 2023).

Customers often experience frustration from automated systems that hinder communication with employees or respond inappropriately to user requests because of their limited conversational capabilities (CFPB, 2023). These limitations create a discrepancy between user expectations and service performance (Yang et al., 2023). Such issues contribute to unfavorable customer behaviors, including resistance, which negatively affects service providers and customers (Akyüz and Mavnacıoğlu, 2023).

Existing literature explores the technical aspects and characteristics of chatbots, such as their interface design, interaction capabilities and AI-based natural language processing (Mariani et al., 2023). Additionally, studies have examined user experiences and preferences, including perceptions of human likeness, trust in chatbots (El Din et al., 2023; Patil et al., 2024; Zhu, 2023) and chatbot visual and conversational design (Li et al., 2021). However, few studies have addressed the barriers that contribute to consumer resistance (Jisham et al., 2024). Understanding resistance and its impact on financial services is crucial to bridging these gaps and mitigating consumer resistance (Dekkal et al., 2023).

The IRT explores consumer resistance as a distinct behavior, not merely as an antithesis of acceptance (Esawe et al., 2023; Santos and Ponchio, 2021). Ram and Sheth (1989, p.6) defined innovation resistance (IR) as “the resistance offered by consumers to an innovation, either because it poses potential changes from a satisfactory status quo or because it conflicts with their belief structure.” Propagation mechanisms (situation) and perception of innovation or consumer characteristics are significant factors that can lead to two distinct types of innovation resistance: passive and active (Heidenreich and Handrich, 2015; Ram and Sheth, 1989).

In passive resistance, individuals are satisfied with the status quo and unconsciously inclined to reject any change imposed by an innovation that may contradict their prior views or existing state, even before evaluating the innovation (Heidenreich and Handrich, 2015; Heidenreich and Kraemer, 2015; Heidenreich and Spieth, 2013; Laukkanen, 2016). This rejection is related to psychological barriers such as image and tradition (Esawe et al., 2023; Ram and Sheth, 1989). On the other hand, active innovation resistance is a deliberate form of resistance formed after a careful evaluation of innovation attributes in relation to consumer expectations (Esawe et al., 2023; Santos and Ponchio, 2021). It is driven by functional barriers, such as risks, use and costs, when individuals face potential conflicts between certain attributes and expectations (Ram and Sheth, 1989; Talke and Heidenreich, 2014).

Previous studies have applied IRT to various innovations including sustainable innovation (Esawe et al., 2023) and driver assistance systems (Chu, 2023). Some scholars have used IRT exclusively (Talwar et al., 2020), while others have supplemented IRT with relevant theories, such as the UTAUT2 model (Migliore et al., 2022) and social support theory (Khaw et al., 2022). Consumer intention and behavior toward innovation can be understood through two lenses: the adoption lens and the resistance lens. The adoption lens (e.g. UTAUT and TAM) examines adoption facilitators, where TAM identifies ease of use and perceived usefulness as adoption drivers (Elsotouhy et al., 2023), and UTAUT considers effort expectancy, performance expectancy, social influence and facilitating conditions (Esawe, 2022). The resistance lens IRT uniquely focuses on functional and psychological barriers that increase resistance, making it highly relevant to understanding resistance to AI chatbots (Jisham et al., 2024). However, studies have suggested that barriers often exhibit asymmetrical effects with incongruities in comprehensiveness and understanding that require further investigation (Leong et al., 2021). Thus, IRT serves as a critical lens for examining the interplay between resistance factors and consumer behavior in contexts such as AI in customer-facing financial services.

