Leveraging Technology to Enhance Access to Financial Advice: Trust, Uptake, and Advisor Adoption

How to cite

d'Astous, Philippe; Voigt, Maximilian. 2026. "Leveraging Technology to Enhance Access to Financial Advice: Trust, Uptake, and Advisor Adoption." AEA RCT Registry. AEARCTR-0018280. https://doi.org/10.1257/rct.18280-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Philippe d'Astous* , Maximilian Voigt

* Primary investigator

Registry ID
AEARCTR-0018280
DOI
https://doi.org/10.1257/rct.18280-1.0
First registered
Last updated
Status
In development
Trial dates
2026-05-20 – 2026-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/18280

Abstract

We study how access to generative artificial intelligence (AI) tools affects the quality of financial advice and consumer investment decisions using a lab-in-the-field experiment. The experiment has two stages. In Stage 1, Canadian certified financial planners (FPs) are randomly assigned to advise hypothetical clients with or without access to a generative AI tool. Client profiles vary by age, gender, and risk preferences. FPs observe asset price developments. The recommend portfolio allocations across five risky assets with known return distributions and a risk-free asset with zero return at two decision points. In Stage 2, customers observe the same asset price development and make portfolio allocation decisions across the same assets. Customers are randomly assigned to one of four treatment arms: no advice, advice from an FP without access to AI, advice from an FP with access to AI, and advice generated directly by a generative AI.

Intervention

In Stage 1—for financial planners (FPs)—we exogenously vary whether the FP giving advice can access a generative AI tool or not. In Stage 2—for customers—we exogenously vary if they can obtain financial advice as well as the source of financial advice.

Experimental design

We conduct a standard portfolio choice experiment as in Camerer and Weber (1998) or Fischbacher, Hoffmann, and Schudy (2017). Financial planners (FPs) as well as customers see the same development of asset prices. While FPs recommend a portfolio allocation for hypothetical customers at two decision points, the customers choose their portfolio allocation over ten periods.

Design details

Randomization method
Computer-generated randomization. Asset-price paths are pre-generated using a fixed random seed. The assignment of FPs to client profiles, paths, and AI access is pre-generated in balanced blocks using computer randomization. Each asset-price path is randomly assigned to a subset of the 18 customer-profile cells, creating a sparse but connected bipartite assignment graph. This ensures joint identification of cell and path fixed effects in an additive two-way framework. Customers assignment to price paths and treatment arms is randomized within profile cells at the time of participation.
Randomization unit
Individual. In Stage 1, the individual financial planner is the unit of randomization for AI access. In Stage 2, the individual consumer is the unit of randomization for treatment arm and path assignment.
Clusters
Not applicable (individual-level randomization). Approximately 1,000 financial planners and 2,000 consumers.
Observations
We target approximately 1,000 financial planners (Stage 1) and 2,000 consumers (Stage 2), for a total of 3,000 participants.
Arms
Stage 1: approximately 500 FPs with AI access, 500 FPs without AI access. Stage 2: approximately 500 consumers per arm (control, FP advice, FP+AI advice, ChatGPT advice).
Primary outcomes
Our key outcome variable is a vector of wealth shares allocated or recommended to be allocated to each of the five risky assets and a risk-free asset at each decision point.
Secondary outcomes
Our secondary outcome variables are (a) beliefs about the expected return of each asset, (b) the decision to access financial advice for customers or to access the generative AI for FPs, and (c) the textual description of the financial advice given. In addition, we elicit a non-incentivized trust measure using the amount sent during a standard trust game.

Cited by

No citation data available for this registration yet.

Keywords