Reducing Fraud with a Machine Learning Model and Ekata Identity Check

All businesses are at risk for chargebacks due to fraud, especially online businesses. The FTC reported that 3.1 million complaints were collected by the Consumer Sentinel Network in 2016. 13% percent of those complaints were related to identity theft, and credit card fraud complaints rose to more than 32%. Card-not-present (CNP) fraud in the U.S. increased 40% in 2016 according to a Javelin Strategy & Research study, and the rising adoption of EMV cards and terminals are driving many fraudsters to open new, fraudulent accounts.

eCommerce companies know that it’s crucial to keep their fraud rate below 1% otherwise they risk losing the ability to accept credit cards, not to mention the lost revenue from having to pay for numerous chargebacks. At Ekata, we found ourselves being subjected to an increasing amount of fraud and chargebacks after launching our consumer product, Ekata Premium, which provides subscribers access to U.S. public records to verify contact details, mobile numbers, bankruptcy history, criminal records, and more to facilitate trusting interactions in today’s sharing economy.

The Problem
Fraudsters were opening new Ekata Consumer Premium accounts with stolen identities in order to steal our Premium data, causing a high number of chargebacks and putting Premium at risk of exceeding the 1% fraud rate.

The Solution
The team created a logistic regression model for Ekata Premium that takes in a number of inputs such as payment behavior, information about the customer, and user behavior on the website, and then combines that data into a score. While the model did help mitigate the fraudulent use of the Premium product and reduce the number of chargebacks, they realized there was still room for improvement.

The Premium fraud team decided to leverage the Confidence Score, which at the time was a newly built feature of the Ekata Identity Check product. The Pro product team was looking to evaluate the new Confidence Score quickly, and Ekata Premium was determined to be the perfect test bed for developing a new version of our logistic regression model, and a great way to understand and improve our offerings.

By using the Identity Check API to continually feed information from our Pro product into the logistic regression model we built for Premium, including the new Confidence Score; the fraud model now provides a clearer picture of our customers, which resulted in the Ekata Premium offer saving money on every order.

To see the full results of implementing Identity Check Confidence Score into Ekata Premium’s model, read the case study here.

Thanks for reading! You might be interested in these posts, too: