How to Improve PPP Lending Efficiency During COVID-19 with Amazon Textract

This is a guest post by Anthony Sabelli, Director of Data Science at Kabbage, a data and technology company that provides cash flow solutions for small businesses.

foreword

Kabbage is a data and technology company that provides small business cash flow solutions. One of the services we offer our clients is to provide them with flexible credit lines through automation. Small businesses connect their real-time business data to Kabbage for fully automated financing decisions in minutes, an efficiency that has driven us to provide more than $16 billion in working capital to more than 500,000 small businesses, including the Paycheck Protection Program (PPP) .

When COVID-19 began, small businesses were forced to close due to the nationwide shutdown of the United States, and we had to overcome multiple technical challenges while dealing with new and changing underwriting standards that became known as the Small Business Administration (SBA). ) the largest federal relief effort in history. Before the PPP, Kabbage had never made any SBA loans. However, in about 2 weeks, the team built a fully automated system for all eligible small businesses, including new customers, regardless of size or reputation, to access government funding.

Kabbage's underwriting has always been based on customers' real-time business data and income performance, rather than payroll and tax data, which are the main criteria for the PPP. The Internal Revenue Service (IRS) does not have an established API to help automate the verification and underwriting process, and we need to fundamentally adapt our systems to help small businesses get their money as quickly as possible. In addition, our team of only a few hundred members joins thousands of seasoned SBA lenders with hundreds of thousands of employees and trillions of USD at their disposal.

In this blog post, we'll share how Amazon Textract https://aws.amazon.com/textract/ helped 80% of Kabbage PPP applicants have a fully automated loan experience and reduced approval time from days to 4 hours Median speed. By the end of the program, Kabbage had become the second-largest PPP lender in the nation (by number of applications), surpassing major U.S. banks (including Chase, the largest bank in the U.S.), serving more than 297,000 small businesses, and serving more than 297,000 small businesses in the U.S. Some 945,000 jobs were retained.

Implement Amazon Textract

As one of the few PPP lenders accepting applications from new clients, Kabbage attributed the increased demand to a large number of small businesses unable to apply for loans from long-established banks, turning to other lenders.

Businesses are required to upload documents such as tax documents, proof of business documents and identity forms, and initially, all loans are underwritten by humans. Personnel must review, verify, and enter values in various documents to substantiate the stated wage calculations, and then submit an application to the SBA on behalf of the client. In just a few days, however, Kabbage received hundreds to thousands of submissions from tens of thousands of small businesses, and the number quickly climbed into the millions. This task needs to be automated.

We need to break it down into parts . Our systems are already very good at automating the verification process – commonly referred to as Know Your Business (KYB) and Know Your Customer (KYC), which allow us to allow applications from new businesses that collectively account for the Kabbage PPP 97% of the number of customers. In addition, we need to standardize the loan calculation process so that we can automate the extraction, verification and review of documents to extract only the appropriate values required for loan underwriting.

To this end, we have compiled loan calculations for different business types, including sole proprietors and independent contractors, which together account for 67% of our PPP customer base, based on specific values from various IRS forms. We launched an initial classifier for key IRS forms within 48 hours. The final piece of the puzzle is accurately extracting the value to issue a loan compliant with the program. Amazon Textract played an important role in overcoming the last hurdle . We went from POC to full implementation in a week and full production in two weeks.

Integrating Amazon Textract into our pipeline is easy. Specifically, we used StartDocumentAnalysis and GetDocumentAnalysis , which allow us to interact asynchronously with Amazon Textract. We've also found that the FeatureTypes table is great for working with tax documents. Finally, Amazon Textract is not only very accurate, it also scales to handle large backlogs. After integrating Amazon Textract, we were able to clear the backlog. This remains a critical step in the PPP process until the end of the program.

Significant impact on small businesses

To put that in perspective, Kabbage customers, including nearly 60,000 new customers, took out nearly $3 billion in working capital loans in 2019. In just 4 months, we have provided more than double the amount of funding (USD 7 billion) to applicants with approximately 5 times the number of new clients (297,000). The average loan size was USD 23,000, the median loan size was USD 12,700, and more than 90% of all PPP customers had 10 or fewer employees, suggesting that businesses are often most vulnerable to a crisis but are seeking financial assistance was ignored. Kabbage’s platform enables it to serve far-reaching and remote areas of the United States, offering loans in all 50 U.S. states and territories, with one-third of loans made to businesses with an average household income of less than $50,000 (by zip code) .

We're proud of what our team and technology have accomplished with so few resources , yet surpassing America's largest banks . Large US banks have an average of 790 employees per bank, while Kabbage has only one. However, our loan volume exceeds those of the big banks, providing more than $7 billion in loans to nearly 300,000 of the smallest businesses in the United States.

the way forward

Kabbage is always striving to find new data sources to enhance our cash flow platform, thereby increasing access to financial services for small businesses. Amazon Textract gave us new capabilities ; we had never extracted value from tax filings before the PPP. This helps us enrich our underwriting model . When helping our clients access capital, this adds another perspective to understand a small business' financial health and performance, and provides greater insight into its cash flow to help build stronger businesses.

in conclusion

COVID-19 has further revealed that, despite making up 99% of all companies, providing half of all jobs, and accounting for half of non-farm GDP, the U.S. financial system is underserved by small businesses. Technology can solve this problem. This requires creative solutions (such as those we build and deliver for PPPs) that fundamentally change customer expectations for future access to financial services .

Amazon Textract is an important feature that has enabled us to successfully become the second largest PPP lender in the nation , providing funding to so many small businesses when they need it most. We've found the entire process of integrating the API into our workflows is simple and straightforward, allowing us to spend more time ensuring that more small businesses, the backbone of our economy, have access to critical capital when it's most needed.

Author of this article

Anthony Sabelli

Head of Data Science at Kabbage

Kabbage is a data and technology company that provides small business cash flow solutions. Anthony holds a PhD in Applied Mathematics from Cornell University and a BA in Applied Mathematics from Brown University. Anthony leads the global data science team at Kabbage, analyzing more than 2 million real-time data connections from small business customers to improve business performance and underwriting models.