Hyper-Personalized Financial Co-Pilots

Hyper-Personalized Financial Co-Pilots

While digital financial advice has made portfolio management broadly accessible, individualization often remains limited to coarse risk categories, and conversational tools struggle to maintain context across interactions. This project investigates hyper-personalized financial co-pilots: conversational AI agents that retain an evolving memory of a client's goals, constraints, and preferences. In an initial paper presented at DESRIST, we derived five design features for building trustworthy co-pilots: robust memory, deep inquiry, education over sales, institutional transparency, and graceful failure. Our working prototype puts these principles into practice using a locally hosted LLM and live portfolio backtesting. First insights from a randomized experiment with 100 participants show that deeper personalization strengthens trust, though conversational interfaces only add value when users actively engage and system boundaries are transparent. Next, we are conducting practitioner interviews and preparing a larger controlled experiment to provide financial institutions with a practical blueprint for conversational advice.