Determinants of intention to use AI for personal financial management among middle-aged and older adults: evidence from Vietnam and China

As thoroughly analyzed in the report from the World Economic Forum, in recent years, increased perception of the capability and readiness of innovative AI-based technologies has fueled a surge in research around artificial intelligence (AI). This technology has been penetrating and revolutionizing various technology sectors while promoting financial enterprises to engage in smart financial technology (FinTech) (Priya and Sharma, 2023). In this context, the capability to engage and interwork with AI-related tools has developed into an essential skill, prompting individuals to upgrade their skills to expand their career opportunities (Schiavo et al., 2024). This has unintentionally created challenges for the elderly's approval of using information systems (Cheng et al., 2023). According to Bai (2018), older adults often face barriers in accessing and using digital financial platforms due to limited digital skills, low self-confidence, and concerns about financial security.

In this context, artificial intelligence (AI) is revolutionizing the financial sector by automating processes, optimizing efficiency, and reducing costs. With its power to analyze extensive datasets rapidly and accurately, AI is becoming an indispensable tool, particularly for older adults. While numerous studies have explored AI applications in areas such as healthcare and health consultation, there is insufficient research on AI's role in personal financial planning and management for these demographic groups. A question posed by Mogaji et al. (2022) is particularly relevant: In what ways can AI be used to improve people's financial capabilities, and how do factors such as financial literacy and information asymmetry influence the acceptance of AI-driven financial services? In the context of an aging population, the United Nations Sexual and Reproductive Health Agency (UNFPA) forecasts that the global population aged 65 and above will double over the next 50 years, reaching 20.7% of the total global population. This demographic transformation not only intensifies the demand for healthcare and social support but also highlights the growing importance of financial well-being among older adults. Although technology adoption among older individuals has increased in recent years (Gudala et al., 2021), they remain frequently overlooked in technology design processes (Chu et al., 2022). Therefore, research models addressing AI adoption must be diversified to accommodate the specific cognitive, emotional, and financial needs of the current older adult population, as the latter group will become tomorrow's seniors. In the study by Dequanter et al. (2022), it was found that older adults experiencing more severe cognitive decline faced greater difficulties in adopting new technologies. The explanation for this finding is that individuals with more severe cognitive decline may experience greater difficulties in accessing the internet and new technologies, in addition to other related barriers. It is essential to enhance their grasp of technology to ensure that they can optimally engage with and experience these advancements.

Previous studies have explored the elements influencing acceptance and intention to use AI in the financial sector, such as digital wallets (Khan and Abideen, 2023) and AI financial customer service (Cheng et al., 2023). However, they have not focused specifically on the domain of personal finance management, leaving a gap in the literature that has been called for by Wang et al. (2025) and Hean et al. (2025). Personal financial management (PFM) is defined as a set of deliberate behaviors and processes aimed at planning, organizing, monitoring, and controlling personal financial resources (such as budgeting, saving, and investment decision-making) to achieve long-term financial goals. This study focuses on the advisory and decision-making dimensions of PFM rather than transactional activities and AI applications; the AI applications examined in this research refer to intelligent systems that assist users in financial decision-making and personalized consultation, rather than basic digital payment tools such as e-wallets. Specifically, the scope of AI applications in PFM is limited to advisory and decision-support tools; therefore, the focus lies not on transactional automation but on systems that deliver diagnostic information (information diagnosticity) and assessment-related value (assessment perceived utility) to support older users in managing complex financial tasks. Additionally, research on the decision to adopt technology among older adults remains limited. This gap is particularly relevant in rapidly aging Asian societies such as China and Vietnam, where demographic and technological transformations are occurring simultaneously. Zhao and Li (2024) highlighted that China is undergoing a significant demographic transition, with a rapidly increasing proportion of elderly citizens and a growing demand for AI-assisted technologies in elderly care. This demographic shift naturally extends to financial well-being, creating a rising need for AI-driven digital financial management support (Fan et al., 2024). Similarly, Cheng et al. (2023) emphasized that older adults generally possess lower financial literacy compared to younger groups in many Asian countries, making financial decision-making even more challenging in the context of digital transformation. In Vietnam, this financial literacy gap among elderly individuals remains evident, underscoring the relevance of investigating AI-based personal financial management (AI-PFM) adoption within this demographic group. Next, most existing studies primarily rely on the technology acceptance model (TAM) and its extended versions (Martín-García et al., 2021; Ma and Lei, 2024; Abdelrahman et al., 2025), which fail to sufficiently address the role of knowledge in technology literacy, particularly AI literacy. Moreover, integrating both the TAM and knowledge-behavior gap (KBG) models is theoretically necessary, as they complement each other in explaining AI adoption behavior. While TAM effectively captures cognitive and attitudinal factors such as perceived usefulness and ease of use, it does not fully address how knowledge and literacy translate into actual behavioral intention, a gap that the KBG model directly explores (Stibe and Dinh, 2024). By combining these frameworks, our study bridges the classical focus on utility perception from TAM with the behavioral conversion mechanism of KBG, thereby offering a more comprehensive explanation of how AI literacy and knowledge drive acceptance and intention to use AI for personal financial management. While numerous previous studies have established that social influence significantly associated with both perceived usefulness (PU) and perceived ease of use (PEOU) in research on technology adoption among older adults (Huang, 2023; Man et al., 2024; Hussain et al., 2025; Zhang et al., 2025a), this study does not revisit these well-documented relationships. Instead, it focuses on examining the link of social influence on AI literacy, acceptance, and intention to use, intentionally excluding its direct effects on PU and PEOU to avoid redundancy with prior empirical findings that have already demonstrated these links with substantial clarity.

Conducting a comparative study across two countries will broaden the scope of this research, offering valuable insights for banks and financial organizations to better promote the adoption of technology for personal finance management among older adults. Furthermore, this research is closely associated with the domains of intelligent customer service, chatbots, and recommendation systems. It explores the characteristics of robots through the lens of algorithms and technologies, contributing to the broader domain of AI applications. The focus of this study is to explore the determinants associated with improvements in knowledge and intention to apply AI technology for personal financial management among older adults in China and Vietnam. By extending the TAM theoretical model combined with the KBG model, it helps to cover more models of artificial intelligence acceptance. Theoretically, the link between AI literacy (AIL), acceptance, and intention to use AI in China and Vietnam has been explored. At the same time, a new variable APU based on TAM was developed and focused on older adults. This study applies AI in personal financial management, pioneering the expansion of research directions to this age group. In the context of the current AI explosion, many aspects of human life have been profoundly changed. Notably, the financial sector has been significantly transformed. Our research has made valuable contributions to companies and financial institutions when researching AI-based financial products for elderly customers. It is necessary to develop assessment perceived utility (APU) to meet practical needs as well as perceived ease of use (PEOU) to support them in easier manipulation and use. For government agencies, in order to mobilize reserve capital from this group of subjects with long-term asset accumulation time, it is necessary to have propaganda methods to support the development of programs to popularize knowledge about AI to improve AIL. Moreover, understanding the behavioral characteristics of older adults is essential when extending the model with external variables. For instance, declining cognitive agility and digital anxiety often reduce their confidence in managing intelligent systems, making AI self-efficacy a critical factor in strengthening their technological confidence (Benge et al., 2023; Zhang, 2023). Similarly, their risk aversion and distrust toward opaque financial technologies heighten the importance of perceived accountability, as clear attribution of responsibility reduces perceived vulnerability and fosters willingness to engage with AI-based financial tools (Enescu and Raileanu Szeles, 2024). Finally, social influence exerts an outsized effect on this demographic, as social reassurance, familial approval, and intergenerational learning can alleviate skepticism and promote both literacy and adoption (Chen and Chan, 2014; Aleti et al., 2023).

While numerous theories have been suggested to account for technology utilization, the Technology Acceptance Model (TAM) is the most popular and proven paradigm for explaining new technology uptake (Kelly and Palaniappan, 2023). Davis (1989) created it to explain the aspects that are associated with people's intentions to apply new technologies. This model identifies factors that are associated with an individual's intention to use new technologies and has become one of the most important theories explaining technology acceptance. These factors include: perceived ease of use (PEOU) and perceived usefulness (PU). Here, PEOU can be referred to as “the degree to which a person believes that using a particular system would be free of effort” and PU is termed as “the extent to which a technology is expected to improve a potential user's performance” due to Davis (1989). TAM has been shown to be a powerful tool for explaining technology adoption in multiple settings, especially within the financial sector (Kelly and Palaniappan, 2023).

Recent research has increasingly questioned the sufficiency of the original Technology Acceptance Model when applied to complex, high-involvement industries such as AI-driven financial services or geriatric populations (Table 1). Researchers have emphasized that the Technology Acceptance Model (TAM), while essential, often overlooks age-related, cognitive, and psychological factors that may be associated with technology adoption. Applying the Technology Acceptance Model (TAM) in older adult populations highlights that perceived usefulness and ease of use are consistently associated with technology adoption. However, seniors often exhibit greater risk aversion and lower confidence in digital systems, making factors such as trust and perceived risk especially critical for understanding their adoption behavior (Martín-García et al., 2021). Incorporating these elements into TAM-based analyses may provide a more accurate and comprehensive assessment of how older users engage with new technologies, particularly in contexts such as digital finance or mobile payments. Investing in the Future: An Integrated Model for Analyzing User Attitudes Toward Robo-Advisory Services with AI Integration augments the Technology Acceptance Model by integrating trust and perceived risk to more precisely reflect users' concerns about transparency and algorithmic reliability in AI-facilitated financial decision-making (Singh and Kumar, 2024). Recent studies employing the Technology Acceptance Model (TAM) within older adult populations indicate that psychological constructs, including self-perception of aging, technology anxiety, and self-efficacy, show significant associations with individuals’ perceived ease of use and perceived usefulness. These findings suggest that the conventional TAM framework may be insufficient to fully capture the determinants of technology adoption among seniors, supporting the inclusion of cognitive, emotional, and experiential dimensions. Such an extension may offer a more nuanced understanding of adoption behaviors in contexts involving digital public services and AI-enabled technologies, emphasizing the critical role of psychological readiness and technological familiarity in relation to older adults' engagement with emerging digital platforms (An et al., 2024).

