In conversation with ETCIO, Dr. V. Chandramouliswaran, Senior Director, Global Financial Crimes and Customer Protection & Chennai Centre Head at PayPal talks about why he feels PayPal is in the risk business and how the company is using a story-based analytics approach to fight frauds and risk in its payment business.“We process approximately 33 million-plus transactions per day, a bit more than $22k of payments per second, and serve over 390 million active customers. We have more than 500 petabytes of data in the systems out of which risk alone uses close to 2 petabytes of data to make decisions in mini seconds. All of us are used to instantaneous approvals and so risk leverages a significant amount of data to make these decisions,” Chandramouliswaran said.
According to him, over three trillion events are added each day to PayPal’s systems. These could be people logging in, transacting, checking something or changing their profile. All of these three trillion events are somehow used to assess the risk factor in payments. But he believes PayPal’s approach to risk and fraud using data is different from what people think about it.
“The pace around modeling is evolving every day. We started 15 years back with some logistics regression to make a decision. Today, we have the most sophisticated algorithms, deep learning, active learning, and multi-class models. We leverage the gamut of these models depending upon the use case that we need to solve. Using these we have pioneered an approach to decision-making through multiple layers,” he explained.
“Let’s say we have a customer who is US-based but has to suddenly travel to Ireland. 24 hours later, he makes a payment in Ireland. A typical model-building exercise will catch it and mark this as fraudulent but there can be so many genuine cases here. It could be someone who has family in Ireland, who is now stationed in Ireland. There are many such use cases but the question is how to identify them,” he added.
PayPal identifies these use cases by a story-based approach. The company has investigators who review such cases and try to understand what is happening here. With their learning from these cases, they create rules to help the model-building process to make more automated decisions.
“We look at the information on customer’s past behavior. Let’s say you use PayPal to buy your flight tickets, this is indicative to us that you are traveling. Now, if the customer makes a transaction from somewhere else, it will not be a surprise for us. Similar patterns like the customer making a payment at the airport or railway station using PayPal gives us an indication that he is traveling. Such patterns help us identify the good transactions and the bad ones. We use a lot of learnings along the way to make that decision,” Chandramouliswaran explained.
“When you talk about analytics, you think and say that you are going to build the coolest model out there. But working with small data points and story-based analysis. People like fast and fancy but true value comes from even working with this kind of counter cases because you want to create amazing experiences for our customers. So, story-based analysis becomes super important on top of what we do from the model-building perspective.” he concluded.
