Automating Financial Review Processes with Ocrolus
Last fall we sent one of our MBAs over to the Caucasus for a month to scout interesting technology startups, one of which was Krisp. We spoke with co-founder Davit Baghdasaryan who previously served time in Silicon Valley before deciding to address an extremely difficult problem that nobody has solved yet – the ability to remove all forms of background noise from conference calls. He’s building an entire business around the growth of what’s being called “the gig economy,” the rapidly growing job market comprised of temporary, part-time, and freelance work.
Here at Nanalyze, we’re intimately familiar with the gig economy because that’s how we’ve structured our entire business. We don’t have employees, we have contractors who operate at arm’s length, giving us the flexibility to add or subtract resources as we search for the elusive pot of gold at the end of the rainbow – achieving profitability as a digital media firm. Another startup that’s benefiting from the growth of the gig economy is Ocrolus.
Founded in 2014, New Yawk startup Ocrolus has taken in $33.5 million in funding to develop a platform that “analyzes financial documents, regardless of format and quality, with over 99% accuracy.” The core capability of the platform is the ability to ingest a variety of documents such as PDFs, scans, or mobile phone pictures, and then return accurate and formatted data. Powered by machine learning and a human-in-the-loop data validation process, Ocrolus plugs directly into customer workflows via an application programming interface (API), eliminating the need for manual data work.
In December of last year, Ocrolus launched a new version of their platform, Ocrolus+, in partnership with two companies: Plaid, a financial data network, and SentiLink, a provider of fraud detection solutions. (You may have read about Plaid recently in the news as they were acquired by Visa for $5.3 billion.) The reason Plaid is making such a stir in the financial community is because of something called “open banking.”
Simply put, open banking is all about letting people access your detailed banking information in exchange for better products and services. In the words of McKinsey & Company:
We define “open banking” as a model in which banking data is shared between two or more unaffiliated parties to deliver enhanced capabilities to the marketplace.
Surprisingly, the majority of American consumers are willing to give up their banking data in exchange for benefits from their bank.
Let’s take mortgages as an example. Going back to our earlier comments on the gig economy, people who have various unpredictable income streams often have a hard time qualifying for financial products such as mortgages. When you don’t have a steady income from a single source, it’s difficult to fill out a traditional loan application. Then there’s the whole issue of self-reporting. Everyone’s probably guilty of fudging their income a bit when applying for a mortgage or car loan. In order to kill two birds with one stone, lenders can look at the ultimate ground truth for borrowers – their bank accounts.
What you spend your money on, how you spend money, the extent to which you avail yourself of the checking overdraft feature, and of course your income, are all data points that can be used to assess creditworthiness. We’ve talked before about the surge in fintech apps offering any number of free services just so they can get access to your banking account information. Plaid is the company that all these fintech apps use to interface with more than 9,600 financial institutions out there once you’ve supplied your credentials. While that may seem like a lot of firms supported by Plaid, in reality, there are thousands of financial institutions around the globe that are still stuck in the stone ages – like AST.
Just the Fax, Ma’am
We use a firm called AST to invest in some dividend growth investing stocks because they offer an economical dividend reinvestment plan (DRIP) that can only be accessed using AST’s platform. Because of this little monopoly, AST is in no hurry to upgrade their archaic operational processes. Want to change the amount of money you invest every month? Download a form and fax it to them – or scan the printed form and then email it to them. It’s a complete pain in the rear.
Let’s say we need to prove to a third party that we’re accredited investors and need to provide them with records of our assets. We can use Plaid to provide access to our bank accounts that are held with major institutions like Chase, but what about our stock holdings at AST? What we would do is download the latest PDF statements from these accounts (AST makes you hold a separate account for each company you own) and then email them to the third party. Or if all we have is a paper statement, we can just take a picture of it with our smartphone and send that over. This is where Ocrolus would come into play. Those PDFs or photos would automatically be ingested into the third party’s back-office system with all the key fields being recorded and understood by machine learning algorithms that could very easily figure out that these were stock holdings. They could even use the latest stock prices and value these holdings in real time. At a much higher level, we refer to this as robotic process automation (RPA).
Ocrolus and RPA
In our previous piece on 7 Startups Using AI for Robotic Process Automation, we talked about how machine learning is taking over back-office functions in just about every industry. One RPA company we looked at was UiPath, a company that claims its software robots can automate 99% of many of the tasks they’re asked to perform, without sleeping or complaining about the quality of the breakroom coffee. Since that article, UiPath has now raised an incredible $1 billion in funding and sits squarely on the CB Insights Unicorn List with a $6.3 billion valuation. Investors are pouring money into UiPath because they’re selling a solution that cuts costs. And when we exit what’s now the longest bull market in the history of mankind, service offerings like UiPath that can cut costs will be in demand even more.
Ocrolus is available on the UiPath platform as an RPA plugin that automates bank statement analysis. All the developer needs to do is configure some connections and suddenly they’re processing financial documents with a 99% accuracy. Given the nature of machine learning algorithms, that accuracy only improves over time as the humans in the loop help point out the algorithms’ shortcomings. The partnership with SentiLink lets Ocrolus start to add additional value to their process – like fighting fraud.
Going back to our previous example of AST, let’s say we didn’t quite have the net worth required to be considered accredited investors. The problem could be solved easily enough by editing the PDF document and fudging the total number of shares for a stock along with the total value of the account. That’s one form of fraud that could be detected by Ocrolus – document tampering. Another form of fraud that’s far more malicious and damaging are synthetic identities.
Founded in 2017, San Francisco startup SentiLink has taken in $14 million in funding to combat synthetic identities, where the name, date of birth, and social security number (SSN) don’t correspond to a single real person. Many of these synthetic identities are operated by sophisticated fraud rings that give each identity its own phone number, email address, device, and address. In order to learn more about how the company identifies fraudsters, we turned to an article published on Medium, a media platform that largely contains absolute drivel, but which occasionally dispenses a gem or two of wisdom.
In a Medium article titled “How to Detect Fraud Rings Using SSNs,” Data Science Manager Alex Zimmer talks about how SSNs contain a structure that most people don’t know about. In 2011, the Social Security Administration switched to issuing randomized SSNs making it easier for fraudsters to create synthetic identities. Once a synthetic identity is used successfully, it will often be reused. Fraudsters often tend to target institutions where they’ve had the most luck before, which means SentiLink can now start to identify patterns. (Maybe they ought to have a chat with the lads over at Group-IB about how threat intelligence might supplement this platform.) In short, SentiLink can then use sophisticated mathematical techniques to start identifying fraudsters that would otherwise pass through systems undetected. It’s exactly the sort of value-add that Ocrolus can use to make their platform more appealing for potential clients. After all, if you’re still using PDFs and paper forms in your day-to-day processes, what’s the likelihood you’re employing any sort of sophisticated fraud detection techniques?
The promise of open banking seems to be both a blessing and a curse. While fintech providers can now provide free services and better assess what products work best for your financial situation, it just becomes another case where your personal information is used by companies to relentlessly peddle things you probably don’t need. What Ocrolus provides is a solution that keeps humans from having to do the dull work of manually scanning through financial documents. They’re also providing tools that help combat fraud, something that will make the world a better place. In emerging markets where banking systems aren’t so technologically advanced, this technology can also democratize the availability of financial products for everyone so that we can all enjoy living lives that revolve around consumerism.