Investor Heterogeneity in Preferences for AI-Based Financial Services: Evidence from Japanese Online Investors

Abstract

This study examines investor heterogeneity in relative priorities for AI-based financial services using a large-scale survey of Japanese online investors. We use data from the 2026 wave of the “Survey on Life and Money,” administered by Rakuten Securities and Kadoya Lab at Hiroshima University. The final analytical sample comprises 14,432 respondents. Respondents selected their first-, second-, and third-preferred AI-based services from 11 options, which were grouped into five functional categories: administrative procedure proxy services, consultation for troubles and emergencies, education and literacy support, information provision and planning support, and advisory and management for investment. Because respondents were required to select their top three preferred services, the dependent variables capture relative priorities rather than absolute willingness to use AI services. Binary probit and ordered probit models show that investor characteristics are associated with relative priorities across service categories, although the estimated marginal effects are generally modest. Information provision and planning support is more strongly prioritized by male respondents, more-educated respondents, and those with greater household financial assets. Advisory and management services are more strongly prioritized by higher-income and more impatient respondents, while risk aversion is negatively associated with this category. Additional robustness checks suggest that these patterns are not explained entirely by unequal category sizes, although option-level results reveal some within-category heterogeneity. These findings suggest that AI-based financial services should reflect investor heterogeneity while recognizing that service categories may contain diverse underlying functions.

1. Introduction

Artificial intelligence (AI) is increasingly being incorporated into financial services, including customer support, information provision, financial planning, portfolio advice, and automated investment management (Shanmuganathan 2020; Han et al. 2023; Polireddi 2024; Gurram 2025; Sai et al. 2025; Vuković et al. 2025). For financial institutions, AI-based services may improve operational efficiency, personalize financial information, and support investors’ decision-making (Polireddi 2024; Gurram 2025; Sai et al. 2025). For investors, these services may be designed to provide information, procedural support, education, consultation, and investment advice (Han et al. 2023; Sai et al. 2025). However, AI-based financial services are not a single homogeneous product category. Different services involve different functions, levels of autonomy, and implications for investor decision-making and risk management.

Understanding investor preferences for AI-based financial services is therefore important for responsible AI adoption in financial markets. If AI services are designed mainly around advanced information provision or automated investment management, they may not fully reflect the priorities of investors who seek consultation, education, or procedural support. Conversely, if AI services are designed only as basic support tools, they may not meet the priorities of investors who seek more advanced information processing, portfolio monitoring, or investment advisory functions. This service-design issue is particularly relevant in financial markets, where AI-based advisory and management services may influence investment decisions and therefore require careful attention to transparency, suitability, explainability, and investor protection (Han et al. 2023; Ridzuan et al. 2024; Vuković et al. 2025; Kulkarni et al. 2025; Nahidi 2026).

Prior studies have examined technology adoption, FinTech use, robo-advisory services, trust in automated advice, digital financial literacy, and individual differences in digital financial behavior (Piotrowski and Orzeszko 2023; Akhtar et al. 2024; Banerjee 2025; Kumari et al. 2024; Aftab et al. 2025; Lal et al. 2025; Santini et al. 2026; Singh and Kumar 2025; Chowdhury et al. 2026). These studies have provided important insights into who adopts digital financial technologies and how individual characteristics are related to technology use. However, many existing studies tend to examine AI or FinTech adoption as a broad outcome. In practice, financial platforms may provide multiple AI-based services that differ substantially in purpose, complexity, and risk implications. Relatively little is known about which types of AI-based financial services investors prioritize within the same financial platform.

This study addresses this gap by examining investor heterogeneity in relative priorities for AI-based financial services among Japanese online investors. We use data from the 2026 wave of the “Survey on Life and Money,” administered by Rakuten Securities and Kadoya Lab at Hiroshima University. Respondents were asked to select their first-, second-, and third-preferred AI-based services from a list of 11 service options. These options were grouped into five functional categories: administrative procedure proxy services, consultation for troubles and emergencies, education and literacy support, information provision and planning support, and advisory and management for investment. Because respondents were required to select their top three preferred services, this study interprets the dependent variables as measures of relative priority among AI-based financial service categories rather than absolute willingness to use AI services. The final analytical sample comprises 14,432 individuals.

We estimate binary probit and ordered probit models to examine how demographic, socioeconomic, and psychological characteristics are associated with these relative priorities. Specifically, we analyze two types of dependent variables for each AI service category: a binary selection indicator and an ordinal priority-ranking indicator. The binary indicator captures whether a respondent selected at least one service in a given category among the three rankings, while the ordinal indicator captures the highest-ranked selection within that category. We also report average marginal effects and conduct additional robustness checks using individual service options and category-size-adjusted outcomes to address the unequal number of options across categories.

This study makes three contributions. First, it provides large-scale empirical evidence on investor heterogeneity in preferences for AI-based financial services. Second, it distinguishes among multiple AI service categories rather than treating AI adoption as a single outcome. Third, it incorporates not only demographic and socioeconomic characteristics but also psychological traits, including risk aversion, hyperbolic discounting, and impatience, which may be relevant for understanding priorities for advisory and management services involving recommendations or automation. In doing so, this study contributes to the literature on digital finance, AI-based financial services, and responsible service design in financial markets.

The results show that investors’ relative priorities differ across AI service categories. Information provision and planning support is more strongly prioritized by male respondents, more-educated respondents, and those with greater household financial assets. Consultation for troubles and emergencies and education and literacy support show different patterns, suggesting that AI-based services may be prioritized not only as advanced information or advisory tools but also as support mechanisms. Advisory and management services are more strongly prioritized by higher-income and more impatient respondents, while risk aversion is negatively associated with this category. Hyperbolic discounting is also positively associated with the probability of ranking advisory and management services as the first priority. However, the estimated marginal effects are generally modest, and the pseudo-R² values indicate that individual characteristics explain only a limited portion of variation in relative AI service priorities. Option-level analyses further suggest that some category-level patterns are driven by particular service options, especially in the information/planning and advisory/management categories. We therefore interpret the findings as evidence of heterogeneous relative priorities, while avoiding claims that the five service categories are internally homogeneous.

The remainder of this paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the data, variables, and empirical methods. Section 4 presents the estimation results. Section 5 discusses the findings and their implications for AI-based financial service design. Section 6 concludes the paper.