The Knowledge Behavior Gap (KBG) model serves as a vital framework that delineates the theoretical pathway of adoption, positing that fundamental knowledge is associated with acceptance and intention before culminating in actual use behavior, thereby challenging the conventional assumption that knowledge automatically translates into action (Stibe and Dinh, 2024; Stibe et al., 2022). Historically, the Technology Acceptance Model (TAM) and its extensions, such as UTAUT, predominantly emphasize technology use and acceptance; however, they often neglect the critical link between adoption and the specific knowledge surrounding the technology itself (Dinh and Stibe, 2024). The KBG perspective has proven applicable in various technological contexts, including the study of user acceptance of Blockchain-Based Solutions (Dinh and Stibe, 2024) and the general acceptance and utilization of Artificial Intelligence (AI) technologies (Stibe and Dinh, 2024). The KBG model is particularly suitable and necessary for the current study, which examines AI adoption for Personal Financial Management (PFM) among elderly individuals in Vietnam and China, because this demographic frequently faces barriers rooted in a lack of knowledge and a perceived fear of making mistakes rather than inherently negative attitudes toward the technology (Broady et al., 2010). Therefore, the integration of KBG allows the research to incorporate AI literacy (AIL) as a foundational antecedent, addressing the initial cognitive stage often missed by TAM, and thereby providing a broader framework for AI acceptance. Combining KBG with TAM is essential because while KBG addresses the pathway from literacy and awareness (AIL to acceptance), TAM contributes the crucial elements of system utility and usability – perceived ease of use (PEOU) and the newly expanded concept of assessment perceived utility (APU) - allowing the proposed framework to examine both the necessary foundational knowledge and the perceived cognitive value users need when evaluating and applying the personalized financial information provided by the AI application (Ortiz-López et al., 2024).

Numerous industries, including retail, catering, and financial sectors, make extensive applications of AI client representatives, in accordance with findings from Wang et al. (2021a), Kar and Dwivedi (2020), Grover et al. (2019). AI-enabled customer service, an intelligent information system, may autonomously interact with users and offer practical guidance, accept complaints, and provide consultancy services (Androutsopoulou et al., 2019). Along with the technical facets of AI customer service, the intrinsic factors associated with AI-based customer service users are crucial and warrant more investigation, particularly given the scarcity of existing research on these elements. Numerous investigations have revealed that user's sex, age, cognitive age (CA), self-efficacy, and other determinants are associated with their initial desire to implement AI technology (Yu and Ngan, 2019; Ghasemaghaei et al., 2019). The emotions, including perceived risk, resulting from users' cognitive processes and interaction with AI customer service, perceived usefulness (PU) and perceived ease of use (PEOU), may be associated with them to gauge the degree of their adoption of this technology and serve as the primary reason of their ongoing UI (Ghasemaghaei et al., 2019; Bhattacherjee, 2001). AI customer service is being applied strongly in the financial sector. Through the SMS financial platform, intelligent customer service was launched by the China Construction Bank in 2012 via SMS service. In 2018, Eximbank Viet Nam launched a ChatBot application for 24/7 customer care. By the end of 2019, more than 13 banks had converted their customer service centers to virtual banking hubs. Additionally, 21 banks planned to open remote branches, as reported by the China Banking Regulatory Commission (CBRC). This has prompted more research into the factors associated with personal financial management behavior and intention to adopt them in the financial service.

According to processing speed theory, cognitive impairment may be interpreted in light of a decline in the brain's ability to process information as people age (Salthouse, 1996). Meanwhile, the elderly will take longer to complete different jobs than younger generation because of inefficient information processing and weak memory, and they may adopt new technologies more slowly due to technological apprehension, reference from Frishammar et al. (2023). Older adults face a digital divide in contemporary culture due to the physical and cognitive limitations brought on by aging, attributable to Choi and DiNitto (2013). The evolution of technology has also improved the lives of older adults by offering assistance and ease. Therefore, bridging the gap between older people and modern technology and assisting them in overcoming the challenge of technological adoption is essential. Previous studies have examined older adults' adoption of technology in areas such as mobile phones, online communities, recreational services, and internet-based public services (Mariano et al., 2021; Chiu and Liu, 2017; Yang and Lin, 2019). These studies aim to enhance the understanding of how older adults engage with emerging technologies and to provide recommendations for service providers. In this study, we examine older adults' adoption of AI technology in the field of financial management.

Technology self-efficacy refers to an individual's belief in their ability to successfully use technology, as defined by Priya and Sharma (2023). Building upon this concept, AI self-efficacy (AISE), developed by Wang and Chuang (2023), AISE is the self-assessment of specific personal abilities related to the use of AI technologies or AI-based products, rather than merely an evaluation of one's general technological skills.

Previous studies have shown that older adults tend to report lower levels of technological confidence, literacy, and perceived usefulness of technology (Baham et al., 2022; Vodanovich et al., 2010). Individuals with low AISE may perceive the use of AI technologies as more complex and stressful (Hong, 2022). Furthermore, Sharma and Priya (2025) found that AI can be associated with higher levels of individuals' confidence in both technological self-efficacy and financial self-efficacy, which subsequently associated with their performance, effort expectancy, and hedonic motivation. In addition, individuals with higher financial self-efficacy are more likely to engage in proactive financial behaviors such as budgeting, saving, and investing (Khan et al., 2022). This aligns with assessing the measures proposed by AI that allow users to control their personal financial health. Importantly, prior research grounded in the Technology Acceptance Model (TAM) and Social Cognitive Theory suggests that higher levels of self-efficacy positively linked to perceived usefulness, as individuals who feel more capable of using a system are more likely to recognize its instrumental value (Agarwal and Karahanna, 2000). This theoretical reasoning can be extended to AI contexts. Individuals with higher AI self-efficacy are expected to feel more confident in interpreting AI-generated recommendations, understanding AI-driven insights, and integrating AI outputs into their financial decision-making processes. Consequently, such individuals are more likely to perceive AI as a useful tool for assessing and managing their personal finances. Within this context, the present study focuses on assessing AI self-efficacy in relation to the application of AI in personal financial management. In addition, the concept of perceived usefulness (PU) has been argued to require expansion toward assessment perceived utility (APU) to more fully capture the cognitive value users experience when understanding, evaluating, and applying technology-generated information (Ortiz-López et al., 2024). Therefore, we hypothesize:

Perceived accountability (PA) refers to the relationship in which a subject is obligated to explain and justify its actions, a forum can ask questions and make judgments, and the subject may face consequences (Bovens, 2007). In the context of AI systems potentially harming users, accountability is particularly important because users must know who is responsible and how error handling mechanisms are established (Diakopoulos, 2014; Selbst and Powles, 2017; Tóth et al., 2022). When users are aware of this responsibility, they tend to view the system as trustworthy and controllable, and are therefore more willing to hold a more favorable perception of a particular entity (Han, 2024). In that context, assessment perceived utility (APU) reflects the extent to which users perceive AI as helpful in assisting with task completion; AI is considered more helpful when it provides accurate feedback, supports effective problem-solving, and is relevant to user needs (Buabeng-Andoh et al., 2018; Shahzad et al., 2024). Furthermore, AI's ability to multitask and provide contextual information is associated with perceptions of its usefulness (Strzelecki, 2023). For older users who tend to be risk-averse and distrustful of systems lacking transparency, clearly defining responsibilities is especially important (Enescu and Raileanu Szeles, 2024) because it can be associated with a higher likelihood of rating AI-provided recommendations as valuable and helpful.

Furthermore, as AI increasingly participates in and even replaces humans in decision-making processes (Baskerville et al., 2020), the autonomy and learning capabilities of systems can be associated with a dispersion of responsibility among multiple stakeholders, supporting consumer concerns (Tóth et al., 2022). Regulations such as Articles 13–15 and 20 of the GDPR require organizations to be transparent about their operating mechanisms and accountable for system decisions, thereby supporting user protection (Binns, 2018). When users perceive that the system and the related organizations have clear responsibilities, trust is associated with greater willingness to accept AI increases (Shahzad et al., 2024; Baek and Kim, 2023). Simultaneously, clarifying the goals and operational logic of AI helps reduce ambiguity and complexity during use (Binns, 2018), thereby being associated with user expectations and attitudes toward the system's usability (Shahzad et al., 2024). Based on this, we propose the following hypotheses:

The concept of literacy has expanded beyond traditional reading and writing to encompass a range of competencies needed in modern society. Various measurement frameworks have been developed to assess specific forms of competencies, such as information literacy, digital literacy, data literacy, media literacy, and technological literacy (Ng et al., 2021). As artificial intelligence (AI) technology becomes increasingly integrated into everyday life, understanding users' abilities to recognize, use, and evaluate AI technology effectively and responsibly becomes especially important (Wang et al., 2022). Accordingly, these researchers define AI literacy as the ability to accurately identify, effectively use, and critically evaluate AI-related products and services while adhering to ethical standards. This construct comprises four components: awareness, usage, evaluation, and ethics. Carolus et al. (2023) continue to inherit and develop this concept within the framework of Meta AI literacy, adding capabilities such as creating AI, self-efficacy, and emotional management to reflect the role of psychological factors in adapting to a rapidly changing technological environment.

