An AI system, like any lending technology or process, must avoid unlawful disparate impact or statistical bias that would be harmful to disadvantaged groups. Fair systems also provide clear explanations to declined consumers so that they can improve their future creditworthiness.
Over the last seven years, Upstart has developed and refined a state-of-the-art fairness testing framework that evaluates every application, loan, and model for bias. This framework encompasses disparate treatment and impact evaluation, including qualitative and quantitative assessments. As part of our service, we provide both lender-specific and platform-wide fairness test results to each lender on our platform each quarter.