2. Literature Review

Artificial intelligence and digital technologies are increasingly being integrated into financial services, including customer support, financial information provision, robo-advisory services, portfolio monitoring, and automated investment management (Shanmuganathan 2020; Han et al. 2023; Polireddi 2024; Gurram 2025; Sai et al. 2025; Vuković et al. 2025). These services are often expected to improve efficiency and personalization, but they also differ substantially in their purpose and level of autonomy. Some AI-based services provide operational or educational support, while others directly assist investment decision-making through portfolio recommendations, trading–timing advice, or automation (Han et al. 2023; Sai et al. 2025). Therefore, AI-based financial services should not be treated as a single homogeneous service category.

Prior studies have examined how individual characteristics are associated with the use of FinTech, digital financial services, robo-advisory platforms, and AI-based tools (Piotrowski and Orzeszko 2023; Akhtar et al. 2024; Banerjee 2025; Kumari et al. 2024; Aftab et al. 2025; Lal et al. 2025; Santini et al. 2026; Singh and Kumar 2025; Chowdhury et al. 2026). These studies suggest that demographic and socioeconomic factors, such as gender, age, education, income, and financial literacy, may be related to digital financial behavior (Banerjee 2025; Kumari et al. 2024; Lal et al. 2025). Psychological factors, including risk attitudes, trust, and time preferences, may also be relevant to how individuals evaluate digital or automated financial services (Aftab et al. 2025; Singh and Kumar 2025). However, less attention has been paid to differences across specific categories of AI-based financial services.

Distinguishing among AI service categories is particularly important in financial markets because different services may involve different implications for investor protection and risk management. AI-based consultation, education, and procedural support may help investors navigate financial platforms or resolve problems, whereas AI-based advisory and management services may influence portfolio choice, trading behavior, or risk exposure. For services involving recommendations or automation, transparency, explainability, suitability, and governance are especially important (Han et al. 2023; Ridzuan et al. 2024; Vuković et al. 2025; Kulkarni et al. 2025; Nahidi 2026). This suggests that financial institutions should consider investor heterogeneity when designing AI-based financial services. The present study contributes to this literature by examining which types of AI-based financial service categories Japanese online investors prioritize, and how these relative priorities are associated with demographic, socioeconomic, and psychological characteristics.

These considerations motivate the inclusion of demographic, socioeconomic, and psychological characteristics in the empirical analysis. Demographic and socioeconomic variables, such as gender, age, education, employment status, income, and financial assets, may be associated with access to financial information, experience with digital platforms, and the ability to use AI-based financial services. Psychological traits may also be relevant because AI-based advisory and management services involve uncertainty, delegation, and intertemporal decision-making. In particular, risk attitudes may be related to willingness to rely on AI-supported recommendations or automated management, while time preferences may be associated with demand for services that provide timely advice, reduce decision costs, or support immediate action. Because the study is exploratory, we do not treat these variables as causal determinants of AI service priorities. Rather, we examine whether they are systematically associated with relative priorities across different types of AI-based financial services.

3. Data and Methods

3.1. Data

This study employed the online panel survey, “Survey on Life and Money,” administered by Rakuten Securities and the Kadoya Lab at Hiroshima University. Rakuten Securities is one of Japan’s largest online security companies, with over 14 million users as of April 2026 (Rakuten Securities 2026). This accounts for approximately 11% of Japan’s population and covers nearly 27.3% of the roughly 51 million online securities accounts nationwide (Japan Securities Dealers Association 2025). These figures highlight Rakuten Securities’ substantial market presence and support the relevance of the findings for a broad segment of Japan’s active online investors.

We analyzed data from the 2026 wave collected in January and February 2026. The survey sampled individuals aged 18–90 who held active accounts and had accessed the company’s website at least once within the previous year. It gathered information on investment preferences, behavior, demographics, socioeconomic status, and psychological traits. The initial 2026 dataset contained 26,534 observations. Because the present study focuses on willingness to use AI-based financial services, observations were excluded only when they had missing values for the willingness-to-use outcomes or for variables used in the regression analysis. Observations with missing willingness-to-pay responses were not excluded because willingness to pay is not analyzed in this study. After listwise deletion based on the variables used in the present analysis, the final analytical sample comprised 14,432 individuals.

Certain variables, including education years and measures of time preferences, were supplemented from earlier survey waves when they were unavailable in the 2026 wave. Specifically, earlier waves were used only to complete missing information for variables that are not observed for all respondents in 2026. This approach increases the analytical sample size but may introduce measurement inconsistencies if some characteristics changed over time. We therefore interpret the estimates as associations and acknowledge cross-wave supplementation as a limitation.

3.2. Variables

3.2.1. Dependent Variable

This study employed two types of dependent variables to capture investors’ relative priorities for each category of AI-based financial services: a binary selection indicator and an ordinal-priority-ranking indicator. These variables were constructed based on the following survey question:

• Q. If the following AI-based financial services were provided by Rakuten Securities, please select the top 3 services you would like to use in order of preference.

• Administrative procedure proxy service (changes to registered matters, etc.)
• Administrative procedures for transferring funds (deposits and withdrawals, inher-itances, gifts, etc.)
• Consultation for troubles and emergencies (issues, complaints, malfunctions, etc.)
• Services that teach general investment rules and theories.
• Services that teach how to use our company’s services and tools.
• Life plan simulation services.
• Investment information provision services (stock information, economic information, etc.)
• Optimal portfolio proposal service.
• Advisory services for trading timing.
• Automated trading services for stock investment trusts, etc.
• Reporting services for investment performance.

Respondents selected their first-, second-, and third-preferred services from the same list of 11 AI-based service options. The options were as follows: (1) administrative procedure proxy services, such as changes to registered matters; (2) administrative procedures for transferring funds, such as deposits, withdrawals, inheritances, and gifts; (3) consultation for troubles and emergencies, such as issues, complaints, and malfunctions; (4) services that teach general investment rules and theories; (5) services that teach how to use the company’s services and tools; (6) life-plan simulation services; (7) investment information provision services, such as stock and economic information; (8) optimal portfolio proposal service; (9) advisory services for trading timing; (10) automated trading services for stock investment trusts and related products; and (11) reporting services for investment performance.

