Credit Card Fraud – Why the Database Matters

This course is a part of Big Data LDN 2019 , a 101-course Paths series from Pluralsight.

Barclays processes circa 30 million-plus payment transactions a day for its 20 million-plus customers and had many fraud detection solutions in place across its various business units. Barclays knew transaction fraud detection required ultra low latency, but not having the ability to seamlessly re-use its large-scale user profile datasets across use cases across its business units resulted in multiple complex, bespoke engineering solutions. These solutions became increasingly difficult to both maintain and evolve, and thus posed a significant limitation to achieving the company’s strategy. As more time is taken during an End-to-End fraud detection process, more risk is introduced. These risks include stand-in processing (STIP), the rise of data consistency issues, which in turn can lead to increased false positives and false negatives for future transactions. Comprehensive expert analysis of the problem revealed that most of these issues could be tracked back to a limiting database deployment and technology. Implementing a new database technology, the payment fraud team at Barclays ended up with a fraud-detection system which solved its problems. The resulting solution could scale the Barclays dataset from 3TB to 30TB-plus over the course of just three years, share fraud rules across platforms, and facilitate machine learning consistently with an aim to achieve a maximum of two digit (<100) millisecond response time for the 99.99 percentile of transactions. Perhaps the database for credit card fraud matters after all?