Harnessing the Power of Machine Learning to Detect Fraud

Machine learning automation is emerging as a powerful risk management tool that mortgage lenders and fintechs can use to identify fraud and improve the accuracy of lending decisions.

I’m fortunate to be on the leading edge of this transformation in the financial services sector. Automated fraud detection specifically is an exciting and impactful field, and Ocrolus has the data, infrastructure, and engineering talent to be a leader in the space. Over the past year, I have led a cross-functional team of engineers working on Detect, a new fraud detection solution we developed to help detect financial fraud on bank statements, paystubs, and W2s. 

Ocrolus Detect leverages our expansive financial dataset and state-of-the-art machine learning algorithms to automatically scan documents for fraud. This output is then converted into contextualized visualizations that highlight exactly which fields and values have been tampered. The visualizations and the API responses offer detailed context so that lenders can make an informed decision on the validity of the document and the intent of the tampering. Detect significantly expands the number of fraudulent documents lenders are able to detect using their existing solutions, and tells a compelling story to the analyst. 

Revolutionizing Fraud Detection

Detect has established Ocrolus as a leader in fraud detection on financial documents. Meanwhile, we’re constantly working to improve. For Detect R&D, we’re focused on improving existing signals, eliminating false positives as much as possible, and expanding the scope of what we are able to detect. 

Our clients can look forward to a new detector focused specifically on detecting issues in alignment that are often used as a signal during manual review by fraud analysts. We’ll also be expanding coverage to scanned documents, other document types, and to generated documents. By detecting templates, we will enable lenders to better determine if a loan application includes fake documents created from downloaded templates. New product features are part of a detailed roadmap we have for continuing to make Detect more robust and increasingly valuable to our customers.

Ensuring Fraud Detection Accuracy

Our machine learning fraud detection algorithms are agnostic to editing software and perform just as well on banks that are well known as on small local banks. Meanwhile, small banks, and paystubs issued by small businesses can pose challenges, especially since these documents often seem to exhibit anomalous patterns. 

It’s virtually impossible to eliminate all false positives due to the diversity of credible documents, nor would we want to. When it comes to document fraud, it’s a lot more expensive to miss one file than to review a few extra, and our visualizations and integrated dashboard are designed to make it easy to confirm or reject the findings. If we were too strict, we could miss true positives, so we aim to strike this balance. 

Ocrolus Detect is designed to save our clients money and time, and we have a high standard of accuracy to support that goal. Detect accurately identified 4x more fraud than our leading competitor and saved an upwards of 30 minutes of review time per application in tests with clients. Reducing loan processing overhead results in a better experience for borrowers and lenders alike.

Detecting Fraud Faster with Machine Learning Automation

Detect is a financial document fraud detection solution that underscores the power of automation. We give lenders the ability to harness the power of high-quality data for lending decisions. Our algorithms leverage our dataset of over 100 million pages of documents from thousands of financial institutions to provide a highly accurate automated fraud detection solution. 

Our algorithms can find changes undetectable to the human eye, as well as changes that would take a human considerably longer to detect. With an automated solution, loans can be halted before they incur further processing expenses, or potential losses. Further, machine learning automation allows operations to scale for higher volume, or for ebbs and flows. 

To learn more about Ocrolus Detect, book your demo with one of our document fraud automation experts today.

Detect at Ocrolus

Personally, Detect has been such an exciting and rewarding project to work on because I love solving problems. I have been able to utilize so many of my strengths — from creativity, to attention to detail, that I haven’t had the opportunity to exercise fully since I left abstract math. 

As a woman in tech, I’ve often felt like there’s less room for me to make mistakes, out of fear of confirming any potential bias. I’ve been reflecting lately on how my resistance throughout my career to “move fast and break things” has actually helped to make me a stronger engineer and a valuable team member. Not everyone can move fast and break things, and that proves why diversity is so valuable to a team. Detect owes a lot of its success to careful design and extensive testing. 

At Ocrolus, I’ve been given autonomy and support to bring my own style and strengths to the problems at hand. Ocrolus has been, and continues to be, a rewarding place to work and to grow. I’m constantly learning, whether it’s by working on interesting problems, learning from colleagues, or being encouraged to work outside my comfort zone. A lot of us at Ocrolus share a love of learning and problem solving that makes for a great environment and is a contributing factor to the quality, value, and impact of our work. 

Learn more about job opportunities at Ocrolus and join great teammates like Sylvia.