The formula and science behind document authenticity and measures of accuracy, defined

January 7, 2019

Delivering beyond-human levels of accuracy in identity verification technology 

When it comes to identity document counterfeit detection in the digital age, most businesses share the same goal of improving accuracy to slash account-opening fraud and boost auto-acceptance. And in turn, ID verification solution vendors need to be able trust their science and automated process, as well as their in-house expert team’s intelligence to label datasets to derive insights. The goal: to deliver beyond-human levels of accuracy using data science, analytics and machine learning.

Defining the metrics surrounding ID verification

As the list of fully-automated authentication techniques grow, the science becomes more sophisticated and robust. It can be easy to feel lost in the complicated technology behind a continuous learning identity verification engine, and unsure of which digital identity verification solution will work best for your business, so here are some further defined metrics to help you understand the identity proofing industry more clearly: 



A balanced approach to fit your business' specific needs

The relationship between false negative rates and false positive rates is an inversely relative one. The key to this give-and-take relationship is to find a balanced approach that best fits your business' specific needs. The opportunity costs associated with false negatives (stopping legitimate prospects) and false positives (letting bad guys through) must be assessed in order to recognize where the most opportunity lies. In other words, as more “good guys” (non-forgeries) are introduced to more friction in order to be authenticated, more “bad guys” (forgeries) will get halted. Inversely, as more “good guys” get through without any friction, more “bad guys” will potentially get through. The management of this relationship is not something overly simple to solve, however, and it's important to note that more and more consumers are willing to go through a bit of friction in return for a safer experience. Data breaches are in everyone’s ears, and even younger generations are becoming sensitive to the handling of their personal information. And so the term “positive friction” has entered our vernacular. (Read the Zogby study on consumer confidence in the digital age.)

When it comes to the ID verification process, allowing a small percentage of false negatives through may create some extra friction, but we can also consider this type of positive friction an investment in research and development in addition to a signal to consumers that their transactions are secure. Mitek’s machine-learning technology is built by a robust team of PhD and MS scientists with expertise in computer vision, data science and deep learning, it’s informed by large datasets of forged and real documents, and the technology is continuously updated with every release.