To facilitate interpretation, the 11 service options were grouped into five functional categories according to the primary purpose of each service. Administrative procedure proxy services include options 1 and 2, which provide procedural support for account-related matters and fund transfers. Consultation for troubles and emergencies includes option 3, which focuses on problem resolution and emergency support. Education and literacy support includes options 4 and 5, which provide investment knowledge and guidance on how to use the company’s services and tools. Information provision and planning support includes options 6, 7, and 11, which provide life-planning simulation, investment information, and investment performance reporting. Although investment performance reporting may also be viewed as a monitoring function, we classify it under information provision and planning support because its primary function is to provide information for evaluating and planning investment decisions rather than directly recommending or executing investment actions. Advisory and management for investment includes options 8, 9, and 10, which involve portfolio proposals, trading–timing advice, and automated trading services.

This classification is intended to organize the 11 options into functionally interpretable service groups. However, the categories contain different numbers of options, and some services may have overlapping functions. Therefore, the category-level variables should not be interpreted as internally homogeneous constructs. To address this issue, we also conduct robustness checks using individual service options and category-size-adjusted outcomes.

Because respondents were required to select their first-, second-, and third-preferred services, the dependent variables should be interpreted as measures of relative priority among AI-based financial service options rather than absolute willingness to use AI services. This forced top-three design also implies mechanical dependence across choices: selecting one service necessarily affects the relative position of the remaining services. Therefore, the category-level outcomes are not independent measures of demand for each category. Instead, they indicate which types of AI-based financial services respondents prioritized relative to other available AI service options.

In addition, because categories contain different numbers of underlying options, categories with more options may have a higher probability of being selected mechanically. We therefore interpret descriptive differences in category selection rates with caution and supplement the main category-level analysis with option-level and category-size-adjusted robustness checks.

We also constructed an ordinal priority-ranking indicator for each category. This variable was based on the highest-ranked option selected within the category. Specifically, it takes the value of 3 if any service in the category was ranked first, 2 if the highest-ranked service in the category was ranked second, 1 if the highest-ranked service in the category was ranked third, and 0 if no service in the category was selected among the three rankings. This approach captures the relative priority assigned to each AI service category while avoiding double-counting when multiple services within the same category were selected.

3.2.2. Independent Variables

To investigate what kind of people prefer to use which AI-based financial service category, the independent variables cover broad individual characteristics. These can be classified into three categories: (1) demographic—gender, age, marital status, parental status, and educational background; (2) socioeconomic—employment status, household income, and household assets; and (3) psychological—risk aversion, hyperbolic discounting, and impatience.

Risk aversion was measured using the respondent’s answer to the question: “When you usually go out with an umbrella, what is the probability of rain?” We interpret this variable as an indirect proxy for general risk attitudes. This measure captures a respondent’s subjective threshold for taking a preventive action under uncertainty, but it does not directly measure financial risk tolerance or portfolio risk preferences. Therefore, the risk-aversion results should be interpreted cautiously. Measurement error in this proxy may attenuate or otherwise affect the estimated associations.

Time-preference variables, such as hyperbolic discounting and impatience, were derived from two survey questions (see Appendix A), following the methodology of Ikeda et al. (2010). In the first question, respondents chose between receiving a reward after 2 or 9 days. In general, individuals tend to prefer the earlier reward (option A), but switch to the later one (option B) when the utility of option B exceeds that of option A. This switching behavior continues as the interest rate progressively increases with each subsequent question. The discount rate for the first question (DR1) was defined based on the interest rate at their switching points. The same procedure is applied to the second question, which involves a more distant time horizon (90–97 days), and a corresponding measure (DR2) was calculated. Respondents who exhibited multiple switching points were excluded from the analysis. Hyperbolic discounting was identified when respondents displayed a higher discount rate at shorter time horizons than longer ones (DR1 > DR2), and such respondents were classified as hyperbolic discounters.

Impatience is then measured as the average of the standardized values of DR1 and DR2, calculated using the following equation:

$$Impatience={\displaystyle \frac{1}{2}}\sum _{i=1}^2\left[\left({DR}_i-E\left({DR}_i\right)\right)/\sigma \left({DR}_i\right)\right]$$

(1)

Table 1. Variable definitions.

Variables	Definition
Dependent Variable
Selection of administrative procedure proxy services among top-three preferred AI services	Binary variable: 1 = selecting at least one AI-based service in this category among the top-three preferred services, 0 = otherwise.
Selection of consultation for troubles and emergencies among top-three preferred AI services	Binary variable: 1 = selecting at least one AI-based service in this category among the top-three preferred services, 0 = otherwise.
Selection of education and literacy support among top-three preferred AI services	Binary variable: 1 = selecting at least one AI-based service in this category among the top-three preferred services, 0 = otherwise.
Selection of information provision and planning support among top-three preferred AI services	Binary variable: 1 = selecting at least one AI-based service in this category among the top-three preferred services, 0 = otherwise.
Selection of advisory and management for investment among top-three preferred AI services	Binary variable: 1 = selecting at least one AI-based service in this category among the top-three preferred services, 0 = otherwise.
Priority ranking of administrative procedure proxy services among AI service categories	Ordinal variable: priority ranking for this AI-based service category (3 = ranked first, 2 = ranked second, 1 = ranked third, 0 = not selected among the top-three preferred services).
Priority ranking of consultation for troubles and emergencies among AI service categories	Ordinal variable: priority ranking for this AI-based service category (3 = ranked first, 2 = ranked second, 1 = ranked third, 0 = not selected among the top-three preferred services).
Priority ranking of education and literacy support among AI service categories	Ordinal variable: priority ranking for this AI-based service category (3 = ranked first, 2 = ranked second, 1 = ranked third, 0 = not selected among the top-three preferred services).
Priority ranking of information provision and planning support among AI service categories	Ordinal variable: priority ranking for this AI-based service category (3 = ranked first, 2 = ranked second, 1 = ranked third, 0 = not selected among the top-three preferred services).
Priority ranking of advisory and management for investment among AI service categories	Ordinal variable: priority ranking for this AI-based service category (3 = ranked first, 2 = ranked second, 1 = ranked third, 0 = not selected among the top-three preferred services).
Independent Variable
Male	Binary variable: 1 = male, 0 = female
Age	Continuous variable: respondents’ age
Age squared	Continuous variable: age squared
Spouse	Binary variable: 1 = having a spouse, 0 = otherwise
Number of children	Continuous variable: number of respondents’ children
Education years	Continuous variable: number of education years
Full-time employed	Binary variable: 1 = employed as a full-time worker, 0 = otherwise
Household income	Continuous variable: the total annual income, including tax, for the household in 2025 (unit: JPY)
Household financial assets	Continuous variable: the total household financial assets
Risk aversion	Continuous variable: respondent’s risk aversion (the answer to the following question: when you usually go out with an umbrella, what is the probability of rain?)
Hyperbolic discounting	Binary variable: 1 = hyperbolic discounter, 0 = otherwise
Impatience	Continuous variable: respondents’ average standardized values of the time discount rates

provides definitions of all variables used for analysis.

