Innovation in Lending
Using Machine Learning and Alternative Data to Improve Credit Outcomes
August 10, 2020
Some 50 million American consumers are so-called thin-file credits — they can’t get traditional access to loans because they have little or no financial history at the major credit bureaus. That doesn’t mean they aren’t good borrowers. It just means that their credit-enhancing behavior — paying rent, utility bills or installment loans — isn’t counting in their favor. Thin-files represent a big growth opportunity for U.S. lenders. Across the globe, the revenue opportunity for lenders reaches $380 billion if we can find a way to bring the unbanked into the mainstream.
This opportunity is not lost on the credit risk and scoring industry, which has been scouring data sources such as public records and proprietary data to find new signals that can improve the art and science of credit assessment. Some vendors claim that adding new data sets of theirs can improve creditworthiness scores for as much as 90% of the U.S. thin-file population.
Sales of so-called alternative data products are booming, as banks and credit unions seek to widen their customer base and combine all this new data with powerful machine learning models that dramatically improve underwriting outcomes. Zest AI customers typically see 15% to 20% increases in approval rates with no added risk or 30% to 50% declines in default rates.
Zest AI is working with its credit data partners to make it easier to use alternative data products in Zest machine learning credit models. With Zest’s ML software, lenders can quickly assess which alternative variables have the most impact on model performance, and plug those into a new ML model.
Knowing how many email addresses a consumer has had or whether they possess a professional license — as a barber or a plumber, for example — can play an important role in whether an application is approved. “If used appropriately,” says Nidhi Panday, senior director of product at Zest AI, "such data actually helps improve outcomes for consumers, when they otherwise would not have been given an opportunity to get credit.”
Let’s say someone owns a business or sits on a board. That can be a positive indicator. On the flip side, there is a broad array of alternative information that can raise questions about an application, such as tax lien judgments and evictions. Primary beneficiaries of more robust data sets are African American and Latinx borrowers who traditionally have lower credit scores from traditional bureaus. Some credit data providers have found that use of alternative data can increase approval rates for people of color by 30%.
To be sure, alternative data sets and the vendors who provide them must ensure the data’s accuracy and ability to withstand the rigors of disputes and disclosures. Accuracy and transparency are particularly important because the Fair Credit Reporting Act requires data providers to be able to show that their data is valid. Traditional credit bureaus have faced criticism as well as complaints from regulators and several state attorneys general for widespread errors with their data. A 2012 Federal Trade Commission report found that one in five consumers found errors on their credit reports, many of which hurt their scores, causing them to fall into weaker borrowing pools, thereby raising their borrowing costs.
Credit data vendors must take great care to ensure their privacy measures meet every standard, and the data vendors that Zest AI works with also need to ensure that Zest won’t be misusing their data and that we take all the necessary measures to certify the data is properly protected, credentialed and used only for permissible purposes. Our track record of working with alternative data sources and a large array of information has put Zest and its partner ecosystem in a better place than most to meet the goal of helping more consumers worldwide enter the credit mainstream.
Photo by Andrea Piacquadio from Pexels
February 26, 2021
February 16, 2021