Episode 84 - Diving into the Deep Data Pool - Feedzai's Saurabh Bajaj and Nick Stanchenko

Fraud prevention today is about how quickly we can separate good customers from questionable ones and, for those doubtful transactions, use the right set of tools and data sources to optimize speed, costs, and fraud losses. When it comes to fraud prevention, it turns out that data is key. No revelation there, but how we manipulate it, gather it, and assure its provenance is undergoing major change. What would be a revelation to fraud managers of a decade ago is the unbelievable amount of data and the wide variety of sources that we have today. It’s a flood. The only means we have to make sense of this data deluge is through algorithmic examination. Rules engines and neural networks are staple approaches. In recent years, application of artificial intelligence and the newer incarnation of machine learning (“let the computer figure it out based on all the data it sees”) has become a hot, and effective, area for fraud prevention. Only machines can find correlations among all that data in order to identify potential fraud. A number of firms focused on the fraud prevention problem employ techniques that gather data and then analyze it in order to provide their customers like e-commerce merchants or financial institutions with a risk score. Companies specializing in device fingerprinting, for example, gather the relevant data (think IP address, mobile IMSI number, device type, OS version, browser software version, etc.) to create a profile or “fingerprint” of that device in order to generate a history of its behavior. Threat Metrix, owned by LexisNexis Risk Solutions, is an example. Behavioral biometric companies may take that data and layer on how the owner actually uses their device, often by looking for keystroke patterns, screen tap rhythms, the angle that the phone is held, and more, in order to build a more nuanced profile that includes how the owner interacts with the device. That richer data then feeds into analysis and risk scoring. Mastercard’s NuData Security acquisition uses this approach. Subsequent bidirectional data sharing can provide these firms with insight into the results of their decisioning. As these firms gain customers, they see more and more devices and develop clearer visibility into the outcome of their work. As a result, it becomes a natural step to pool or federate the data they see from all of their customers. There’s an expectation that a card account, for example, will be seen at multiple merchant clients of the fraud solution provider. These repeat interactions will improve fraud detection for all when the cardholder is a bad actor or speed the transaction of a trusted one. Data consortia where multiple financial institutions and merchants pool their fraud and chargeback data also exist. Ethoca is a prime example. The deeper the data pool the better, provided, of course, there’s the ability to analyze it all. Massive analytical capability is the foundation for artificial intelligence and machine learning. In the fraud prevention space, Feedzai is a firm that applies its analytics power to data sourced from multiple provider and techniques. Feedzai, like others providers who have attained a critical mass of customers, has also invested in federation of their data to improve, for everyone, its fraud prevention results. In an earlier episode, we spoke with Feedzai CEO Nuno Sebastiao to get us grounded in how AI and ML apply to fraud prevention. In this discussion with Saurabh Bajaj, Feedzai’s Head of Product and Nick Stanchenko, product manager for Feedzai’s Risk Ledger, its data federation program, we go further. Saurabh catches us up on Feedzai's growth and then take a look at how Feedzai works and at the data sources it uses. Nick addresses federation, its value, and the light integration required.