3.3. Descriptive Statistics

Descriptive statistics are reported in

Table 2. Descriptive statistics.

Variable	Mean	Std. Dev.	Min	Max
Dependent Variable
Presence of WTU for the administrative procedure proxy service	0.283	0.451	0	1
Presence of WTU for the consultation for troubles and emergencies	0.392	0.488	0	1
Presence of WTU for the education and literacy support	0.355	0.478	0	1
Presence of WTU for the information provision and planning support	0.726	0.446	0	1
Presence of WTU for the advisory and management for investment	0.685	0.465	0	1
Intensity of WTU for the administrative procedure proxy service	0.593	1.038	0	3
Intensity of WTU for the consultation for troubles and emergencies	0.868	1.200	0	3
Intensity of WTU for the education and literacy support	0.714	1.075	0	3
Intensity of WTU for the information provision and planning support	1.555	1.174	0	3
Intensity of WTU for the advisory and management for investment	1.506	1.214	0	3
Independent Variable
Male	0.734	0.442	0	1
Age	51.681	11.230	19	83
Age squared	2797.073	1159.987	361	6889
Spouse	0.655	0.475	0	1
Number of children	1.138	1.122	0	12
Education years	15.125	2.153	9	21
Full-time employed	0.604	0.489	0	1
Household income	7,529,171.286	4,443,009.245	1,000,000	20,000,000
Household assets	29,224,812.916	30,139,830.408	2,500,000	100,000,000
Risk aversion	0.555	0.234	0	1
Hyperbolic discounting	0.112	0.316	0	1
Impatience	0	0.960	−0.805	1.985
Observations	14,432

. The proportions of respondents who selected each AI-based financial service category among their top-three preferred services were as follows: 28.3% for administrative procedure proxy services, 39.2% for consultation for troubles and emergencies, 35.5% for education and literacy support, 72.6% for information provision and planning support, and 68.5% for advisory and management for investment. Information provision and planning support and advisory and management for investment were the most frequently selected categories. However, these categories each contain three service options, whereas consultation for troubles and emergencies contains only one option. Therefore, these descriptive differences should be interpreted with caution because unequal category sizes may mechanically affect selection probabilities. The mean priority-ranking scores, measured on a scale from 0 to 3, were 0.593 for administrative procedure proxy services, 0.868 for consultation for troubles and emergencies, 0.714 for education and literacy support, 1.555 for information provision and planning support, and 1.506 for advisory and management for investment.

In terms of demographic characteristics, 73.4% of respondents were male, and the average age was 52 years. Additionally, 65.5% were married, with an average of more than one child, and the mean number of education years was 15. Regarding socioeconomic status, 60.4% of respondents were employed full-time, with average household income and financial assets of approximately 7.5 million and 29.2 million JPY, respectively. As for psychological traits, the mean level of risk aversion was 0.6. Moreover, 11.2% were classified as hyperbolic discounters, and the average level of impatience was close to zero.

3.4. Methods

This study employed binary probit and ordered probit regressions to examine which investor characteristics are associated with relative priorities for different categories of AI-based financial services. The binary probit models were used for the category-level selection indicators, which equal 1 if the respondent selected at least one service in the category among the three rankings and 0 otherwise. The ordered probit models were used for the ordinal priority-ranking indicators, which take values from 0 to 3 according to the highest-ranked service selected within each category. The estimation specifications are as follows:

$$\begin{array}{c}{Y}_i= {\beta }_{0 }+ {\beta }_{1 }{Male}_i+ {\beta }_{2 }{Age centered}_i+{\beta }_{3 }{Age centered squared}_i+{\beta }_{4 }{Spouse}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{5 }{Number of children}_i+{\beta }_{6 }{Education years}_i +{\epsilon }_i\hfill \end{array}$$

(Model 1)

$$\begin{array}{c}{Y}_i={\beta }_{0 }+{\beta }_{1 }{Male}_i+{\beta }_{2 }{Age centered}_i+{\beta }_{3 }{Age centered squared}_i+{\beta }_{4 }{Spouse}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{5 }{Number of children}_i+{\beta }_{6 }{Education years}_i+{\beta }_{7 }{Full-time employed}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{8 }{Log of household income}_i+{\beta }_{9 }{Log of household financial assets}_i+{\epsilon }_{i }\hfill \end{array}$$

(Model 2)

$$\begin{array}{c}{Y}_i={\beta }_{0 }+{\beta }_{1 }{Male}_i+{\beta }_{2 }{Age centered}_i+{\beta }_{3 }{Age centered squared}_i+{\beta }_{4 }{Spouse}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{5 }{Number of children}_i+{\beta }_{6 }{Education years}_i+{\beta }_{7 }{Full-time employed}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{8 }{Log of household income}_i+{\beta }_{9 }{Log of household financial assets}_i\hfill \\ \hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em} +{\beta }_{10 }{Risk aversion}_i+{\beta }_{11 }{Hyperbolic discounting}_i+{\beta }_{12 }{Impatience}_i+{\epsilon }_i\hfill \end{array}$$

(Model 3)

where $Y_i$ denotes either the binary selection indicator or the ordinal priority-ranking indicator for one of the five AI-based financial service categories for respondent $i$. The demographic variables include gender, age, age squared, marital status, number of children, and education years. The socioeconomic variables include full-time employment status, household income, and household financial assets. The psychological variables include risk aversion, hyperbolic discounting, and impatience.

