Automated Document Analysis Works Best When Paired With Human Oversight

With digitization and automation as the order of the day for two years and counting, it’s unusual to hear about how important human decision makers remain in business lending.

The trend could hold over the next several years as lenders and their technology partners find the ideal balance between purely digital decisions and those needing a human mind.

Saying that finance is entering “a golden age of automation” Sam Bobley, co-founder and CEO of document automation platform Ocrolus, told PYMNTS that, “What makes us most unique is that we have a human-in-the-loop component to our software.”

The “human loop” kicks in after Ocrolus ingests documents, using automation to read and validate as much information as possible.

“Then any data fields that we can’t automatically confirm we slice into smaller tasks and route through a quality control process where Ocrolus employees perform quality control,” he said, “and then algorithmic checks make sure our workers did the work correctly.”

He explained that for lenders, the pandemic worsened a historic “game of supply and demand” when refinancing skyrocketed, fueled by favorable interest rates. The result was operational turmoil as mortgage lenders were faced with processing incoming documents — often working from home. The resulting backlog found many patching together virtual private networks and other stopgaps, while others began tapping into the human loop potential for greater speed and accuracy.

It was a turning point for the concept.

“It’s universally accepted today that machines alone are not sufficient for reading complex documents. It’s exciting to see other companies, investors, and even some of the big players in the industry are now buying into this concept of human in the loop.”

See also: Ocrolus Gives Flawed SMB Credit Scoring System A Cash Flow Based Upgrade

Benefits of a Scalable Back Office

The human loop is a last line of defense, but also somewhat antithetical to the “golden age of automation” mantra. Lenders that staffed up to clear their application backlogs now face the issue of not enough work for staffers brought on during the refi boom.

The middle ground is targeted use of application programming interfaces (APIs) to clear as much work as possible via automation before calling for human backup.

“The pitch we’ve been able to make the lenders is, ‘If you could offload that, you can be a lot more aggressive with marketing channels and experimenting about ways to get new volume in without having to worry about the flex up/flex down of your back-office staff.’”

Ocrolus does this with a four-step process that ingests and classifies documents, the “capture engine” that extracts vital data from documents — checked by the human loop — the Detect phase where fraud signals are sought in document contents, the Analyze function that looks at aspects of the borrower’s finances, and a new service called Network to help small businesses.

“Now we can say to you, ‘Here is the revenue of this florist that just applied for a small business loan,’” Bobley said. “‘By the way, here’s how the florist stacks up on a percentile basis against all of the other florists that Ocrolus has seen historically, and here’s how it stacks up against all the other florists that we’ve seen in [a given geographic area].’”

See also: Automation to Work to Bring Much-Needed Credit Relief to SMBs

Evolution of the Human Loop

Combining the human loop with a fresh take on APIs and automation for document processing, business lending promises to become faster and more rigorous over next three years or so. For example, cashflow analytics are revealing new efficiencies tailored for a changed employment landscape where borrowers have post-pandemic profiles.

“There’s a group of borrowers who are traditionally employed and relatively straightforward to calculate their income,” he said. “Then there’s a long tail of self-employed Uber drivers, personal trainers, people with rental properties, all sorts of different variations of employment profiles, where performing document analysis and then performing income analytics on top of the data is quite a valuable tool set that we want to continue investing in.”

Returning to his “golden age of automation” vision, Bobley said, “For now we only need to do small business and mortgage, because those are very big markets and we’re just scratching the surface.”

“The 24-month horizon for us is trust lending, and sprint as much as we can on the lending path,” as they eye other verticals that can benefit from the human loop in lending.

See also: Ocrolus, Blend Announce Partnership to Automate Lending