Why Machine Learning Can Boost Your Lending Automation

The quest to increase automation sits atop every lender’s digital agenda. “How can you help us increase the number of loans we auto-approve or deny?” is a question we get on every Zoom call with banks and credit unions. The term “automation” means a lot of things to a lot of people, and it’s often applied to a plethora of digital activities such as account creation and management, and customer experience, so I’ll just clarify that, for Zest AI, automation means faster and better credit underwriting decisions, one of the most important calls a lender can make.

Zest-built models are designed to accelerate trusted loan decisions because they use more data and better math than traditional scorecards and linear models. It’s not some generic or pre-packaged model you’re using: These are custom models trained on your data so they’re designed to be more predictive than generic scorecards. On average, our customers see a 21% increase in approvals, holding risk constant, when they switch from legacy scoring to ML. Lenders looking to reduce risk typically achieve a 28% decrease in charge-offs, holding approvals constant. When you have a better credit model, you can automate more approvals and improve your take rate with a faster response time. Greater automation lets you grow faster with the same operational resources. You take your Zest score and combine it with your existing credit policies and systems to auto-approve and auto deny-applicants safely and responsibly.

Every lender wants something different from their automation play. The main benefits of auto-decisioning are speed, scalability and consistency. Credit unions tend to focus on the impact that automation can have on member experience, with goals to reduce friction and freeing up team members to provide higher-touch service to the edge cases that require a greater attention. Banks tend to be more focused on how increased automation can help them process a larger volume of applications while effectively managing their operational expenses. Specialty auto lenders want to drive speed to decision as a competitive advantage. They know if they can return an approval quickly and with confidence dealerships will send them more volume due to their consistently fast response.

No matter what type of lender you are, increased automation will give you more consistency because, by definition, you’re limiting the number of applications that go through manual review. I was recently talking to an experienced head of lending for a midsize credit union who was concerned about maintaining consistency across his growing underwriting team. He admitted that inconsistency could potentially open them up to a fair lending violation. That’s a serious issue automation can fix.

Let’s look at an extremely conservative scenario of a lender that gets 100,000 applications per year and manually reviews 70% of them. We estimate it would take that lender an average of 11 minutes to respond to an application and it would need nine to 11 underwriters to support that volume. If you flipped that ratio to 70% automation, you could cut your average response time and operational overhead in half. Our piece of the process typically runs about three to five seconds, and that’s generally if we’re not in a huge rush.

Not everybody’s metrics are the same and you may have fewer applications, and that’s perfectly fine if your application volume is lower. That just means you’ll require fewer underwriters to support that volume, and automation will decrease that further.

For those organizations already at a fairly high degree of automation, say 50%, you may have a more lofty goal of 80%. Working in your favor is likely an organizational comfort with auto-decisioning, which means you can achieve your goals faster. Here, again, you get a sizable decrease in average response time and your people can do more manual reviews.

The greatest obstacle to increasing automation is neither technology nor methodology, it’s having the organizational trust to move forward knowing that automation is not going to add portfolio risk. Every organization is going to have a different mentality toward automation. There’s no magic sweet spot for all. Some want to go from 20% to 80% next year. Some are cool with being under 50%. If you’re manually reviewing all applications right now, something like 20% or 30% auto-decisioning may seem like a pretty lofty goal in the near term.

What is your goal for increasing automation? We’re happy to work with you on a more detailed analysis of the impacts that automation could have on your lending operation. Drop us a note at hello@zest.ai and we can walk you through what it will take to automate without adding risk.

And if you already know you want to automate and just want more insight into how ML automation fits into your lending process and IT environment, read my cleverly titled post How AI And Automation Fit Into Your Lending Process.