This discussion explores Machine Learning apps in financial services.
Sramana Mitra: Let’s start by introducing our audience to yourself and Featurespace.
Dave Excell: I’m the Founder of Featurespace.
Sramana Mitra: What is Featurespace?
Dave Excell: Featurespace is a machine learning company on understanding and outsmarting risks. We use machine learning to look at different financial transactions in different contexts and optimize to see which ones are genuine transactions. We try to make those transactions as efficient as possible so that we help genuine customers avoid their transactions getting declined.
Sramana Mitra: Let’s do a few customer use cases. It sounds like you work in different segments of finance. Pick different use cases from different segments and let’s go through a few use cases. As you do that, please highlight the machine learning aspect of each of those applications and use cases.
Dave Excell: One of the examples is for a credit card issuer where you may have a relationship with a bank and where they’ve issued you with a credit card. One of our use cases looks at the authorization of the transactions that come through to that card issuer when you go to a shop or an online website and try to pay for something using that card.
Our machine learning algorithm is deployed as part of that authorization flow to look at the information about the transaction and the history in terms of other transactions that you’ve done in the past or that merchant has submitted in the past to determine whether or not that is a legitimate transaction.
Sramana Mitra: What is the machine learning aspect of this?
Dave Excell: When you look at the machine learning aspect, it’s doing two things. It’s looking at how to learn what is the legitimate activity of that cardholder or that customer. These are things like the typical merchants where they shop, when they shop, and how do they spend their money.
Also, we try to understand where we see transactions that don’t fit our profile for that cardholder. When do we think that those transactions are suspicious or different enough that we think they should be blocked without trying to disrupt the actual customer. It’s stopping fraudulent transactions, but still enabling the customer to use their card when they need to be able to do that.
Sramana Mitra: When you start a customer up on your platform, what is the dataset that the customer is starting on? Is it all the customers’ native dataset or is there other dataset that is being brought in or other heuristics to feed the algorithm?
Dave Excell: For our deployment, it’s just using the customer’s own data. To make sure that the system is effective from day one, we’ll use some historic information from that customer to essentially be able to bootstrap the algorithm so they’re not starting from no knowledge.