Fraud Detection and Prevention
Fight Fraud with TigerGraph
Fraud is Expensive and Becoming More Prevalent
Fraud is a growing problem in all industries with ecommerce losses alone exceeding $57.8 billion in 2017. For every dollar involved in fraudulent transactions, the financial services industry spends $2.67 dealing with chargebacks, fees, interest and labor. Healthcare fraud is estimated in tens of billions per year.
Fraudsters Are Becoming Increasingly Sophisticated
Fraudsters are getting more sophisticated over time creating a network of synthetic identities combining legitimate information such as social security or national identification number, name, phone number and physical address. Legacy fraud detection systems are largely based on analysis of behavior of an individual business entity such as a customer, citizen, device, doctor or healthcare provider and find unusual patterns in that behavior.
Fraud is perpetrated using fake accounts created with synthetic identities. Each individual fraudster account looks and behaves much like a legitimate account, making detection much harder with traditional fraud detection solutions. Detecting fraud requires going beyond individual account behavior, analyzing relationships among groups of accounts or entities over time often combining information from third party sources. Traditional fraud solutions built on relational databases were not designed to address this challenge.
Why TigerGraph, a Native Parallel Graph Database for Fraud Detection?
Detect Fraud With Deep Link Analytics
Take the example of fraud in healthcare, related to opioid addiction treatment centers. A provider is prescribing a large number of patients to one specific opioid or drug addiction treatment center. Traditional fraud solutions can’t find anything unusual based on the information available for the doctor as well as the treatment center.
TigerGraph’s deep link analytics taps into the enterprise knowledge graph from a third-party source – such as Thomson Reuters or Dunn & Bradstreet – to find all known administrators for the drug treatment center and their current and previous addresses listed in the public domain. One of the previous addresses for the administrator is very close to the address for the physician prescribing patients to the drug treatment center. In order to detect this hidden relationship between the doctor and the administrator, TigerGraph executes a query that goes eight hops across claims for patient visits, opioid treatment center claims by the same patients and combining that with the third-party knowledge repository data such as addresses and phone numbers to find the collusion. Traditional fraud solutions built on relational databases struggle to incorporate new data sources from third-parties due to rigid schema and require computationally intensive database joins, rendering deep link analysis infeasible.
Detect Fraud With Real-Time Analytics
Fraud detection is time sensitive: every passing minute, hour and day that fraud goes undetected results in increasing losses for your organization as well as for your customers or citizens. TigerGraph is purpose-built for real-time fraud detection to address this challenge. Let’s consider the example of China Mobile, the world’s largest mobile service provider serving over 900 million subscribers. China Mobile is using TigerGraph to detect phone-based fraud in real-time by analyzing calling patterns of pre-paid subscribers. Subscribers are alerted in real-time of a potential fraudster call, with high probability fraud calls being redirected to the call center for China Mobile for investigation.
Improve Fraud Detection With Machine Learning
Less than 1% of the total call volume for telecom or claims data for healthcare and government benefits or payment transactions for financial services are fraudulent. This means that the machine learning models do not have sufficient training data with confirmed fraud activity to learn and improve accuracy of fraud detection.
TigerGraph’s native parallel graph is purpose-built to address this challenge. Consider the example for phone-based fraud detection at China Mobile, where TigerGraph creates over 118 features for each phone in real-time by analyzing relationships among subscribers over time, identifying a good phone, owned by a regular customer, and a bad phone, suspected to be a fraudster.