As we marvel at the ability of new technologies to change the way we work and do business, we sometimes miss one of the more fundamental effects of technological advances. They very often change the way we think.
Consider one of the most nettlesome problems for online merchants: The costly problem of false declines, also called “false positives.” False positives are a direct result of the fear of fraud. Legacy fraud prevention systems often raise red flags on unusual orders and out of an abundance of caution, an online merchant will hold the order back, including in cases when the order was legitimate.
The declined order costs the merchant a sale, distorts what the merchant knows about the conversion performance of the item and leaves a consumer insulted and hardly inclined to give the merchant another try.
But the rapidly emerging field of guaranteed fraud protection has changed the game dramatically when it comes to false positives, as was recently pointed out by Signifyd Head of Marketing Stefan Nandzik writing for payments publication The Paypers.
Give false positives some love
Nandzik explained in his piece declaring that it was time to give false positives some love, that the latest disruption in the fraud space makes it clear that the goal of fraud protection isn’t to prevent every incidence of fraud.
“In fact,” Nandzik wrote, “in order to achieve true fraud protection, merchants or their fraud vendors must be willing to ship orders that they are convinced are fraudulent. They need to test the bounds of what is fraudulent and what is legitimate. They need to go up to the line representing fraud and then step over it.”
How can that be? How can merchants afford to ship fraudulent orders? How do fraud managers explain to their bosses that they are intentionally shipping bad orders? How do their bosses explain it to board members and investors?
It all comes down to guaranteed fraud protection, an innovation that relies on big data, machine learning and human expertise to shift the liability away from merchants and onto fraud-prevention providers. Under the guarantee model, fraud protection providers agree to pay chargebacks and other fraud costs on any order they approve that later turns out to be fraudulent.
Ideally fraud-prevention providers see transactions from thousands of merchants and have a rich and evolving data set that allows them to spot fraudulent orders — and also legitimate orders that might have some troubling signs. Signifyd, for instance, has likely seen 60 percent of a merchant’s customers before they conduct their first transaction with that merchant. In other words, the customer has made purchases at other merchants in Signifyd’s 5,000-merchant network.
But fraud isn’t static. Fraudsters change tactics at an alarming pace, trying to outsmart even the smart machines that provide protection today. Therefore, guaranteed fraud protection systems need to occasionally ship bad orders to keep up with the sinister methods of sophisticated fraud rings.
“How else are smart machines going to learn all the permutations of fraud — or all the the different looks a legitimate order can display?” Nandzik wrote in The Paypers. “Think of the tactic as the fraud-prevention equivalent to a vaccine to prevent disease. A vaccine contains the very antigen that causes the disease. It’s how the body learns what to combat.”
Think about it: If you’re an online merchant and you never see a chargeback, you’re no doubt too restrictive in the orders you ship. How many of those declined orders were actually false positives? Here’s a clue: Business Insider reported on a year’s worth of declined orders industry-wide, concluding that U.S. ecommerce retailers lost $8.6 billion to false declines in 2016. Digital Transactions followed up with a report citing LexisNexis data that showed that false positives increased to 35 percent of rejected orders in early 2016, up from 25 percent during the same period the year before.
Shipping suspect orders is a form of R&D
In business terms, then, retailers should think of shipping suspect orders as a form of research and development, at least if they have deployed guaranteed fraud protection. Every order teaches the machine valuable lessons, none more valuable than what a fraudulent order looks like. If a system withholds every questionable order, the system will constantly be pushed in the direction of becoming more conservative. Every significant data point fed into the fraud model will be the result of a bad order, which constricts the pipeline of outgoing orders.
Without testing apparently fraudulent orders, the system never gets a signal that, “Hey, what looked like a questionable order was actually legitimate. These orders should go through.”
High-quality guaranteed fraud protection systems are constantly testing the boundaries between fraudulent and legitimate. It’s a practice, as Nandzik wrote, that allows retailers to discover the positive in false positives.
Photo by iStock.
Contact Mike Cassidy at firstname.lastname@example.org; follow him on Twitter at @mikecassidy.