Real-time Fraud Detection & Prevention

CORTEX aiFRAUD is an end-to-end integrated AI-based solution that provides a comprehensive real-time fraud detection and prevention using all available structured and unstructured data. From a detection perspective, the accurate identity of an individual or fraudulent actor is determined through a combination of machine learning-based models, scoring and rules. Where relevant, intelligence from past discovery and investigation processes are also used to recognize fraudulent activity patterns and highlight potential payment submission improprieties for further investigation by the investigations team. In order to discover fraudulent patterns, a rich set of AI capabilities is leveraged to identify non-compliance by retrospectively reviewing historical data, analyzing patterns and anomalies to identify individuals or organizations that might be developing fraud schemes. As illustrated in Figure 21 other aspects of a comprehensive fraud detection and prevention solution includes quick response to criminal patterns, activities and intentions as well as proper investigations into suspicious activity that will support the compilation of evidence and provide the thorough analysis required to build more compelling cases for prosecution and recovery, or denial of payment.

Although CORTEX aiFRAUD solution makes use of a toolbox of rich analytical and decision- making capabilities, a machine learning-based approach is typically preferred above a pure rules-based approached for the following reasons :

  • Machine learning concerns with algorithms that can learn from data such as multivariate statistics, automated predicted analytics, and deep learning
  • There is a need for scalable and computationally efficient prediction models
  • Typical rule based engines do not make use of information from multiple attributes at the same time
  • Traditional ideas of finding patterns through hand crafted, careful querying, does not scale to large data sets
  • Increases in accuracy can lead to significant savings
  • Fraudsters are becoming increasingly smarter and adaptive. There is a need for cost-effective solutions that can model complex attack patterns not previously observed

Depending on the industry and application, fraud detection models are typically deployed on the following three levels :

Transaction level

  • Employs state-of-the-art machine learning and statistical models to flag fraudulent behaviour up-front
  • More sophisticated algorithms after transaction is complete

Account level

  • Monitor level activity to identify abusive behaviour
  • Abusive pattern includes frequent payments, suspicious profile changes

Network level

  • Monitor account-to-account interaction
  • Frequent transfer of money from several accounts to one central account

The CORTEX aiFRAUD solution typically utilizes different machine learning techniques for discovery and prediction of fraud :
Supervised learning : Predict Fraud

  • Collect historical transactions
  • Learn from past examples of fraud
  • Predict fraud (in real-time)

Unsupervised learning : Discover Fraud

  • Segment transactions
  • Investigate potentially new fraud

In one use case a major medical scheme administrator realized that they had a significant problem in terms of fraud amongst their schemes. It was estimated that around 10% of all contributions were lost due to fraudulent activity by service providers and members. Some data mining highlighted significant “fraud/abuse” across the three schemes. This was after the existing fraud detection has been “passed”. The 3 schemes represent approximately 15% of the total members of this administrator. If the results are extrapolated the outcome is clearly much more significant.


Relevant industries : Financial services, Healthcare & Life Science, Retail & eCommerce, High Tech & Telecommunications, Transportation, Media & Entertainment, Education, Legal, Travel & Tourism, Resources & Mining, Utilities, Manufacturing, Construction, and the Public Sector.