One of the major challenges in scaling digital finance solutions is handling old school paper. Even with all the automation and digitization, documents still need to be processed and in many cases, humans need to be looped in.
Sam Bobley is the co-founder and CEO of Ocrolus, which works with firms like BlueVine, Brex, and PayPal to analyze financial documents in any format.
Sam joins me on the podcast to talk about the challenges of balancing speed and accuracy in data extraction in fintech and how the Payroll Protection Program really highlighted the leaders in the space based on their ability to scale up quickly. Sam shares his view on what it takes for the financial industry to move beyond paper to a fully automated ecosystem. He also shares some insights on where he thinks the market is headed as well as Ocrolus’ product road map.
Sam Bobley is my guest today on the Tearsheet Podcast.
My name is Sam Bobley. I’m the co founder and CEO of Ocrolus, and we are a fintech infrastructure company that analyzes financial documents of any format or quality with perfect accuracy.
When we first started building Ocrolus, we were very surprised that there was no technology in the market that could analyze financial documents with a very high degree of accuracy. There were dozens of products out there in the optical character recognition, OCR, data extraction and data capture space, but almost all of them were insufficient. When it came to accuracy, they cap out at roughly 80% or 85% accuracy plus or minus, depending on the complexity of the document.
We saw an opportunity to build a unique solution that combines machines and human intelligence to process every document we see with outstanding accuracy. And once we realized that we had a new mousetrap to solve this problem, we started realizing how many amazing market opportunities there were for the company. We kind of stumbled into the fintech lending space as our focal point.
The challenge for lenders is the balance between speed and accuracy. There are a few different ways that lenders can obtain financial information from borrowers. There are digital connections offered by companies like Plaid, Finicity and Yodlee, where a borrower can credential in access to their bank account and port information digitally over to the lender. That’s fantastic. And when it works well, it’s a very seamless process.
But in many cases, a lender isn’t able to get that digital connection. And they have to review documents. In mortgage, for example, there are hundreds of pages of documents that are submitted for every loan application. And the lender has the task of reviewing those documents, essentially page by page, line by line, to figure out all of the pertinent financial information and to make a decision. And as part of the process, the lender has a trade off decision: how quickly do they want to go through the documents versus spending more time to go through the documents more diligently and focusing on the accuracy. With Ocrolus, we created a solution that allows a lender to both optimize the speed and accuracy of the review process.
Standardization — data normalization — is one of the biggest challenges. Bank statements and pay stubs can come from hundreds or thousands of different financial institutions. Each has its own format and layout. It’s difficult to make all sense of all that data in a normalized format. And then when you throw in the digital connectivity — data being ported in directly from financial institutions — you get this kind of hodgepodge of different, diverse data streams.
What Ocrolus helps lenders do is normalize the data: data from documents of any format or quality, as well as data directly imported from a financial institution, regardless of where it comes from. Ocrolus can help the lender normalize the data into cash flow scores, income calculations and analytics that help determine the borrower’s financial health.
Digitization of financial services
We’re definitely making progress. I think we’re at least getting from paper to the PDF. In terms of going full digital, I think it’s going to take longer. We are now seeing some forward momentum towards open banking — we believe that consumers should own their data and should be able to port their data digitally to a lender or to a bank. Ultimately, data portability and fast and accurate review of data ultimately help the end consumer in terms of better price transparency and accessing credit products at a lower cost. I think in order to get to that full financial nirvana, where data can be ported from one system to another, there are a lot of complexities to get there. Ocrolus focuses on being the bridge — we want to provide all of the digital options where possible.
There’s a regulatory component of the government mandating standards around open banking. And secondly, there’s a technical component, and particularly in the US where the financial landscape is so fragmented (there are literally thousands of financial institutions), it’s very difficult for all of those institutions to all of a sudden have modern technical architecture in order to share their data in an appropriate way. There are going to be a number of steps in order to ultimately get to a world where data can be ported from one system to another.
We power the fintech lending ecosystem. We have more than 100 customers in the fintech lending space. We started in the small business lending space. Our core product is a bank statement analysis product that performs cash flow scoring for small businesses. And we have many of the leading fintech small business lenders on the platform — folks like PayPal, Square, Brex, Bluevine, and Enova. Most of the big players are now using Ocrolus as part of their underwriting automation.
We used the small business lending market as the launching pad to do the same thing in other asset classes. So after that, we built out an income verification product where we could analyze pay stubs and W2s. We launched with several of the leading consumer fintech lenders, like SoFi and LendingClub. And then even more recently, we said, hey, the world’s changing and traditional financial institutions are now moving towards building digital lending flows. So our goal was to do the exact same thing on a larger scale with mortgage lenders and traditional banks. We’re earlier in that part of our journey, but the macro tailwinds around process automation, as well as modernizing credit decisioning, are really playing to our favor. Today we’re focused on launching our product suite with mortgage lenders and banks, and continuing to grow our fintech lending customer base, as well.