For older adults, declining cognitive agility and digital anxiety often diminish their perceived ability to control and operate intelligent systems (Benge et al., 2023). Consequently, strengthening AI self-efficacy becomes essential because it is associated with greater confidence in interacting with AI-driven financial applications, thereby mitigating the fear of technological inadequacy that frequently deters technology use among the elderly (Zhang, 2023). Although the three relationships between AI knowledge and PU, PEOU, and AI acceptance have been clarified simultaneously, the author's research scope mainly focuses on ordinary users, without specifically considering older adults, who are groups that often face barriers in awareness and belief when accessing financial technology (Bai, 2018). In addition, Ortiz-López et al. (2024) argued that the concept of PU should be expanded to APU to more fully reflect the cognitive value that users feel in the process of understanding, evaluating, and applying information provided by technology. The expansion of the research model of Schiavo et al. (2024) to older adults, while integrating the concept of APU, is necessary to better understand how AI knowledge influences perceived utility evaluation, technology acceptance, and perceived ease of use in the field of AI-based financial technology. Therefore, we hypothesize:

Social influence “refers to the degree to which an individual perceives that people who are important to them think they should use a particular technology” (Ajzen and Fishbein, 1977; Venkatesh et al., 2003). In particular, social support has been identified as particularly important in keeping older adults engaged as their technology learning progresses. (Tsai et al., 2017). Social influence plays a disproportionately strong role among older adults, whose technology adoption decisions are often guided by social reassurance, familial approval, and intergenerational learning (Chen and Chan, 2014). In financial technology contexts, observing trusted peers or family members successfully using AI-based tools can alleviate skepticism and uncertainty, and may be associated with both literacy development and acceptance (Aleti et al., 2023). Within the framework of the study, since AI is a technology, social influence can be understood as the extent to which a user's social network believes that using AI chatbot-based financial management consulting applications aligns with and is compatible with group values. Hence, we hypothesized:

The study of Venkatesh and Davis (2000) has evolved a conceptual design to determine the societal influence and cognitive process determinants that can affect customer acceptance of advanced technology. Following a thorough review that included focus groups, relevant literature, and theories of technology acceptance, as stated by Lu et al. (2019), a metric was created to gauge consumers' acceptance and openness to employing AI service robots, which includes social influence factors. Anthropomorphism, social influence, facilitating condition, performance efficacy, hedonic motivation, and emotion are the six main factors associated with of consumers' readiness to incorporate deploying AI service robots for service delivery, according to Lu et al. (2019), who conducted several qualitative and quantitative studies. In conclusion, the study presents the following hypothesis:

The confirmed findings and popular theoretical models, including UTAUT (Venkatesh et al., 2003) have also established a statistically significant association between the social influence factor and the objective to use AI. People's use in terms of digital systems is associated with social impact (Hsu and Lin, 2008). The level at which people close to the user believe they are advised to use the new product or technology is perceived as social influence, based on the research of Venkatesh et al. (2003). For older adults, social influence plays a particularly critical role, as they tend to rely on the opinions, recommendations, and behavioral cues of significant others–such as family members, friends, or colleagues–when deciding whether to adopt new technologies. Empirical studies have supported this relationship. Lu and Tsai-Lin (2024) found that older adults' adoption patterns of eHealth technologies were primarily associated with by social influence and facilitating conditions. Similarly, Bozan et al. (2015) demonstrated that social influence was significantly linked to elderly users' acceptance of healthcare technologies.

In the context of using AI for personal financial management, this effect becomes even more pronounced. Older individuals often face challenges such as low confidence in technology use or heightened concerns about potential risks. Observing peers or family members successfully utilizing AI-driven financial tools can be associated with reduced concerns, the normalization of technology use, and stronger behavioral intention to adopt. Considering the established relationship between social influence and intention to use AI, we propose the following hypothesis:

Information diagnosticity refers to the understanding or comprehension of a product and how it works that is enhanced by applying information from a specific source (Filieri, 2015). When reviews provide diagnostic cues that enable users to infer product quality and performance, they are perceived as more helpful (Weathers et al., 2015). At the same time, AI chatbots that provide information with high diagnostic accuracy are associated with an improved overall user experience, easier access to the required information, and may link to better decisions (Nah et al., 2023). Therefore, we believe that when older adults search for a product with a specific desire, AI can provide highly relevant, informed information, such as features, benefits, and comparisons with similar products. As a result, they can easily absorb and make more informed financial product choices. From these arguments, we hypothesize:

Information quality is interpreted as information that is reliable, contains knowledge value, is relevant, and useful. According to Zwain (2019), information quality plays an important role when considering users' perceived value in electronic systems. High levels of perceived information quality are associated with individuals' systematic evaluation of utility (Bhattacherjee and Sanford, 2006). When information is lacking in completeness or clarity, users are left with increased uncertainty, which may link to skepticism and reduced satisfaction with the content (Laumer et al., 2017). Therefore, the following hypotheses are formulated:

The concept of information understandability was first mentioned by Wang and Strong (1996), who stated that it is the ability to read, interpret, and comprehend, as well as the language, semantics, and vocabulary expressions used by the evaluator. Building on this, Filieri and McLeay (2014) defined the ability to understand information as the extent to which information can be easily received, read, and understood. Research by Kushwaha et al. (2021) also shows that friendly AI interfaces that provide specific step-by-step information help users understand and perform operations more accurately, thereby supporting the perception of the usefulness of the technology. In particular, Li et al. (2025) also revealed that for older adults, who tend to be cautious when using technology, the clarity and comprehensibility in the way explainable artificial intelligence (XAI) presents financial information may be associated with stronger trust, confidence, and sense of control, thereby supporting PU. On this basis, this study expands the concept of PU into APU to more fully reflect the cognitive value that users perceive in the process of understanding, evaluating, and applying information provided by AI (Ortiz-López et al., 2024). In this context, IU is considered a fundamental cognitive component within the concept of assessment perceived utility (APU). IU does not merely reflect the clarity of presented information but also represents users' ability to cognitively process, evaluate, and confidently apply AI-generated outputs in decision-making situations. When information is structured in an understandable, transparent, and logically explained manner, users may be more likely to assess the relevance, reliability, and practical usefulness of AI recommendations. Therefore, IU functions as a key mechanism through which users transform accessible information into perceived assessment value, strengthening APU by supporting comprehension, reducing uncertainty, and supporting informed judgments, particularly among older users who require higher levels of clarity and interpretability when interacting with financial technologies. Accordingly, this study suggests that:

Many previous studies have confirmed that PEOU is positively related to PU in TAM (Eneizan et al., 2022). Specifically, Bahaw et al. (2025), in their study on the acceptance of online learning systems, found that when learners perceive the technology as easy to learn, flexible, and time-saving, they tend to evaluate the system as more useful in improving learning efficiency. Meanwhile, Mustofa et al. (2025), in their study on technology acceptance behavior in educational and business environments, also indicated that PEOU has a significant positive correlation with PU, while PU plays a mediating role in its association with behavioral intention to use technology. Notably, Liesa-Orús et al. (2022) employed the TAM model to study an older group and confirmed that perceived ease of use and perceived usefulness of digital technology (ICDT) are two factors associated with technology acceptance intention in this group. On this basis, this study will expand the concept of PU into APU as proposed by Ortiz-López et al. (2024) to more comprehensively reflect how older adults evaluate the ease of interaction processes and make decisions when using technology. Therefore, we propose the hypothesis:

Users could avoid engaging with any financial technology service or option if they experience difficulties or perceive it as challenging to use (Mohammad et al., 2023). Previous studies have illustrated that physiological and psychological factors are associated with the low level of technology adoption among older adults, such as poor vision, declining memory, and limited physical dexterity (Motamedi et al., 2021). In addition, older adults often show little interest in new technologies because these technologies fail to fully consider their strengths and limitations, thereby supporting their perceived ease of use. A number of previous research papers have proposed various conceptual frameworks to explore the factors linked to customers' acceptance of new technologies, such as their perception of ease in using them, which has been emphasized in the work of Davis et al. (1989), Schiavo et al. (2024), Bui et al. (2025). As highlighted by Mois and Beer (2019), emphasize that Perceived Ease of Use helps transform technology from a complex system into a more accessible and user-friendly tool. Moreover, greater ease of using electronic financial tools associated with older adults' attitudes toward them, and is associated with greater adoption of these technologies (Choi et al., 2024). Hence, we hypothesized:

Researchers define perceived ease of use (PEOU) as an individual's belief in using technology effortlessly (Davis, 1989). PEOU is a crucial factor associated with users' decisions to adopt AI. It refers to the extent to which an individual believes that using a specific AI technology will be easy (Shahzad et al., 2024). When users perceive applications as easy to use, they are more likely to feel comfortable and satisfied, factors that support their intention to adopt and continue using them (Albashrawi and Motiwalla, 2017). Moreover, the integration of AI-based technologies can be associated with greater user comfort by providing personalized experiences, predictive insights, and efficient and agile support (Lopes et al., 2025; Ameen et al., 2021). According to Shahzad et al. (2024), if users consistently receive accurate and useful information through interactions with AI, they are more likely to perceive the system as easy to use. Emphasizing that perceived ease of use is an essential factor for technology adoption, it is associated with user satisfaction and, consequently, linked to their intention to use the technology. Based on these insights, we propose the following hypothesis:

In the study conducted by Ortiz-López et al. (2024), the concept of assessment is examined from the classical perspective proposed by Tyler (1950), which assumes that perceived usefulness within the Technology Acceptance Model (TAM) reflects the extent to which users believe that technology is associated with improved performance. According to Davis (1989), when middle-aged and older adults perceive artificial intelligence (AI) as useful for assessment purposes, they are more likely to adopt such technologies for financial management. Based on the view that formative assessment and summative assessment represent two distinct dimensions (MacLellan, 2001), Ortiz-López et al. (2024) expanded the original TAM construct of perceived usefulness into two specific constructs corresponding to each assessment type, consistent with the modeling approach suggested by Moore and Benbasat (1991). The development of perceived usefulness of formative assessment (PUFA) and perceived usefulness of summative assessment (PUSA) is also grounded in the distinction between formative and summative evaluation introduced by Scriven (1991), thereby establishing two parallel constructs that function in a complementary manner.

