Table of contents:
Summary
- Lenders typically use NAICS or SIC codes for industry identification, which is time consuming and can be highly inaccurate.
- Determining the correct industry for a business is an important part of the risk assessment process, but previously was not automatable.
- Credibly uses generative AI and machine learning to search for relevant information online and automatically determines industry sectors.
- Credibly has filed a provisional patent application for this proprietary innovation.
Setting the stage
In any underwriting process, accurately determining a business’s industry is an important part of assessing risk and affordability for that business. In addition, for reporting purposes, each file must be assigned a code from the North American Industry Classification System (NAICS) and the Standard Industrial Classification (SIC). However, these systems have thousands of codes to choose from, making it both time-consuming to assign and highly subjective.
Previously there was no automated means of assigning an accurate code to a business based upon the information provided on a financing application. Instead, a manual review process was required to research the business and infer which industry it was actually in. While it worked most of the time, this process added time to the file review and increased the risk of human error.
Catalyst for change
To automate industry assignment in underwriting, Credibly first reduced thousands of NAICS and SIC codes down to approximately 80 risk-adjusted industry sectors based on the nature of the business and the types of margins these industries would typically have.
Credibly then developed a proprietary search engine using GenAI and machine learning to gather accurate information about a business. With this context information, GenAI matches the business with the most relevant sector and subsequently, a NAICS and SIC code. The model finally assigns a risk score to each file, allowing underwriters to quickly determine which files need additional review.
How it works
Once a business submits their financing application with basic business information–such as their business name–Credibly’s algorithm will automatically scrape the results of a Google search for the business as well as the business’s website, if found.
Depending upon the type and amount of information available, the algorithm can expand the search with additional context information or verify existing information through further searches. This step often results in multiple outputs, which the model then ranks by relevance.
Credibly then uses prompt engineering to guide a GenAI model to summarize the information from the merchant’s website and search results. Finally, to ensure accuracy, a predictive model uses the GenAI summary of the information gathered online to classify the merchant into an industry. Occasionally, if the model is not confident in its answer, an underwriter will be alerted to manually review the file.
Business Impact
Credibly’s industry determination model is currently being used on 85% of files, which has already reduced file review time by up to 5 minutes per file. As hundreds of files a day are processed automatically, this innovation represents a significant cost and time saving.
Early results are pointing to an estimated 10% improvement in profit and a lift in accuracy from 62% in a fully manual process to 85% with generative AI.
Credibly has also filed provisional patent applications for the innovative methods developed to automate this process. These automations have far-reaching implications for improving accuracy and speed across the entire underwriting process.
