How AI can help lenders weather the storm

More data and better math can help lenders make better predictions and shift strategy quickly

‍

As lenders consider how to manage the emerging economic crisis coming out of the COVID-19 pandemic, many are looking for better data. They recognize that they're going to have to re-evaluate existing loans to mitigate potential losses and — in the future — become more selective about how they extend credit. Having the best tools at hand to help navigate economic shifts is important not just now, but also for the future. Read on to see how AI and machine learning can help lenders limit current losses while enhancing their future profitability.

‍

With better predictions

Machine learning helps you make better use of the data you already have. When using a traditional model for underwriting, you might have 30 or 40 data features to use. By using a machine learning model for underwriting, you can collect and fully incorporate all available data into your decisioning. Instead of being confined to 30-40 variables, machine learning expands a given feature set through correlations, which helps paint a better picture of your borrowers.

Machine learning is able to track non-linear behavior data, such as the number of inquiries a consumer makes. Extremely low numbers of inquiries could represent a higher risk borrower. That person may have a limited history and therefore limited performance data, such as an 18-year-old just starting to build a credit history. As the number of inquiries grows, risk typically decreases, since there's more data available to paint that more accurate picture of an applicant. And sometimes, if the number of inquiries reaches a certain amount — perhaps 40 or 50 — that can actually raise red flags about a person’s risk profile, deeming them a higher risk applicant. Traditional modeling techniques cannot capture this type of non-linear behavior, meaning making decisions off of it puts your institution at risk of the loan defaulting.

Improvements in predictability can positively impact performance. The better predictions generated by machine learning can reduce net charge-offs 20 to 30 percent for new portfolios when holding an approval rate constant, or improve growth by about 10 to 20 percent when holding the charge-off rate constant.

‍

With advanced monitoring

We are seeing the importance of model monitoring with the current economic volatility from COVID-19. Predictions can be lost if you don’t have advanced monitoring. Model monitoring ensures accuracy and validity, to help you stay the course.

It’s also important because of the sheer amount of data machine learning models assess makes it possible for some interactions to go unnoticed. Machine learning can test models as they are being trained to see whether their signals change from the training period to when they are put into use. This ensures stability and reliability.

Machine learning can quickly alert you to changing behavior. In early March, one client got an alert just around the time the impact of COVID-19 first began to become widespread. In the past few weeks one group, whose incomes had remained stable, had begun to accumulate more debt. This could represent an uptick in the percentage of available credit being used, presenting significant risk exposure to a given lender. This type of rapid signaling can only really be found with machine learning. Through advanced monitoring, this client was able to take time to strategize their response and take action.

‍

With fast response and documentation

In times of economic uncertainty, there are bound to be changes in your borrowers which means that you may need to create a new model. The traditional way of creating a new model is time consuming, and at times like this — you can’t afford to wait a year.

That’s where machine learning can help. It takes a year or more of work down to weeks. With more data points, there is less impact per feature. And with automated documentation, it enables your teams to get to market faster.

Zest AI's machine learning models are compliant with existing banking regulations. When you make credit decisions using Zest AI’s technology, they are delivered in a way that’s fully explainable. The machine learning technology also delivers automated documentation that is compliant with applicable banking regulations. While AI and machine learning can’t stop the volatility we’re about to experience, it can help to mitigate its impact on lenders portfolios.