The Simility Blog
Driving Improvements in Fraud Prevention As Threats Grow in Sophistication
Sharon LuceroAugust 02, 2019
As the world digitizes, organizations and their customers are increasingly exposed to internet fraud. It’s the price we are paying for unfettered online growth. But it doesn’t need to be this way. Anti-fraud solutions have evolved over the years, from simple rules to the kind of advanced machine learning-powered solutions delivered by Simility. We can chart the progress of them using a simple automotive metaphor.
Today’s next-generation fraud prevention tools are akin to the new breed of driverless cars hitting the roads. For those who may be uncomfortable handing over control, the key to widespread adoption will be explainability and empowerment.
The fraud journey
Online fraud is already causing business losses on a grand scale. It’s not just the chargebacks and lost products that are costing them. If fraud prevention measures add too much friction, they can lead to lost sales, customer attrition and high operational overheads stemming from too many manual reviews. But go too far the other way and fraud costs become unsustainable.
The goal is automated fraud solutions that catch 100% of unauthorized transactions without impacting the customer experience for legitimate users.
Just as the automobile industry started life only producing cars with manual transmissions (stick shift), anti-fraud tools began many years ago with simple rules and blacklists. Analysts had to work with black and white–listed identities, and pick out specific data points to compare with values already associated with fraud.
Next, simple linear models emerged to help better spot fraud. For example, large value transactions and multiple transactions within a short timeframe could trigger alerts. The resulting machine and data-assisted rules these systems generated could be likened to automatic transmission: providing a helping hand to those in charge.
The industry then evolved to more sophisticated models and feedback loops, with intelligence created and shared across multiple business entities — not just gathered at a single point in time but contextual and dynamic. Black box machine learning models emerged to help make predictions, albeit with limitations: most notably that they don’t allow humans to understand how they arrive at their decisions.
Confidence in the future
Over these years, fraud has grown in sophistication thanks to automated tools capable of cracking open user accounts with ease, anonymizing techniques, and dark web sites on which huge volumes of breached identity data are traded. At the same time, data and how it is processed has become more complex, just as the systems of engineering underpinning modern cars have evolved.
So where do we stand today? There’s a huge opportunity to drive value for the organization by investing in advanced, behavior-based fraud solutions that leverage machine learning to provide highly accurate, automated decisioning. Yet just as driverless cars have their detractors, there may be some who are yet to be convinced by this new way of doing things.
To get them on board we need to take a similar approach, based around explainability and empowerment. Empower the motorist and explain how the driverless car works, to allay any fears over safety. And empower fraud teams with clear-box machine learning which explains how a fraud score is calculated, allowing them to tweak settings and streamline reviews.
With organizations expected to spend billions on fraud solutions each year, let’s make sure it’s on best-in-class, data-driven solutions like Simility’s Adaptive Decisioning Platform.