Fraud Detection Screening: What is Accuracy?

We have to be careful about how we talk about accuracy when screening an applicant. In a simplistic sense, it is the only important thing in fraud detection. Bells and whistles don’t matter if you aren’t getting the right answers. But it depends on what you consider to be accurate. Is it accurate to declare a document to be authentic even if it doesn’t belong to the applicant? That is what many of our competitors do.

In determining fraud, there are multiple aspects of accuracy:

1. Is the document authentic?

2. Does the document belong to the applicant?

3. Does the document satisfy your criteria for recency and look back?

Certainly, if the document isn’t authentic, then the document is fraudulent.  But items 2 and 3 can also be signs of fraud.

Our competitors only evaluate authenticity. They boast about their high accuracy ratings, but what good is a high accuracy rate with regard to authenticity if the document may not even belong to the applicant or if the documents presented are months old?

Authentication Accuracy

Some fraud detection platforms like to claim how accurate they are, highlight how many documents they’ve processed, and stress the importance of sample size as a basis for their claimed accuracy rates.  Sample size is in fact one of the elements to accurately detect fraud, but to be great at detecting fraud, there are other critical factors that play a role. At Docuverus, we have sampled over half a million documents, real and fake. This is far more than is needed to ensure the 99.98% accuracy from our analysis.

Analyzing sample documents is how we determine what real looks like versus fake. Simple things like misalignment of columns of numbers can be a tell-tale sign of a fake. But it takes much more than just a visual inspection of a document.

Accurate detection of fraud is predicated on an analysis of the document’s metadata. Metadata analysis is critically important as a means to detect fraud, but Docuverus takes a very different approach to analyzing metadata than our competitors.  While they are satisfied with looking at the basic metadata of a document, we go deeper.  Docuverus utilizes a proprietary Multidimensional Metadata Analysis, which instead of just evaluating documents on a single layer, actually looks at metadata in multiple layers.  Where single-layer analysis is akin to using an X-ray to detect an ailment, the Docuverus Multidimensional Metadata analysis is like using an MRI, exposing a level of detail and clarity that would evade a simple X-ray scan.  A single-layer metadata approach can and will detect fraud and it will even catch most fraud.  But it will not catch all fraud. Simply put, that Multidimensional Metadata analysis is how Docuverus catches the most sophisticated forms of fraud – fraud that on the surface appear to be valid, but that can be caught by detecting irregularities several layers deep.

 

Contextual Accuracy in Fraud Detection Software

Another key to delivering the highest accuracy is to have the ability to contextualize the document being evaluated, which only Docuverus can do.  What is contextualization?  Simply put, it’s being able to determine the type of document being presented, so that we know, for example, if the document is an ADP paystub, a Paychex paystub, a Bank of America bank statement, or a Wells Fargo bank statement.  This kind of document classification relies on machine learning and computer vision, and because Docuverus is the only company that incorporates an advanced, self-developed vision analysis into its service, Docuverus has the unique ability to be able to tell exactly what type of document is being evaluated.  Based on what’s detected, machine learning is then used to compare the document presented not to all other documents evaluated, but to the specific class of document identified.  This contextualization is not only one of those extra fraud detection layers that makes Docuverus 99.98% accurate (higher than all of our competitors), but it’s also one of the keys to reading document contents. 

Reading document contents should be table stakes in our business. But it’s not. We find ourselves as the only players in the market that, after thoroughly authenticating a document,  can read the contents, contextualize it, and further determine whether the document contents a) make sense and b) applies to the applicant. If a fraud detection platform only tells you if something is real or fake, it could be missing the most basic forms of fraud.

Accuracy in Advanced Document Checks and Automatic Income Calculation

Studies show that leasing staff spend hours of time each month manually reviewing and calculating income, and despite all that time spent, calculation errors are rampant.  Other studies have shown that over 40% of applicants are screened with incorrectly calculated income, and a miscalculation error could result in qualifying an applicant who shouldn’t have – or denying an applicant who should have qualified.

Best-in-class fraud detection has to verify an applicant's income (down to the penny), and contextualization is one of the key prerequisites required to do this.  But automatic income calculation is just the start.  To accurately screen for fraud a system must also detect if paystubs and bank statements are the most recent, if documents are consecutive, if documents are sufficient according to a standard you can define, and detect if the documents presented belong to your applicant, which is yet another way it can detect fraud that more simplistic models cannot.  These features are not just nice to have; they are keys to accurately calculating income, and more importantly, they identify other forms of fraud like using someone else’s documents, using old documents, or cherry-picking documents that are the most favorable to the applicant.

Docuverus income calculation is also smart, in that it knows that the calculated income is only relevant if the documents are valid; if they’re not, our system automatically alerts you that the calculated income is what the applicant is trying to fool you into believing.

Final Thoughts on Accuracy When Selecting Fraud Detection Software

It took us over 2 years to develop and refine our algorithms and our SOCR document reading technology, which explains why no other company has been able to offer this level of functionality. While others invested in marketing and sales, we invested in developing the best fraud detection platform available. Believe us when we tell you that this is no easy task to accomplish, and it’s a key differentiator in our battle against fraud.

When you look for your partner in fraud detection, make sure you understand what they mean by accuracy, how they back up their claims, and how that translates into saved time, money, and headaches for you.