When lending, bet on the trend, not a point in time
Credit risk models built with Zest AI’s technology have already scored more than 6.1 million loan applications.
Around the six million mark, we decided to write down the biggest lessons we’ve learned about where we think consumer lending can go when using AI and machine learning. Here’s what we found.
Lenders should bet on the trend, not a point in time
Part of a lender’s hesitancy to approve loans down the credit spectrum is the inability to trust a traditional credit score’s risk assessment to be accurate, especially when it comes to the middle band of applicants. Those folks who often fall into a “gray area” on the credit spectrum still need a loan at the end of the day, so what is a lender to do? Use AI to help make their decision, of course.
Two borrowers with the same score could be at completely different levels of risk under the surface. One applicant with a 650 score might be on their way to a 700, while the other may be heading in the opposite direction. A lender can’t spot these distinctions by using generic scores that rely on only a handful of data points pulled at a specific point in time. What a lender needs is a richer set of insights — which also includes as many trade lines as feasible — to better assess the credit profile of applicants and help determine the different risk profiles of these two borrowers based on their trended data.
AI-automated underwriting generates its advantage over generic scores by incorporating hundreds more variables than traditional models can even handle. And when AI-automated underwriting can tap into thousands of correlations between data points, it provides a more nuanced view of consumers that can lead to increased approvals and fewer losses.
How does trended data work in real time?
Here’s an illustration of what we’re talking about. Meet John and Jane, two individuals who just applied for a loan at your financial institution. Looking at their credit scores based on a traditional credit model, John has a score of 700, and Jane has a score of 620.
Now, a traditional credit model would look between these two applicants and almost certainly approve John over Jane. But is John really the applicant with less risk?
Not exactly.
Taking a look between these two applicants with AI-automated underwriting, a lender can use more data — especially trended variables — to uncover the red flags that indicate John is a riskier applicant than his traditional credit score would indicate.
Based on trended data and using AI-automated underwriting, a lender can conclude that John’s risk profile actually indicates that his credit score should be 620. John’s financial behavior is far more risky than traditional scoring would have a lender believe.
Let’s take a look at Jane’s trended data with AI-automated underwriting.
Using insights pulled from Jane’s credit file, the AI-automated underwriting technology can extract a lot of information, even though she has fewer accounts. But, compared to John, she has more high-quality, current accounts such as her mortgage. Based on her activity, Jane’s risk profile indicates that her credit score should be a 700, not 620.
Results can vary, since every financial institution has different risk tolerance levels and policies. However, when these two loans are put back to back based on AI-automated underwriting, Jane is a lower-risk loan approval compared to John.
One question we often get: How do you get trended data for down-spectrum borrowers without a lot of credit history?
If there’s no history, we’re not going to make it up with synthetic data. That would be far from compliant. But AI-automated underwriting can easily accommodate lenders interested in going deeper down the credit spectrum by ingesting cash flow data from rent, cell phone bills, utility bills, and first-party savings or checking accounts. You can get a lot of predictive signals from looking at how people use their money. Even more, the industry is already moving in this direction because it produces better models than simplistic generic scores.
