Innovation in Lending
Build vs. Buy: What are the Costs of Getting AI Lending Wrong?
May 17, 2021
As more banks and credit unions focus on loan growth, leaders recognize new technology is needed to support new tactics. To adapt and thrive, lenders are considering machine learning-based underwriting to boost approvals with no added risk, provide a superior digital experience with auto-decisioning, and respond more quickly to market volatility. cv
It’s an age-old dilemma but for ML underwriting, the decision is particularly nuanced. Many factors are overlooked which is why those who decided to build their own end up with an ML model they can’t deploy. In fact, AI project failure rates are at 53%, according to a Gartner report.
ML underwriting requires a new build vs. buy framework
In short, a general technology framework that works for an AI chatbot build vs. buy decision won’t serve you well for ML underwriting. Yes, the typical factors (costs, timeline, resources) still apply but the risk, regulatory, costs of getting it right or wrong, and deployment considerations raise the bar and require a new lens.
So, what are the costs of getting it wrong?
The risk and regulatory considerations for MLunderwriting demand a new way of looking at the build vs. buy decision. To highlight the increased exposure to risk differences, let’s look at an AI chatbot. If your DIY AI chatbot makes wrong predictions and delivers incorrect messages, the impact is a bad customer experience. While less than ideal, it’s definitely not catastrophic to the business.
The same can not be said for DIY AI underwriting. Deploying an ML model that hasn’t been properly de-biased exposes the company to expensive fines from the Consumer Financial Protection Bureau (CFPB), negative publicity, and considerable brand damage. In fact, the CFPB is poised to renew tough industry oversight, placing more pressure on banks and credit unions to make sure their models meet fair lending standards.
Financial institutions also need to factor in the risks of a toxic portfolio. An ML model that isn’t built and validated correctly can lead to inaccurate approvals of borrowers whose inability to pay won’t come to light until it’s too late. For leaders building their first AI capability, calculating how much a year of bad loans would cost needs to be included in your decision process.
Download build vs. buy guide
The stakes are high and without the proper explainability and validation expertise and experience, the DIY approach is inherently riskier. Making the wrong decision could have severe consequences to the long-term success—or even viability—of the business. To ensure success, check out our latest guide to get an overview of available options in the marketplace, a complete list of factors to consider, and an understanding of the benefits of getting ML underwriting right.
Zest AI team
June 2, 2021