Go to market
We’re fortunate now that at this stage — the company has been around for seven years and we’ve raised many millions of dollars in venture capital — we now have a very mature sales and marketing organization where we go after customers from every channel possible. But in the earlier days, we didn’t have that. We got in contact with many of our early fintech lending customers through warm introductions through our network. We have fantastic investors like QED Investors, Fintech Collective, Oak HC/FT, and others who helped us get in touch with these decision makers.
I think why we’ve been successful is the product sells itself. When we get in contact with these fintech lenders, we encourage them to do an A-B test to compare the manual review process that was the traditional way that they were analyzing documents, versus our API solution where they could send us the documents and we just send them clean structured digital data back.
Initially, it was more of a product or operations person. I think the most obvious value proposition for Ocrolus is the opex, just reducing the cost and increasing the speed of doing this work. Over time, as our company has gotten bigger and our technology has gotten more sophisticated, we’ve really evolved to targeting the chief risk officer.
About 18 months ago, we brought on a guy named David Snitkoff, who was the former co-founder of Orchard Platform. He was really a role model for me when I was first entering the fintech space and learning about fintech lending. He sold Orchard to Kabbage, and he went on to run data strategy and analytics at Kabbage. And then about 18 months ago, we were fortunate enough to get him to come over to Ocrolus and join and build out our analytics function.
Because we take every document and turn it into highly accurate data, it puts us in an awesome position to then layer on fraud detection, cash flow scoring, and additional business intelligence on top of the data — that component of our business has become much more mature over time. And as the analytics component has become more mature, it’s changed the conversation completely. We’ve now been able to demonstrate with certain fintech lender customers a 20% lift in predicting default, and up to a 3x to 3.5x improvement in our ability to detect fraudulent applications. When you can elevate the conversation to managing risk and really impacting portfolio performance. It’s a different ballgame.
I think the PPP program really highlights the benefits of an infrastructure technology. When PPP first hit, banks and lenders were figuring out how they would process an unpredictable amount of applications. In the normal course of lending, the lender usually has at least a rough idea of how many applications might come through the doors over the course of a month. Obviously, it can fluctuate plus or minus a bit, but with PPP, they didn’t really have a good sense of how many applications would come through the door. So you have this classic problem of trying to match supply and demand. What many of the big banks did is, without exaggeration, they hired hundreds or thousands of temporary workers to come in and sift through PPP-related documents, documents like 941s and 944s, payroll reports and other documents that small business owners were required to submit as part of the PPP application.
They had massive backlogs. The big banks were taking days or weeks or months to review applications. Many of the small businesses, particularly the main street businesses, your pizza store, or your florist or nail salon, weren’t getting attention from the big banks because the big banks were catering to their large customers. In contrast, there were a handful of fintech lenders who leaned in to participate in PPP.
In the first PPP, we launched with customers like Cross River Bank, Square and BlueVine and several others. And in particular, Cross River Bank became a major success story for us. In 2020, Cross River actually became a top three PPP lender nationally in terms of PPP loans originated. They outperformed many of the big banks. Folks like TD Bank, Citizens, US Bank, PNC did fewer loans than then Cross River did. Cross River had Ocrolus in the back end, providing this elastic function to flex up and flex down and automatically handle any application that would come in. We were able to process many of the Cross River loan applications, the transfer bank loan applications within eight to 12 minutes, which allowed them to fund within 24 hours. Iit was exciting for the company from a corporate social responsibility perspective to be able to help small businesses across the country, keep the lights on and it was just a really cool and rewarding experience to go on to Twitter and see small business owners talking about the great success they had applying to fintechs, like Cross River and Square and BlueVine and others, and having really smooth borrower experiences.
We continued to do work for the PPP program in 2021. We launched with several other fintechs, including Womply, who became one of the largest PPP lenders, nationally. We achieved massive scale with those guys, and believe it or not, by March or April, in the tail end of the PPP program in 2021, Ocrolus was processing something like 65% to 75% of all PPP loan applications nationally. We ended up doing it at a very large scale. And I think the value of automation and human and loop processing versus traditional manual page by page review was really exemplified.
We’re deep in the mortgage space. Our initial products, our bank statement and income verification products for fintech lenders, were only covering a few document types: bank statements, pay stubs, W2s, etc. In the mortgage world, there are quite literally hundreds of different document types that a borrower may be asked to submit. There are a lot more nuances and edge cases and different things that you need to handle. We’re pretty far down the path already. We support dozens of different document types.
We have a few dozen mortgage customers already on the platform and successful. But there’s a lot more investment to do in terms of really polishing and perfecting our mortgage product. And that’s the focus point of the business. I think that the timing is great in the sense that the mortgage system was really stressed in 2020 with similar problems, like difficulty matching supply and demand and having to process loan applications in a work from home environment due to the pandemic. And because of the various stresses on the mortgage system, it’s clear that coming out of the pandemic, many of the mortgage lenders are prioritizing process automation and really figuring out ways to add efficiency and scale to their back office. And we are right there to meet the market and to help introduce further automation in the mortgage and banking space.