Anyone who’s owned a car of a certain age knows the classic conundrum: When the vehicle wheezes and sputters, is it better to junk the thing and start anew, or is the problem nothing that a few upgrades here and there can’t solve?
Large ecommerce enterprises have been faced with a similar dilemma in recent years when considering the best way forward with their fraud-protection efforts. Many of these retailers are big operations with considerable money sunk into their legacy fraud management teams and systems. They’ve spent years developing and modifying the rules underpinning the barriers they’ve erected to keep fraudsters out.
Their fraud systems are tightly integrated with other vital systems that have been added over years of retailing.
The prospect of ripping out and replacing existing systems is painful, expensive and more than a little bit scary. Sure, the old thing is far from perfect — missing too many fraudulent orders, declining too many legitimate orders — but who knows how a new system would perform? Can’t we just decide that good enough is good enough for now?
The problem, of course, is that the world of fraud is evolving at hyper-speed. Fraudsters are constantly changing tools and tactics in an effort to outrun online retailers’ protection.
Gartner in its “Begin Investing Now in Enhanced Machine-Learning Capabilities for Fraud Detection,” published in July 2017, noted:
“New fraud attacks are happening with such speed and sophistication that rule-based and traditional risk scoring models cannot keep up; false positive rates are increasing while detection rates decrease.”*
The fraud protection industry, of course, is not sitting still while everything changes around it. New models of protection, based on advanced machine learning and shifting liability from merchants and to fraud-prevention providers are establishing a new market segment of guaranteed fraud protection. For now, the providers of guaranteed fraud protection are fairly young and relatively small as companies.
Guaranteed fraud protection vendors provide nimble flexibility
But the best among those young and small companies provide something legacy players don’t: a fresh and nimble approach to fraud that provides a predictable — and often lower — total cost of fraud for retailers who’ve been buffeted by a growing fraud threat.
So, what to do? (Hint: “Nothing” is the wrong answer.)
Some would argue the best course for established retail enterprises with legacy, rules-based fraud prediction systems is to break out of the either-or, black-or-white, one-or-zero mindset when it comes to fraud protection. Why not add an advanced machine learning fraud solution to an existing rules-based system?
Gartner, in its report, points out that fraud leaders need to, “Define and articulate the use cases, metrics and business objectives related to an investment in advance fraud analytics to ensure that the selected solution is the right technology for the job.”*
Put another way: It is always wise to understand the job you want to do before you chose the tools to do it. Gartner offers some sound advice in just how to figure that out.
“A 90-day proof of concept (POC) with an unsupervised machine-learning solution will be the best way to determine the effectiveness of the approach in identifying anomalies for a particular use case that had not been detected by predictive strategies,” Gartner wrote in “Begin Investing Now in Enhanced Machine-Learning Capabilities for Fraud Detection.”*
Learn more here about how large enterprise retailers can best scope out a proof of concept approach to advanced machine learning fraud protection and read about how Jet is using Signifyd to protect itself from fraudsters.
Makes sense. Why not try the new, different thing while keeping the familiar in place? That way, the existing predictive fraud tool is enhanced by the advanced machine-learning tool. Otherwise, older systems can grow less useful as their fraud decisions are reinforced over time without factoring in the changing nature of fraud.
Consider false positives, for instance. A machine-learning, guaranteed fraud protection model means fraud-prevention providers can choose to ship some orders that are almost certainly fraudulent. Ecommerce retailers are protected by the 100 percent financial guarantee and the provider deals with millions of orders, meaning the cost is spread widely. In fact, the cost is in the interest of research and development.
By shipping an apparently fraudulent order, the provider learns whether its machine’s decisions are the correct ones. If the order turns out to be legitimate, that data is fed back into the model and the machine becomes more precise, reducing the number of false positives.
Traditional fraud solutions, which look backwards at past activity and outcomes, are reinforced in only one direction. When a fraudulent order is shipped, they learn one more thing not to do. They inevitably become more conservative over time, increasing the number of false positives.
Investing in the new while keeping the old is the path to the future
Not only does pairing the old with the new provide a check on traditional systems’ one-way learning, it means large retail enterprises don’t have to rip out their old systems entirely and replace it with something new.
Instead, enterprises can invest gradually in the new wave of fraud protection. Given that providers of guaranteed fraud prevention are cloud-based software-as-a-service companies, ecommerce retailers have maximum flexibility with scaling and changing or ending relationships with modern fraud-protection vendors.
That allows even the largest retail enterprises to act like a startup, to succeed or fail fast by trying new solutions — starting small and growing new initiatives; expanding the successes and shutting down the failures.
In the end, embracing a measure of advanced machine-learning fraud protection now, without leaving old fraud tools behind, gives the best of both worlds to large ecommerce enterprises scrambling to keep up with the rapidly changing world of online fraud.
* Gartner “Begin Investing Now in Enhanced Machine-Learning Capabilities for Fraud Detection,” Tricia Phillips, Avivah Litan, Danny Luong, 14 July 2017
Photo by iStock
Contact Mike Cassidy at firstname.lastname@example.org; follow him on Twitter at @mikecassidy