Pretty Soon, Your Loan Application Will Be Reviewed By A Robot

OBSERVATIONS FROM THE FINTECH SNARK TANK

Efficiency is the name of the game in banking. Itʼs amazing, then—and not in a positive way—that banks’ use of lending technology is lacking.

For all the hype in the press about artificial intelligence (AI), just 13% of banks have deployed AI technologies in their credit and lending processes according to a new report from Cornerstone Advisors (commissioned by ZestAI) titled Achieving High-Performance Lending: The Impact of AI on Lending Efficiency.

The Coming Boom In AI-Driven Lending

Despite the low use of AI in lending today, big changes are coming. More than half of the banks surveyed plan to invest $100,000 or more in machine learning-driven credit models over the next three years, with 12% planning to spend $500,000 or more. AI-driven auto-decisioning will increase because of banks’ need for:

• Efficiency. The efficiency gains for automated decisioning are substantial. According to Cornerstone’s benchmark data, the number of loan applications reviewed per underwriting FTE per month is 3.5 times greater among banks that use automated decisioning than those that don’t.
• Effectiveness. Machine learning models can increase credit access by more accurately identifying applicants who are likely to repay loans and to reduce the number of people given loans that they are unlikely to repay.

AI Brings Familiar Challenges

Skeptics point out that AI will present lenders challenges like:

• Explainability. Understanding how and why a model arrives at a particular decision or prediction can be difficult. Lenders must ensure explainability (i.e., why a model produces the results it does) in their machine learning models to comply with regulatory requirements, address customer inquiries and maintain trust.
• Transparency. The black box nature of many machine learning models has focused attention on model transparency as a critical threshold question for both lenders and regulators.
• Data availability and quality. Accuracy, security, and data availability are challenges with machine learning models. Machine learning models rely on large volumes of high-quality data for training and validation, and banks and credit unions are often challenged to access and aggregate diverse and relevant data sets.
• Model validation and bias mitigation. Models must be thoroughly evaluated to ensure they do not perpetuate biases, discriminate against certain demographics or exhibit unfair behavior. Establishing robust validation processes and addressing bias concerns are essential to ensuring fair and responsible lending practices.

The Benefits of AI-Driven Lending

These aren’t new challenges, however—they’re the same challenges banks face today with manual (i.e., human) loan underwriting. Leveraging machine learning models will help banks:

• Improve credit risk assessment. By incorporating alternative data sources and nonlinear relationships, machine learning models can generate more accurate and comprehensive credit risk assessments, enabling lenders to make more informed lending decisions and better manage risk.
• Create more efficient collections and loss mitigation. Focusing on collections and loss mitigation becomes crucial in a down lending environment. Machine learning models can analyze borrower behavior, payment patterns, and external factors to identify early warning signals of potential delinquency or default.

As Visa concluded in a recent study:

“Lenders that automate decisions were able to auto-decision roughly 50% to 60% of application volume, which drove a more consistent level of decisioning speed and cost reduction while maintaining a stable level of risk performance.”

AI Can Help Ease The Small Business Credit Crisis

AI will do more than just benefit banks—it will help expand minority small business owners’ access to capital. According to the Federal Reserve:

“Black- and Hispanic-owned businesses have a harder time accessing credit from traditional sources because they’re evaluated as higher credit risks. These risks are determined in part by lower owner wealth, lower business revenue, and insufficient credit histories.”

At the heart of the small business fair and accessible solution is data—which is often incorrect, missing or misleading.

According to an analysis from Uplinq, relying strictly on FICO scores for small business loan underwriting can unfairly impact protected class business owners.

The data and technology firm’s analysis found that expanding the set of data inputs used to evaluate protected class small businesses—including data from accounting systems, payment and e-commerce systems, core banking platforms and other sources—could increase the number of good (i.e., non-loss) loans accepted by 215%.

Small business owners’ access to affordable credit can’t be solved by regulatory change. It will improve when the lenders—banks, fintech, and other non-bank lenders—see improved profitability from the use of AI and expanded data sources.