Since formative and summative assessment concepts originate primarily from the educational domain, their application in other contexts remains relatively limited. Within the scope of this study, PUFA is defined as users' immediate evaluation of the continuous feedback provided during the use of AI to support financial decision-making. In contrast, PUSA refers to users' overall evaluation after financial decisions have been implemented based on AI-generated feedback.

Regarding the role of PUFA, users may perceive AI as useful in supporting financial problem-solving and information retrieval. AI tools such as ChatGPT and Gemini can generate context-specific responses based on users' financial inputs while delivering feedback in a rapid and continuous manner. These characteristics support greater users' perceived usefulness of AI throughout the financial decision-making process (Shahzad et al., 2024). Conversely, PUSA reflects the perceived usefulness derived from AI-generated outcomes, which may support users in evaluating the effectiveness of financial decisions more quickly and accurately. AI-generated analytical outputs can be presented in various visual formats, including charts, tables, and comprehensive reports. Furthermore, AI can function as a financial assistant or a personalized financial planning tool for users (Choi and Kim, 2023; Amato et al., 2024; Zhu et al., 2023a).

Nevertheless, Ortiz-López et al. (2024) argue that assessment processes do not operate independently between formative and summative forms but instead represent a combination of both, whereby assessment is conceptualized as an aggregation of its constituent components (Dolin et al., 2018). This perspective implies that, within technology acceptance models, PUFA and PUSA may be integrated into a single construct representing assessment perceived utility (APU). This construct is conceptualized as a formative variable, in which the aggregation of component indicators provides a comprehensive measure of assessment usefulness (Simonetto, 2012).

Accordingly, APU reflects users' perceptions of the value provided by AI both during the financial decision-making process and in the summarized outcomes following decision implementation. When users perceive AI as delivering practical benefits across these two stages, they are more likely to develop positive attitudes and demonstrate a willingness to accept the use of AI in personal financial management. Therefore, we hypothesize:

The sophisticated interaction between user perception, need for control, and platform reliability forms the foundation of the connection between user behavior and perceived privacy (Cheng et al., 2023). Control privacy in using AI for personal finance management refers to the ability to consciously decide the way your information is gathered, treated, archived, and removed. Owing to this power over data, users can subscribe, change, or withdraw data rights, and are regarded as involved participants in their digital journey rather than merely passive data subjects. Control and the idea of “consciousness” are combined (Cheng et al., 2023). Malhotra et al. (2004) and Foxman and Kilcoyne (1993) assert that privacy awareness encompasses beyond just understanding potential hazards. Moreover, it entails understanding business data usage protocols and the implications they have. In the case of a platform's data policies meeting users' anticipated privacy standards, they feel more secure and are more prone to accept using it, as referenced by Cheng et al. (2023). Consequently, we present the following hypothesis:

Security is the foremost concern of users when adopting any technology in numerous research studies. Among these, the relationship between user behavior and privacy perception, especially in the context of emerging technologies such as AI, is based on the complex interaction between users' awareness, their tendency to seek control, and their trust in platforms. Privacy is a fundamental human need and a basic right (Merhi et al., 2019). It serves as a foundation, enabling individuals to safeguard their self-awareness, autonomy, and dignity in a society rapidly transitioning to digitalization. According to Foxman and Kilcoyne (1993) and Malhotra et al. (2004), privacy perception is not merely an awareness of potential risks. It also encompasses a clear understanding of companies' data practices and their consequences. Consumers expect entities handling personal data to implement robust data security measures and transparent privacy policies (Mullan et al., 2017; Raza et al., 2017; Hidayat-ur-Rehman et al., 2022). A lack of control over personal information may be associated with heightened privacy concerns. Mobile device users often express privacy concerns when interacting with online products or services (Sutanto et al., 2013). With the development of AI technology in the financial sector, sophisticated security measures have been introduced, such as biometric authentication and advanced fraud detection systems (Jáuregui-Velarde et al., 2024; Khan et al., 2023). These innovations not only support greater protection against cyber threats but also associated with higher users' awareness of security, thereby supporting greater trust, while manufacturers and service providers can be bound by a legal framework to ensure adequate privacy and informed choice (Li et al., 2021). Therefore, we hypothesize:

An individual's tendency to adopt and engage with new information technology is referred to as their openness to innovation according to Agarwal and Prasad (1998) and Lee et al. (2007). This refers to “the extent to which an individual assimilates new ideas and makes independent innovative decisions regardless of the transmitted experiences of others,” according to Midgley and Dowling (1978). In other words, it describes how quickly and easily consumers accept new ideas, with innovation being associated with how readily they adopt new technologies or products.

From an information technology perspective, individuals with high personal innovativeness tend to hold positive views toward technological innovations and are more capable of dealing with the difficulties associated with adopting new technologies (San Martín and Herrero, 2012; Agarwal and Prasad, 1998). They are typically open-minded, adventurous, and risk-tolerant, enabling them to embrace chosen innovations even when there is a risk of failure (Baumgartner and Steenkamp, 1996; Rogers, 1983). Furthermore, their willingness to participate in rare and bold experiences or activities shows their tendency to embrace innovation or seek out new things, on account of Patil et al. (2020).

Prior studies have consistently confirmed the positive association of consumer innovativeness with behavioral outcomes. Consumer innovativeness has been shown to be associated with the intention to try new products (Al-Jundi et al., 2019) and demonstrates a straightforward association with acceptance (Shaikh et al., 2020). Individuals who exhibit high innovativeness are more inclined to adopt and incorporate AI-based financial applications (AFAs) into their daily routines (Roongruangsee and Patterson, 2023). The research of Zhu et al. (2023b) also highlighted the tendency of highly innovative older adults to hold more favorable attitudes toward digital payments and to feel a greater sense of control when using them. There remains a gap in the literature regarding the role of personal innovativeness in the link between acceptance and intention. Therefore, we hypothesized that:

The reputation of a brand name is considered an extrinsic cue, meaning it is an attribute linked to the product itself rather than its physical characteristics (Zeithaml, 1988). Brand reputation reflects how consumers view a brand's reliability and quality, often being associated with their perception of a product's value based on its brand name (Rindell and Iglesias, 2014). This perception is also shaped by the consumer's social status, with product quality being assessed based on the brand's perceived prestige (Martínez et al., 2014). When making purchasing decisions, consumers consider brand reputation to reduce perceived risks associated with a product, thereby being associated with buying decisions (Martínez et al., 2014). Moreover, studies have demonstrated that consumers generally prefer brands that align with their personal values (Bian and Forsythe, 2012; Snyder and DeBono, 1985) as well as those known for their strong credibility (Belén del Río et al., 2001). In certain contexts, brand reputation may be associated with consumer commitment and intentions than overall satisfaction (Delgado-Ballester and Munuera-Alemán, 2001).

Previous analysis indicates that brand reputation is significantly associated with consumers' intentions to adopt AI in financial management. Hua et al. (2023) demonstrated that exposure to a well-established and reputable service linked to greater trust among elderly users, which in turn is associated with their intention to adopt it. This study expects a higher likelihood of AI adoption for financial management among consumers who perceive a brand as highly reputable, whereas those who view the brand as less reputable will have lower intentions to do so. Thus, we hypothesized:

The likelihood of using AI chatbots is associated with higher levels of user acceptance (Stibe et al., 2022). Acceptance is a critical factor in shaping how users engage with and perceive intelligent technologies, particularly among middle-aged and elderly individuals. Moreover, some scholars argue that a favorable attitude toward these technologies is strongly linked to the decision to adopt them (Vijayasarathy, 2003; Priya and Sharma, 2023). Additionally, several prior studies have investigated this phenomenon among older populations, focusing on their acceptance of computer technology (Mitzner et al., 2010), their perceptions of and readiness to use remote care technologies (Wang et al., 2021b), and their adoption intentions regarding home medical systems (Jo et al., 2021). Thus, we put forward this hypothesis:

To assess the anticipated correlation in this study, we administered an online survey to gather primary information from two Asian nations: Vietnam and China. The questionnaire underwent pre-testing with 20 respondents (10 from Vietnam and 10 from China) and was then modified to validate the content. All the scales were either created in or interpreted as Vietnamese and Chinese (utilizing a translation/back-translation methodology). For this reason, the wording of some points has been changed to make the questions clearer. To ensure that participants possessed the minimum level of financial and digital literacy required to meaningfully evaluate AI-based personal financial management tools, a structured screening procedure was applied before respondents could access the main questionnaire. The screening section consisted of five eligibility questions, covering two domains: personal financial management knowledge and AI awareness/intention. First, participants were presented with a short definition of personal financial management: “Personal financial management refers to managing one's assets and income to achieve specific financial goals such as saving, investing, spending, and protecting wealth. It involves activities such as budgeting, controlling expenditures, investing in profitable instruments, managing debt, and preparing for future events such as retirement or emergencies.” Based on this definition, respondents were asked: (1) “Do you consider yourself knowledgeable about personal financial management?” with answer options Yes/No. A second question assessed whether individuals actively apply financial management practices: (2) “Do you allocate your personal or household financial resources according to a planned approach?” (Yes/No). Next, three questions assessed respondents' level of AI awareness and readiness: (3) “Are you aware of AI tools such as ChatGPT, Gemini, ChatSpot, etc.?” (Yes/No). (4) “Do you have a basic understanding of how AI tools operate?” (Yes/No). (5) “Are you willing to use AI (e.g. chatbot or recommendation system) to support personal financial management?” (Yes/No). Only participants who answered “Yes” to all five screening items were permitted to proceed to the full questionnaire. Respondents who selected “No” on any item were automatically redirected to an exit page and excluded from the dataset. This approach ensured that the retained sample consisted of older adults who had sufficient exposure to both financial practices and AI technology to provide informed responses regarding AI adoption in personal financial management.

