Mobile banking offers convenient services and reduces physical barriers for users, while also lowering operating costs and improving competitiveness for banks. The integration of Artificial Intelligence (AI) transforms mobile banking into intelligent mobile banking, enhancing user experience through personalized and intelligent services. For example, AI-powered chatbots can use human-like language and avatars to improve user interaction and service efficiency. Understanding users' continued intention to use these AI-enabled mobile banking apps is crucial for their successful development and adoption. While Technology Acceptance Model (TAM) and Unified Theory of Acceptance and Use of Technology (UTAUT) have been used to study mobile banking adoption, the Expectation Confirmation Model (ECM) offers a more suitable framework for understanding continued usage, focusing on the effects of expectation confirmation, perceived usefulness, and satisfaction. This study addresses the research gap by examining the influence of AI characteristics—perceived intelligence and perceived anthropomorphism—on users' continuance intention through the lens of the ECM. The central research question is: Do intelligence and anthropomorphism affect users' continuous adoption intentions towards mobile banking apps through the ECM, and if so, how?

The Expectation Confirmation Model (ECM), rooted in Oliver's expectation confirmation theory, posits that confirmation of expectations, perceived usefulness, and satisfaction influence continuance intention. Confirmation refers to the congruence between users' expectations and the actual performance of the system. Perceived usefulness reflects users' assessment of the system's benefits, while satisfaction represents users' overall affective response. The ECM has proven robust in explaining continued usage in the mobile banking context. Existing research shows that AI characteristics, specifically intelligence and anthropomorphism, significantly influence users' adoption decisions of AI-based applications. Intelligence refers to the AI's ability to exhibit efficient and autonomous behavior, while anthropomorphism refers to the AI's human-like behavior. This study incorporates these AI characteristics as antecedent variables within the ECM to explore their influence on user continuance intention in the context of AI-enabled mobile banking apps.

This study employed a survey research method using a random sampling approach to collect data from 365 Chinese users with experience using AI-enabled mobile banking apps. A questionnaire was developed using a back-translation approach and pretested to ensure validity and clarity. The minimum sample size was determined using G*Power software. Data collection was facilitated by a professional online questionnaire company. The questionnaire measured perceived intelligence, perceived anthropomorphism, confirmation, perceived usefulness, satisfaction, and continuance intention using established scales adapted for the mobile banking context. A seven-point Likert scale was used for all items. Several ex ante and post hoc statistical tests were employed to mitigate common method bias (CMB), including Harman's single-factor test, the full collinearity test, and a marker variable approach. Partial Least Squares Structural Equation Modeling (PLS-SEM) with confirmatory composite analysis was used to analyze the data, chosen for its suitability for complex models, exploratory research, and ability to handle multicollinearity issues. The analysis involved examining the measurement model (reliability, validity) and the structural model (path coefficients, R², predictive power using PLSpredict). A mediation analysis was conducted using Zhao et al.'s approach to investigate the mediating roles of the ECM constructs.

The PLS-SEM analysis supported all ten hypotheses. Confirmation was positively related to perceived usefulness and satisfaction (H1, H2). Perceived usefulness was positively related to satisfaction and continuance intention (H3, H4). Satisfaction was positively related to continuance intention (H5). Perceived intelligence was positively related to perceived anthropomorphism, confirmation, and perceived usefulness (H6, H7, H8). Perceived anthropomorphism was positively related to confirmation and perceived usefulness (H9, H10). The R² values for anthropomorphism, confirmation, perceived usefulness, satisfaction, and continuance intention were 0.305, 0.53, 0.571, 0.735, and 0.652, respectively, indicating significant explanatory power. PLSpredict analysis confirmed the predictive power of the model. The mediation analysis revealed that all ten mediation paths were indirect-only mediations, indicating that intelligence and anthropomorphism influenced continuance intention primarily through their effects on the ECM constructs.

The findings demonstrate that AI characteristics, particularly intelligence and anthropomorphism, play a significant role in shaping users' continued intention to use AI-enabled mobile banking apps. These features influence users' expectations and their perceptions of the app's usefulness and overall satisfaction. The mediating roles of confirmation, perceived usefulness, and satisfaction highlight the importance of aligning user expectations with actual system performance, providing valuable services, and creating positive user experiences. The results support the integration of AI features into the ECM for a more comprehensive understanding of mobile banking app adoption. The high R² values and the confirmation of predictive power through PLSpredict indicate the strong explanatory and predictive power of the model.

This study makes significant theoretical and practical contributions. Theoretically, it extends the ECM by integrating AI characteristics, providing a more nuanced understanding of user continuance intention in the AI-enabled mobile banking context. Practically, the findings offer valuable guidance for app developers to design and develop AI-enabled mobile banking apps that cater to user needs and expectations, leading to increased user satisfaction and continued adoption. Future research could explore the generalizability of these findings across different cultures and contexts, utilize longitudinal data to establish causal relationships, incorporate moderating variables to enhance model explanatory power, and examine the influence of personality traits and AI service agent quality.

This study has several limitations. The sample was limited to Chinese users, potentially affecting the generalizability of the findings. The cross-sectional design limits the ability to establish causal relationships. The relatively small sample size, while exceeding the minimum required, may have introduced some bias. The model could be further enhanced by incorporating moderating variables such as personal innovativeness and trust. Finally, this study did not consider other factors that could potentially influence continuance intention such as AI service agent service quality.