174 / 2024-04-22 16:19:13

How Can Human Save a Robot? The Role of Trust in Robo-advising

AI,trust model,hybrid robot,robo-advisor

Automated investment managers, usually referred to as robo-advisors (RA), have gained widespread popularity in recent years. Because of their low expenses and wide accessibility to vast retail investors, the market of RA has soared and is expected to reshape the market of asset management. However, for general retail investors, survey evidences show that a considerable portion of them still do not trust algorithms (Rossi and Utkus, 2020). To this issue, some existing research try to find solutions to improve users’ adoption by understanding the factors that drive the use of RAs (Belanche et al., 2019; Hodge et al., 2021; Dietvorst et al., 2016; Ge et al., 2021). For those RAs in use, evidences indicate that they do not systematically outperform human advisors (Muller, 2023). To this issue, a set of literatures try to understand the technological and operational factors that explain unsatisfactory performance of RAs (Jung et al., 2019; Liu et al., 2023). Although trust is crucial in financial service delivery process, relatively little attention has been given to the mechanism of investors’ (dis)trust to RA, and how investors’ (dis)trust could affect the service delivery process.

In this paper, we explore the mechanisms of investors’ (dis)trust to RAs, its potential consequences on service delivery, and to what extent a hybrid RA (i.e., a RA with human assistance) could serve to improve investors’ trust relative to a RA. Based on the model of trust proposed by Mayer et al. (1995), we crafted two controlled experiments to compare investors’ trust among three different roles: human advisor, RA, and hybrid RA. In Study 1, 115 subjects with investment experience are recruited from an online platform. They are asked to assume the role of an individual investor and that they are going through the online “account-opening” process with the help of an advisor. They will interact with one randomly assigned advisor in online chats, and the advisor will guide them answer questions about their investment goal, financial status, and personal risk preferences. The conversation section ends with investors receiving a risk score which, to minimize potential confounds, is kept constant for all participants. Then the participants come to the second part of Study 1, where they need to indicate their trust to the advisor and how genuine their responses are when they respond to advisor’s financial status questions. Study 2 is a 2´3 between subject design with two independent variables; one is the performance outcome of the initial consultation, which has 2 levels (profit or loss by 8%), and the other is the advisor roles. A new set of 227 participants are recruited and randomly assigned to one of the six conditions. The participants assume that they need to make investment decisions given the result from an initial consultation with their advisor (profit or loss by 8%). In the online chat part, the investors receive the advisor’s recommendations on the stocks that they are interested in. Then in the following sessions, the investors indicate their trust to the advisor, the likelihood of following the advisor’s advice, and their personal risk preferences.

We have two major findings. First, investors bear significantly lower level of trust on a RA relative to a human advisor. Such lower level of trust is mostly due to the lower ratings on the ability and benevolence of a RA. Moreover, investors’ distrust has consequences. In the account opening stage, RAs are less likely to elicit genuine answers from investors about their risk preferences and financial status. In the following-up stage, investors have lower level of trust on RA and thus are less likely to follow subsequent investment suggestions made by it comparing to a human advisor when the initial consultation makes a profit, while no difference when initial consultation incurs a loss. Second, a hybrid RA can help alleviate the distrust of RAs to some extent. In general, investors’ trust on a hybrid RA is significantly higher than RA and is comparable to human advisor. Such trust can help improve the cooperation between investors and advisors. In the accounting opening stage, a hybrid RA can elicit genuine answers from investors at a level that comparable to a human advisor. In the following-up stage, investors’ trust to a hybrid RA is comparable to human advisor no matter its performance is good or bad, but investors seem to be more likely to follow investment suggestions from a human advisor when the initial suggestions turn out to be good. By further grouping the investors by their risk propensity, we find that investors with higher risk propensity score are willing to trust hybrid robot advisors at a level similar to human advisors regardless of the performance.

These findings advance our understanding of the trust between investors and RAs. First, we provide additional evidences on investors’ distrust to RAs; such distrust is mainly due to the lower level of perceived ability and benevolence. Second, our findings suggest that investors’ distrust on RA can lead to non-negligible consequences, including non-genuine responses and reluctant to follow their investment suggestions. These consequences could contribute to the inferior performance of RAs. Third, our findings suggest that hybrid RAs could potentially serve as a solution to investors’ distrust to RA. They could bring more benefit when used in accounting opening process for all investors, and when used in investment consultation process for investors with higher risk propensity. In our experiment, the hybrid RA is essentially a “conceptual” one; although it claims with access to human assistance, there is actually no physical assistance provided. All these findings may contribute to the practice in financial service industry.