At the country level, we distributed 407 questionnaires in Vietnam and 306 questionnaires in China, for a total of 713 questionnaires submitted for data collection. The respondents were evenly distributed by gender, with females (46.2%) and males (48.5%) in Vietnam and males (46.2%) and females (53.8%) in China. As for the traits within the group sampled, the respondent's participants' ages varied from 45 to 60 years old and above. We provided respondents with information about the purpose of the study and confidentiality to mitigate non-response bias.

The survey was distributed using carefully selected social media platforms, email, and personal networks to reach the target demographic, following established procedures in research on technology adoption among older adults (Lee et al., 2024). In Vietnam, Facebook and Zalo groups aimed at older individuals interested in digital literacy or technology adoption were targeted, whereas in China, WeChat groups for older users had a comparable purpose. The survey invitation was distributed through email to professors and PhD researchers at prominent universities to promote participation from intellectually engaged older individuals. A non-probability sampling approach, specifically a purposive (judgmental) sampling technique with screening-based eligibility criteria, was utilized to facilitate the recruitment of respondents with varying levels of digital literacy and ensure thorough demographic representation. This multi-channel strategy corresponds with prior online data collection methods for geriatric demographics.

The research was conducted in accordance with internationally recognized ethical standards for studies involving senior participants (Reyes et al., 2023; Kabacińska et al., 2020). Participation was entirely voluntary, and informed consent was obtained electronically before the survey began. Participants were clearly informed of the study's objectives, anonymity, confidentiality, and their right to withdraw at any time. The questionnaire was designed using clear, age-appropriate language, larger fonts, and a structured layout to reduce cognitive load and visual fatigue. The instrument underwent pilot testing with older individuals before formal distribution to improve accessibility and understanding. All responses were collected anonymously via Google Forms (Vietnam) and Wenjuanxing (China) and were securely stored with restricted access for the research team. These procedures ensured the safeguarding of participant autonomy, privacy, and data integrity throughout the study. The demographic characteristics of the participants are provided in Table 2.

The disproportionate number of respondents with higher educational attainment is an inherent characteristic of online surveys addressing advanced technological topics. Older adults with lower education levels tend to exhibit limited digital literacy, reduced exposure to AI-related terminology, and lower willingness to participate in online questionnaires. In studies examining technology adoption among older adults, it is widely documented that online surveys tend to attract respondents who are more highly educated and possess stronger digital literacy. Individuals with lower educational attainment or limited technological skills are often less able to access, navigate, or complete web-based questionnaires, which naturally leads to samples that overrepresent digitally capable participants (Ausserhofer et al., 2024; Schroeder et al., 2023). This pattern has also been noted in recent research conducted across Asian contexts, where older adults with higher education levels and more frequent exposure to digital tools are disproportionately present in studies involving online data collection (Kim et al., 2023). Accordingly, the concentration of highly educated respondents in this study should be interpreted as a characteristic of the population segment realistically reachable through online survey methods, rather than an unintended sampling distortion.

To validate the theoretical model (Figure 1), scales were extracted from the literature on information understandability, information diagnosticity, information quality, AI self-efficacy, perceived accountability, AI literacy, social influence, perceived privacy control, personal innovativeness, brand reputation, perceived ease of use, assessment perceived utility, acceptance, and intention. Every item was transcribed from English to Vietnamese and Chinese to verify that respondents interpreted them correctly (Harkness et al., 2003). Participants replied to questions with a five-point Likert scale with labels (strongly disagree/disagree/neutral/agree/strongly agree scale). Based on the literature, the following variables and items were used: Information Understandability with three items were modified from Iranmanesh et al. (2024a).

Information diagnosticity was measured using three items sourced from Iranmanesh et al. (2024a). Information quality was measured using four items derived from Changalima et al. (2024). AI self-efficacy was assessed using an eight-item scale adapted from Wang and Chuang (2023). A two-item scale from Iranmanesh et al. (2024a) was used to assess perceived accountability. The scale for AI literacy, consisting of 10 items, was modeled after Carolus et al. (2023). Social influence was assessed through a six-item scale sourced from Venkatesh et al. (2012). Perceived privacy control with four items was extracted from Salah and Ayyash (2024). Drawing on the scale developed by Ciftci et al. (2021), personal innovativeness was measured using four items. Brand reputation with three items was derived from Roberts and Dowling (2002). The construct of perceived ease of use was assessed with five items based on Iranmanesh et al. (2024a). Assessment perceived utility with 14 items were referenced from Ortiz-López et al. (2024). Acceptance with three items was adapted from Dinh and Stibe (2024). Intention items were measured using a three-item scale adapted from Dinh and Stibe (2024).

We used CFA to evaluate the items adapted from the mixed-source scales recommended by Hair et al. (2010). The model fit and item loading yielded satisfactory results. A very good model fit was revealed with the criteria: 1 to 5 for χ²/df; above 0.90 for GFI, TLI, NFI, and CFI (Kline, 2005); below 0.08 for RMSEA (Bollen, 1989). Table 3 presents the CFA findings.

The approach of Partial Least Squares Structural Equation Modeling (PLS-SEM) is favored for an intricate study framework that encompasses constructs at a higher level that are both reflective and formative alongside moderating variables (Crocetta et al., 2020). Because the path model and hypotheses are based on hypothesized relationships, the study estimates these associations and evaluates the model's explanatory and predictive performance, thereby providing managerial guidance (Shmueli et al., 2019; Hair et al., 2019). After considering these explanations, we used structural equation modeling with SmartPLS4 software (Cheah et al., 2023) to test our model. Conversely, PLS-SEM in SmartPLS 4 enables researchers to estimate the results based on the hypothesized relationships and perform bootstrap multigroup analysis (Ringle et al., 2023). Additionally, as a method for investigation, it provides formative and reflective indicators for shaping as well as data screening.

Using a single method to collect data can bias the relationships between variables and compromise the validity of the analysis (MacKenzie and Podsakoff, 2012). To address this potential issue in our study, which collected data at a single time point, we conducted two diagnostic evaluations to check for common method bias (CMB). First, Harman's single-factor test (MacKenzie and Podsakoff, 2012) demonstrated that the largest single factor accounted for 45.56% of the variance, which is well below the 50% threshold. Second, the full collinearity test (Kock and Lynn, 2012) yielded variance inflation factor (VIF) scores ranging from 1.356 to 2.51, which are safely under the critical value of 3.33. Consequently, these results confirm that CMB did not significantly impact our study's findings.

As presented in Table 4, we first examined the reliability values (outer loadings), which should exceed 0.70 according to Hair et al. (2021). After analyzing the data, we found that all items had values above 0.70, which met the stipulated conditions.

Next, we assessed the reliability and convergent validity of the latent variables using Cronbach's Alpha (CA), composite reliability (CR), and Average Variance Extracted (AVE). To satisfy academic benchmarks, the values for CA and CR should exceed 0.70 (Hair et al., 2021), while the AVE must surpass the 0.50 threshold (Fornell and Larcker, 1981). As demonstrated in Table 4, all constructs successfully met these criteria. Specifically, all CA and CR values were above 0.70, and all AVE scores exceeded 0.50, thereby confirming adequate reliability and convergent validity for the model.

To evaluate discriminant validity, we analyzed both the heterotrait-monotrait (HTMT) ratio and the Fornell-Larcker criterion. First, regarding the HTMT ratio presented in Table 5, researchers suggest a strict threshold of 0.85 (Clark and Watson, 1995; Kline, 2023). However, a more lenient threshold of 0.90 is widely accepted for constructs that are conceptually similar. This higher limit is justified because closely related constructs naturally share a higher degree of overlapping variance, making a strict 0.85 cutoff overly restrictive (Gold et al., 2001; Henseler et al., 2015; Thompson et al., 2008). In general, all HTMT values in this study successfully fell within these acceptable limits. Second, the Fornell-Larcker criterion was fully satisfied because the square root of each construct's AVE was greater than its correlations with any other constructs (Fornell and Larcker, 1981), as detailed in Table 6. Consequently, these combined results confirm that no issues with discriminant validity were present.

In evaluating the hypotheses, the associations among the variables in Table 7 were analyzed through t-test analysis and path coefficient p-values. To bolster the robustness of the results, the authors implemented a nonparametric approach by employing the bootstrap technique to analyze the reliability of the hypotheses (Becker et al., 2022). The findings demonstrated that the majority of the proposed hypotheses were supported, as evidenced by t-test statistics exceeding 2.3 and p-values under 0.05 (Rasoolimanesh et al., 2016).

The findings presented in Table 7 shed light on several key relationships within the proposed model (Figure 1) among older adults in Vietnam. First, AI self-efficacy was positively related to both Assessment Perceived Utility (H1a: β = 0.178, p < 0.01) and Perceived ease to use (H1b: β = 0.362, p < 0.01). Similarly, perceived accountability showed a positive link with system simplicity (H2c: β = 0.120, p < 0.05). However, no significant statistical evidence was found for its association with Assessment Perceived Utility (H2a: β = 0.048, p > 0.05) and acceptance (H2b: β = −0.063, p > 0.05), showing an insignificant role in shaping user perceptions. AI literacy showed significant associations with Assessment Perceived Utility (H3a: β = 0.236, p < 0.01), acceptance (H3b: β = 0.158, p < 0.05), and perceived ease of use (H3c: β = 0.320, p < 0.01). Meanwhile, social influence was positively linked to AI literacy (H4a: β = 0.674, p < 0.001) and acceptance (H4b: β = 0.149, p < 0.01). However, its association with intention was not statistically significant (H4c: β = 0.044, p > 0.05), illustrating that social influence may play a limited role in driving direct behavioral impacts. The study also highlights the function of information diagnosticity, which strongly links with Assessment Perceived Utility (H5: β = 0.128, p < 0.05). In contrast, Information Quality (H6: β = 0.068, p > 0.05) and information understandability (H7: β = −0.053, p > 0.05) did not exhibit meaningful statistical evidence. The perceived simplicity of use proved to be an important determinant, significantly linked with Assessment Perceived Utility (H8a: β = 0.324, p < 0.001) and acceptance (H8b: β = 0.218, p < 0.001). Nonetheless, its association with intention was not significant (H8c: β = 0.086, p > 0.05). Furthermore, Assessment Perceived Utility strongly predicted both acceptance (H9a: β = 0.260, p < 0.001) and intention (H9b: β = 0.400, p < 0.001). Interestingly, Perceived Privacy Control showed no significant statistical evidence with either acceptance (H10a: β = 0.008, p > 0.05) and intention (H10b: β = 0.048, p > 0.05), suggesting that privacy concerns may not weigh heavily in this context. Finally, acceptance was strongly linked to intention, supported by a highly significant result (H13: β = 0.249, p = 0.000).