Usage barriers (UBs) are emerging because of compatibility and complexity issues (Leong et al., 2021). The former increases when innovation contradicts consumers’ current usage experience and established routines (Ram and Sheth, 1989). The latter involves: innovation as an abstract idea (is it easy to understand?) and implementation complexity (is it easy to use?) (Laukkanen, 2016; Talwar et al., 2020). Higher compatibility and complexity lead to higher resistance and rejection (Ram, 1989; Jana, 2022). In customer-facing financial services, chatbot limitations, such as repetitive loops or reliance on complex large language models, hinder meaningful conversations and complicate obtaining clear and reliable answers conversation (CFPB, 2023; Yang et al., 2023). Previous studies have confirmed a positive relationship between UB and CIR (Bhatnagr et al., 2024; Leong et al., 2021). Thus, chatbot attributes such as compatibility and complexity remain critical factors influencing chatbot resistance (Jana, 2022; Jisham et al., 2024). Based on these arguments, the following is proposed:

A value barrier (VB) is perceived when innovation has extraordinarily little or no added value (price-to-performance tradeoff) compared with its substitutes (Ram and Sheth, 1989). Consumers consider the innovation they use as reference points, and if the new one lacks superior value, then they are unlikely to change their habits or routines; consequently, they will not consider switching (Laukkanen, 2016) and may even resist it (Leong et al., 2021). In financial services, rigid chatbot protocols that fail to accommodate the diversity and privacy inherent in individual queries fade the perceived value (Chaouali et al., 2024). In addition, chatbots provide insufficient assistance and hinder interactions with human agents, which may lead to additional costs for consumers (CFPB, 2023). Such limitations contribute to resistance, as consumers prefer the superior benefits of interacting with human agents over chatbots (Jana, 2022). Prior studies have suggested a positive correlation between VBs and CIR (Baklouti and Boukamcha, 2024; Bhatnagr et al., 2024). Based on this, the following is proposed:

Consumers perceive risks as uncertainties that could threaten their adoption (Esawe, 2022). Perceived risks such as privacy breaches, impersonation, phishing fraud and security vulnerabilities (Huang et al., 2021; Leong et al., 2021; Xie et al., 2024) significantly shape and drive consumer resistance to chatbots (Chang and Hsiao, 2024). Consumers who perceive chatbots to be less secure than traditional methods prefer interpersonal interactions with human agents (Jisham et al., 2024). Poorly designed chatbots or a lack of customer support can exacerbate these concerns, leading to a loss of customer trust and heightened resistance (CFPB, 2023). Additionally, fear of errors or third-party privacy violations during chatbot usage further increases apprehension (Chaouali et al., 2024; Cheng et al., 2018; Laukkanen, 2016; Santos and Ponchio, 2021). While some studies confirm a positive effect of risk barriers (RBs) on CIR in financial services (Behera et al., 2023; Bhatnagr et al., 2024; Jana, 2022), Khanra et al. (2021) find an insignificant effect. Therefore, the following is proposed:

Consumers may perceive tradition barriers (TBs) when an innovation clashes with existing traditions, values, beliefs, norms and culture (Kleijnen et al., 2009; Ram and Sheth, 1989). This barrier is also observed when consumers prefer traditional ways of interacting (Laukkanen, 2016). When deviating from an established routine, clients often feel frustrated because of anxiety about a perceived loss of control (Mani and Chouk, 2018), leading them to prefer familiar services, as the interactions with chatbots markedly contrast with traditional interaction methods that many customers associate with valued interpersonal engagement (Chaouali et al., 2024). If interacting with a chatbot leads or forces consumers to change routines or clash with their culture, then they are likely to display increased resistance toward its adoption (Jana, 2022). Prior studies have suggested a positive correlation between traditional barriers and CIR (Baklouti and Boukamcha, 2024; Bhatnagr et al., 2024). Therefore, the following is proposed:

Innovation origins, such as country of origin, product category or brand, give innovation its own identity, which can lead to a negative image associated with innovation (Laukkanen, 2016). Another reason is consumers may perceive it as difficult to use (Ram and Sheth, 1989). These barriers positively affect resistance toward digital payment systems (Sivathanu, 2019), mobile payment services (Behera et al., 2023; Khanra et al., 2021) and neo-banking (Bhatnagr et al., 2024). In the context of chatbot, if consumers have a negative image about chatbots or believe they are difficult to use, then their resistance will be stronger (Jana, 2022). Therefore, the following is proposed:

According to Kaur et al. (2021, p. 1747), CIR is “the behavior toward the adoption and usage of any innovation that results in maintaining the status quo and resisting any deviances from the current beliefs.” Consumers who exhibit resistance to innovations are inclined to criticize and actively oppose innovation, which hampers its success (Kleijnen et al., 2009). Word-of-mouth (WOM) is a significant factor in innovation and can lead to NWOM if customers have an unpleasant experience (Kaur et al., 2021). These resistance consumers can have a negative impact on not only their immediate social connections but also society (Jana, 2022). Consequently, consumers who resist chatbots actively attack and oppose them, affecting their success and influencing the adoption decisions of other consumers. According to Jana (2022), consumer resistance leads to NWOM. Therefore, the following is proposed:

Consumer resistance toward innovation may be linked to conservative attitudes, such as a preference for cognitive closure and anti-hedonic approaches, resulting in brand loyalty, aversion to new options and preference for nostalgic products. Additionally, CIR is often driven by functional and psychological barriers (Leong et al., 2021). These barriers create dissatisfaction and mistrust, leading to resistance and ultimately NWOM (Méndez-Suárez and Danvila-Del-valle, 2023). Previous studies highlight that IRT barriers can increase CIR (Behera et al., 2023; Bhatnagr et al., 2024; Jisham et al., 2024), whereas resistant consumers may actively spread NWOM to discourage others from adopting disruptive innovations (Jana, 2022; Talwar et al., 2021).

This study builds on Jana's (2022) findings on chatbots in financial services, proposing that CIR serve as an intermediate mechanism linking IRT barriers to NWOM in the chatbot context. By investigating how resistance translates barriers into adverse word-of-mouth, this study enriches IRT by exploring a nuanced behavioral dynamic. Thus, this study proposes the following hypotheses:

Figure 1 schematizes this study’s conceptual framework.

This study targeted users of the chatbot services offered by Egyptian banks. As there was no accessible sampling frame, we used purposive (judgment) sampling to ensure the inclusion of participants with relevant experience (Bougie and Sekaran, 2019). Participants were recruited via social media platforms such as Facebook and Twitter using a structured questionnaire hosted on Google Forms. Three postgraduate students with expertise in market research were trained by the researcher to facilitate the distribution and data collection processes. Rigorous quality controls were implemented, including single-response restrictions and a screening question to restrict access to customers who had no prior experience using financial services chatbots, maintaining the focus of the study.

The final data collection spanned six weeks between January and February 2024, yielding 312 responses. After 18 incomplete questionnaires were excluded, a final sample of 294 responses were retained for analysis. This sample size surpassed the minimum sample size of 138 estimated using G*Power v3.1 software with specific parameters (effect size = 0.15, α = 0.05, power = 0.95 and number of predictors = 5) (Erdfelder et al., 2009). Male and female participants were 60.9% and 39.1%, respectively, and the average age of the respondents was 21–30 years, which is considered tech-savvy. Approximately 24.1% had used chatbots for more than two years (Table 2).

We adopted multiple items to measure each latent variable from previously validated scales in the literature, resulting in 31 items (Table 3). Each item was rated on a seven-point Likert scale, ranging from “strongly disagree” to “strongly agree.” To avoid translation bias, a back-translation method was used, as all constructs were translated from English to Arabic, the primary language in Egypt (Behr, 2017). The questionnaire consisted of two parts: the participants’ personal information and questions related to the endogenous and exogenous variables used in the study.

Initially, a pre-test was conducted involving a committee of three professionals and four academic experts to improve the instrument and ensure adequate face validity. Subsequently, a pilot test was conducted with 59 chatbot service users to identify areas for improvement in the questionnaire and assess the factor loadings for all the latent constructs, which ranged from 0.706 to 0.982.

