Automation can reduce disparities in loan approvals

The Paycheck Protection Program included guidance from the SBA to prioritize loans to businesses owned by socially and economically disadvantaged individuals. Yet an October 2021 research paper entitled “Racial Disparities in Access to Small Business Credit: Evidence from the Paycheck Protection Program,” published by a group of New York University professors, found a disparity in the approval of PPP loans to Black-owned businesses. These differences in approval rates are not explained by factors like pre-existing bank relationships or applicant behavior.

The interesting finding of the research was evidence that small banks that had automated their lending processes, thereby reducing human involvement, had higher rates of PPP lending to Black-owned businesses. Because of this, the paper’s authors suggest that preference-based discrimination occurs — and automation can make lending more equitable. 

This is part of a larger conversation about the role of artificial intelligence in lending. After all, the application should produce results that are consistent with the bank’s policies or underwriting requirements of an entity like the SBA. If, as this study suggests, implicit bias is at the root of discriminatory lending practices, how can automation and AI solve these problems?

Examine bias in credit approvals

Many banks think of lending automation as an application that can process loan applications using the same criteria that a human would use. To some extent that’s true: The bank’s current risk assessment becomes the basis for any automated model. 

Lenders may feel that they can advocate for a borrower who may not otherwise have the necessary creditworthiness. After all, they know how to “work the system” to get a loan approved when bias in the system itself seems to be in play, either through the secondary market or loan committee. Yet this relies on humans making the decision to push for a borrower, and cannot correct intentional or unintentional discrimination.

Automation can remove variables from a loan approval that may need to be collected by the bank to comply with regulations but should have no impact on the credit decision. Think: Processing a loan without using data points like name, address, race or gender. Underwriters or loan officers may inadvertently be using this information to make a decision about creditworthiness; lending automation would only process the data in the loan application.

Use AI to spot patterns

AI can also be used to determine if a bank’s loan approval criteria has led to underrepresentation among groups of borrowers. Removing human involvement only solves one layer of the issue if the criteria itself is leading to inequitable lending. The objective criteria that a bank has in place for its in-house loans may have an exclusionary impact, creating barriers that are impossible to overcome for some borrowers.

By using historical data around loan approvals, AI can identify if racial minorities or women are underrepresented in loan approvals. The analysis can be used to identify the variables within the bank’s loan policy that are causing the underrepresentation. This can help the bank understand how to modify its approval criteria to extend credit more equitably, through the use of additional objective variables that will keep approvals aligned with the bank’s lending goals.

Make equitable lending a priority

In a press release regarding the record number of SBA traditional loans originated in 2021, SBA Administrator Isabella Casillas Guzman said, “While progress has been made, our data also tells a deeper story: Historic inequities in accessing capital persist, and we must do more to lower the barriers of entry to opportunity for all our entrepreneurs.”

Guzman’s statement reflects a national conversation about the role of systemic racism. Much as community banks may want to believe that they are abiding by fair lending practices, they need to acknowledge that biases may creep into the approval process.

Introducing automation can be the first step for banks committed to examining ways that their loan policies and lending practices may be inherently unfair. The key is finding a fintech company with an AI-based solution that has a specific mission of rooting out patterns of biased lending behavior and simultaneously working to reduce those biases. Such vendors are out there, ready and able to help banks achieve more equitable lending practices.