How the credit industry is embracing AI

In broad strokes, credit and risk are impacted by AI. They are areas of the economy, drawing together disparate data sources to make million-dollar decisions. Firms fight tooth-and-nail for new data sources and analysis techniques, as even seemingly small advantages can quickly reshape the market in their favor. The opportunity posed by AI is undeniable and the credit industry is starting to take notice byt by exploring the latest generation of AI models. While credit professionals begin to integrate generative AI tools into their businesses –many still hold concerns about how to manage the risks that can be introduced by AI.

 

On June 12th, S&P Global Market Intelligence called together nineteen senior credit, risk and investment professionals to talk about how generative AI is changing their industry. To allow the participants to speak candidly, the meeting was conducted under Chatham House rules, and as such none of the discussion will be attributed by name. But the conversation provided a surprising view of how new breakthroughs in large language models are changing the daily business of credit.

 

The largest single barrier to automation is the sheer size of many decisions. When extending a $20 million line of credit, it’s not enough to have good reasoning and well-curated data; a firm also needs a system of accountability if things go wrong. As one attendee put it, “you can’t fire a machine.” As a result, there’s still widespread hesitancy within the industry about integrating generative AI systems into the big-ticket deals that define much of the industry, and specifically in a decision making capacity.

 

At the same time, large AI models are already serving a similar function in fraud-monitoring, where firms like Visa and Mastercard use AI to flag small-dollar transactions as potentially fraudulent. Those systems are run by different firms in a different sector of the industry, but it’s hard to avoid the parallels. In each case, the model is ingesting vast amounts of data (including internal information that is proprietary to the firm) and using it to assess which deals should be scrutinized. But while such systems are commonplace in consumer finance, they’ve been more difficult to incorporate into the higher levels of the credit system, even in supplementary roles.

 

Current projects tend to have more modest goals. One firm is exploring the use of AI to generate credit memos, a regulatory requirement that consumes a significant amount of workforce hours. While the memos themselves typically follow a set formula, gathering the necessary information can require digging through PDFs and other documents. It’s an ideal task for contemporary LLMs that specialize in retrieval-augmented generation (or RAG), in which a model draws on external data to respond to queries. If RAG models could reliably produce credit memos, it would lift a huge burden on credit firms, freeing up more resources for the core task of assessing credit risk.

 

The industry has also shown some appetite for incorporating AI into the research process, far upstream of the decision to extend credit or invest. Credit professionals rarely have time to assess all the opportunities available to them, so an AI filter that surfaces the most appealing deals would be extremely useful. As one participant put it, simply reducing a backlog of 100 opportunities to the 30 most interesting could completely change the attention that can be given to a single deal. This would be similar to how AI systems are used for content moderation by companies like Meta, using AI systems as a precursor to human review. It could also be seen as the reverse of the fraud-monitoring system, highlighting the best credit opportunities rather than flagging the worst. In either case, there is significant expertise outside the industry that could be brought in to oversee implementation. S&P Global Market Intelligence’s own CreditCompanion ™does just this, by synthesizing insights across a vast array of credit research to aide in discovery.

 

As in most industries, there is real anxiety about how the growth of AI will affect human workers. If AI systems are ever trusted to handle credit decisions alone, the entire credit industry could be automated away – an alarming prospect for any industry. But as credit firms slowly adopt new technology, a more humane possibility is emerging, with generative AI systems handling routinized, quantitative work while human analysts focus on the qualitative aspects of credit analysis. As automated systems get better at surfacing and synthesizing financial data, analysts’ first-hand knowledge of their industry will become all the more valuable. It’s a more likely future than wholesale automation – and for many of the professionals at the event, it was an exciting future to contemplate.

 

---

To learn more, please visit S&P Global website: www.spglobal.com