The human mind has amazing capabilities. For example, the average fraud prevention specialist at cleverbridge manually reviews between 26,000 and 40,000 ecommerce transactions every year. They look at those orders and make a decision about their validity in a very short time period. To make the right decision, they have to remember if the order fits the pattern of a fraud attack, an affiliate attack or a bot attack. Along with these attacks, fraud screeners have to evaluate each transaction according to a set of rules that are specific to each product or vendor:
- Is it a valid transaction if product x is over $100.00 and is placed by a first-time customer using a freemailer account?
- What happens if product y is from same vendor but costs over $900.00?
- Does it make a difference if a customer places an order for product z from Thailand or France?
In order to accept valid orders and deny fraudulent ones, a fraud prevention specialist has to keep all of these potential variables in their mind while performing manual reviews. They have to make as much revenue for vendors while providing the end user a seamless purchasing experience. The fraud prevention specialists are the line of defense between the good guys and the bad guys. Occasionally, because they are human, they will make a mistake.
But what if we could take the human variable out of the equation?
Tabitha Stang, Global Manager of Fraud at cleverbridge, has asked herself this same question over the years. After attending a presentation from a company talking about its 0 percent manual review rate using statistical techniques, Tab started to wonder if this was possible at cleverbridge and how to make it a reality. She graduated with a Master of Applied Statistics from Colorado State University and went to work solving this problem: how to create a machine that thinks like human but is superior to a human?
That presentation inspired Tab to focus on the idea of using a predictive model to reduce cleverbridge’s manual review rate to 0 percent (while still preventing fraud & keeping false positives down). “Predicative modeling is the process of creating, testing, and validating a model to best predict the probability of an outcome,” explained Tab. “In our case, an outcome is whether an order is fraudulent or not.”
Tab is taking all the expert knowledge of her fraud prevention specialists and using that to create a model, using potential fraud variables, that will make the correct decision every time — even if it looked at millions and millions of orders. By replacing the human element, the occasional mistake is removed from the equation. Fraud prevention specialists will now get to focus on more complex fraud projects, developing the model, and creating specialization opportunities. This change empowers our team to find and fill gaps and take charge of their future.
Tab has spent most of her waking hours working on this project. She was lucky enough to find another person at cleverbridge who could be able to speed up the process – Soukaina Messaoudi, who has a Master’s Degree in Informational Quality from University of Arkansas at Little Rock.
Tab and Soukaina have spent numerous hours creating, testing and retesting various models to develop a tool that could react and screen like one of our human experts. On March 26th, Tab will present the process to industry peers and experts in Las Vegas, in a presentation titled “Using a Predictive Model to Reduce Manual Review.”
As fraudsters have evolved over the years, they learned what most fraud prevention teams are looking for on any given order. Fraudsters are smart, but we have to be to smarter. The EMV switch is going to drive fraudsters online and out of the stores. Fraud prevention tools have to be advanced today to stop the onslaught of card-not-present (CNP) fraud tomorrow.
Using a machine to accurately predict the purchasing behavior of a fraudster is one of the ways we can keep up, and hopefully, surpass their advantages. If a fraud prevention team can work proactively, instead of reactively, they can look to reduce fraud (not just manual review rates) to 0 percent for any given company. If we can predict what a fraudster is going to do before they actually do it, we will have entered the Age of Precrime. Okay, so I am not advocating for that type of Minority Report style of prevention, but using a predictive model keeps fraud prevention teams everywhere half a step behind the fraudsters, instead of the 5-20 steps behind that fraud prevention specialists often find themselves.
The human mind is amazing. Its ability to process even the most minutiae of information and make parallels and comparisons is what has driven innovation for so long. We are now nearing a fork in the road between the human mind and human technology. The human variable will never be out of the fraud prevention equation entirely, but if we can evolve the equation and utilize more technological variables, our ability to predict and prevent fraud is endless.
The time for preventing fraud with predictive analytics has arrived. Putting in the time and effort now will have an exponential cost savings in the end — less orders reviewed, less fraud, less chargebacks and a better customer experience. The singularity has arrived, and I, for one, welcome our new technological overlords.