This study followed the recommendations of Podsakoff et al. (2003) to mitigate common method bias (CMB). Procedural remedies included using a concise, straightforward survey with validated scales to minimize ambiguity and redundancy, providing clear instructions, explaining objectives and assuring respondents confidentiality and anonymity. To reduce response bias, the dependent and independent variables were spatially separated. Statistically, Harman’s single-factor test confirmed factor accounted for more than 50%, with the first factor accounting for only 36.58% of the variance, thus supporting the robustness of the findings.

We used partial least squares structural equation modeling (PLS-SEM) and used SEMinR to evaluate the measurement model and estimate the structural model following the two-stage approach (Hair et al., 2021). This study uses PLS-SEM as an alternative to covariance-based SEM because of its advantages in terms of fewer restrictive assumptions, making it widely used in experimental research (Sarstedt et al., 2021). Specifically, PLS-SEM is preferred when the study’s primary objectives are confirmatory and explanatory modeling objectives (Benitez et al., 2020).

A two-stage approach was used to analyze the data in this study (Sarstedt et al., 2021). The first stage involved evaluating the measurement model by examining indicators and internal consistency, as well as validity (convergent and discriminant). The second stage consisted of assessing the structural model by examining collinearity issues, the significance and relevance of the model’s relationships and the model’s explanatory and predictive power (Hair et al., 2021).

In the first stage, to validate the measurement model, we assessed the reliability, convergent validity and discriminant validity of the constructs (Hair et al., 2021). Table 3 presents the results, which indicate that Cronbach’s (α) values ranged from 0.810 to 0.899, composite reliabilities (CR) ranged from 0.875 to 0.929 (Fornell and Larcker, 1981), average variances extracted (AVE) ranged from 0.636 to 0.766 (Kwong and Wong, 2016) and standardized factor loadings ranged from 0.782 to 0.888 (Hair et al., 2019). All these values exceeded the recommended thresholds, indicating sufficient reliability and convergent validity (Sarstedt et al., 2021).

To analyze the discriminant validity of the latent variables, we used the Heterotrait–Monotrait ratio (HTMT) criterion (Henseler et al., 2015). Table 4 presents the HTMT values, which confirm the discriminant validity as the highest HTMT value is 0.679, below the threshold value of 0.85 (Benitez et al., 2020).

In the second stage, we evaluated collinearity issues to ensure the absence of high correlations among the constructs in the structural model. According to Table 4, all VIF values were ≥ 3–5, suggesting that collinearity was not a critical issue in the structural model (Hair et al., 2019).

Next, we used a bootstrap routine with 10,000 iterations to evaluate the relevance and significance of the path coefficients (Hair et al., 2021). Hair and Alamer (2022, p. 7) suggest that “path coefficients (β) in the structural model ranging from 0 to 0.10, 0.11–0.30, 0.30–0.50, and > 0.50 are indicative of weak, modest, moderate, and strong effect sizes.” Table 5 displays the results, indicating that H1, H3, H4 and H6 are supported, as evidenced by the t-test values (t-values ≥ 1.96; and p < 0.001). Specifically, TB (β = 0.333, CI 95% [0.212, 0.434]; moderate effect size), UB (β = 0.249, CI 95% [0.125, 0.363]; modest effect size) and RB (β = −0.201, CI 95% [0.109, 0.292]; modest effect size) all exhibited a positive and significant influence on CIR. However, H2 and H5 are not supported. VB (β = 0.066, CI 95% [−0.032, 0.153]; modest effect size) and IB (β = 0.009, CI 95% [−0.080, 0.099]; weak effect size) have no significant effect on CIR. Finally, CIR (β = 0.171, CI 95% [0.050, 0.274]; moderate effect size) positively and significantly influenced NWOM.

The in-sample predictive power (explanatory power) of the model was assessed using R² and f². Hair et al. (2021) stated that “R² values of 0.75, 0.50, and 0.25 can be considered substantial, moderate, and weak, respectively.” The R² values for CIR = 0.454 can be considered moderate, whereas the R² value of NWOM = 0.25 is weak. It is worth noting that the addition of the mediating effect and barriers direct effects on NWOM to the model increased the R² value of NWOM to 0.38.

Moreover, the results in Table 5 imply that the rank order of f² corresponds directly to the rank order based on the path coefficients. The f² value of CIR → NWOM (0.325) implies that CIR has medium effect size on NWOM, whereas the f² values of TBs → CIR (0.112), UBs → CIR (0.065) and RBs → CIR (0.051) imply that they have small effect sizes on CIR. On the contrary, the VBs and IBs have no effect size on CIR (Hair et al., 2021).

The predictive power of the model was assessed using the PLS_predict procedure recommended by Hair et al. (2021), and the direct antecedents approach was adopted (Shmueli et al., 2019), in which both the antecedent and the mediator would consider in the PLS_predict as predictors of outcome constructs (Danks, 2021). Initially, the prediction error was evaluated, revealing significant skewness (Figure 2). Consequently, the mean absolute error metric is deemed more suitable (Danks and Ray, 2018). The analysis of NWOM revealed that the PLS path model had a lower out-of-sample predictive error across all indicators compared to the baseline naïve LM model. Specifically, the MAE values for the PLS model were lower than those of the LM model for all NWOM indicators: NWOM_1 (PLS: 0.687 vs. LM: 0.721), NWOM_2 (PLS: 0.592 vs. LM: 0.637), NWOM_3 (PLS: 1.13 vs. LM: 1.143), and NWOM_4 (PLS: 0.92 vs. LM: 0.964). These results suggest that the model possesses strong predictive power (Shmueli et al., 2019).

The study introduced CIR as a mediator in the relationship between IRT barriers and NWOM. The results in Table 6 indicate that the direct effect from UB to NWOM and RB to NWOM is not significant. Therefore, we conclude that the relationship between UB and NWOM and RB and NWOM is fully mediated by CIR, supporting H7a and H7c. Additionally, the direct effect from TB to NWOM is significant. Thus, we conclude that CIR partially mediates the effect of TB on NWOM, and there may be other factors that mediate this relationship, supporting H7d. Finally, the direct and indirect effects from VB to NWOM and IB to NWOM are not significant. Therefore, we conclude that CIR has no mediating role in the relationship between VB and NWOM and IB and NWOM.

This study investigates the functional and psychological barriers contributing to CIR toward AICAs in financial services and their implications for NWOM. Grounded in Innovation Resistance Theory (IRT), the study introduces CIR as a mediator between IRT-related barriers and NWOM. The findings highlight how usage, risk and TBs influence resistance, which mediates the relationship with NWOM. The findings reveal key insights into consumer resistance in this sector.

UBs significantly and positively influence CIR. This indicates that CIR increases when chatbots are difficult to use because of inconveniences, lags and unclear steps (Yang et al., 2023). These barriers align with TAM, and UTAUT emphasizes the importance of ease of use. To address this, chatbot providers must ensure transparent and readily accessible protocols for all consumers. This result is supported by the study results of Elok Behera et al. (2023), Chu (2023), Khanra et al. (2021) and Baklouti and Boukamcha (2024).

Contrary to prior studies that emphasized VBs (Behera et al., 2023; Chu, 2023; Leong et al., 2020, 2021), VBs had an insignificant influence on CIR. This result was presumed to be because chatbots in financial services often provide similar functionalities, are perceived as “complementary” and come with an inherent expectation of convenience, reducing the importance of value trade-offs (Jana, 2022). Immediate concerns, such as usability, trust and perceived risks, overshadow abstract evaluations of value, especially in markets such as Egypt, where trust and cultural norms heavily influence consumer behavior (Elsotouhy et al., 2023; Khanra et al., 2021). Providers should focus on addressing these pressing concerns, rather than solely emphasizing the chatbot’s value proposition. This finding aligns with prior studies (Jana, 2022; Khanra et al., 2021; Mani and Chouk, 2018).

