Ocrolus Uses AI to Automate Document Capture, Improve Accuracy, and Reduce Fraud for Lenders

In a Nutshell: Small business, mortgage, and consumer lending firms compete on a dynamic playing field where technology often gives firms the edge to grow market share. Ocrolus uses AI-based computer vision and natural language processing models to automate digital document capture and analysis for digital lenders. Its core mortgage model performs at 96% accuracy across 106 document classes, with human-in-the-loop validation handling the rest. Intelligent document automation from Ocrolus helps digital lenders gain efficiencies for managing risk and avoiding fraud.

Documents are the life’s blood of the lending industry because every lender needs to collect them at some point in a loan’s lifecycle. As a lending firm grows, so does its need to accurately classify, capture, detect, and analyze documents in various formats and conditions.

While some routine lending only requires a few bank statements, more complex loans require much more complex documentation. For example, a typical mortgage may involve hundreds of pieces of information, which can take many formats depending on the contributor.

The tried-and-true method for dealing with all this paper and digital information is to put a team of people in a room with a mission to sort it all out. However, as digital document capture technology advances, lending firms in all verticals turn to Ocrolus for intelligent document automation.

Ocrolus helps lenders attain best-in-class accuracy with document classification that combines machine learning with human validation. That generates faster capture with better accuracy while preserving information vital for compliance.

Then, while file tampering detection and algorithmic validation technology protect against potential fraud, Ocrolus produces clean, normalized data for deep insights into cash flow and income.

Beyond the obvious efficiencies in moving information around and drawing conclusions from it, actionable insights from Ocrolus empower lending teams to underwrite with more confidence. The lending decision is quicker because the assessment of the borrower’s ability to pay is more precise.

“In the traditional world, lenders have teams of people staring at documents in a costly, error-prone process that’s hard to replicate, scale, and increasingly hard to justify,” said David Snitkof, SVP Growth at Ocrolus. “We automate that completely.”

The mission at Ocrolus is to work with all types of lenders, including small-business, mortgage, consumer, and auto lenders, to help them make high-quality decisions with trusted data and unparalleled efficiency.

Ocrolus works with lenders at all development stages, from early-stage fintech companies to firms with decades of history or more.

Onboarding through the platform’s easy-to-integrate, modern API may require as little as a day. Typical API implementations require a few weeks of working together.

A dedicated implementation team at Ocrolus works hand in hand with customer teams to pinpoint implementation objectives, overcome legacy infrastructure challenges, and provide support from start to finish. Smaller-scale integrations through a web dashboard are also available.

“We claim some of the biggest lenders around as our customers, but we also have fintech customers that started doing business six months ago,” Snitkof said. “I love speaking with all our clients, but I especially love that latter group because if you start building your systems with AI, automation, and scale in mind, that will serve you well.”

The first step in any Ocrolus workflow is automated document and data classification. Ocrolus looks at PDFs, scans, smartphone images, and all other document types and yields a labeled and indexed output that quickly surfaces incorrectly submitted or missing information.

Intelligent capture selects the extraction or character recognition method according to document type and structure. Fraud protection detects signals based on document origin and inspects for signs of tampering after creation.

Data analysis uses a proprietary transaction tagging algorithm to identify income sources, payments, recurring transactions, and overdrafts.

“Our goal is to help lenders automate the entire process, taking what used to be a cumbersome, document-driven workflow and making it automated, intelligent, and trustable,” Snitkof said.

Of course, there’s always a document that stumps even a state-of-the-art machine-learning system like Ocrolus. Indeed, no AI document platform can perform with 100% accuracy and 100% coverage.

But instead of pushing those numbers as high as possible through technology and calling it a day, Ocrolus injects a universal failsafe into the system through human-in-the-loop validation. The goal is to achieve a symbiotic relationship between human expertise and machine intelligence.

Intervention occurs at the initial classification stage when Ocrolus routes documents with imperfect confidence to data classification and verification specialists for labeling and quality control.

That sets up several wins. The first is for accuracy because the human-in-the-loop process guarantees a resolution or a follow-up request with the lending client. The second is for the Ocrolus AI and machine-learning models because those human interventions generate super-high-quality training data to improve over time.

Ocrolus incorporates third-party document AI platforms from Google, Amazon, and OpenAI. But that massive training database from human-in-the-loop intervention — more than 250 million pages worth — enables it to enhance its in-house computer vision and natural language processing capabilities.

The third win is for fraud prevention. The human-in-the-loop model enhances the ability of the system to respond to signals indicating potential document fraud.

For example, Ocrolus can dive deep into the guts of a document to determine whether someone has edited a field, such as an account number, balance, date, or contact information. It can even detect whether someone opened a document in editing software.

“We can tell when someone adds a font different from the one a financial institution typically uses to produce a document,” Snitkof said. “We can sometimes even identify an earlier version of a document embedded in the file before the fraudster did their evil magic.”

At that point, the lender’s fraud analyst team can view the document in the Ocrolus dashboard. Ocrolus highlights where it was tampered with to give the lender a sense of what the fraudster might have been up to.

It all adds up to saving time and money, which works as favorably for lenders as it does for other businesses. Ocrolus enables faster, more accurate lending decisions and corresponding bottom-line improvements.

Ocrolus customers learn the human element isn’t just a factor during the document classification stage. Given the dynamism of the financial marketplace, Ocrolus goes out of its way to communicate with customers and elicit product feedback for future iterations. A comprehensive support channel helps.

“The more customer input we receive, the better,” Snitkof said. “We have a great user-experience research function where we interview clients and prospects, watch them do their jobs, and seek to understand their pain points to tailor what we build to their use cases.”

The payoff at Ocrolus is more efficient mortgage underwriting, fewer bad loans, more right-sized credit lines, and more fraud identified. Ocrolus offers high-quality, trusted financial data and valuable insights to advance lending decision-making.

“That shows the value we’re creating as an ecosystem,” Snitkof said.

Evidence that it’s working includes a record number of new clients acquired in Q1 2023. Even in an uncertain market with higher interest rates and lower lending volume, the strongest and most thoughtful lenders invest in automation, scale, and better decision-making to manage risk and serve customers more effectively.

Lenders plug Ocrolus insights into their credit risk models to enhance the predictiveness and applicability of those models. They receive a much more accurate picture of the customer’s financial condition.

“The smart ones are taking the time to do that,” Snitkof said. “Lenders want to make good loans with a precise handle on the risk.”