Every lender has a fraud problem, but AI-powered detection is here to help

If you’re a lender, you have a fraud problem! Fraud is an unfortunate reality of every single lending business, because if your product is money, someone will try to steal it. As a potentially major component of a lender’s P&L, loan losses from fraud can be a costly issue. In fact, every $1 lost to fraud now costs U.S. financial services firms $4.23, according to LexisNexis

Just like fraud, documents are often a constant across many lending application processes from mortgages to small business lines of credit and beyond. And while fraud has the potential to negatively impact the profitability and efficiency of a lending operation, it can be mitigated through the intelligent application of automation, fraud detection technologies, and advanced analytics. 

Document fraud in lending

Let’s begin by reviewing how lenders collect and assess documents. Legacy methods, especially the manual review of documents, can increase the risk of fraud going undetected, as many alterations are invisible to the naked eye. Various technologies, ranging from straightforward pattern recognition to advanced machine learning and AI, can go deeper into the digital layers of a document and identify modifications, anomalies, and the fingerprints of malfeasance. 

One may assume that fraud only occurs in complex materials, but evidence of tampering can be found in even the most common documents used by lenders. Having reviewed literally hundreds of millions of documents over the past few years, Ocrolus has used this massive dataset to train its models to identify some of the most common ways documents are altered, including: 

  • Altered date fields – This type of document fraud is often found in bank statements that are legitimately those of the prospective borrower. For example, a lender might ask applicants for three months of statements to assess financial health and cash flow. Let’s say a potential borrower doesn’t have the best numbers from that time frame. An applicant might take its own statements from a previous period when finances were better and change the dates to those requested so it appears they are who they say they are, do belong to that financial institution, that the statement is real, and the information found within is accurate. 
  • Modified transactions – Another prime example of fraud we see in financial statements is altered transaction data. Applicants may edit the size or source of a deposit to make income or revenue appear larger or more legitimate than it actually is. This more complex alteration requires additional edits, sometimes hundreds, throughout the document to make sure numbers reconcile and the formatting appears legitimate. For example, an applicant might add thousands of dollars to an account balance and then alter every single transaction amount in the document to make the numbers tie out. Alternatively, an applicant might alter the text of a transaction to make an internal transfer look like revenue from a legitimate customer.
  • Fake, generated materials – Similar to financial statements, people sometimes edit real documents from a legitimate source, but what happens if someone doesn’t have certain types of documents they need, like paystubs? That’s where fake document generators come in. These websites provide realistic paystubs that an applicant can purchase and submit to make their loan application appear complete. 
  • Falsified identities – Identity theft is another prevalent type of fraud in lending applications. This may come in the form of someone stealing an existing person’s information, such as social security or driver’s license number, or taking it a step further than that by combining fragments of real and fake personal information to fabricate a new, fictitious identity. 

Mitigating and preventing fraud

There is a balance needed in detecting and preventing these types of fraud. Lenders need to be able to effectively detect fraud without adding too much friction to the application process or letting ‘false positives’ ensnare legitimate customers in an overly sensitive filter. Lenders could interview each applicant or require two years of statements before approval and prevent most fraud, but nearly all borrowers would quickly flee such an onerous process!

Automation helps bridge this gap with the ability to detect fraud or tampering that is below the threshold of human perception. By diving deep into the digital ‘guts’ of a document, advanced software can alert lenders to alterations and inconsistencies. 

Whether flagging that 100 different fields in a statement have been edited, uncovering the digital fingerprint a paystub generator left behind, spotting inconsistencies in the personal information provided on an application, or other alterations, AI-powered fraud detection software can identify a wide variety of document tampering. 

While mitigating fraud starts at the individual applicant level, preventing fraud at scale can only happen when organizations have strong systems, policies, and operations in place. Lenders need to continuously gather and review important information, such as borrower patterns and related outcomes, to develop an effective learning system. Based on this data, organizations can continually enhance detection practices, adjust their thresholds, and improve analyst review processes for more confident, efficient decision-making. 

In doing so, lenders can find the balance between effectively fighting fraud and maintaining a low-friction application process for their customers. 

  • David Snitkof

    David Snitkof is the SVP of Growth at Ocrolus and has a successful track record of developing analytical systems, teams, and businesses from the ground up. He was most recently Head of Analytics and Data Strategy at Kabbage, where he led a highly successful global analytics organization and the development of new data products. Prior to that, he was co-founder of Orchard, a pioneering data, analytics, and transaction platform that accelerated the growth and institutionalization of online lending during a time of massive growth. Prior to Orchard, David held various analytical, product development, and risk management leadership roles at American Express, Citigroup, and Oyster.com.