At Strata, VoltDB, provider of an in-memory, translytical database, showcased a new database function that will enable real-time predictive fraud detection and prevention. With User Defined Function (UDF) support for SQL, VoltDB v7.6 enables enterprises to accelerate the identification and prevention of fraud before a transaction is complete.
The new capability will be in production at a number of institutions soon and is being piloted now at one of the largest banks in China.
“They have been in production for a year now and picked VoltDB because they were looking for something that would allow them to run fraud models at the scale they need for retail transactions,” said John Hugg, VoltDB founding engineer, at Strata. “They were attracted to the simplicity of VoltDB and that it could do the work in a way that was very manageable and cost-effective for them, and was one that they could replicate to other institutions. They can take the piece of software they built with VoltDB and deploy it for sale after sale.”
For all fraud use cases, whether it is Fintech, retail, or telecommunications, there is a component that is real time, and while that is a fuzzy word, said Hugg, for VoltDB that means determining whether a transaction should go forward in the moment, which is followed by a component where it is examined in retrospect and in aggregate and with the application of more sophisticated analysis.
“The question is how sophisticated you can be before the first transaction, and that is where VoltDB comes in,” said Hugg. “We come in to improve your rate of detection and lower the false positive rate in between the swipe and the authorization and that is at the forefront of fraud detection and can involve machine learning models that are generated on the big data side, and loaded into VoltDB so they can be applied in milliseconds.”
Many systems have low mean latency but VoltDB stands out because it can provide repeatable performance within the real world constraints it is given and still return an answer as quickly, said Hugg.
Helping companies get up to speed more quickly and test the technology before deployment in production, VoltDB has also recently launched an Enterprise Lab Program. The program provides a zero-cost license for non-production use whether it is a big bank, telco provider, or a service provider, and whether they need to get access to the database to add real real-time capabilities to transform existing applications, or build new fast data applications from ground up.
As VoltDB interacts with the large telcos and financial institutions, it has found that many have innovation labs where they like to do proofs of concept and build out solutions to see how they can use technology to innovate and be faster than their competitors. That is what the Enterprise Lab Program is targeted at, said Madhup Mishra, head of product management, VoltDB. “It allows organizations to be able to deploy our full featured enterprise capabilities inside their labs.”
When VoltDB considered what a next-generation processing system entails, it looked at what has changed in the 21st century, said Hugg: “More memory, commodity hardware, clustering in the cloud, and certainly parallel CPUs—you have to be parallel in the 21st century to get the performance out of the hardware—and we designed an architecture for that.”
The use case for VoltDB of event processing and reacting to events and making decisions came as more people started using it, and wasn’t from the start something VoltDB intended to do, said Hugg. Because VoltDB moved to the stored logic model in the server and over time has added more and more amortized work on the server, the materializations and indexes, people started using VoltDB for this mixture of analysis in milliseconds and immediate decision making.
“Our goal is to get as much sophistication and power into a real-time decision as possible—in the action,” said Hugg. “What we want to do is put more capability into those milliseconds than anyone else,” said Hugg. “If you know the kinds of questions you are going to be asking and you want to ask those questions 500,000 times a second we can build a system that can pre-materialize a lot of that work, that can push the work of computing those values to the continuous writes so that whenever you do the read, the answer is instant.”
For more information, go to www.voltdb.com.