Three things lenders need to know before adopting machine learning

I’ve worked with many industry leaders who successfully introduced the idea of AI/ML into their businesses. Most of them already have an idea of the potential positive impact this change will bring, but don’t know how to navigate the corporate bureaucracy to get to a yes on investing in the technology.

Fundamentally, all stakeholders in your lending business want the same thing — positive business impact — but how do you get AI/ML into production so you can start generating that impact?

Here are the top three things I’ve learned that might help you influence your organization to adopt AI-automated underwriting.

‍

AI underwriting works, the question most people have is around how significant the impact will be on their business

In my experience working with lenders across the globe and asset categories, the minimum benefit I’ve seen from machine learning and AI is a 15 percent reduction in net credit losses. Yes, 15 percent, and that is the smallest benefit I’ve seen.

If you take all of the gains from increasing approval rates instead of reducing losses, I’ve seen improvements in a range from eight to 12 percent. So AI will deliver significant benefits to your lending business, either by reducing credit losses or increasing originations.

There are several ways that you can gauge the impact AI can have on your business. One classic method is to run a proof of concept, but a lender can also see benefits from data optimization and automation, and workflow management. The proof of concept path involves building an AI model, comparing the performance to your current model, and analyzing the incremental business impact that the AI can deliver. You can do this on your own using widely available open source modeling packages, or you can also hire a technology partner that brings deep expertise and experience in AI underwriting. Either way, you will ultimately discover that AI credit decisioning will add significant value to your lending business.

‍

The real challenge is getting the AI model into production

There are several challenges to getting AI models into production, including IT integration and regulatory and compliance concerns. In my experience, we have found suitable, low-friction IT solutions that are agnostic to whether a lender prefers to run its solution on-premises or in the cloud.

IT should not be a considerable blocker if your organization is genuinely committed to adopting AI.

If you were to ask me for my opinion, the real challenge is around model explainability to meet model risk management requirements and satisfy existing compliance and regulatory frameworks, specifically fair lending and adverse action requirements. To capture the full power of AI, you will want to find technology solutions that provide full explainability for even the most sophisticated machine learning algorithms, including deep neural nets. The math to accomplish this is complicated, but there are solutions. If you have a model that you want to get into production but require explainability tools, then you should consider a technology partner that has addressed this problem.

‍

AI adoption requires significant change management

If you are the CEO of your lending company, then you can likely marshal the required change to drive AI adoption. If you are not the CEO and want to influence your organization to adopt AI, then you’ll need a good strategy.

In my experience, having a VP-level internal champion is necessary. This person will need to be able to influence up to the CEO and maybe the board and down to the user level, cutting across all areas of the organization including credit, analytics, IT, legal, compliance, and operations. If this is you, then great. You can get going. If this is not you, then you need to find a VP-level champion with strong influencing skills.

My best example of successful AI adoption occurred when a junior analyst brought his idea of transformational organizational change through AI to his senior vice president. The trust embedded in the fabric of that organization — even at a junior level — allowed the key stakeholders all the way up to the president of the organization to align behind adopting AI credit underwriting. That lending business is now enjoying close to 30 percent reduction in net credit losses.

At Zest AI, we believe every organization is ready for AI credit decisioning. At the same time, we recognize that change is difficult, but we're ready to help you through the process.