Model 1 includes demographic variables only. Model 2 adds socioeconomic characteristics. Model 3 further incorporates psychological traits. We mainly focus on Model 3, while Models 1 and 2 are reported in Appendix B to assess whether the estimated associations are broadly consistent across alternative specifications.

Age was centered around its sample mean before constructing the squared age term. This reduces the mechanical collinearity between age and age squared and makes the age coefficients easier to interpret. Household income and household financial assets were log-transformed to reduce skewness and approximate normality. We also checked for multicollinearity by calculating pairwise correlations and variance inflation factors (VIFs) for all independent variables.

Because probit and ordered probit coefficients are not directly interpretable in probability units, we also report average marginal effects. For the binary probit models, the marginal effects indicate the average change in the probability of selecting each service category. For the ordered probit models, we report marginal effects for the probability that a category is ranked first. These marginal effects are used to assess the magnitude of the estimated associations, not only their statistical significance.

We also conduct three robustness checks. First, we estimate option-level models using the 11 original service options to examine whether the category-level findings are driven by specific options. Second, we construct category-size-adjusted outcomes, defined as the number of selected options in each category divided by the number of options in that category. Third, we examine a rank-ordered logit specification as an additional robustness check for the forced-ranking structure of the survey question. These analyses are used to assess the sensitivity of the main interpretation rather than to replace the category-level probit and ordered probit models.

Because the data are cross-sectional, the results should be interpreted as associations between investor characteristics and relative priorities for AI-based financial service categories, not as causal effects. In addition, because respondents were required to select their top three preferred services, the dependent variables do not identify whether respondents would use AI-based financial services in absolute terms. Rather, they indicate which types of AI-based financial services respondents prioritized relative to other AI service options.

4. Estimation Results

Table 3. Regression results on relative priority for each AI-based financial service category using Model 3.

Variable	(i) Administrative Procedure Proxy Service		(ii) Consultation for Troubles and Emergencies		(iii) Education and Literacy Support		(iv) Information Provision and Planning Support		(v) Advisory and Management for Investment
	Selection	Priority	Selection	Priority	Selection	Priority	Selection	Priority	Selection	Priority
Male	0.034	0.054 **	−0.218 ***	−0.226 ***	−0.160 ***	−0.154 ***	0.081 ***	0.090 ***	0.030	0.070 ***
	(0.027)	(0.025)	(0.026)	(0.024)	(0.026)	(0.024)	(0.027)	(0.022)	(0.026)	(0.022)
Age centered	0.000	−0.001	0.017 ***	0.016 ***	0.002 **	0.001	−0.007 ***	−0.006 ***	−0.003 **	−0.004 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Age centered squared	0.000	0.000	0.000 **	0.000 *	0.000 **	0.000 **	−0.000 **	−0.000	−0.000	−0.000 *
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Spouse	−0.054 *	−0.041	−0.049 *	−0.042	−0.002	−0.012	0.050 *	0.017	0.066 **	0.066 ***
	(0.030)	(0.028)	(0.029)	(0.027)	(0.029)	(0.027)	(0.030)	(0.024)	(0.029)	(0.024)
Number of children	−0.001	−0.002	−0.007	−0.008	0.000	0.004	0.007	−0.003	0.001	0.004
	(0.012)	(0.012)	(0.012)	(0.011)	(0.012)	(0.011)	(0.012)	(0.010)	(0.012)	(0.010)
Education years	−0.011 **	−0.012 **	−0.012 **	−0.014 ***	−0.006	0.001	0.023 ***	0.017 ***	−0.002	−0.001
	(0.005)	(0.005)	(0.005)	(0.005)	(0.005)	(0.005)	(0.006)	(0.005)	(0.005)	(0.005)
Full-time employed	0.021	0.037	−0.015	−0.007	−0.007	−0.020	−0.013	−0.008	−0.023	−0.024
	(0.029)	(0.028)	(0.028)	(0.026)	(0.028)	(0.026)	(0.029)	(0.024)	(0.029)	(0.024)
Log of household income	−0.019	−0.033	−0.108 ***	−0.102 ***	−0.063 ***	−0.046 **	0.021	0.044 **	0.100 ***	0.080 ***
	(0.022)	(0.021)	(0.021)	(0.019)	(0.021)	(0.020)	(0.022)	(0.018)	(0.022)	(0.018)
Log of household financial assets	0.004	0.004	−0.009	−0.008	−0.093 ***	−0.093 ***	0.074 ***	0.070 ***	−0.022 **	−0.014
	(0.011)	(0.011)	(0.011)	(0.010)	(0.011)	(0.010)	(0.011)	(0.009)	(0.011)	(0.009)
Risk aversion	−0.044	−0.005	0.092 **	0.069	0.081 *	0.058	0.019	0.029	−0.146 ***	−0.125 ***
	(0.048)	(0.047)	(0.046)	(0.044)	(0.047)	(0.044)	(0.049)	(0.039)	(0.047)	(0.040)
Hyperbolic discounting	−0.043	−0.023	−0.040	−0.036	0.015	0.009	0.017	−0.047 *	0.049	0.079 ***
	(0.035)	(0.034)	(0.034)	(0.032)	(0.034)	(0.032)	(0.036)	(0.028)	(0.035)	(0.029)
Impatience	0.002	0.001	−0.007	−0.010	−0.022 **	−0.018 *	−0.004	0.001	0.034 ***	0.024 **
	(0.012)	(0.011)	(0.011)	(0.010)	(0.011)	(0.011)	(0.012)	(0.010)	(0.012)	(0.009)
Constant	−0.165		1.886 ***		2.281 ***		−1.383 ***		−0.667 **
	(0.311)		(0.297)		(0.300)		(0.314)		(0.303)
/cut1		−0.011		−1.818 ***		−1.961 ***		1.572 ***		0.540 **
		(0.299)		(0.275)		(0.284)		(0.255)		(0.254)
/cut2		0.258		−1.532 ***		−1.631 ***		2.088 ***		0.961 ***
		(0.300)		(0.275)		(0.284)		(0.255)		(0.254)
/cut3		0.649 **		−1.183 ***		−1.132 ***		2.730 ***		1.562 ***
		(0.300)		(0.275)		(0.284)		(0.255)		(0.254)
Observations	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432
Pseudo-R-squared	0.00110	0.000900	0.0250	0.0163	0.0111	0.00703	0.00919	0.00512	0.00489	0.00291
Log likelihood	−8594	−13,059	−9423	−15,386	−9281	−14,899	−8390	−19,728	−8950	−19,496

Note: Robust standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

reports the estimation results based on Model 3, while the results for Models 1 and 2 are presented in Appendix B. The columns labeled “Selection” refer to the binary probit specifications, whereas the columns labeled “Priority” refer to the ordered probit specifications.

