Scott Zoldi: We’ve developed, since the year 1992, technology that allows us to do that real time assessment of fraud risk using technologies like Storm, NOSql so that we can persist data in a summarised way and retrieve it very efficiently. That is one of the major areas for FICO and many of the products that we differentiate in is having analytics that run at very low latency and very high throughput to deal with the velocity of big data.
Sramana Mitra: I’m still trying to take this conversation in a slightly more use case-oriented way. Take what you described in terms of the technology and put in the context of a consumer use case and flip it around and show us how that happens. I’ll lead you into a use case that I encounter often. I travel quite a bit for pleasure.
If I’m going to Europe, I often book performances like theatres. When I’m doing that online, oftentimes, a transaction that I do with my credit card will get declined. I get a notification that if this is a transaction that is legitimate, I’ll have to verify it. That’s what I’m experiencing on the consumer side. Help us understand how your Big Data analytics help make this kind of use case possible.
Scott Zoldi: When you get that notification where they want to make sure it’s you, at the end of the day, that’s a score that’s being produced by one of these Falcon systems. The way it works is, you go and make your online transaction. You put in your credit card details. That credit card detail goes from that merchant to an acquirer bank, and the acquirer bank will look at it and pass it along to the issuing bank.
The issuing bank is going to make a decision. The decision they have to make is whether they allow that transaction to move forward. When it’s at the issuing bank, that’s the time frame which we have to make that decision within tens of milliseconds. They’re not doing that by looking at, necessarily, your transaction history. They leverage this streaming analytics model.
As an example, some of the technology that we use in these models utilize collaborative filtering technologies. These are technologies that produce archetypes of purchasing behaviour that are specific to each and every customer. You may have a very strong interest in travel and you have a travel and entertainment profile in terms of the types of places you go and the types of merchants you visit. That can all be learned in one of these collaborative profiles, which essentially have archetypes and distribution of your transaction patterns. They are learned by Bayesian models but they are a unique view of who you are as a customer. We use that to make propensity decisions about whether or not this transaction is likely something that you would have done based on things you’ve done in the past or based on people that are like you and have a similar archetype distribution.