Regarding the effects of AI self-efficacy, the analysis showed a significant positive relationship with assessment perceived utility, supporting H1a (β = 0.154, p < 0.05). Similarly, AI self-efficacy was significantly and positively linked to perceived ease of use, which validated H1b (β = 0.357, p < 0.001). In contrast, the hypotheses related to perceived accountability yielded mixed outcomes. Statistical evidence indicated no significant relationship between perceived accountability and assessment perceived utility (H2a: β = 0.091, p > 0.05) and acceptance (H2b: β = 0.069, p > 0.05), resulting in the rejection of both H2a and H2b. However, perceived accountability demonstrated a significant positive link with perceived ease of use, thereby confirming H2c (β = 0.243, p < 0.001). Similar to the findings in Vietnam, AI literacy was significantly associated with assessment perceived utility (H3a: β = 0.230. p < 0.001), Acceptance (H3b: β = 0.207, p = 0.001), and Perceived Ease of Use (H3c: β = 0.295, p < 0.001). Social interaction factors were also positively linked to AI literacy (H4a: β = 0.784, p < 0.001) and acceptance (H4b: β = 0.190. p = 0.001), while societal factors were related to intention (H4c: β = 0.138, p < 0.05). Moreover, hypothesis H5 was not supported, indicating that information diagnosticity showed no significant statistical evidence with assessment perceived utility (H5: β = 0.100, p > 0.05). Similarly, neither information quality (H6: β = 0.117, p > 0.05) nor information understandability (H7: β = −0.011, p > 0.05) demonstrated meaningful statistical evidence on assessment perceived utility. Furthermore, perceived ease of use exhibited significant corresponds with assessment perceived utility (H8a: β = 0.311, p < 0.001) and acceptance (H8b: β = 0.157, p < 0.05), but not with intention (H8c: β = 0.078, p > 0.05). Several statistically meaningful links were identified, including between assessment perceived utility and acceptance (H9a: β = 0.323, p < 0.001), as well as intention (H9b: β = 0.311, p < 0.001). Other non-significant findings include perceived privacy control with acceptance (H10a: β = −0.032, p > 0.05) and intention (H10b: β = 0.088, p > 0.05). Moreover, acceptance was significantly linked to intention (H13: β = 0.149, p < 0.05). These results correspond with the findings observed in Vietnam.

To further evaluate the structural model, R Square values were calculated, ranging from 0.615 to 0.756, suggesting a moderate to strong degree of explanatory power (see in Table 8).

In addition, the model also includes interaction variables to assess how two factors moderate the association between acceptance and intention. Both personal innovativeness and brand reputation do not moderate the association between acceptance of AI-based personal financial management and intention to use AI for personal financial management among older adults.

Specifically, in the case of Vietnamese older adults, the results indicate that personal innovativeness does not moderate the association between acceptance and intention (PIxA → I, H11: β = 0.024, p > 0.05). Similarly, brand reputation shows no moderating role in the correspondence between acceptance and intention (BRxA → I, H12: β = −0.063, p > 0.05). Therefore, H11 and H12 were not supported for the Vietnamese sample.

In China, the moderation results show a similar trend compared to Vietnam. Personal Innovativeness does not moderate the appropriation between Acceptance and Intention (PIxA → I, H11: β = −0.045, p > 0.05). The same applies to Brand Reputation, which does not moderate the relation between Acceptance and Intention (BRxA → I, H12: β = −0.014, p > 0.05). Therefore, H11, H12 are also not confirmed in the Chinese sample (See Table 9).

The findings of this study reveal that the determinants of AI adoption for personal financial management (PFM) differ substantially between Vietnam and China. These cross-country variations underscore the importance of examining sociocultural and institutional mechanisms rather than treating country as a background variable. By drawing on Hofstede's cultural dimensions, institutional theory, and aging-related behavioral characteristics, we offer a possible explanation for why certain predictors show stronger associations in one country than the other.

First, AI self-efficacy (AISE) is positively associated with assessment perceived utility (APU) in both China (H1a: β = 0.154, p < 0.05) and Vietnam (H1a: β = 0.178, p < 0.05). It also yields results similar to those reported by Ortiz-López et al. (2024) in their study on the correlation between Mobile Self-Efficacy (MSE) and assessment perceived utility (APU) and offers deeper insights compared to previous studies, such as those by Baham et al. (2022) and Vodanovich et al. (2010), which indicated that older adults generally have lower levels of confidence, technological literacy, and perceived usefulness (PU) of technology. Besides, AI self-efficacy (AISE) exhibits a positive association with perceived ease of use (PEOU): Vietnamese people (H1b: β = 0.362, p-value = 0.000) and Chinese people (H1b: β = 0.357, p-value = 0.000), expanding the understanding compared to the previous study by Zhang et al. (2023). This finding suggests that in Eastern cultural contexts such as Vietnam and China, older adults who possess confidence in using AI-integrated systems tend to feel more assured and perceive themselves as more competent, which may be associated with lower perceived difficulty of engaging with technology.

Second, the study found that the relation between perceived accountability (PA) and the assessment perceived utility (APU) of using AI for personal financial management among older adults was not clearly observed in China (H2a: β = 0.091, p > 0.05) and Vietnam (H2a: β = 0.048, p > 0.05). However, this finding contrasts with that of Cornell et al. (2011), in which process accountability (PA) was shown to significantly connect with perceived usefulness (PU) in the context of technology applications in business environments. This inconsistency may stem from differences in the conceptualization of perceived accountability and process accountability. Since PA remains a relatively new construct, and at the time this study was conducted, most existing research tended to focus more on the process accountability (PA) related to perceived ease of use (PEOU) rather than perceived usefulness (PU). Furthermore, the lack of prior studies examining the perceived usefulness of formative assessment (PUFA) and summative assessment (PUSA) in financial contexts highlights the need for further research to clarify observed patterns. Furthermore, this study shows that the association of perceived accountability (PA) with acceptance decision (A) in personal financial management is not statistically significant, with all p-values greater than 0.05 in Vietnam and China. This result is completely contrary to the previous studies (Adam, 2022; He et al., 2019). In those studies, higher perceived accountability was found to be positively associated with users' willingness to adopt technology, as individuals who felt responsible for their financial decisions were more likely to rely on AI systems to ensure accuracy, transparency, and ethical standards. In contrast, the current finding suggests that when users perceive a higher level of accountability, they may become more cautious or hesitant to adopt AI tools, possibly due to concerns about errors, data misuse, or the difficulty of justifying AI-generated decisions. The study also revealed a novel finding regarding the link of perceived accountability with perceived ease of use (PEOU), with a significant positive connection observed in Vietnam (H2c: β = 0.120, p < 0.05) and in China (H2c: β = 0.243, p < 0.001). The finding indicates that when users perceive a higher level of accountability from the system provider, including transparent operation, responsible data management, and clear communication, they are more likely to perceive the system easier to use. This result is consistent with the study of Baek and Kim (2023), which argued that trust in ChatGPT is primarily associated with by reliability, transparency, and accountability, which are associated with greater user acceptance and engagement.

Third, AI literacy was positively related with assessment perceived utility (APU) in both China (H3a: β = 0.230, p < 0.001) and Vietnam (H3a: β = 0.236. p < 0.05). This indicates that when users have a deeper understanding of technology, they tend to place higher value on the usefulness, value, and reliability of AI applications in finance, which corresponds with the role of technological knowledge in shaping positive perceptions of usefulness and is related to the acceptance of new technology (Salah and Ayyash, 2024). This similarity may be interpreted in light of the fact that Vietnam and China are both in the stage of developing and orienting their technology strategies. Moreover, this research examines the positive association between AI literacy and acceptance, which was significant in both countries: Vietnamese users (H3b: β = 0.158, p-value = 0.010) and Chinese users (H3b: β = 0.207, p-value = 0.001). This outcome corresponds with the results made by Schiavo et al. (2024), showing a positive correspondence of AI competence on the acceptance of AI use in personal financial management among older adults. AI literacy also denotes individuals' competence and expertise in AI-related knowledge and skills, which may be related to older people's comprehension and acceptance of AI (Kang et al., 2022). In addition, the study also shows a positive connection of AI literacy to Perceived Ease of Use (PEOU) in Vietnam (H3c: β = 0.320, p-value = 0.000) and in China (H3c: β = 0.295, p-value = 0.000). This pattern may be interpreted in light of the fact that AI knowledge is linked to greater users' autonomy, confidence, and sense of control when interacting with technology, which may correspond with perceiving the use of AI as easier. This mechanism is consistent with previous studies on digital finance and FinTech, in which technological knowledge is considered a factor that enhances operational competence and linked to lower barriers (Zeng et al., 2025; Salah and Ayyash, 2024). Furthermore, both countries emphasize the development of digital competence and the dissemination of technological knowledge (Zhao, 2024), which offers a possible explanation for the positive and similar perceptions of AI's ease of use in the two contexts.

