AI credit scoring raises hopes for access and concerns over bias
AI-powered lending firms are using broader financial information to assess borrowers who may be poorly served by traditional credit scores. Young people and recent immigrants can struggle to obtain credit because they lack established borrowing histories, while fraud can damage otherwise strong records.
Upstart and Zest AI offer lenders alternative assessment tools. Upstart says its models examine more than 2,500 variables and could enable more affordable lending across demographic groups. Mortgage specialist Sarah Tucker sees potential for more individual assessments of affordability and risk, provided consumers understand how decisions are made and can challenge errors.
Oxford professor Sandra Wachter warns that models trained on past decisions could reproduce existing social inequalities unless firms apply regulations and establish safeguards. Upstart says it incorporates automated bias evaluations into model training. Noggin HQ takes a different approach, analysing current-account transactions without AI because of concerns about embedded bias and models potentially inferring protected characteristics.
Established credit agencies are also adopting the technology. Experian has partnered with OpenAI to let users check credit scores inside ChatGPT, Equifax says its Amplify AI platform supports credit assessment and fraud detection, and TransUnion is integrating Google’s Gemini models into its analytics. Consumer comfort with such uses of financial data remains uncertain.