Artificial intelligence and machine learning have found their way into lending. But can we trust robots with loan origination quite yet?
Upstart — a five-year-old online lender — bets you can (and you must).
In fact, more than 25% of Upstart’s current loans are fully automated, Bank Innovation has learned.
“We have made this progress in the last four to five months; back in November none of our loans were fully automated,” said Upstart’s founder and CEO Dave Girouard. “We started out these tests for extremely low-risk, smaller loans, but over time we will begin to broaden our scope. By the end of the year we expect the majority of our loans to be 100% automated.”
The lender utilizes AI and machine learning algorithms to analyze consumer data points, including education history, employment history, social media, and even web behavior (during the online application process), in order to make underwriting decisions. The company has originated almost $700 million to date, from about 60,000 loans.
“We have been continually upgrading our credit model as we get more data, and we are making significant advances around automation and customer authentication,” Girouard told Bank Innovation.
Full automation, the Holy Grail of lending, is on the road-map of all online lenders out there, but most still keep the human element. OnDeck, for example, provides an online application portal, but has a human underwriter review those applications before a final decision, according to an OnDeck customer service rep.
Upstart raised $32.5 million last March in yet another round of funding in order to continue the development of its machine learning algorithms. The company also began licensing its tech to other FIs back in March.
“Most fundamentally, we believe that every flavor of lending will be based on machine learning and artificial intelligence in 10 years, and it’s hard to imagine any successful lender of any type without having built or partnered with someone, who provides this technology,” Girouard said.
Machine learning is the future of underwriting, and banks agree. But who will be running the show?
“Machine learning is absolutely the future of underwriting,” Sandeep Sood, VP of software engineering at Capital One, told Bank Innovation previously. “But the startups that are doing that today, they take, what I call a fairly shallow data — like social graphs — and make these wild claims that those superficial elements are really the best predictive pieces to figure out if someone should be qualified or not.”
Girouard claims, however, that banks don’t have much advantage over startups in terms of data access.
“Over 80% of our applicants provide us with credentials to their banking accounts; they do that for many reasons, but mainly so they can verify their accounts and transfer money,” he said. “But we have data on their income, we have built a proprietary database of virtually all employers in the U.S., and have access to customers’ bank statements; so I don’t believe we lack data at all.”