Fourth, this study investigates the findings that support a significant association between social influence and AI literacy in both countries, including Vietnamese older adults (H4a: β = 0.674, p < 0.001) and Chinese older adults (H4a: β = 0.784, p < 0.001). While previous research has mainly focused on privacy risks, acceptance, and intention to use AI (Xu et al., 2021; Inan et al., 2022), the present results show a statistically significant association between social influence and the level of understanding and competence in AI. Especially in the context of mainstream financial technology, where workplace access and use of AI often rely on shared experiences within the social group, social influence may be related to technology acceptance behavior and may also be linked to the level of knowledge, offering a possible explanation for differences in AI competence and awareness (Srivastava et al., 2024). Regarding the statistical findings, there is a significant connection between social influence and acceptance in both Vietnam (H4b: β = 0.149, p < 0.001) and China (H4b: β = 0.190, p < 0.001). They access information and opinions from society, such as community support and encouragement from children, which may be related to greater trust and acceptance of this new technology. This result is consistent with previous research highlighting the important role of social influence in promoting technology adoption among older adults. Specifically, Wang et al. (2024) found that social influence, expressed through perceived image, shows a statistically significant association with users' willingness to use technology. Lin et al. (2024) showed that social influence, together with AI ethics, related to self-efficacy, which subsequently is linked to trust in AI, performance expectancy, and the intention to adopt AI caregiving robots. Likewise, Ma et al. (2021) reported that social influence is strongly associated with older adults' behavioral intention to adopt new technologies. In addition, the research results show that the relation between social influence and AI usage intention differs significantly between Vietnam and China among older users. Specifically, in China, the social influence factor has a statistically significant association with AI usage intention (H4c: β = 0.138, p < 0.05), while in Vietnam, this relationship is not statistically significant (H4c: β = 0.044, p > 0.05). China's significantly stronger collectivistic culture and higher power distance (Hofstede, 2011) help explain why social influence shows a stronger association with acceptance and intention among older Chinese respondents, but not among Vietnamese participants. In collectivistic societies, individuals often rely on social norms, peer recommendations, and family endorsement when evaluating new technologies. For older Chinese users who typically maintain close intergenerational living arrangements, social reassurance and social proof function as important risk-reduction mechanisms. This is consistent with prior findings in Asian cultural contexts where collectivism amplifies the weight of normative pressures (Zhang et al., 2025b; Chen and Chan, 2014). In contrast, Vietnamese respondents demonstrated a weaker link between social influence and AI usage intention. Although Vietnam is also collectivistic, the country's rapid digital transformation and higher exposure to Westernized consumer behavior among urban older groups may offer a possible explanation for the weaker reliance on social norms. Vietnamese older adults increasingly make technology adoption decisions based on personal utility evaluation and experiential knowledge, aligning with emerging trends in Southeast Asia where individual experiential learning moderates social norm dependency. Furthermore, differences in uncertainty avoidance may provide additional insight. China traditionally exhibits higher uncertainty avoidance levels, which may be associated with older adults relying more strongly on trusted social referents when engaging with technologies involving financial decisions. This explains why social influence and perceived accountability show stronger associations in China. Meanwhile, Vietnam's comparatively lower uncertainty avoidance allows older adults to tolerate greater ambiguity when experimenting with AI-based tools, making diagnostic information and individual evaluations (APU) more salient.

Following, the information provided by AI is related to assessment perceived utility (APU), which is classified into three categories: information diagnosticity (ID), information understandability (IU) and information quality (IQ). The study found that Information Diagnostics of AI was positively related to assessment perceived utility (APU) in Vietnam (H5: β = 0.128, p < 0.05). This aligns with previous studies that demonstrated the positive association of information diagnosticity with perceived usefulness (Iranmanesh et al., 2024b). In contrast, in China (H5: β = 0.100, p > 0.05), the opposite result may be interpreted in light of specific policy orientations toward AI development and technology protectionism aimed at supporting domestic innovation. The results indicate that the connection of Information Quality (IQ) and Information Understandability (IU) to assessment perceived utility (APU) did not show statistical significance in either China or Vietnam. Both Hypothesis 6 and 7 showed p-values greater than 0.05, suggesting that these relationships were not supported. This finding is consistent with the study of Iranmanesh et al. (2024b), which also reported a non-significant result. However, as discussed in Hypothesis H1a, this study calls for further research to clarify these factors and their associations with assessment perceived utility (APU) rather than PU. Although Vietnam and China are neighboring countries that share many cultural similarities, they differ substantially in terms of economic scale and technological advancement. This finding may suggest that informational factors related to AI may not be strongly related to economic or technological disparities when cultural foundations are relatively similar. Institutional theory suggests that technology adoption is shaped not only by cultural factors but also by the maturity of the regulatory, technological, and financial ecosystems (Scott, 2001). China's highly advanced fintech infrastructure, robust AI governance frameworks, and state-led digital ecosystem may be associated with stronger institutional trust in technology. This institutional trust may offer a possible explanation for the perceived importance of accountability mechanisms and social influence pathways. The prominence of tech giants (e.g. Ant Group, Tencent) and long-established digital finance usage among seniors also may be related to stronger institutionalized norms around technology use. Vietnam's institutional environment remains developing, with lower fintech penetration among older populations and less formal integration of AI accountability frameworks. Consequently, Vietnamese older adults may rely more heavily on cognitive evaluation of information quality, understandability, and diagnosticity when assessing AI usefulness. This aligns with research suggesting that in emerging markets, user trust is more strongly constructed through direct evaluation of informational features rather than institutional assurances (Mullan et al., 2017). Future research could yield more comprehensive insights by comparing nations with greater cultural heterogeneity, such as those in Asia and Europe.

Moreover, considering the different factors and the basis of this relationship, we found a positive association between perceived ease of use (PEOU) and assessment perceive utility (APU) in both countries, achieving favorable and acceptable results in both China (H8a: β = 0.311, p < 0.001) and Vietnam (H8a: β = 0.324, p < 0.001). However, this is contrary to Ortiz-López et al. (2024), which indicated that the relationship between perceived ease of use (PEOU) and assessment perceived utility (APU) was not supported in the education sector, especially for continuing master's students. In contrast, it is accepted in the finance sector with individuals who have mostly stopped studying or are retired. The study also identifies a positively linked between perceived ease of use (PEOU) and acceptance among older adults in Vietnam (H8b: β = 0.218, p = 0.000), suggesting that systems perceived as easy to use and intuitive are more likely to be adopted, which is the same as the study of Bui et al. (2025). Similarly, elderly individuals in China also showed a statistically significant association between perceived ease of use (PEOU) on acceptance (H8b: β = 0.157, p < 0.05). This finding is consistent with previous studies, such as Schiavo et al. (2024), which highlight the critical role of perceived ease of use in promoting acceptance in the context of technology adoption. In addition, hypothesis H8c was rejected because the effect of perceived ease ofuse (PEOU) on intention to use AI among older adults in Vietnam and China was not statistically significant (H8c: β = 0.086, p > 0.05) and (H8c: β = 0.078, p > 0.05). This contrasts with the findings of Zhang et al. (2023), in which PEOU had a positive relation to the intention to use AI when studying a group of pre-service teachers. Similarly, the study by Pang et al. (2024) on ChatGPT usage behavior also confirmed that PEOU was associated with usage behavior through perceived usefulness and technological compatibility. This difference may be interpreted in light of the demographic characteristics and technological proficiency of older adults. According to Liesa-Orús et al. (2022), older age makes the use of technology more difficult, and therefore their attitudes toward technology tend to be more negative. This implies that when the ability to use and confidence in using technology are low, the ease of use factor is no longer a direct driver of behavioral intention.

Subsequently, this study delves into the profound relationship between intention to use AI, perceived utility of AI, acceptance of AI, Brand reputation, Personal Innovativeness, and perceived privacy control. First, this research also shows a statistically significant association with assessment perceived utility (APU) and acceptance among older adults for personal financial management and planning: in Vietnam (H9a: β = 0.260. p < 0.001) and in China (H9a: β = 0.323, p < 0.001), indicating that users who perceive clear benefits from the technology will have higher motivation to accept it. This is especially true for older adults in areas such as personal financial management. This finding is consistent with previous studies and corresponds with technology acceptance theories, including the Technology Acceptance Model (TAM) (Martín-García et al., 2021; Cheng et al., 2023). Next, the research results show that perceived utility (PU) is related with intention to use AI among older adult people in both Vietnam (H9b: β = 0.400, p < 0.001) and China (H9b: β = 0.311, p < 0.001), showing that when users clearly perceive the benefits that AI brings to their financial lives such as saving time, optimizing spending plans, or supporting investment decisions, they tend to report stronger intentions to apply AI in personal financial management (Iranmanesh et al., 2024b). This is consistent with the technology acceptance model of Davis (1989), in which APU is identified as one of the factors associated with the behavioral intention to use technology. This result is also similar to the study of Ortiz-López et al. (2024), when the author confirmed that APU is positively associated with the intention to apply mobile devices for learning assessment activities. This similarity shows that, in both contexts, when users perceive that technology brings practical benefits to their work or personal life, they tend to report stronger intentions to accept and use (Mustofa et al., 2025).

Even though there is still a correlation between Vietnam and China, the connection between perceived privacy control and acceptance among older adults was not statistically significant; therefore, H10a was rejected: in Vietnam (H10a: β = 0.008, p > 0.05) and China (H10a: β = −0.032, p > 0.05). That is, Perceived Privacy Control is often considered an important factor in technology acceptance, but for this age group, this factor may not be strongly associated with the decision to use AI in financial management in both countries. This result is consistent with the findings of Li et al. (2024), who argued that older adults are more concerned about the usefulness and security of technology than about privacy issues. This suggests that among older users, the motivation to adopt technology is largely pragmatic, as they are willing to use a system when it provides clear and practical benefits in daily life. Moreover, when privacy is not viewed as a major concern, its association with technology acceptance may be limited. This is consistent with Ho et al. (2023), who found that privacy-related considerations play only a minor role in older adults' decisions to adopt new technologies. Similar results were also seen in the relationship between perceived privacy control and intention to Use AI for Personal Finance Management in both Vietnam (H10b: β = 0.048, p-value = 0.298) and China (H10b: β = 0.088, p-value = 0.112). The link between perceived privacy control and intention to Use AI was not statistically significant, which contrasts with previous studies such as Seiler and Fanenbruck (2021) and Belanche et al. (2019), which affirmed that perceived privacy control is positively associated with the intention to adopt financial technology. The difference may be interpreted in light of the technological characteristics and financial behaviors of the elderly group, as they often have a low level of familiarity with digital technology and are easily hindered by security barriers, leading them not to view the ability to control data as a key factor. Instead, they place their trust in the safety and system design that follows the “privacy-by-design” principle, considering it the main guarantee for reliability and readiness to use technology (Jnr, 2024).

