A common theme during the recent SAS and FICO user conferences was how to use Big Data to make fraud decisions faster, more accurately and without impacting the customers in any negative way.
Big Data is basically about 3Vs: Volume, Velocity and Variety of data to gain veracity and value in fraud management. Volume and Velocity are nothing new: fraud management products have long been capable of analysing terabytes of data in billions of transactions - in real time.
What's really new for Fraud Management about Big Data is Variety: Using all types of new information to make better decisions with lower false positive rates. The new data sources that are increasingly used in Fraud Management are:
- Social network data. Has this user been writing about committing fraud on Facebook? After seeing how dumb some criminals can be, this data source is pretty important.
- Geolocation of mobile devices. The fraud management system should warn ahead of time if a user has been in the same location as the ATM when he/she used her ATM card to empty her bank account)
- Identity and Access Management systems logs. The fraud management system should warn ahead of time if the authentication system in front of my customer facing system see any evidence of the user logging in from a risky geography or from a new device before the user emptied their bank online by making unauthorised transfers to a mule account)
- Textual and unstructured data. The fraud management system should warn ahead of time if, for example, a medical provider or insurance adjustor is always using the same combination of terms of "suture removal" or "rear hit accident" in suspicious contexts or just in an excessively repeated way)
- Link analytics data. Using link and entity graphs beyond just Googling and link analytics risk score that can be injected into a holistic fraud score is something that can help identify fraud rings and collusion much faster.
Posted by Andras Cser