Zest AI Releases New Race Prediction Model to Reduce Systemic Bias in Lending


LOS ANGELES, March 1, 2022 /PRNewswire/ — Zest AI, a leader in AI-powered lending software, today announced the launch of Zest Race Predictor (ZRP). This open source machine learning algorithm estimates an individual’s race/ethnicity using only their full name and home address as inputs.

The ZRP can be used to analyze racial equity and outcomes in critical areas such as healthcare, financial services, criminal justice, or anywhere there is a need to attribute race or ethnicity to a population dataset when race/ethnicity data is missing. The financial services industry, for example, has struggled for years to achieve fairer outcomes amid accusations of discrimination in lending practices. A better criterion can help reverse this legacy of bias.

ZRP improves on the most widely used racial and ethnic proxy method, Bayesian Improved Surname Geocoding (BISG), developed by RAND Corporation in 2009. In several tests against BISG, ZRP was able to correctly identify African Americans 25% more often, identify 35% fewer African Americans compared to non-African Americans and 60% fewer whites compared to non-whites.

“Zest AI started developing ZRP in 2020 to improve the accuracy of our customers’ fair loan analytics using more data and better math,” says Mike deVere, CEO of Zest AI. “We believe ZRP can significantly improve our understanding of the disparate impact and disparate treatment of status-protected borrowers.”

“I have used the ZRP output myself and found that it provided results consistent with our predictions, in the context of predicting the run of PPP borrower business owners,” says Sabrina Howel, assistant professor of finance at NYU Stern. “Getting good running estimates is critical to facilitating fair lending practices in America, and by making their tool open source and available for free, Zest’s app is an important step toward that goal.”

“We know from our 2014 study that BISG leaves a lot of room for improvement,” says Dr. Marsha J. Courchane, Vice President and Head of Financial Economics Practice, Charles River Associates. “We are delighted that Zest has taken the initiative to apply modern data science methods to develop a better stroke estimator, and we look forward to further validating this work.”

A more accurate run prediction will help the entire lending ecosystem:

  • Lenders will be better able to identify unfair outcomes to improve models.
  • Regulators will have a better tool to enforce fair lending rules that promote equity in access to credit products that could help people of color achieve better credit scores.
  • Borrowers will benefit knowing that their race and ethnicity are more accurately reflected alongside their credit history.

To learn more about the Zest Race Predictor, please visit www.zest.ai/zest-race-predictor or download the code from the ZRP Github page here: https://github.com/zestai/zrp.

About Zest AI

Zest AI software helps lenders make better decisions and better loans, which increases revenue, reduces risk and automates compliance. Since 2009, it has made fair and transparent credit available to everyone and is today the leader in more inclusive underwriting software. The company is headquartered in Los Angeles, California. Learn more about www.zest.ai and join us on Twitter at @Zest_AI or Zest AI Knowledge Blog.

CONTACT: [email protected]



About Author

Comments are closed.