Regarding the complex relationship between acceptance and intention, the moderating roles of personal innovativeness and brand reputation were found to be statistically insignificant in both Vietnam and China (H11: p > 0.05; H12: p > 0.05). Higher personal innovativeness was expected to strengthen the relationship between acceptance and intention, given its established role in digital transformation (Cavalcanti et al., 2022) and financial technology acceptance (Shaikh et al., 2020). However, this moderation did not materialize among older adults using AI-based financial management tools. One possible explanation is that many older adults, although open to innovation, still fail to translate that openness into behavioral intention when their AI literacy, social influence, and assessment of perceived utility remain limited. Similarly, regarding brand reputation, the hypothesized moderating effect on the relationship between acceptance and intention was not supported in both countries. This finding contrasts with previous studies, which emphasized that a strong brand reputation can enhance trust, reduce perceived risks (Rindell and Iglesias, 2014; Martínez et al., 2014), and increase consumers' intention to adopt new technologies (Hua et al., 2023) particularly in financial contexts. A possible explanation is that AI-based financial services are still relatively new and under development, meaning that even well-established brands may not yet have built sufficient familiarity or emotional trust to significantly influence consumers' acceptance and intention to adopt these technologies.

Finally, acceptance had a significant positive relate to the intention to use AI for personal financial management and planning (H13). This hypothesis was supported in both countries. Specifically, the effect was significant for Vietnam (H13: β = 0.249, p < 0.001) and China (H13: β = 0.149, p < 0.005). This finding indicates that when older adults feel comfortable, confident, and ready to integrate AI technology into their financial lives, they not only form a temporary intention to use it but also develop a sustainable acceptance tendency toward new technology (Rahman et al., 2023). This is also consistent with the findings of Han and Kim (2020), in which the older age group showed stronger emotional responses compared to the younger group, and they tended to form sustainable acceptance and intention to use. In summary, the above results confirm previous studies, such as the relationship between the perceived usefulness of mobile device use and the intention to use it (Ortiz-López et al., 2024), and the association between adoption and intention to use emerging blockchain technology (Dinh and Stibe, 2024), and extend research findings on AI usage intention into new relationships in the financial sector and provide a comparative analysis between older adults in Vietnam and China.

This study contributes to research on aging in economics and finance by examining older adults' behavioral intentions to use AI for personal financial management, particularly in the context of Industry 4.0 and rapid global aging. According to projections by the World Health Organization, the proportion of the global population aged 60 and above is expected to reach 22% by 2050 (da Paixao Pinto et al., 2021).

While much of the existing research on technology adoption by older adults has centered around healthcare and medical applications (Rubeis, 2020; Porkodi and Kesavaraja, 2021), other studies have primarily focused on their adoption of technologies such as intelligent social media networks, recreational platforms, and Internet public services (Mariano et al., 2021; Kim et al., 2021). Building on this foundation, this study proposes an integrated framework that combines the technology acceptance model (Davis et al., 1989) and the knowledge behavior gap model (Dinh and Stibe, 2024). This model emphasizes the connections between knowledge, acceptance, and intention to use technology. Older adults face difficulties in accessing digital services and available knowledge resources, and in many cases, they may become socially excluded (Valkonen and Kujala, 2024).

Additionally, this study is among the first to investigate AI applications for personal financial management among older adults, laying the groundwork for future research in expanding AI use in economics and finance. The study also offers an initial foundation for developing AI systems that are genuinely tailored to the needs and lived experiences of older adults. As noted by Rubeis et al. (2022), deploying technology to support healthy aging in place may unintentionally reinforce existing inequalities. This risk becomes more pronounced when the diverse needs and resources of individuals or groups are insufficiently recognized, when their characteristics are reduced to oversimplified or stereotypical assumptions, or when structural barriers limit equitable access. Consequently, the design and functional requirements of such technologies demand a more nuanced and comprehensive understanding of the multiple factors that shape older adults' usage contexts. The findings of this study contribute valuable data, further refining and supplementing research models such as TAM and KBG. Integrating multiple models offers a novel direction for future studies to explore the impact of technology on older adults and encourages a deeper investigation into how AI can address their financial needs.

This study also evaluates and extends the assessment perceived utility (APU) variable, derived from perceived usefulness in the TAM model, as proposed by Ortiz-López et al. (2024). By examining information characteristics such as diagnosticity, quality, and understandability, this study analyzes how these factors shape older adults' perceptions and acceptance of AI for financial management. Furthermore, it aligns with calls to prioritize older adults as research subjects (Bernal et al., 2024). Most research in this domain is still in its infancy, relying on observations, focus groups, and interviews, which underscores the need for further study.

Finally, in an AI-driven world, older adults face an increased risk of social exclusion due to the digital divide (Rosales and Fernández-Ardèvol, 2020). Unlike racial or gender biases, age-related bias in AI has received little attention in the literature. It is crucial to address how ageism manifests in AI, where aging in a digital world perpetuates inequities and marginalization (Chu et al., 2022). This study seeks to bridge the gap between older adults and modern technology, promoting equality and improving their quality of life. The cross-national design, conducted in Vietnam and China, enables a comparative analysis of age-related differences in technology adoption within economic and financial contexts. This approach enhances the generalizability of the findings and provides deeper insights into the structural relationships among the proposed variables.

This study highlights the practical implications for key stakeholders including fintech providers, financial educators, and government bodies in Vietnam and China, to promote the adoption and effective use of AI Chatbots in Personal Finance Management (PFM) among older adults.

Fintech providers targeting older users should prioritize designing AI Chatbot tools in the field of Personal Finance Management (PFM) with intuitive interfaces and strong security features to enhance usability and build user trust. In addition, simple and user-friendly functions, such as voice commands or step-by-step guidance, can shorten the learning curve for elderly users. In particular, AI Chatbots can serve as “personal financial assistants”, supporting users in various aspects of financial management. Rather than merely providing information or payment support, Chatbots can proactively remind users of payment schedules, analyze spending patterns, suggest effective saving strategies, or recommend safe investment channels based on individual risk profiles. With their natural conversational interaction, AI acts as a companion, helping older adults feel supported, guided, and more confident when making financial decisions.

Financial educators also play an essential role in promoting and guiding elderly individuals to use AI Chatbots in personal financial management. They can organize training programs, tutorial videos, or online courses to help users better understand how Chatbots can assist in financial planning, monthly expense management, small-scale investment decisions, and retirement preparation.

Governments should establish practical and safety-oriented policies to encourage older adults to adopt AI Chatbots in Personal Finance Management (PFM). Specifically, they should issue clear regulations on data protection and privacy to ensure that users feel secure when sharing personal financial information; set up support centers or hotlines to provide technical assistance and timely guidance for elderly users; and introduce financial incentives such as subsidies, tax reductions, or development support for businesses that create user-friendly Chatbots. In addition, insurance or financial compensation mechanisms could be implemented for Chatbot-based transactions to strengthen user confidence and create a safe, trustworthy environment for older adults as they engage with this technology.

In Vietnam, the government can focus on the Information Diagnosticity factor by enhancing the transparency, reliability, and usefulness of information provided by AI Chatbots. This can be achieved through data quality verification standards, source validation mechanisms, and standardized financial advice guidelines to ensure that elderly users trust the accuracy of Chatbot information and make more effective financial decisions. Meanwhile, in China, the government should leverage the social influence factor by promoting public communication campaigns and community-based initiatives that enhance the positive social perception of using AI Chatbots in PFM. These programs could encourage celebrities, financial experts, or social organizations to share positive experiences with Chatbots, thereby creating social momentum and motivation for older adults to adopt and use this technology in managing their personal finances.

This study has some limitations. First, the sample contains a relatively high proportion of respondents with higher educational backgrounds and stronger digital literacy. This imbalance is common in online research involving older adults, as individuals with lower education or limited technology skills are less likely to access, navigate, or complete web-based surveys. As a result, the findings should not be interpreted as fully representative of the general older population in Vietnam and China, but rather of the subgroup that is realistically reachable through online data collection and already capable of engaging with digital tools. Furthermore, the antecedents that affect intention are complex. Although AI self-efficacy, perceived accountability, and AI literacy play vital roles, factors such as trust and support also influence the intention to apply AI in personal financial management. Therefore, future research will consider more antecedents to enrich and improve existing research. Another limitation is that the research outcomes may not generalize to other financial AI technologies beyond tools that provide diagnostic information, evaluation support, and personalized financial recommendations. As developers continue to introduce an increasing number of financial AI applications, future research should expand its scope to other types of applications to compare adoption levels and impacts on acceptable and personal financial management behavior. Furthermore, because the data were collected using a cross-sectional survey, the estimated path coefficients should be interpreted as statistical associations rather than evidence of causal direction. The design does not establish temporal precedence among the examined constructs and cannot rule out reverse relationships or omitted-variable explanations. Future research could employ longitudinal, experimental, or multi-wave designs to evaluate the temporal ordering and potential directionality of these relationships more rigorously. These new directions can help build a more in-depth understanding of technology and a highly focused intention to adopt, considering the broader range of influences and contexts.