AI Will Take Over Credit Decisions… And That’s Good

When the first Model T drove out of the factory in 1908, it was pretty clear something big was happening. If you had asked someone then whether cars would replace horses for transportation by 2010, even without the benefit of hindsight the answer would be “Duhhh.” Like the assembly line before it, artificial intelligence (AI) will transform nearly every industry by making powerful technology ubiquitous. Yet the talking heads are still asking: will AI take over heavily regulated high stakes tasks like credit underwriting? (Spoiler: yes, it will, and it already is.) This will be among the topics hotly debated this week at Money20/20 in Las Vegas.

Written by Douglas Merrill - Originally published at this link, October 24, 2017.

AI is heartless, and that may be one of its best traits. Humans’ hearts — and their unconscious biases — make poor decisions. Judges in Baton Rouge give defendants harsher sentences when the local football team unexpectedly loses a game. Imagine how significant these human biases could be for deciding who gets a loan and who doesn’t.

The real question isn’t “Will AI take over lending?”; it’s “Why hasn’t it already?”. Why isn’t all credit underwriting automatic, fast, and human-free?

It takes time for everything around an innovation to catch up to it. The Model T may have come in 1908, but the first drive-in gas station in Kansas wasn’t erected until 1923. This kind of infrastructure change outside of the largest American cities was an indication that the world was on an inevitable path to change. The Model T made it possible to drive anywhere, from the busy streets of Detroit to the plains of Kansas. Mass automobile production changed the game when it made automobiles available everywhere.

The same thing is true for credit. It’s going to be a while before financial institutions’ have the capacity to support real-time AI-based decision making, so we’re still quite a ways from robot overlords like the Terminator or HAL taking over (though we could probably build a HAL pretty easily with today’s tech). Most of the entities talking about their all-powerful AI are doing pretty simple stuff. And if Janelle Shane’s neural network fortune cookie test is any indication, the math needs a lot of development before it's ready for prime-time (unless you can make sense of the wise words, “When death comes, it will have no sheep.”).

In addition, most financial institutions lack the data scientists and infrastructure needed to implement advanced machine learning. Even if those firms are well-equipped, until recently, the tools have not existed for model risk management — machine learning has a tendency to obscure its reasoning. In credit decisioning, explainable AI is crucial for compliance with adverse action, fair lending, regulatory filings, and internal stakeholder support.

FinTech firms — such as our company, ZestFinance — are now addressing these concerns with machine learning tech that brings institutions up to speed quickly and provides powerful explanatory analysis of the way it makes decisions. With these hurdles overcome, there is little reason to believe that credit decisioning will remain heavily manual and reliant on elementary math and data.

Here’s my prediction: by the year 2030, artificial intelligence will have replaced humans in underwriting. We’re not there yet, but the AI train has already left the station. Now lenders have to decide whether they want to be on the train or get hit by it. AI is poised to improve credit decisioning and make it more accessible and fair. Manual credit underwriting is going to be a relic of a bygone era, and that is a great sign for borrowers and lenders alike.