Table 4. Average marginal effects for relative priority for each AI-based financial service category using Model 3.

Variable	(i) Administrative Procedure Proxy Service		(ii) Consultation for Troubles and Emergencies		(iii) Education and Literacy Support		(iv) Information Provision and Planning Support		(v) Advisory and Management for Investment
	Selection		Selection	Priority	Selection	Priority	Selection	Priority	Selection	Priority
Male	0.012	0.010 **	−0.082 ***	−0.059 ***	−0.059 ***	−0.030 ***	0.027 ***	0.031 ***	0.011	0.024 ***
	(0.009)	(0.005)	(0.009)	(0.006)	(0.009)	(0.005)	(0.009)	(0.007)	(0.009)	(0.008)
Age centered	0.000	−0.000	0.006 ***	0.004 ***	0.001 **	0.000	−0.002 ***	−0.002 ***	−0.001 **	−0.001 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Age centered squared	0.000	0.000	0.000 **	0.000 *	0.000 **	0.000 **	−0.000 **	−0.000	−0.000	−0.000 *
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Spouse	−0.018 *	−0.008	−0.018 *	−0.011	−0.001	−0.002	0.016 *	0.006	0.023 **	0.023 ***
	(0.010)	(0.005)	(0.011)	(0.007)	(0.011)	(0.005)	(0.010)	(0.008)	(0.010)	(0.008)
Number of children	−0.000	−0.000	−0.003	−0.002	0.000	0.001	0.002	−0.001	0.000	0.001
	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.003)
Education years	−0.004 **	−0.002 **	−0.005 **	−0.004 ***	−0.002	0.000	0.008 ***	0.006 ***	−0.001	−0.000
	(0.002)	(0.001)	(0.002)	(0.001)	(0.002)	(0.001)	(0.002)	(0.002)	(0.002)	(0.002)
Full-time employed	0.007	0.007	−0.006	−0.002	−0.003	−0.004	−0.004	−0.003	−0.008	−0.008
	(0.010)	(0.005)	(0.010)	(0.007)	(0.010)	(0.005)	(0.010)	(0.008)	(0.010)	(0.008)
Log of household income	−0.006	−0.006	−0.040 ***	−0.027 ***	−0.023 ***	−0.009 **	0.007	0.015 **	0.035 ***	0.028 ***
	(0.007)	(0.004)	(0.008)	(0.005)	(0.008)	(0.004)	(0.007)	(0.006)	(0.008)	(0.006)
Log of household financial assets	0.001	0.001	−0.004	−0.002	−0.034 ***	−0.018 ***	0.024 ***	0.024 ***	−0.008 **	−0.005
	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.003)
Risk aversion	−0.015	−0.001	0.035 **	0.018	0.030 *	0.011	0.006	0.010	−0.052 ***	−0.043 ***
	(0.016)	(0.009)	(0.017)	(0.011)	(0.017)	(0.008)	(0.016)	(0.013)	(0.017)	(0.014)
Hyperbolic discounting	−0.015	−0.004	−0.015	−0.009	0.005	0.002	0.006	−0.016 *	0.017	0.027 ***
	(0.012)	(0.006)	(0.013)	(0.008)	(0.013)	(0.006)	(0.012)	(0.010)	(0.012)	(0.010)
Impatience	0.001	0.000	−0.003	−0.003	−0.008 **	−0.004 *	−0.001	0.000	0.012 ***	0.008 **
	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.002)	(0.004)	(0.003)	(0.004)	(0.003)
Observations	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432	14,432

Note: Robust standard errors are in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. For the binary probit models, marginal effects indicate changes in the probability of selecting each category. For the ordered probit models, marginal effects indicate changes in the probability that each category is ranked first.

reports the corresponding average marginal effects for Model 3. For the binary probit models, the marginal effects indicate changes in the probability of selecting a category among the top-three preferred services. For the ordered probit models, the marginal effects indicate changes in the probability that a category is ranked first.

The results should be interpreted as associations with relative priorities rather than determinants of absolute willingness to use AI-based services. The pseudo-R² values are low across specifications, indicating that individual characteristics explain only a limited portion of the variation in AI service priorities. Therefore, the discussion below focuses on broad patterns and effect magnitudes rather than treating statistical significance as evidence of strong predictive power.

Overall, demographic, socioeconomic, and psychological characteristics are associated with relative priorities for different AI-based financial service categories. As shown in Appendix B, the coefficient signs in Models 1 and 2 are broadly consistent with those in Model 3. The statistical significance of some variables disappears after additional controls are introduced, suggesting that the associations observed in the simpler specifications may partly reflect socioeconomic and psychological characteristics included in the fuller model. The detailed results are discussed below by category of independent variables.

4.1. Demographic Characteristics

The results indicate that gender is associated with different relative priorities across AI service categories. Male respondents are more likely to prioritize information provision and planning support, while female respondents are more likely to prioritize consultation for troubles and emergencies and education and literacy support. The average marginal effects suggest that being male is associated with a 2.7-percentage-point higher probability of selecting information provision and planning support and a 3.1-percentage-point higher probability of ranking it first. In contrast, being male is associated with an 8.2-percentage-point lower probability of selecting consultation for troubles and emergencies and a 5.9-percentage-point lower probability of selecting education and literacy support. Male respondents are also more likely to rank advisory and management services as their first priority, although the binary selection coefficient for this category is not statistically significant.

Age is associated with several service categories, but the estimated marginal effects are small. Older respondents are more likely to prioritize consultation for troubles and emergencies and less likely to prioritize information provision and planning support or advisory and management services. Because the age coefficients are estimated using centered age and centered age squared, they should be interpreted as local associations around the sample mean rather than simple linear age effects.

