Digital Twin Robo-Advisor Adoption Dataset V_2
Description
This dataset contains respondent-level data from a randomized online experiment on digital twin robo-advisors and conversational AI in investment services. The study examines how different robo-advisor designs influence investor trust, privacy concern, perceived personalization, adoption intention, willingness to follow advice, and hypothetical allocation behavior. The experiment uses a 2 × 2 between-subjects design. Respondents were randomly assigned to one of four fictional robo-advisor conditions: (1) standard robo-advisor with plain dashboard interface, (2) standard robo-advisor with conversational AI interface, (3) digital twin robo-advisor with plain dashboard interface, and (4) digital twin robo-advisor with conversational AI interface. In all conditions, the recommended portfolio remained constant, allowing the study to isolate the effects of personalization architecture and interface style. The digital twin conditions describe a continuously updated financial profile built, with user permission, from factors such as goals, risk tolerance, income pattern, spending behavior, savings rate, liabilities, household needs, tax context, and projected life events. The conversational AI conditions present the recommendation through a chatbot-like advisory interface. The dataset includes responses to manipulation checks, trust measures, privacy concern measures, perceived usefulness and personalization items, adoption-related items, hypothetical investment allocation responses, and demographic controls. It is intended for research on fintech adoption, robo-advisors, AI-enabled financial services, trust in automated advice, privacy trade-offs, and digital personalization in wealth management. The dataset supports the study: “From Robo-Advisors to Financial Digital Twins: A Randomized Experiment on Conversational AI, Trust, Privacy, and Adoption.”
Files
Steps to reproduce
Create an online survey in Qualtrics or a comparable survey platform. Build a randomized 2 × 2 between-subjects experiment with four fictional robo-advisor conditions: (a) standard robo-advisor with plain dashboard interface, (b) standard robo-advisor with conversational AI interface, (c) digital twin robo-advisor with plain dashboard interface, and (d) digital twin robo-advisor with conversational AI interface. Keep the recommended portfolio constant across all four conditions: 60% global equity ETFs, 35% bond ETFs, and 5% cash equivalents. This ensures that only interface style and personalization architecture vary across treatments. In the digital twin conditions, describe the system as using a continuously updated financial profile, with user permission, based on factors such as risk tolerance, goals, income pattern, spending behavior, savings rate, liabilities, household needs, tax context, and projected life events. In the conversational AI conditions, present the recommendation through a chatbot-style advisory interface. In the plain-interface conditions, present the recommendation as a standard dashboard or platform screen. Recruit adult respondents with at least basic familiarity with investing or personal finance. Randomly assign each respondent to only one of the four conditions, using even allocation across cells where possible. After exposure to the assigned vignette, administer the post-treatment questionnaire measuring perceived personalization, trust, privacy concern, perceived usefulness, adoption intention, willingness to follow advice, and hypothetical allocation behavior. Collect demographic and background variables, including age, gender, education, income, and investment experience. Export the raw survey data to spreadsheet or statistical software format. Clean the data by removing incomplete submissions, straight-lining cases, duplicate responses, failed attention checks if used, and obviously invalid entries. Recode treatment indicators into binary variables for digital twin presence and conversational AI presence, and compute scale scores for the main constructs. Reproduce the main analyses by comparing mean responses across the four groups and estimating regression or ANOVA models for the effects of digital twin design, conversational AI, and their interaction on trust, privacy concern, adoption intention, and related outcomes.
Institutions
- Ca' Foscari University of VeniceVeneto, Venice