Representative Engagements

CrossCheck was engaged by a credit union to evaluate whether AI model outputs created elevated fair lending risk within the underwriting process. The review included analysis of AI model score disparity patterns, comparisons between the credit union’s instant approval thresholds and actual credit outcomes, and statistical testing of whether differences in model outputs across protected classes were significant. We also assessed average AI scores by prohibited basis group to determine whether the observed score differences warranted further investigation from a fair lending perspective.

To better understand how the AI testing results aligned with the lender’s existing underwriting framework, we used regression analysis to evaluate the extent to which traditional underwriting factors explained the AI score. In addition, we performed separate disparity testing on applications that qualified for instant approval and instant denial outcomes, to identify whether the credit union’s subsequent decision segmenting introduced or amplified risk. The analysis provided the client with a clearer understanding of how AI model outputs influenced credit decisioning, where disparities were concentrated, and what monitoring steps could strengthen ongoing governance of AI-assisted underwriting.