Education years are positively associated with information provision and planning support. The average marginal effects indicate that one additional year of education is associated with a 0.8-percentage-point higher probability of selecting this category and a 0.6-percentage-point higher probability of ranking it first. Education years are negatively associated with administrative procedure proxy services and consultation for troubles and emergencies. These results suggest that more-educated investors are relatively more likely to prioritize information-oriented services, while less-educated investors may place relatively greater priority on procedural or consultation support.

4.2. Socioeconomic Status

Full-time employment status is not significantly associated with relative priority for any AI-based financial service category in Model 3. Household income, however, shows clearer associations. Higher household income is negatively associated with consultation for troubles and emergencies and education and literacy support, while it is positively associated with advisory and management for investment. The average marginal effects suggest that a one-unit increase in log household income is associated with a 4.0-percentage-point lower probability of selecting consultation for troubles and emergencies and a 2.3-percentage-point lower probability of selecting education and literacy support. In contrast, it is associated with a 3.5-percentage-point higher probability of selecting advisory and management services and a 2.8-percentage-point higher probability of ranking this category first.

Household financial assets are positively associated with information provision and planning support and negatively associated with education and literacy support. The average marginal effects indicate that a one-unit increase in log household financial assets is associated with a 2.4-percentage-point higher probability of selecting information provision and planning support and a 2.4-percentage-point higher probability of ranking it first. Household financial assets are negatively associated with education and literacy support and with the binary selection of advisory and management services, although the latter association is not statistically significant in the priority-ranking specification. These results suggest that income and financial assets are associated with different forms of AI service priority, but the magnitudes are modest.

4.3. Psychological Traits

Psychological traits are associated with some AI service priorities, especially advisory and management for investment. Risk aversion is negatively associated with both the selection and first-priority ranking of advisory and management services. The average marginal effects indicate that a one-unit increase in the risk-aversion proxy is associated with a 5.2-percentage-point lower probability of selecting advisory and management services and a 4.3-percentage-point lower probability of ranking this category first. Risk aversion is also positively associated with the selection of consultation for troubles and emergencies, although this association is weaker in the priority-ranking specification.

Hyperbolic discounting is not significantly associated with most service categories, but it is positively associated with the probability of ranking advisory and management services first. Impatience is also positively associated with advisory and management services and negatively associated with education and literacy support. The average marginal effects suggest that the magnitudes of these associations are relatively small. These findings are consistent with the possibility that present-oriented investors may place greater relative priority on AI-based services involving trading advice or management support. However, because the data are cross-sectional and the psychological variables are measured imperfectly, these results should be interpreted as associations rather than evidence of behavioral mechanisms.

4.4. Robustness Checks

We conducted additional robustness checks to address concerns about the aggregation of the 11 service options into five categories and the unequal number of options across categories. First, we estimated option-level models using the 11 original service options. These results show that some category-level patterns are driven by particular options within each category. For information provision and planning support, the positive association with male respondents is mainly observed for investment information provision and performance reporting, while the association with education years is concentrated in life plan simulation. For advisory and management services, the associations involving risk aversion, hyperbolic discounting, and impatience are mainly observed for trading-timing advisory services. These results suggest that the five service categories are useful for summarizing functional differences but should not be interpreted as internally homogeneous constructs.

Second, we estimated models using category-size-adjusted outcomes, defined as the number of selected options within each category divided by the number of options in that category. The results are broadly consistent with the main category-level findings. In particular, the positive associations of information provision and planning support with male respondents, education years, and household financial assets remain evident, and the negative association between risk aversion and advisory and management services is also observed. These findings suggest that the main category-level patterns do not appear to be explained entirely by unequal category sizes.

Third, we examined a rank-ordered logit specification as an additional robustness check for the forced-ranking structure of the survey question. This analysis is used to assess whether the main interpretation is sensitive to the fact that respondents selected ranked alternatives from a fixed list. Overall, the robustness checks support the main interpretation that investor characteristics are associated with different relative priorities across AI-based financial services. At the same time, the option-level results highlight the need to interpret category-level findings cautiously.

Detailed tables for these robustness checks are not reported to avoid overburdening the manuscript, but the results are summarized here because they are directly relevant to the reviewers’ concerns about category aggregation, unequal category sizes, and the forced-ranking structure.

5. Discussion

This study examined investor heterogeneity in relative priorities for AI-based financial services using a large-scale survey of Japanese online investors. The results show that demographic, socioeconomic, and psychological characteristics are associated with different relative priorities across AI service categories. However, these associations should be interpreted cautiously. The dependent variables capture relative priorities among a fixed set of AI-based service options rather than absolute willingness to use AI services, and the pseudo-R² values indicate that individual characteristics explain only a limited portion of the variation in these priorities. The average marginal effects are generally modest, even when statistically significant.

The findings support the view that AI-based financial services should not be treated as a single homogeneous product category. Investors appear to assign different relative priorities to procedural support, consultation and emergency support, education and literacy support, information and planning support, and advisory or management services. At the same time, the option-level robustness checks show within-category heterogeneity, especially for information/planning and advisory/management services. Therefore, the five categories should be understood as useful functional groupings rather than internally homogeneous constructs.

One important finding is that information provision and planning support is more strongly prioritized by male respondents, more-educated respondents, and those with greater household financial assets. This category includes life-plan simulation, investment information provision, and investment performance reporting. These services are relatively information-intensive and may be more attractive to investors who are better positioned to process and use financial information. The positive associations with education and financial assets suggest that more-educated or resource-rich investors may prioritize AI as a tool for information processing, monitoring, and planning support. However, this result should not be interpreted as evidence that education or assets causally increase demand for such services. Rather, it indicates that these investor characteristics are associated with stronger relative priority for information-oriented AI services.

The results also show a different pattern for consultation for troubles and emergencies and education and literacy support. Female respondents are more likely to prioritize these categories, while higher household income is negatively associated with both categories. Education years are also negatively associated with consultation for troubles and emergencies, and household financial assets are negatively associated with education and literacy support. These findings suggest that AI-based services may play different roles for different investor groups. For some investors, AI may be valued primarily as an advanced information or planning tool; for others, it may be valued as a support mechanism for resolving problems, understanding investment rules, or using financial platforms. From the perspective of financial service design, this distinction is important. If AI-based financial services are developed mainly as advanced information or advisory tools, they may not sufficiently address the service priorities of investors who emphasize consultation, education, or operational support.

