Smith School’s AI Collaboration with Google Sets New Standard in Credit Risk Evaluation

In 2024, Smith Master of Quantitative Finance (MQF) students, supported by the Smith Enterprise Risk Consortium (SERC), participated in an experiential learning project (ELP) aimed at rating the effectiveness of bank and fintech company credit risk management capabilities and their credit outlook for various consumer and commercial asset classes. 

This spring, MQF students at the University of Maryland’s Robert H. Smith School of Business partnered with Google to deliver a groundbreaking approach to examining credit data and financial statements.

Led by Professor of the Practice and SERC Director Clifford Rossi, students collaborated with SERC Advisory Council member and Google Compliance and Risk Management Director Pedro Morales, as well as Google Analytics specialists Fuad Alam and Rikesh Patel. 

Together, they leveraged the company’s suite of AI tools, including Gemini, AI Studio and Notebook LM, to analyze over 25,000 pages of public financial documents and develop ratings across banks of different sizes and fintech companies. Through a rigorous prompt engineering process augmented with expert judgment, the students built a set of ratings for each bank and fintech company in the sample over eight quarters.

The students trained their models on a repeatable basis to ensure the consistency of the ratings over time and further validated them against the GFC period for several banks that failed, as well as a group that did not fail.

They found that their tool could effectively rank-order the firms in both benign and stressful credit periods, as well as identify differences in outlook across loan types over time and between banks and fintechs.

According to Alam, this project represents “uncharted territory from a technical and conceptual perspective,” pushing the tools to their limits and achieving a breakthrough in recreating a financial analyst agent using AI models.

Patel, in particular, was impressed by the students’ ability to quickly familiarize themselves with the suite of tools and synthesize years of data to drive insights.

“There’s something about students who are really open-minded and looking to soak it all in like sponges,” says Patel. “It was refreshing to see their willingness and ability to take on a new challenge at a time when AI is a revolutionary landscape, and we need to integrate it into all the work that we are doing.”

The project, albeit a tangible example of how AI tools can be leveraged by investors, regulators and institutions to analyze large quantities of data for credit performance and outlook, further demonstrates the need for financial and risk domain expertise to fully extract the tool’s capabilities. 

Morales, who headed the Large and Foreign Banking Organizations analytics team at the Federal Reserve Bank of New York earlier in his career, was impressed with the students’ thoroughness and ability to grasp complex concepts quickly. This was especially evident in how they incorporated the suite of Google tools to glean financial insights that generally are more apparent to seasoned industry professionals, he says.

“If you use AI in the ways that we’re trying to promote, it’s amazing how quickly someone can learn specific topics or concepts,” says Morales. “I observed that with the students. The way they were able to explain their reasoning and understanding of these topics did not mimic someone who just came out of college, but rather someone who knows what they’re doing in this space.”

On July 15, the students will present the ELP once again, this time to Google senior leadership at a higher education and AI event held at the company’s New York office. 

During the event, they will showcase the culmination of their efforts, which can ultimately positively impact the industry by helping identify warning signs of potential adverse financial events, as their model continues to be refined.

“There’s no book that the students could read on how to do this, but they were still able to generate seriously useful insights,” says Alam. “If we are able to scale something like this at an industry level, it could change the game in terms of how we look at reading financial disclosures in general.”