When considering potential AI bias in consumer lending, it is helpful to think about the role of screening technology. Automated decision-making tools (ADMTs) play a role in screening potential customers: Lenders try to use screening technologies to extract signals of how well a loan will perform based on information about an applicant, but this opens the door to bias in selection.
In the article “Thinking About AI Bias Like an Economist” published in Consumer Financial Law Quarterly Report, Managing Director Ling Ling Ang and Consultant Daniel Brown describe the economic context of screening in lending and how that relates to potential discrimination, present considerations regarding the technology and implementation of ADMTs, explain certain fairness metrics and related caveats, and discuss related legal and policy considerations. Throughout their article, Dr. Ang and Dr. Brown highlight tradeoffs between precision of prediction and fairness considerations, as well as the potential for ADMTs to embed biases from past decisions previously made by human decision-makers. Machine learning, at its core, is statistical, and thinking about AI bias builds on decades of economic research on discrimination.
Published by the Conference on Consumer Finance Law (CCFL) in the Consumer Finance Law Quarterly Report: https://www.ccflonline.org/report/
Key Takeaways
- Consumer lending clients should view AI bias in the context of screening technology: Model design affects profitability and fair lending risk.
- ADMTs can improve speed, consistency, and predictive accuracy, but they can also replicate or scale bias embedded in historical data or poorly aligned targets.
- Removing protected characteristics from a model may not eliminate discrimination risk because other variables or combinations of variables can act as proxies for protected class status.
- There is no single fairness metric that resolves AI bias concerns, and commonly used criteria such as independence and sufficiency often cannot be satisfied at the same time.
- Because AI regulation, enforcement, and litigation are evolving, institutions using ADMTs should proactively document models, test for fair lending risk, and maintain audit-ready governance.
How NERA Can Help
NERA experts help clients evaluate the economic, statistical, and regulatory implications of ADMTs and AI-enabled screening technologies. In consumer lending and other consumer-facing markets, our experts can assess model alignment with legitimate business objectives, evaluate potential disparate impact or proxy risks, and analyze the tradeoffs between predictive accuracy, profitability, and fairness. Drawing on experience in consumer protection, financial services, litigation, regulatory investigations, and large dataset analysis, NERA can help clients understand how ADMTs affect consumers, business outcomes, and compliance risk.
NERA also supports clients in developing defensible, audit-ready approaches to AI governance and model evaluation. Our experts can assist with:
- Fair lending testing
- Model documentation
- Damages estimation
- Consumer harm analysis
- Survey research
- Economic justifications for business practices or regulatory positions
As legal and policy frameworks for AI continue to evolve, NERA combines rigorous economic analysis, practical experience with regulators and technology markets, and expert testimony experience to help clients navigate enforcement, litigation, and strategic compliance challenges related to algorithmic bias and consumer protection.
