The marriage of machines and humans is often the driver of business innovation, but for the startup Ocrolus Inc. it is more than that – it is the core of the final products it offers.

With a human-in-the-loop infrastructure, the company transforms documents into actionable data with great accuracy, according to Sam Bobley (pictured), co-founder and chief executive officer of Ocrolus. The startup primarily operates with financial services by “reading” tons of pages of bank statements, pay stubs, invoices, tax forms and other documents that fintech lending companies need to process to decide on customer orders.

“As much of the heavy lifting as we can do with machines, we do, but whatever we can’t do automatically, we slice into smaller tasks and intelligently route those tasks to humans to perform verification,” Bobley said. “We then layer in algorithmic checks to make sure our humans did the review correctly, [the] customer gets perfect results, and that same perfect output is used in a feedback loop to train our machine learning models to get smarter and smarter.”

Bobley spoke with Dave Vellante, host of theCUBE, SiliconANGLE Media’s livestreaming studio, for a digital CUBE Conversation. They discussed the evolution of Ocrolus services, how the startup’s technology complements the robotic process automation system, the company’s pricing model, and its prospects for the market.

From Medicaid to loan applications

Although Ocrolus focuses today on financial services, it has not always been so. Founded in 2014, the company had as its first bet the healthcare industry, using artificial intelligence to process Medicaid applications. However, it quickly realized that its technology was valuable to lenders to help them automate the underwriting process.

“Our thesis is if we can take out all of the necessary evils, like document review, and allow underwriters to focus on the actual analysis of financial health, it’s a win-win and creates a really fantastic complementary relationship between us and our customers,” Bobley said.

The strategy proved to be correct. Lending companies have a greater volume of documents to process than nursing homes and attorneys in the universe of the Medicaid applications, which made it possible for Ocrolus to scale and increase its revenue quickly.

“And then the other thing that happened is kind of, as we were getting deeper and deeper into the space, the fintech space as a whole started growing massively,” Bobley explained. “So, we kind of had the perfect storm of a product market fit plus the market growing that allowed us to really ramp significantly and grow revenue.”

Building an efficient platform first went through the process of realizing that optical character recognition, or OCR, and machine data capture could not do the job completely, since, in general, OCR can process financial documents with no more than 80% to 85% of precision, according to Bobley. The machines struggle especially with semi-structured and unstructured documents in which the format is unpredictable, as well as with lower quality images.

“So, pretty early on, we recognized that we needed human intervention in the process in order to achieve perfect accuracy every single time and also to create training data to constantly teach our machine learning models to get and drive additional automation,” he explained.

Filling in the gaps left by RPA

Ocrolus’ human-in-the-loop infrastructure is complementary to robotic process automation (RPA) because it fills in the gaps left by this technology, according to Bobley.

“RPA companies like UiPath Inc. and Automation Anywhere Inc. typically provide a horizontal toolkit to allow banks and lenders to automate much of the mundane work like, for example, collecting information from emails or doing onboarding for a new employee or different types of tasks that a manual worker would have done, but can be automated with relatively simple code,” he explained.

However, what happens in RPA workflows is that it gets stuck with tasks that cannot be fully automated.

“A robot might be trying to complete an intend lending flow, but when a bank statement is submitted as part of that flow, the robot can’t parse it,” Bobley said. “So, what they do instead is they route it to an underwriter who performs a manual analysis, [place] key information into a back-office system that a bank is using, and that information then gets handed back to a robot and continues the automation.”

Ocrolus’ role in this process is exactly to plug the gaps that used to be manual. Thus, a robot sends documents like bank statements, pay stubs or tax files to Ocrolus and it runs its human-in-the-loop process and returned structured output directly to a robot, which proceeds to the next step in the flow.

“In summary, the combination of robotic process automation and human-in-the-loop, which is what we’re doing, creates true end-to-end automated flows, rather than RPA by itself [that] might get you 80% of the way there,” Bobley pointed out.

For this, Ocrolus has software integrations with UiPath and Automation Anywhere. Although RPA is not as prevalent in its core business, fintech, this technology is common in other areas, such as mortgage loans and traditional banks, which are Ocrolus’ new targets.

“And we’re also expanding use cases,” he said. “Historically, small business lending was the core of our business; more recently, we’ve moved into consumer, auto, mortgage, additional asset classes. And, as we’ve gotten deeper with financial institutions, we’ve seen even more opportunity to partner and coexist with a broader RPA players.”

Ocrolus’ pricing model is usage based, and customers pay either per application, per document or per page. In this way, it is possible to subscribe for the processing of one document or millions of documents per month. Other important characteristics are scalability and flexibility, which make the company able to flex up and down to meet the demand.

Watch the complete video interview below, and be sure to check out more of SiliconANGLE’s and theCUBE’s coverage.

Photo: SiliconANGLE