Back in March, amid the economic fallout from the COVID-19 pandemic, Congress passed a $2 trillion relief package—the Coronavirus Aid, Relief and Economic Security (CARES) Act. It initially included $349 billion in forgivable loans for small businesses that were hampered by shutdowns to prevent the spread of the novel coronavirus.
But as businesses scrambled to apply for the loans under the Paycheck Protection Program —meant to incentivize companies to keep employees on the payroll—a major complication emerged. How could traditional lenders, like banks and community development financial institutions, possibly keep up with the influx of applications, which quickly skyrocketed into the millions?
Cognistx, a Pittsburgh-based startup located along Liberty Avenue, devised an intelligent solution: an automated system that uses machine learning to sift through the applications in mere milliseconds. CEO Sanjay Chopra said that the aim is to get those applications rolling and disburse the money as quickly as possible because the funding is sink-or-swim for so many small businesses.
The system saw its first test in Florida’s Miami-Dade County over the summer. There, county commissioners used coronavirus relief funding to set up a $25 million micro-loan program to award low-interest loans of up to $30,000 to businesses with less than $2 million in sales and fewer than 25 workers. While not directly tied to the federal Paycheck Protection Program, the RISE Miami-Dade Fund may serve as a model for the broader coronavirus loan ecosystem.
“Little did we realize…we’d have 50,000 applications in two to three weeks,” Chopra said. “Everyone wants a loan to survive, [so] we asked what we could do, like a [Customer Relationship Management tool] that can look at all of the applications.”
Typically, it can take anywhere from 30 minutes to one hour to process a single loan application, Chopra said. That’s because a human must manually rifle through stacks of supporting documentation, from lease agreements to sales disclosures, tax returns, payroll, and then some.
To boil it all down into something more manageable, Chopra’s team built a system that uses natural language processing—a subset of machine learning closely related to linguistics that helps computers to understand human words, rather than just numbers or code.
That, in turn, helps the system parse massive amounts of unstructured data, or information that is not organized in some predetermined fashion. A tax return, for instance, has a very regimented series of input fields, making the form routine and simple to pull data from, while a letter describing the company’s personal story of financial hardship may contain useful data that is more difficult to isolate.
Another layer of the system helps to ensure that the loans were doled out to businesses across Miami-Dade County in an equitable manner by examining the zip codes of each business address.
Since July 6, Cognistx has helped the RISE Miami-Dade Fund process more than 2,000 small business loan applications, according to Jagriti Pandey, a product management intern at Cognistx. Once the applications make their way through the automated system, a team of human reviewers makes the final call on which businesses will receive funding, keeping the people in ultimate control.
Now, Cognistx is expanding that work to include a credit union in Miami-Dade County for other types of loan applications, from student loans to mortgages, to potentially another round of PPP. Chopra said that, in theory, any traditional lender could implement the system, including banks—but it’ll be a challenge to bring those clients on board.
“It’s harder to integrate new fintech technology with antiquated bank systems, and to some extent, an antiquated mindset,” he said. “Domino’s Pizza can show you where your pizza is from oven to box…the same needs to be true for people who have applied for loans.”
Photo Credit: Cognistx
Knowledge is power!
Subscribe for free today and stay up to date with news and tips you need to grow your career and connect with our vibrant tech community.