Have you noticed that your credit card sometimes declines transactions or flags fraud alerts? This conversation will help you understand the backstory of that workflow and potentially trigger other ideas in that domain.
Sramana Mitra: Let’s start with introducing our audience to yourself as well as to FICO.
Scott Zoldi: I’m the Chief Analyst Officer at FICO. I’ve been with FICO for 17 years. Today I have responsibility for all the analytics products and technologies that we use within our products. FICO, as a company, is 60 years old, and is well known for the analytics that we deliver to make better decisions for companies. The main thing that we offer is the FICO score. It’s a score that’s used in lending decisions as well as fraud detection.
For each and every transaction, we make a prediction of whether or not it’s fraudulent. We develop a lot of expertise in different verticals and take a lot of data and extract insights and turn them into scores that can be used for decisions.
Sramana Mitra: I’d like to ask you to drill down on a couple of points based on what I heard. Firstly, I’d like to understand, from an analytics point of view, what are the key parameters that you are collecting data around that helps you do fraud detection and pattern recognition?
Scott Zoldi: I’ll mention two different ways. With respect to our Falcon product, we are basically seeing two-thirds of the world’s debit card and credit card transactions. The first flavour of Big Data is having a huge set of data roughly of the size of a couple of petabytes that we leverage to build these models. There’s a lot of variety in that data in terms of the countries and the consumers of those cards.
The first part of the Big Data piece is to properly leverage all that data for the creation of insights about where fraud is occurring, what are the merchants where fraud is more prevalent, what regions, and what times of days. All of that is a process where we can leverage technology such as Hadoop or Spark to consolidate and condense that data into key fraud features. That’s a bit more traditional use of big data.
The area that I generally have a lot more interest in is actually around the streaming analytics technology that we developed. Streaming analytics is essentially working on the “V” of Big Data – velocity. Data is coming at you at an incredible rate. You have a narrow window to make a decision and to drive a meaningful score. In fraud, when you go into your cash register and use your credit card, we have a very narrow window of about 10 milliseconds to get that transaction, retrieve data about your history, and make a decision based on a score. That score would range from 1 to 999 where the larger the score, the more likely it is to be fraudulent.