Machine learning and artificial intelligence have made waves in the lending space recently, but human underwriters are by far not out of the picture.

That’s according to OnDeck’s Chief Financial Officer Howard Katzenberg.

“What we want to do is be great at both the algorithmic and analytic part, as well as [maintain] a human element,” he told Bank Innovation. “Small businesses are diverse, so you always want to have that judgment element as well, particularly for larger loans, and our hybrid underwriting capabilities allow for that.”

At the initial stages of a loan application, “100% of our applicants get algorithmic decisions,” Katzenberg explained. The algorithm then “spits out” one of the three decisions: the applicant is either declined; approved and can move on to the booking stage; or pending and needs further investigation. “In this case, the algorithm picks up on something, or the loan value is larger than our average, or maybe it’s a random test of the model,” he said.

Currently, about one third of OnDeck’s loans experience a manual intervention. The small business lender, launched in 2006, currently has $1.2 billion in loans under management, according to second quarter 2017 earnings filings. The number’s up 25% compared to the year prior.

Lenders, both traditional and online, are increasingly looking to machine learning and automated underwriting in order to reduce costs and the time it takes to approve borrowers. Online lender Upstart, for example, is looking to automate “most” of its loans by the end of the year, the company told Bank Innovation recently. However, such practices have received criticism from traditional players, fearing that the process may fail to eliminate risky borrowers.

The marketplace lender is now focused on continually improving its credit models and introducing new product offerings to customers. “For our term loan product, for example, we are now testing a service where, if before a customer wanted to pay off an OnDeck loan early, they were responsible for the remaining interest,” he explained. “As a result, some of the higher quality borrowers would not book their loan after approval, so we are changing that, and testing the product now.”

The company is also looking to bring in more banks onto its platform, similar to the deal it made with JPMorgan Chase back in 2015.