We spent four months building a robo advisor users barely touched. Here’s what the failure taught us about assumptions, UX research, and validation.