Do you ever wonder how other financial institutions are managing credit risk amidst the uncertainty created by the pandemic? LendIt and Brighterion collaborated on a study of lenders to ask if they were using AI to manage credit risk, what they’re using now and what their plans are for future investment. We learned surprising things.

We will forever remember 2020 as the year fueled by a global pandemic. The aftershocks will continue to make waves for years to come, especially across the economic and financial landscape. This LendIt Fintech survey helps us understand how organizations are thinking about applying artificial intelligence (AI) to mitigate risk in light of the economic shakeup, and where they see the biggest opportunity for better credit risk management.

The current situation is what Tom Bittlemann, Senior Specialist, Product Development at Mastercard, would characterize as a “tail risk event”, a rare occurrence that carries massive impact. He says that for lenders, this type of event calls for innovative technologies that help balance portfolio risk while creating a positive borrower experience.

Bittlemann has come away with five key findings from the study, which he details in our report Using AI to manage credit risk: lenders report on current and future AI use. Let’s look at some of what lenders told us.

Takeaway #1 – 50% of financial institutions and lenders use AI for credit risk

One-half of the financial institutions and lenders (FIs) who responded to the survey already use AI today for a range of functions, and another 25 percent plan to in the future. In fact, 88 percent of respondents say they plan to invest even more into AI in the next 2-5 years.

Bittlemann wasn’t surprised by this finding as Brighterion’s lending customers are seeing major benefits from using AI: improved customer experience, earlier risk detection and tens of millions of dollars in savings annually.

Takeaway #2 – AI/Machine learning usage is on the rise, but rules-based systems are predominantly used to manage credit risk

“Rules-based systems” topped the list of technologies used today to manage credit risk, with “AI/machine learning” coming in second, followed by “manual reviews” and “other” technologies.

Bittlemann points out that rules-based decisioning systems alone lack adaptability, especially in periods of uncertainty. “You can’t capture every rule for managing credit risk during a pandemic. Credit risk management systems that are designed by human experts and leverage AI models will be more resilient. For instance, we need to layer in contextual knowledge of how the recent CARES Act and forbearance programs will impact model features and make adjustments to data that better represents our current reality.”

He also says that Credit Bureau data may be skewed for the same reason. With late mortgage payments protected by forbearance programs, lenders don’t really get a clear picture. “Credit bureau scores are a good lagging indicator of historical behavior, but don’t give a clear picture of what’s happening in real-time like transactional data can.”

The advantage of AI during uncertain times, Bittlemann says, is real-time risk scoring. While historical trends may put economic data in context, only real-time scoring can look at individual trends as they happen.

Takeaway #3 – 32% of FIs fear black box models that are hard to interpret and explain

According to the FIs we surveyed, the biggest concern about adopting AI is the lack of transparency around what they perceive as “black box” solutions. This can be a sensitive topic, as understandably lenders need to know how their technology arrives at decisions.

“That’s always a question we get from financial institutions: ‘Can you tell us why the model arrived at these scores? I need to report back to my chief risk officer and regulators and tell them key factors in each credit decision ranked by importance,’” Bittlemann says.

Model explainability has become increasingly necessary given the potential for adverse outcomes as AI adoption increases. At Brighterion, we score hundreds of billions of events every year and provide a layer of explainability. These scores are used by thousands of banks globally. Not only are we providing users with predictive scores, we are also helping them understand the reasons behind those predictions. This allows credit risk managers to managetheir portfolios on a one‑to‑one level and develop more personalized strategies to improve their borrowers’ experiences.

“AI practitioners are coming to the realization that we need to be more conscious about how we develop models and that complex does not mean better in every case,” Bittlemann says. “Using AI to manage credit risk is a very good example of that. Mastercard and Brighterion’s approach to building models is to be accurate and explainable.”

“FIs, lenders and others in the payments ecosystem are asking constantly for that technology because it’s not offered everywhere, and we are doing it. That inspires a lot of confidence in what we’re doing,” he says.

Our explainability techniques make it easier for FIs to manage their model governance and meet regulatory requirements. By taking a thoughtful and collaborative approach to AI, we build solutions with our partners that they can deploy with confidence.

Takeaway #4 – 26% of FIs are concerned about AI implementation difficulty

We know that building an AI model takes months, if not years, for most organizations. Many try to do it in-house and they aren’t successful. As a result, there’s a perception that adopting AI is a difficult task. But that doesn’t have to be the case.

When building a model with Brighterion AI, we use our proprietary AI Express process and Smart Agents technology to develop custom models. AI Express is a collaborative process that takes only six to eight weeks in which our team of data scientists and subject matter experts help you define your objective, required data and desire outcomes. From there, our data scientists leverage a full stack, state-of-the-art machine learning toolkit to develop features, generate models and deploy the best solution.

As a result, AI Express produces a deployment-ready model to present to the customer in less than two months.

Takeaway #5 – Lenders are equally divided on how they would use AI for credit risk in the future

Here’s an interesting finding that really shows the diversity amongst lenders: 30 percent would use AI for loan origination for new customers, 30 percent would use it for credit risk monitoring and 27 percent would use AI to optimize collections. The other 13 percent said, “other uses” or “none of the above.”

This finding identifies common concerns among lenders. They want to understand the likelihood of a customer defaulting on a loan and how that impacts profitability; what credit limit should be assigned to a customer and the right time to take action; and how they should optimize collection strategies with these factors in mind.

Bittlemann reminds us that risk factors need to be measured alongside revenue drivers like interest income, interchange revenue and fees. That’s where AI is particularly useful as it can inform critical decisions throughout the customer lifecycle, improving the overall customer relationship.

Brighterion is partnered with some of the largest financial institutions across the globe to apply AI for credit risk management. Using this technology, lenders can enhance their credit risk practices across the customer lifecycle.

Learn more about these and other findings in our report Using AI to manage credit risk: lenders report on current AI use and future investment.

Tom Bittlemann manages Mastercard’s AI Express for Credit Risk product, partnering with the largest banks across the globe to develop cutting edge AI solutions. Tom is a trailblazer in the AI space, leveraging his deep expertise in machine learning and systems engineering to craft new approaches to challenging problems in the credit risk space. He and his team are pressure testing these ideas, putting them into practice and saving banks tens of millions of dollars annually.