Balancing Art & Science in Lending Decisions

Federal Reserve Bank of New York data shows that nearly one-third of business borrowers file an application with an online lender, a dramatic increase from 19% just two years earlier. Traditional financial institutions remain the more popular choice, but those odds are not trending in their favor. Speed of the credit decision is the most oft cited reason for this shift, pointing to members’ continued desire for convenience and the new market players’ success in harnessing automation to meet this demand.

Fintech firms have approached lending as a science, generally speaking, applying artificial intelligence to streamline what has long been a labor-intensive exercise for both underwriter and applicant. With the wealth of available data constantly expanding and algorithms growing more refined, the question becomes, to what extent can these decisions become purely formula driven?

At the other end of the spectrum are those who view lending as more art than science, considering qualitative inputs like local context and an applicant’s background. After all, the “5 Cs of Credit” (Collateral, Capital, Capacity, Conditions and Character) were coined for a reason – and some of these are more ripe for AI than others. Arguably, credit unions serve as the standard bearers for bringing art, and heart, to the process in the form of incorporating community and personal factors.

As is often the case, the optimal approach likely resides somewhere between these two broadly drawn extremes. The operative question then is how to identify the right balance and construct an infrastructure to support it.

Keeping a Human in the Loop

Numbers can only tell so much, but they do reveal a lot. There are plenty of opportunities for credit unions to leverage technology to improve member experience as well as operating efficiency. An obtainable near-term focus is embracing science to address workflow and analytics, automating the steps inside the institution. One good example would be monitoring members’ cash flow using near-real-time data, enabling credit unions to anticipate borrowing needs and/or repayment challenges, rather than merely reacting to them.

In responding to loan applications, the most straightforward requests can almost certainly be decisioned without human intervention. Even then, however, it’s important to keep a human in the loop. Some institutions adopt a policy of letting algorithms approve but not decline applications in order to avoid compliance issues down the road. It may require no more than a brief manual confirmation, but regulators will expect to see documented proof that criteria for declines are being consistently applied.

A guiding principle should be “triage.” Credit unions may have different definitions of artificial intelligence and varying appetites for applying it, but across the board it can be deployed to automate workflow and deliver actionable data to trained professionals at the right time to inform efficient decision making.

Similarly, each credit union’s unique culture will invariably play a role in its decision on how to deploy AI. Naturally, there is no one right answer. Just as credit unions have different risk profiles, whether driven by the communities they serve or other strategic variables, comfort levels will vary on how much latitude to delegate to computers before bringing humans into the loop. Given members’ clear expectations for speedy responses, however, it’s difficult to imagine a successful scenario in which the credit union meter doesn’t tilt at least a bit further in the direction of automation.

Shining Light on the Gray Areas

Credit unions should place their emphasis on using automation to address the gray areas, sifting out the noise of clear-cut approvals and declines and serving up the relevant info to enable human decisions on the remainder. Whether that gray area should comprise 70% of loan volume or 30% is a matter of credit union preference and strategic debate. (Many online lenders have pushed their boundaries even below these thresholds.) It all comes down to comfort levels; AI parameters can be calibrated to accommodate a given institution’s credit policy, for instance.

Traditional lenders like credit unions still hold important advantages over online disruptors in the form of existing member relationships and the perception of better borrowing rates. These advantages are at risk of erosion, however, as online players build out full product suites and provide concierge-like services to facilitate easy applications. Startups like Chime that launched with deposit offerings are now adding lending, while payment players like Square, PayPal and Stripe expand into lending and deposit gathering. A portion of online lenders’ higher rates can be attributed to serving lower credit quality borrowers, but as they build scale this gap is likely to narrow as well. Credit unions should likewise be doing their part to narrow these competitors’ areas of technology advantage.

Realistically, neobanks and online lenders can increase the human component in their operating models at least as easily as traditional lenders can adopt AI. The difference between winning and losing the race will come down to finding a balance best suited to your credit union’s field of membership and market strategy. There’s ample room for art and science to coexist in lending; credit unions need only affirm their risk appetite and incorporate appropriate levels of AI-based decisioning in response to what will enhance their existing risk culture. After all, credit unions are armed with far more data – even if their decisions incorporate other less numeric considerations.

Gary Lewis is Director, Lending Sales for ProfitStars, a division of Jack Henry & Associates based in Allen, Texas.