Why is Proactive Fraud Management Important, and What Makes it Possible?
Unsupervised machine learning, account level detection, and the secret to staying ahead of fraud.
Proactivity and reactivity in and of themselves are neither good nor bad approaches to any given scenario. Sometimes, it is better to take the initiative early, other times it’s better to wait and see. When it comes to fraud, however, there is no question as to which is the better option — proactivity is always the right approach. Why? Because a reactive approach to fraud means the damage has already happened, and that’s never good. Proactivity means you’re taking action before the attack comes.
Proactive fraud management happens at two levels — the conceptual, and the tangible.
Embracing the Concept of Proactive Fraud Management
At the conceptual level, proactive fraud management is a kind of psychological battle waged against fraudsters. It’s an attempt to “get into the head” of malicious actors, to try and understand what they’re going to do next, and where and how they’re going to do it. Proactive fraud prevention is a sort of mash-up of weather forecasting, profiling, and good old-fashioned detective work.
As of this writing, the Women’s World Cup is in progress, and we can find an additional metaphor there for the face-off between good and bad actors on the fraud playing field. Consider a penalty kick scenario — the shooter, and the goalkeeper. Each knows something of the other, each brings a wealth of knowledge, skills, and resources to the battle, and each is trying to outmaneuver the other before the shot is ever even taken. As the shooter, you’re looking to best leverage your strengths to best exploit the goalkeeper’s vulnerabilities. As the goalkeeper, you’re faced with a dramatic choice — dive first, and hope you guess right, or wait to dive until you see the ball come off the shooter’s foot.
For those battling fraud, diving afterwards is too late — as it generally is for goalkeepers as well (look no further than the current debate around VAR and penalty kicks in this year’s Women’s World Cup to understand the significance of the issue). Reactivity allows an attack to take place, and harm to occur. Success is not measured by attacks prevented, but by how quickly the damage can be constrained. Not so with a proactive approach. Proactivity means neutralizing attacks before they can launch. Before damage can occur.
The Tangible Application of Proactive Fraud Management
At the tangible level, advanced proactive fraud management is best achieved through a potent combination of unsupervised machine learning (UML), and account level detection.
Unsupervised Machine Learning
Unsupervised machine learning has the advantage of not requiring historical labels, lengthy training times, or frequent retuning. With an advanced fraud management solution informed by deep domain expertise — one which offers a rich library of proprietary fraud features, and enables both auto and custom feature engineering — you can build and deploy high-performance models quickly and efficiently, and start receiving actionable insights right away. Most importantly, advanced UML algorithms can surface correlated patterns and reveal coordinated activities across seemingly disparate accounts and actions, making it possible to spot burgeoning attacks while still in development. DataVisor’s dCube is an example of this kind of solution.
Account Level Detection
Account level detection is the other key component of this process. To understand its importance, let’s consider a fraud use case — loan stacking, or, specifically, fraud stacking. Loan stacking is getting multiple loans simultaneously; fraud stacking is doing so with no intention of paying those loans back. Fraud stacking is generally achieved through techniques such as identity theft, account takeover (ATO), and the creation of synthetic identities. Once a fraudster has possession of multiple legitimate-seeming accounts, they can use them to fraudulently obtain loans. Reactive, transaction-level approaches are helpless to manage these types of attacks, because they don’t “see” fraud until a loan is defaulted upon. At that point, it’s already too late — the money is gone. Account-level detection, on the other hand, focuses on the accounts themselves — how and when they’re created, and, most importantly, the context in which they’re created. Using a solution like DataVisor’s dCube, you can detect fraudulent activity at the point of account creation; for example, were a large number of accounts created on the same day, all with similar email addresses? Were the application sessions all spaced apart at identical intervals, and did they all go through the identical sequence of web pages to submit their applications? Signals like these indicate coordinated fraudulent activity, and surfacing these signals early enables you to proactively take action on the account before it can be used in a malicious attack.
The Drive to Stay Ahead
Data and technology have always been part of strategy — in sports, and in fraud. Professional athletes today undergo VR immersion training to enhance performance, even as their managers and trainers leverage massive volumes of sabermetrics-esque data to make adjustments to swings, strokes, strides, and more. Data management continues to change the fraud game as well, and the winners are those who best use data to stay ahead. This is what proactive fraud management is all about — staying ahead of fraud. As DataVisor Co-Founder and CEO Yinglian Xie recently wrote in the June 2019 Digital Fraud Tracker:
“When the choice is between eliminating fraudulent accounts the moment they’re created, or repairing damage perpetrated by fraudulent accounts that are allowed to linger, the choice is clear. To protect their organizations, their customers, and their data, financial institutions must adopt proactive, AI-powered fraud solutions that can detect and deter new and unknown attack types and prevent damage before it happens.”
To learn more about proactive fraud management, please request a demo to experience dCube, from DataVisor.
