ml-vs-rules

4 Reasons Why Fraud Prevention Needs to Move Beyond Rules Based Engines

By Dr. Pedro Bizarro, Chief Data Science Officer @Feedzai

Commerce has evolved from its many different forms – from in-store purchases and mail order businesses to online market places and omni-channel commerce. Digital downloads and instant fulfillment services like Uber are transforming the way people consume goods and services. Online shopping has slowly eroded the traditional brick and mortar space of shopping malls. Consumers’ retail trips in November and December, the two biggest shopping months out of the year, dropped from 35 billion visits in 2010 to 17.3 billion visits in 2013, according to a report from Cushman & Wakefield.

In spite of rapid rise and adoption of modern commerce, fraud prevention is still based on archaic methods that are rigid, resource constrained and time consuming. Rules-based engines were thought of as the holy grail of fraud prevention. Partly due to the very simple fact that they worked … in some cases. Rules are effective to detect simple, non-changing, known patterns such as validating black lists or performing velocity checks. However, rules-only systems have a very hard time to distinguish true risk from normal behavior: is spending $400 in a hotel in Rio de Janeiro a sign of a cloned card being used abroad or is it simply an executive going about her business? It really depends on the history of that card, card holder, terminal or hotel, and of what just happened before that suspicious transaction. Thankfully, with the advent of real-time systems, big data, and machine learning, hundreds or thousands of fields can be combined to build predictive models that can detect fraud more efficiently and accurately. Rules simply cannot do justice to the magnitude of data and variety of ever-evolving fraud that exists today.

Businesses, while aware that they need to adapt, are wary of new tools and technologies that promise better, more sophisticated ways to identify and prevent fraud. The procrastination, while justified, limits the foresight that companies can proactively develop to not only manage risk but also gain competitive advantage.

Here are 4 reasons why companies have to start thinking beyond rules-based engines.

  1. Rules-based engines require active management: Rules are effective only when they are actively monitored and managed by dedicated fraud teams. This means fraud teams need to be staffed to review manual review queues, rejects and chargebacks and recalculate thresholds that can then be codified in a rule. This makes fraud prevention reactive in a fast moving business environment.
  2. Rules-based engines are one size fits all: Like hamsters on a wheel, rules are trained to do what is told without adding intelligence. They follow a binary view of whether the rules criteria are met or not. They do not dynamically adjust themselves for normal behavior or for seasonal fluctuations. This results in more false positives and customer insults.
  3. Information silos ignore the benefits of collective intelligence: Evolving ecommerce landscape has multiple touch points from mobile, to online to cash less commerce. Rules that are based on a single channel behavior do not provide a holistic view of consumer activity across multiple channels. 68% of fraud today is cross channel – Having the ability to build customer profile across channels provides risk models more data to base their decisions on.
  4. Rules are prone to human errors and bias: Rules are classic examples of Garbage-In-Garbage-Out. Incorrect and poorly coded rules increase manual review queues and continue to result in high fraud rates.

While rules based engines serve their purpose in certain environments, their ability to adapt to the modern world of big data is limited and restricted by design.

Machine Learning can help overcome the limitations of rules-based engines. By studying the pattern of good transactions and comparing it to fraud attacks, machine based algorithms can process data with greater speed and efficiency than humans can. The result is better detection with fewer false positives and lower manual reviews.

To learn more about machine learning and its application to fraud prevention, download the latest Feedzai e-book – A Primer on Machine Learning for Fraud Prevention.