The findings for advisory and management for investment are particularly relevant to the risk and investor-protection context. Higher household income is positively associated with priority for advisory and management services, while risk aversion is negatively associated with this category. In addition, impatience is positively associated with advisory and management services, and hyperbolic discounting is positively associated with the priority-ranking specification for this category. These associations suggest that investors who place stronger priority on AI-based advisory and management services may not simply be more cautious or risk-averse investors. Rather, some psychological traits related to present-oriented decision-making are also associated with stronger priority for these services. Because this category includes optimal portfolio proposals, trading–timing advice, and automated trading services, its design may have important implications for investor protection (Han et al. 2023; Ridzuan et al. 2024; Vuković et al. 2025; Kulkarni et al. 2025; Nahidi 2026). Financial institutions should therefore consider transparency, explainability, suitability checks, and appropriate safeguards when providing AI-based advisory or management functions.

These results have practical implications for financial institutions adopting AI-based services. A one-size-fits-all approach may be insufficient because investors appear to prioritize different AI functions depending on their characteristics. For investors who prioritize information-oriented services, AI tools that support portfolio monitoring, performance reporting, market information, and life planning may be useful. For investors who prioritize consultation or education, AI services may need to provide simple explanations, platform-use support, and accessible troubleshooting functions. For investors who prioritize advisory and management services, institutions may need to combine convenience with stronger governance mechanisms, especially when services involve portfolio recommendations, trading advice, or automation. In this sense, responsible AI adoption in financial markets requires not only technical efficiency but also user-sensitive service design.

The findings also contribute to the literature on digital finance and AI adoption by showing the importance of distinguishing between different types of AI-based financial services. Many discussions of AI adoption treat AI use as a single outcome (Akhtar et al. 2024; Banerjee 2025; Kumari et al. 2024; Aftab et al. 2025; Santini et al. 2026; Singh and Kumar 2025). However, the present results suggest that investors’ priorities differ substantially across AI service categories. This distinction is especially important in financial markets, where different AI functions may involve different levels of complexity, autonomy, and investor risk. By analyzing relative priorities across multiple service categories, this study provides evidence that investor heterogeneity should be considered when designing and evaluating AI-based financial services.

Several limitations should be noted. First, the dependent variables capture relative priorities among AI-based financial service options rather than absolute willingness to use AI services. Because respondents were required to select their first-, second-, and third-preferred services, the results do not identify investors who would reject AI-based financial services altogether. Second, the forced top-three design creates mechanical dependence among choices. Although we address this issue through robustness checks, the main probit and ordered probit models should be interpreted as category-specific associations rather than independent demand models. Third, the 11 service options were grouped into five functional categories, but the option-level analysis shows within-category heterogeneity. Therefore, the category-level results should not be interpreted as evidence that all services within a category are homogeneous.

Fourth, the data are cross-sectional, and the findings should be interpreted as associations rather than causal effects. Although the models control for a broad set of demographic, socioeconomic, and psychological characteristics, unobserved factors—such as prior experience with AI tools, trust in financial institutions, investment knowledge, or perceived service quality—may also be related to investors’ relative priorities. Fifth, some variables, including education years and time-preference measures, were supplemented from earlier survey waves when unavailable in 2026. This approach improves sample coverage but may introduce temporal measurement inconsistency. In addition, listwise deletion may create sample-selection concerns if excluded respondents differ systematically from the analytical sample. Sixth, risk aversion is measured using an umbrella/rain-probability question. This variable should be interpreted as an indirect proxy for general risk attitudes rather than a direct measure of financial risk tolerance, and measurement error may affect the estimated associations. Finally, the sample consists of active online investors using Rakuten Securities, so the findings may not be directly generalizable to non-investors, offline investors, or the general population.

Overall, the results suggest that AI-based financial services may create value in different ways for different investor groups. Information and planning tools may be more strongly prioritized by investors with higher education and greater financial assets, while consultation and education-related services may be relatively more important for other groups. Advisory and management services may attract investors with distinct socioeconomic and psychological characteristics, raising important issues for service design and investor protection. These findings underscore the need for financial institutions to design AI-based financial services that reflect investor heterogeneity rather than assuming uniform demand across all users.

6. Conclusions

This study examined investor heterogeneity in relative priorities for AI-based financial services using a large-scale survey of Japanese online investors. Rather than treating AI adoption as a single outcome, we distinguished among five functional categories of AI-based financial services: administrative procedure proxy services, consultation for troubles and emergencies, education and literacy support, information provision and planning support, and advisory and management for investment. Because respondents were required to select their first-, second-, and third-preferred services, the dependent variables capture relative priorities among AI-based financial service categories rather than absolute willingness to use AI services.

The results show that investor characteristics are associated with different relative priorities across AI service categories. Information provision and planning support is more strongly prioritized by male respondents, more-educated respondents, and those with greater household financial assets. Consultation for troubles and emergencies and education and literacy support show different patterns, suggesting that AI-based financial services may serve not only as advanced information or advisory tools but also as support mechanisms. Advisory and management services are more strongly prioritized by higher-income and more impatient respondents, while risk aversion is negatively associated with this category. Hyperbolic discounting is also positively associated with the probability of ranking advisory and management services first. However, the estimated marginal effects are generally modest, and individual characteristics explain only a limited portion of variation in relative AI service priorities.

Additional robustness checks using option-level outcomes and category-size-adjusted outcomes suggest that the main patterns are not explained entirely by unequal category sizes. At the same time, the option-level analysis reveals within-category heterogeneity, especially for information/planning and advisory/management services. These findings suggest that the five categories are useful for organizing AI-based financial services, but they should not be interpreted as internally homogeneous constructs.

Overall, the findings suggest that financial institutions should consider investor heterogeneity when developing AI tools for information provision, education, consultation, advisory services, and automated management. In particular, AI-based advisory and management services may require careful attention to transparency, explainability, suitability, and investor-protection mechanisms, given their potential influence on investment decisions. Future research could extend this study by examining actual usage behavior, willingness to pay for AI-based financial services, and the long-term effects of AI service adoption on investment behavior and financial well-being.