Scott Zoldi: These are technologies that are facilitated in Big Data because we have the ability, with the Storm-based architectures today, to build a model that reaches out to a NoSQL database, retrieves a data record associated with your previous transaction history. That data record is not of the transactions you did in the past but variables or summarization of your behaviours. We deal with a small latency in terms of the NoSQL database pull. We update the variables that are contained there and they’re re-updated. Those variables are sent to a model. For a fraud model, it’s a neural network-based model that produces that score. The bank takes that score and they apply their rules. They they will make the decision. That’s what get sent back to the acquirer and the merchant, which then gets back to you. If it’s declined, they’ll provide you some challenge questions or ask you to phone in.
Sramana Mitra: The kind of clustering and behavioural analytics that you’re describing, does it have heuristics to extrapolate other rational behaviours? For example if I book theatre tickets in London three months before my travel, by the time the trip happens, does your fraud algorithm know that since I have theatre tickets in London at this time, it’s legitimate that I’m spending on other things in London or is it going to trigger some sort of an exception due to a foreign transaction usage?
Scott Zoldi: One of the things that is typically in these models is an understanding of other purchases you’ve made. Some of those purchases may correlate to future activities. If you have an airline ticket, we should hopefully expect or not be overly concerned if we see transactions in that country. That is information that can get caught in these fraud models.
Even in a more sophisticated way, what we are starting to understand is based on your purchase behaviour, are you behaving like you’re on business or are you behaving like you’re on vacation? Are you staying at a typical American-branded hotel which might be typical for a certain class of consumers that hold the credit cards that you do, or are you going to the boutique or local places? That also gets learned by the analytics.
It can understand if the types of behaviour that you’re exhibiting are consistent or inconsistent with both you as a consumer and also nature of what you are doing. We leverage that in the credit card space. We actually have mobile apps that do very similar things. I’m in London right now taking this call. I can tell you that I haven’t seen any sites. With just where my phone has been and the way I’ve been using my phone tells a bleak picture of how much fun I’m having in London because I’m doing business. I’m moving back and forth to our major clients. I’m at our office. That’s other information that we collect and we can correlate with what’s happening once you arrive to make a better decision from a fraud perspective.