RBs, including concerns about privacy, fraud, uncertainty, chargebacks and transaction errors, significantly affect CIR. Addressing these risks through robust security measures and transparent communication is crucial for reducing consumer apprehensions (Huang et al., 2021; Leong et al., 2021; Xie et al., 2024). This result is supported by the study results of Behera et al. (2023), Chu (2023) and Jana (2022) and, in contrast, with the results of Khanra et al. (2021).

Traditional barriers emerge as the strongest determinant of CIR. This result is supported by the study results of Behera et al. (2023), Leong et al. (2020) and Sivathanu (2019) and, in contrast, with the results of Chu (2023) and Jana (2022). These results indicate that perceptions of incompatibility with users’ experiences, values, and norms–as well as the absence of human interaction–significantly contribute to CIR. Such barriers are consistent with the UTAUT, which emphasizes the role of social influences in shaping resistance. These barriers align with UTAUT, emphasizing the importance of social influences.

Surprisingly, IBs had an insignificant impact on the CIR to chatbots. This result was inconsistent with that reported by Behera et al. (2023), Chu (2023), Jana (2022) and Khanra et al. (2021) and in line with the results of Cheng et al. (2018) and Leong et al. (2020). This insignificant result can be explained by the fact that consumers have a positive impression of chatbots because of their ease of use and usefulness (aligned with the TAM and UTAUT). This positive image may be attributed to respondents’ tech-savviness and their perception of chatbots aligning with their self-image (Chu, 2023). In addition, government support for financial services in the banking sector may contribute to this positive image.

Moreover, the results indicate that CIR significantly influences NWOM. This result aligns with Jana (2022), indicating that resistance, characterized by feelings of irrelevance, mistrust or perceived inefficiency, fuels NWOM behavior, which encompasses actively voicing dissatisfaction, disseminating negative remarks, participating in adverse conversations and cautioning others against availing of these services. Service providers should focus on improving chatbot relevance, usability and trustworthiness to reduce the CIR and its subsequent impact on NWOM.

The mediating analysis reveals that CIR fully mediates the effects of usage and RBs on NWOM and partially mediates the effect of TBs, suggesting that these barriers lead to resistance, which in turn drives NWOM behavior. The findings align with Jana (2022) regarding RBs but differ from those of Talwar et al. (2021). The variances in NWOM conduct concerning RBs can be ascribed to an amalgamation of cultural, technological, regulatory and market-specific determinants, with Egyptian consumers potentially augmenting risk tolerance or a more lenient perspective of nascent financial technologies. The partial mediation of TBs suggests the presence of additional mediators that are not captured in this study, consistent with Kaur et al. (2021) but differing from Jana (2022). The non-significant impact of value and IBs on NWOM may reflect that consumers do not fully perceive value trade-offs and perceive innovation as aligning with their self-image because of tech-savviness. These variations emphasize the importance of cultural and technological contexts in shaping CIR and NWOM relationships.

This study contributes to the existing literature on CIR regarding AICAs by analyzing CIR toward chatbots in the context of financial services, thereby reducing gaps in the relevant literature. This expansion enriches IRT’s applicability beyond the traditional innovation context. Moreover, the current investigation extends the scope of IRT research by exploring the outcomes of CIR in the form of NWOM. This integration extends beyond traditional metrics (e.g. adoption intention, use intention or continued intention to use) and situates CIR within broader societal impacts, enriching the explanatory scope of IRT.

The mediating role of CIR between functional and psychological barriers and NWOM provides a nuanced understanding of the resistance dynamics. Usage and RBs are fully mediated by CIR, whereas TBs exhibit partial mediation, suggesting varying pathways of influence. These insights emphasize the critical role of CIRs in shaping consumer responses to innovative technologies, offering a foundation for future studies investigating the mediating and moderating roles of CIRs across different contexts.

The results revealed the limitations of TAM and UTAUT, emphasizing the need to include resistance-related factors. Barriers, such as usability issues, risk perceptions and preferences for conventional approaches, provide an alternative lens to enhance the explanatory power of these models. The integration of resistance elements may resolve conflicting outcomes in technology adoption studies and offer a balanced view of both facilitators and inhibitors.

Recognizing that IRT barriers have varying influences on CIR and NWOM, financial service providers can adopt targeted, data-driven approaches to overcome these barriers. To overcome traditional barriers, service providers must position chatbots as complementary to traditional interactions, rather than substitutes (Chang and Hsiao, 2024). Personalizing AICA interactions to increase human touch can further alleviate concerns regarding impersonal interactions (Mariani et al., 2023). Additionally, using hybrid service models to integrate AICA with existing customer service channels allows consumers to effortlessly transition from chatbot interactions to live-agent support, when necessary, thereby alleviating frustration and fostering positive engagement.

To address UBs, service providers should implement a user-centric design that prioritizes ease of use, clear instructions and intuitive interfaces. This could involve training programs for customers to interact with chatbots effectively and transparently to communicate their functionalities. Furthermore, addressing RBs requires proactive measures, such as robust data privacy protocols, transparent communication regarding data protection measures and clear guidelines for error resolution. These steps can mitigate privacy and security concerns and foster greater confidence in AICAs.

Despite this study’s conclusion that value and IBs do not significantly impact CIR or NWOM, it remains imperative to improve consumers’ perceptions of AICA in financial services. Financial service providers should carefully craft marketing campaigns that emphasize the advantages of AICA, such as convenience, time-saving and improved control over financial matters (Talwar et al., 2020). Moreover, targeted campaigns should focus on showcasing successful real-life applications of chatbots in financial services to transform chatbots’ perceptions from mere technological innovations to practical financial instruments, thereby underscoring their tangible value to consumers.

Given the mediating role of CIR, reducing consumer resistance can indirectly mitigate NWOM (Jana, 2022). Educational initiatives, such as interactive tutorials, comprehensive FAQs sections and user feedback mechanisms, can enhance user confidence and satisfaction. Furthermore, encouraging positive word-of-mouth through loyalty incentives and satisfied customer testimonials can counteract NWOM (Méndez-Suárez and Danvila-Del-valle, 2023).

Finally, financial service providers should adopt an agile framework for continuous monitoring and adaptation. Tools such as social listening and sentiment analysis can track real-time discussions regarding services, enabling prompt responses to user complaints and preventing negative experiences from escalating into NWOM (Talwar et al., 2021). Providers should implement a dynamic strategy that adapts to evolving consumer needs and expectations to sustain trust, minimize resistance and ensure the long-term success of AI-powered solutions in financial services.

This study has a few limitations. First, it uses an IRT lens that focuses only on the barriers that influence consumer behavior. Future investigations could assess both the inhibitors and motivators of technology adoption that influence consumer resistance through frameworks such as the Behavioral Reasoning Theory, providing a more balanced view of consumer behavior. Second, this study focused on barriers specific to technological characteristics. Consumer barriers such as individual attitudes, emotions and cognitive biases can also play a significant role in shaping CRI. Future studies should integrate technological and consumer barriers into the holistic view of resistance. Third, while this study targets emerging markets, it highlights the necessity for research on the bottom of the pyramid populations, where technological access and consumer behavior may differ significantly. Exploring resistance to financial innovation among bottom of the pyramid consumers in developing contexts could offer valuable insights into affordability, trust and accessibility. Finally, this study focused on chatbots as a financial service innovation, yet other evolving financial technologies such as “buy now, pay later warrant investigation because of their potential to elicit distinct consumer resistance types”. Future research could include these innovations to better understand consumers’ responses.

Table 7 summarizes the study conclusion and implications.

Statements on conflict of interest: The authors declare no potential conflicts of interest concerning this article’s research, authorship or publication.

Data accessibility: The data sets supporting this study’s findings are available upon reasonable request from the corresponding author.