Artificial Intelligence, Online lenders

The future of lending: intuition is out, intelligence is in

  • There is great potential for machine learning in lending.
  • A recent study showed it cut losses by 75 percent and boosted approval rates by 173 percent.
The future of lending: intuition is out, intelligence is in

This article was contributed as part of Tearsheet’s new Thought Leaders contributor program. Dave Girouard is the founder and CEO of Upstart, the first lending platform to leverage artificial intelligence and machine learning to price credit and automate the borrowing process. 

“There are times when you’re better off having the sales force go play golf than make new loans,” said JPMorgan Chase CEO Jamie Dimon during that firm’s earnings call on January 15th, seeking to assure analysts that his firm would proceed with caution when it came to late-cycle lending.

Of course, prudent lending makes intuitive sense when most experts believe a recession is likely in the next year or two.

But it begs a different question: when exactly does imprudent lending make sense?

Famed investor and general provocateur Peter Thiel has a favorite interview question for founders: “What important truth do very few people agree with you on?” The question is centered around his thesis that all great startups have discovered something that most of the world dismisses.

Lending is an industry where most people believe one thing: Aside from modest efficiency improvements, technology can’t alter the fundamentals of lending. Some lenders — those staffed by teams with strong analytics and disciplined processes — tend to do better than others. Others will be sloppy or greedy or both, and those lenders will not fare as well. But ultimately, lending is lending, and as reliably as gravity, it will all come crashing down at the next recession.

But I believe that the company that wins in the end will have an opposing belief: The emergence of machine learning in lending will entirely reshape the banking and broader credit industry in the next 10 years.

This means that nearly all lending will have sophisticated machine learning and broad use of (what is today called) alternative data at its core — in customer acquisition, credit decisioning, fraud, verification processes, servicing, and cross-selling. The economic gains available via ML within the credit decision alone are so vast that they will remake an entire industry.

ML and alternative data lead to much higher approval rates, lower loss rates, and reduced operating expenses. Our recent study shows an ML-driven solution could cut losses by 75 percent and boost approval rates by 173 percent for banking giants in the US. If this is true, then it goes without saying that lenders who adapt will thrive and gain market share. Those who don’t will slowly slide into obsolescence.

What will ML-based lending replace? Intuition-based lending. Mr. Dimon’s crystal ball on the next recession is just the tip of the iceberg: predictive value of a credit variable, preferred use of funds, documentation requirements, prepayment behavior, best time to call a delinquent borrower. There are a thousand decisions lenders making every minute based on experience, anecdotes, and intuition that are limited if not outright flawed.

It’s worth noting that there will never be a recession-proof lending platform. There is an inevitable relationship between recession, unemployment, and the ability and propensity for borrowers to make good on their debts. ML-based models won’t necessarily predict recessions, but they will accommodate the current state of the economy in their decisioning and will react to changes in the economy faster than others will. As surely as autonomous vehicles are safer than human-driven cars, self-driving loans will outperform those powered by intuition.

Why do so few people agree with this? There’s a cascading set of reasons why the industry is skeptical about a fundamentally better lending mousetrap:

  1. Somebody would have already done it: change is hard, and change at the core of credit is really hard. The data science, the regulatory hurdles, the financial risk – add these up and it’s no secret why the pre-eminent lending institutions in the US (banks) haven’t taken on this challenge. Furthermore, the availability of sophisticated ML techniques and inexpensive cloud computing have only come to fruition in the last handful of years. And let’s be honest – most tech entrepreneurs would prefer a ticket to Fyre Festival to innovating in an industry as heavily regulated as consumer lending.
  2. 5000 years of lending disasters say otherwise: It doesn’t take many NextCards to create deep skepticism that those touting new and improved lending models are mere snake-oil salesmen. Lending incompetence can hide in the shadows for a long time. Likewise, a more enlightened form of lending needs to prove itself for an equally long time.
  3. FICO works:  It’s been around since 1989 and it does the job, so why mess with it? In reality, FICO is extremely limited in its ability to predict credit performance because it’s narrow in scope and inherently backward looking. In fact, it’s the decades-long comfort and that familiarity banks (and capital markets) have with FICO that have suppressed innovation in credit decisioning in the US.
  4. The R word: When TV mob boss Tony Soprano asks his underling Silvio what businesses are immune from recession, Silvio replies ‘certain aspects of show business and our thing’. Sounds about right. On the other hand, we can all agree that lending is one of the industries most susceptible to recession. And what are the commonly accepted defenses lenders can arm themselves with in the face of recession? Experience, anecdotes, and intuition. Note that technology and data science aren’t on that list.

Artificial intelligence and machine learning have sucked so much oxygen out of the room that you’d be hard pressed to find a FinTech startup that doesn’t extoll their virtues. But saying ‘we do ML’ isn’t sufficient, as these technologies require significant investment and commitment to have meaningful impact, and doing it in a lab setting won’t get you there. They can’t be added to the fringes of your product – they have to be your product. And in truth, you can’t build an ML-based lending model from scratch without taking on significant risk — financial, regulatory, and reputational alike.

The power and potential of AI/ML can be both bewildering and intimidating to consider. But we’re optimistic they’ll one day soon bring more affordable credit to every corner of the world. Feel free to bet against this — you’ll be in great company.

Online lenders, Podcasts

TS Lending Podcast 3: How SMB lenders scale — JPM Chase gets into POS lending — Amazon thinking about lending?

  • Dan and Zack talk about the techniques top SMB lenders use to grow at scale.
  • Also, decentralized credit scoring? Dan explains what it is and how it's being used.
Tearsheet Editors | March 20, 2019
Online lenders, Podcasts

Using fintech to improve financial wellness with Salary Finance’s Dhiren Master

  • Employees with financial worries aren't as productive or engaged.
  • Salary Finance works with employers to help their people get out of debt and start saving.
Tearsheet Editors | March 18, 2019
Online lenders

How online lenders attract and service the evolving SMB borrower

  • SMBs always need money to finance their businesses.
  • But how they shop for loans and their education about finance has changed.
Michael Deleon | March 13, 2019
Online lenders

Forget about digital marketing, SMB lenders double down on partnerships for growth

  • Direct and digital marketing play a role in growing underwriting for top SMB lenders.
  • But it's partnerships that provide the real key to growth.
Michael Deleon | March 06, 2019
Online lenders

Scooch on over Affirm: JPM Chase enters purchasing financing market

  • JPM Chase plans to enter the point of sale lending market.
  • With 'My Chase Plan', cardholders will be able to finance purchases of $500 and above.
Michael Deleon | March 05, 2019
More Articles