GenAI Revolutionises Credit Underwriting: From data to credit memo creation

The advent of generative artificial intelligence (GenAI) has opened new chapters in the world of LLMs.

By Mohit Modi and Viral Vora

Artificial intelligence (AI) is all good, but generating a credit memo is beyond its realm, purists might argue. Not anymore. The entire credit decisioning value chain — underwriting, memo and all — is in the throes of a transformation.

To be sure, credit underwriting has been perhaps the most underrated of all credit decisioning processes.

Adoption of technology in credit underwriting has lagged other spheres despite the many benefits it offers ­— from standardising the process to instituting guardrails and controls in the credit decision process to bringing in industry data as a comparative reference point against which credit proposals are viewed.

Technology also enables development and implementation of industry-specific models that factor the nuances of an industry in the credit decision process.

Advances in technology have integrated credit underwriting platforms with origination systems, allowing seamless data collection from public sources and simplifying document data extraction. This integration helps credit officers focus on analysing credit proposals rather than gathering information. Digitised data and automation have also standardised financial computations, such as financial spreading. Embedding workflows within underwriting platforms ensures that the entire lifecycle of the underwriting process is tracked, providing historical insights and enabling portfolio analytics.

Technology extends beyond decisioning, with continuous monitoring of portfolios offering early indications of borrower stress, allowing for proactive interventions. The real game-changer, however, is generative artificial intelligence (GenAI), which is revolutionising credit assessment, especially the credit memo creation process.

GenAI opens new vistas

The advent of generative artificial intelligence (GenAI) has opened new chapters in this storyline.

The real potential of GenAI is in leveraging large language models to generate parts of credit memos, which capture details of a company under consideration, including industry structure, business details, financials, and management capabilities. By working with large global banks, some platforms now use GenAI to assist in portfolio credit assessments.

However, leveraging GenAI in this context presents two main challenges: hallucinations and information loss. Hallucinations occur when the GenAI model produces content not present in the input data, while information loss happens when it overlooks or misinterprets relevant details. To address these issues, GenAI solutions must incorporate strong guardrails to identify and correct hallucinations, and safety nets to ensure no critical information is omitted.

Another challenge lies in the varying relevance of metrics across industries. For example, a pharmaceutical company may prioritise metrics like drug pipeline status, while a telecom company may focus on metrics like average revenue per user. Designing GenAI systems that can dynamically adapt to different industries is key to overcoming this challenge.

GenAI-embedded credit underwriting can boost productivity

We can take such a solution to the next level by embedding GenAI capabilities and making it a human-in-the-loop platform.

Credit risk models generally evaluate the borrower’s rating across quantitative and qualitative parameters. Quantitative parameters use ratios and financial inputs generated via financial spreading for computing the financial risk. Qualitative parameters require the analyst to score the borrower using non-financial parameters. These cover risks such as business risk, industry risk and management risk. The process requires the analyst to understand the risk and do a subjective assessment of the borrower. For this, the analyst is required to read and understand large documents such as annual reports, earning transcripts and management presentations. GenAI can significantly simplify this process.

The credit analyst can upload documents related to the entity through the GenAI Module. The Platform leverages its prompt libraries for identifying the prompts applicable for each GICS sub-industry. Based on the sector and the industry, the platform sends borrower-specific prompts along with the documents uploaded, to a purpose-built GenAI engine, which prepares appropriate responses to questions.

The prompts in the prompt library are tagged to individual qualitative credit risk parameters. This helps the analyst to get contextual inputs from the uploaded documents and enables the analyst to score qualitative parameters.

Once the assessment is completed, the analyst can generate a comprehensive, integrated report that has the financial spread, credit rating and GenAI data. This extensive report will help credit committees deliberate and decide on credit actions and assist the credit analyst in quickly preparing credit appraisal memos. In the process, the credit analyst spends far more time on value added activities such as risk and credit assessment rather than on searching relevant information from hundreds of pages and preparing the first draft of the memo.

The advent of GenAI promises to unleash the next wave of transformation in the credit assessment process, taking forward the gains from adoption of technology so far.

An idea that sounded straight out of science fiction is now a reality.

Mohit Modi is Head of Strategy and Corporate Development, and Viral Vora is Senior Director, Data Analytics, CRISIL Limited.

Views are personal, and do not represent the stand of this publication.