Rethinking Your Fraud Approach: Velocity Variables and Stream Analytics
In our last installment of “Rethinking Your Fraud Approach,” we examined one tactic—the application of network analysis—for using data you’ve already collected to mitigate fraud against your business and your customers. Today we discuss velocity variables and stream analytics—just one more way you can leverage your first-party data for better fraud detection and prevention.
Velocity of use is simply the number of times a certain data element is used in transactions or attempted transactions within a given timeframe. The time period of interest will vary with the type of data and transaction being examined, but the primary goal is to spot unusual trends that may indicate fraudulent activity. Examples of common velocity variables include:
- The number of times a credit card or account number is used per day
- The frequency a physical address, email address or phone number is used for new account creation or credit application in a month or year
- The number of open accounts for a given customer over a chosen interval
Stream analytics refers to examining changes to these velocity variables in real-time to make decisions or take actions. This analysis of “data in motion” is critical if financial service providers are to detect and prevent sophisticated attacks by today’s fraud rings, which can repeatedly use stolen PII over a period of minutes or hours, not weeks or months. This activity extends beyond merely making unauthorized purchases to outright identity theft, identity morphing, and account takeover.
To be an effective indicator of fraud, the organization must properly configure and fine-tune the velocity variables to be examined. As with link variables for network analysis, the organization can define and store velocity variables on the backend by specifying the data elements, velocity thresholds, and time intervals of interest. The organization can then create alerts to indicate when a velocity significantly deviates from its baseline and trigger an investigation.
In addition, velocity variables and 3rd-party data can be combined through machine learning techniques to better assess the overall risk of a person or transaction. We’ll talk more about applying machine learning in an upcoming post.
At Enova Decisions, we’ve packaged 15+ years of know-how in fraud detection and prevention with technology solutions that let you reduce fraud losses by up to 75% in the first few months. We can even ingest data from your existing stream analytics to develop a predictive model to make your fraud detection even more effective.