Answers To Legit Concerns Credit Unions Have About AI Lending
March 24, 2021
There’s a powerful shift taking place in the leadership of lending technology. Smaller financial institutions such as specialty auto lenders and credit unions are now adopting machine learning to power their newest models faster than large banks. We don’t have conclusive stats on this, but based on the adoption of ML underwriting models in our customer pipeline, the credit unions are leapfrogging in innovation. The reasons why are clear: Machine learning-built models are better at predicting risk than traditional credit scores. That gives credit unions what they’ve always wanted: the ability to say yes to more members and serve their communities more deeply.
Our experience suggests that ML can benefit institutions of all sizes. The results are every bit as powerful for smaller lenders as larger ones, if not more so. Nonetheless, we often encounter a healthy skepticism and solid questions about the risks. We thought we’d share the concerns we hear most often from CU executives and other smaller lenders and provide some answers that might begin to put those fears at ease.
I’m not sure where to start
From all our meetings with prospective clients, we get the sense that most folks are not so much fearful of machine learning as they are overwhelmed. A 2018 Fannie Mae survey, for instance, found that nearly 1 in 5 credit union executives (20%) said not knowing where to start was among the most prominent reasons they had not adopted ML.
That’s completely understandable. Not only is ML a new technology, but the marketplace is fraught with hype. The truth is that ML can help you achieve impressive results by allowing you to use far more powerful models to identify the most creditworthy borrowers. But it doesn’t have to disrupt your business. Just think about an ML model as a turbocharged version of the traditional credit decisioning scorecard that you use today.
Zest just published this in-depth guide to AI adoption. Something you might want to check out.
I don’t have enough data
Another worry that small lenders especially have is a lack of quality data to build a model. Nearly a third of all credit union executives cited it as a significant hurdle in that 2018 Fannie Mae study. The reality? ML is nothing more than making better use of the information you already have by capturing the complex interactions among hundreds of standard credit data points. Your model doesn’t have to be Google-sized to generate real predictive power. If you have at least 100,000 records or 1,000 defaults over a several-year period, you probably have enough data to start. Plus, Zest can augment your data with the FCRA-compliant credit files of lookalike customers.
You don’t need to go too far afield or get too exotic with data, either. More data is better, but traditional sources, such as bureau and application data, are usually all you need. While alternative sources such as social media “likes” or Spotify playlists capture the imagination, they don’t necessarily add that much signal. They can also run afoul of fair credit and fair lending laws.
I need a fancy IT infrastructure to put my ML model into production
For most clients, the complexity of integrating an ML model into their existing loan operating systems and technology infrastructure is also a significant concern. The Fannie Mae survey, for example, found that 43% of credit union executives rated this among their two most significant roadblocks to ML adoption. That may be because the institutions that pioneered ML model development had struggled to get them into production because of the technical challenges and demanding compliance requirements or push-back from an already swamped IT department.
Today that is no longer the case. New tools and more powerful automated approaches have accelerated the model development process. Institutions have greater flexibility, too. The latest models and model development tools work in the cloud or on-prem, offering various deployment options. We’ve found that close collaboration with a client service team can expedite the model adoption and deployment to take no more time than implementing a traditional scorecard.
Close collaboration with a client service team can expedite the model adoption and deployment to take no more time than implementing a traditional scorecard.
I don’t have any data scientists to explain the model results to me
The arms race for top tech talent is intensifying – not surprising when nearly a quarter of financial executives across all institutions cited a lack of necessary skills as a significant impediment to AI/ML adoption. The biggest banks might be the only financial institutions that can compete for talent with Google, Facebook, and well-funded fintech companies. Several of the biggest banks have announced plans to spend billions of dollars building data science and modeling teams. Most smaller institutions simply don’t have the resources to make such an investment.
Here, an outside partner who can fully support the model development and maintenance processes, or collaborate with your existing team, can be hugely beneficial. So, too, can the proper technology. The latest model-building technology renders even the most complex models fully interpretable, so you won’t need a roster of whiz kids to explain it. Interpretable models mean you should be able to rely on easily digestible readouts from your model that can explain how it arrived at every credit decision for every borrower – and provide accurate adverse action reasons as well.
AI Underwriting Anxiety: I’m worried about managing change
Change is never easy, but AI doesn’t have to be the boogeyman. The key will be to educate employees about the benefits and explain how new processes will look. AI brings automation which frees teams to focus on more personalized services for members. As Jenny Vipperman, the chief lending officer at VyStar Credit Union, once said, “The role of the underwriter doesn’t go away. It just becomes something new, just like it became something new 12 years ago.”
For lending teams, AI provides better predictive abilities to avoid toxic borrowers and identify previously overlooked creditworthy buyers. For example, one lender achieved a 22% increase in portfolio approvals while holding risk constant. Lending teams will also benefit from automation productivity gains by expanding their auto-decisioning thresholds.
With AI improving risk assessment accuracy, underwriters can deliver faster decisions to help teams meet member experience goals. Their day-to-day drastically improves by spending more time on a smaller percentage of applications that require a more personal touch.
The good news is that ML no longer needs to be a black box, and automation can expedite the documentation and validation of underwriting models.
I don’t think my board and the regulators will sign off on AI lending
One concern credit union executives have is that this new technology may not be 100% safe to use; models might go awry or perpetuate disparate impact and racial or gender bias. Although this received the lowest number of responses in the Fannie survey – only about 9% of all financial executives and 6% of credit union executives rated it a significant hurdle – it perhaps has the most critical consequences. Simply put: failure here is not an option.
The good news is that ML no longer needs to be a black box, and automation can expedite the documentation and validation of underwriting models. Since you can fully explain, document, and validate the way your model arrived at every lending decision, you’ll be able to satisfy all internal risk reviews as well as regulatory compliance requirements.
Stop debating and start implementing
Credit unions have a rare opportunity to seize the high ground of innovation in automated decisioning by incorporating ML into their core business operations. The software has caught up with the concerns that have historically slowed its adoption and are poised to transform how lenders of all sizes market, underwrite, and service their customers. Credit unions have the chance to embrace this game-changing opportunity; they shouldn’t be left behind.