New York City-based Ocrolus has raised a $4 million Series A financing in April led by Bullpen Capital with participation from QED Investors, Laconia Capital Group, ValueStream Ventures, and RiverPark Ventures. The startup claims to help analyze bank statements and other financial documents with 99% accuracy using Optical Character Recognition (OCR) technology that converts images into text.
The startup’s two products, PerfectAudit and Medicaid-Genius, are modernizing the financial review process by analyzing scans and images of financial documents speedily and at a lower cost. By doing so, Ocrolus has become one of the many companies launched with the aim to replace wearisome, error-prone and imperfect human tasks with AI-driven processes for a variety of industries.
Launched in 2014, CEO Sam Bobley and President, Victoria Meakin started the company to expedite the Long-Term Care Medicaid application process. The application, which requires a submission of over five years of bank statements to qualify, is reviewed manually in what can be considered one of the most time-consuming and tedious processes. Ocrolus was formed to automate this healthcare eligibility process but its founders quickly learned that the same technology can be used for a variety of industries.
Numerous industries such as banks, accounting, auditing, insurance to the lending industry need statements to be converted into digital data and are now paying Ocrolus for its services.
The Story Behind Captcha
While the OCR technology has existed for several years, Ocrolus claims it has been ridden with shortcomings because of its reliance on the document format and quality. “Our biggest differentiator”, CEO Bobley says, “is that we’re able to read bank statements from every financial institution regardless of the layout, format or document quality with 99+% accuracy”.
An unlikely event inspired the team to explore this technology: “When you’re typing in a CAPTCHA (Entering text from image while accessing a site to prove that you’re not a robot)” Bobley says, “you may not realize it but you may have been performing work for Google, helping them transcribe old books and magazines into digital text.
When Google started a project called The Google Books Initiative, they wanted to digitize its archive of old books and magazines to make them searchable. However, relying on OCR alone proved unsuccessful and inaccurate. Google realized that they needed human intuition to read what the computer couldn’t. What Google, then, did was snip the text that couldn’t be confirmed into tiny, bite-sized images and released it as CAPTCHA so that, Bobley says, “people like you and me could verify the data for Google.”
Ocrolus applied the same concept to financial statements and devised a system which follows a three-pronged approach. First, they utilize automated processes to extract all the information from the documents. Next, what can’t be confirmed is snipped into tiny, bite-sized images similar to CAPTCHAS and sent to what Ocrolus calls, “crowd workers” for verification. This is followed by the use of ‘algorithmic reconciliation’ where, Bobley says, “the validated transaction value is taken and ensured that it is sync with the beginning and ending values of the account”. This system of checks and balances allows the startup to claim that it is 99% accurate.
Human v/s Machine Verification
Ocrolus’ elite OCR technology is not reliant on type or quality of paper and doesn’t require the scanned documents to be in order. It therefore finds a large use-case in investigation by lawyers, forensic doctors, the police force, or by the government to accelerate the speed of service.
The platform, however, according to its founders is intended for those who need to review high-volume bank-accounts manually. While the company has government clients, Ocrolus’ biggest (and wealthiest) users are lenders who issue small-business or personal loans. Here’s where Ocrolus sweeps in with its promise of speed and accuracy through automation.
How it works: The users are asked to upload images of documents, pdfs on Ocrolus’ web platform. The user is alerted once the data is processed and s/he can then filter it through “count, date, description, amount, recurring transactions”, etc. Ocrolus also has a feature that points you to exactly where the data was pulled from in the source document.
Along with its OCR-powered offering, the startup also provides the option to integrate PerfectAudit, a product the company launched in 2016. The PerfectAudit API automates cash-flow analysis for lenders. Bobley elaborates that this API can be integrated in to their client’s homepage such that if anyone applies for, say, a loan, the documents directly reach Ocrolus. Ocrolus then processes the data and updates its client with the results. They can then use this summarized, processed information to determine if the loan should be issued.
The obvious advantage to Ocrolus’ clients and the reason for the quick adoption of its services is, like AI, the reduced cost and speed which automation brings. Bobley says that companies charge between 85 cents and four dollars per page for manual work while they do it between 35 to 70 cents. They also manage to deliver results quickly, the least turnaround time, depending on volume, being two hours.
The company has reached this level of expertise slowly. They started by learning, say, how to provide reconciliation of transactions and have now developed a feature to indicate fraud – to alert users about discrepancies in bank statements. But its most impressive work is in the healthcare industry – especially in trying to modernize Medicaid eligibility determination.
The possibilities for Ocrolus are immense. Moving forward, the company plans to take its advanced analytics to read additional document types such as invoices, brokerage statements, tax statements, and not be limited to bank documents.
If the future is digital, an error-free and ubiquitous OCR technology can make it smoother and easier to achieve it.