Discrimination in lending has long been a problem, shutting minority groups out of the home buying process.
ZestFinance, the artificial intelligence software company focused on the credit market is trying to change that with ZAML Fair, a new software tool that aims to reduce the instances of biases and discrimination in lending.
Similar to a dimmer knob, the artificial intelligence based tool lets lenders tune models to achieve fairness by reducing the impact of discriminatory credit data without affecting performance. ZAML Fair, which is built into ZestFinance’s main ZAML platform ranks credit signals by how much they lead to biased outcomes. It will then automatically create a new model that has more fairness attached to it. Lenders can choose to minimize the impact certain discriminatory factors have on determining if a borrower is creditworthy including income and traditional credit score. “Models are by nature very biased,” Douglas Merrill, founder and Chief Executive of ZestFinance told Forbes. “The ability to make decisions that are biased is an epidemic.”
WUNC, the National Public Radio member, The Center for Investigative Reporting’s Reveal Show and the Associated Press recently teamed up tostudy millions of Home Mortgage Disclosure Act records and found African Americans and Latinos are denied conventional mortgage loans at rates that in some cases are much higher than what their white neighbors are given. The study found that across 61 cities in the U.S. disparities are particularly bad. Individuals applying for mortgages in rural areas were denied more often than those looking to purchase a home in an urban area.
ZestFinance was founded in 2009 by Merrill, the former CIO of Google, and a team of former Google employees with the mission of making fair and transparent credit available to everyone. It began as a lender but pivoted into modeling, applying AI to develop accurate and explainable credit risk models.
“People should be treated fairly but until now there was no way for banks to do the right thing because they couldn’t understand their own models well enough to know what variables if any cause discrimination,” said Merrill. He said banks deal with it by removing any offending variables but that could hurt performance. The tool gives them the ability to remove variables and at the same time gauge how it will impact its portfolio.
The new tool works on traditional linear models and most machine learning model regardless of how complicated it is. The machine learning model is run through ZAML Fair to ascertain if there are any differences across protected classes and if there is, what variables are causing those differences. The lender can increase and decrease the influence of the variables to lessen bias and increase accuracy.
ZestFinance said several unnamed lenders that tested the fairness tool produced models that reduced the disparity between minority and white approval rates. If the tool was applied nationwide, ZestFinance said it could remove 70% of the mortgage approval rate gap between Hispanic and white borrowers, amounting to 172,000 new homeowners a year. It would close the gap between blacks and white borrowers by more than 40%. The results are based on applying the ZAML Fair algorithm to traditional credit models. When used with machine learning models the company expects the reductions in biased to be even greater.
While Merrill didn’t name names, he said the tool is getting a lot of attention from lenders. “ We are seeing a lot of interest and are running at or near production capacity,” said the executive “Everyone wants to do the right thing it’s just hard if you don’t know what to do to do your best.