Americans currently lose $50 billion a year to a variety of fraudulent practices, according to estimates from the Financial Fraud Research Center at Stanford University. These practices include theft of credit cards and personal financial information, unauthorized checks, forged documents, tax evasion, and the manipulation of mortgages, corporate financial statements, securities trading, and computerized banking. Financial services firms that fail to prevent this fraud also suffer significant damage to their reputations.
Financial fraud is a rapidly growing, multi-billion-dollar ‘business’, with Juniper Research predicting that online fraud alone will climb from $10.7 billion in 2015 to $25.6 billion in 2020. A reason for this is that the amount of data that needs to be processed and analysed now is overwhelming. The solutions that have been designed to automatically verify, analyse, and audit transactions to detect and prevent the fraudulent activity cannot handle the volume.
Automatic fraud prevention requires sophisticated approaches. These include:
- Statistical and multi-channel analysis
- Models and probability distributions
- Comparisons with user profiles
- Algorithmic analysis
- Data clustering and classification
- Artificial intelligence and machine learning
The one common element is that performing these activities in real-time or near real-time on extremely large datasets requires high performance and highly scalable technologies. It must also be accomplished while firms are simultaneously tackling other crucial and processing-intensive activities, such as ensuring regulatory compliance. For a deep dive into this topic, read “Powering Financial Fraud Prevention with In-Memory Computing,” a new white paper by GridGain Systems, a leading provider of open source in-memory computing solutions for the financial services industry.
Technologies used for fast data analysis
To attempt to detect and prevent fraud in today’s data-intensive environments, most financial firms rely on a variety of technologies, including:
Big data The first step in using financial data for fraud prevention is preparing the data for analysis. Big data technologies provide ways to organise large datasets into multiple pools and connect them for fraud detection and analysis.
Apache™ Hadoop® with MapReduce Financial transactions typically execute within milliseconds. The fraud detection technology must analyse a transaction, validate it, and check all available data pools without impacting overall transaction performance.
Complex Event Processing (CEP) with data streaming This technology involves looking at multiple incoming data streams and using artificial intelligence (AI) to identify potential fraud.
Data partitioning and parallel processing clusters In high-volume transaction systems, checking one transaction at a time for fraud is not an option. Systems with data partitioning and parallel processing across clusters make it possible to process multiple transactions simultaneously and distribute the processing load across the cluster – while maintaining data consistency.
Scalable data architecture Large financial institutions are experiencing 20-30% data growth year-over-year, and they can’t risk running out of space. A highly scalable data architecture should enable firms to keep adding additional storage without impacting performance.
The key to performance: In-memory computing
While all these technologies contribute to fraud prevention systems, they cannot overcome the inherent performance limitation of writing data to disks, typically the slowest activity in a modern computing solution. To overcome this limitation, many financial services firms have turned to in-memory computing to meet the stringent performance requirements of fraud prevention programmes. In-memory computing is faster than any storage-based computing method.
The performance and scalability revolution being led by in-memory computing is the result of a number of factors occurring at the same time. A primary driver of the uptake in in-memory computing is that the cost of memory has dropped roughly 30 percent per year since the 1960s. It is now affordable to equip clusters with terabytes of RAM to support big and fast data projects. This has led to a maturation in the in-memory computing market. Products used to require extensive customization to cobble together a solution that provided little more than the bare bones functionality of storing data in RAM. Financial services companies can now enjoy full-featured in-memory computing platforms that offer ACID compliant transactions, transparent scalability, high availability, and enterprise grade security. Some in-memory computing solutions now include SQL support and a full range of APIs, enabling developers to leverage existing code bases and get up and running quickly. Financial services companies that have implemented in-memory computing in their fraud prevention systems have reported processing transactions roughly 1,000 times faster than disk-based solutions.
Thanks to in-memory computing, financial services companies such as Barclays, Citi, Sberbank, and others are seeing a measurable difference in transaction performance at scale. For example, Sberbank, the largest bank in Russia and the third largest in Europe, has 130 million customers and ever-increasing transactions volumes. Its legacy systems were not able to sustain the pace of transaction processing to ensure an optimal customer experience. While testing in-memory computing solutions with distributed parallel processing, the bank was able to demonstrate one billion transactions per second in a test environment using only 10 Dell® blades with a combined memory of one terabyte and a total cost of only $25,000.
The in-memory computing solution Sberbank settled on provides several other required capabilities including:
- Machine-learning and analytics
- Ease of deployment
- Hardware independence of cluster components
- A rigorous level of transactional consistency
- The ability to conduct integrity checking and rollback on financial transactions
Inundated with ever-increasing amounts of data to process and analyse for potential fraud, financial institutions and other companies need highly scalable solutions that meet the performance requirements for real-time and near real-time fraud prevention. Fortunately, today’s mature in-memory computing solutions now offer a full range of mission critical features and an affordable way to achieve the performance, scale, and comprehensive approach required to stop fraud in its tracks.
Nikita Ivanov - Founder & CTO, GridGain