Barclays processes approximately 30 million to 40 million payment transactions per day for its 20 million customers. It also has many fraud detection solutions across its business units. Barclays understood that transaction fraud detection needed ultra-low latency. However, the inability to seamlessly re-use large-scale user profiles across different use cases across its business units led to multiple complicated, bespoke engineering solutions. These solutions were becoming increasingly complex to maintain and to evolve and posed a major limitation in achieving the company's strategy. More risk is created when End-to-End fraud detection takes longer. These risks include STIP (stand-in processing), data consistency issues that can increase false positives and negatives for future transactions, as well as stand-in processing. A thorough expert analysis revealed that the majority of these problems could be traced back to a limited database deployment and technology. Barclays' payment fraud team developed a fraud detection system that solved all its problems by implementing a new database technology. The result could increase the Barclays database from 3TB to 30-plus in three years. It also allowed for fraud rules to be shared across platforms. Machine learning was made possible with a goal to have a maximum of two digits (100) response time for transactions up to the 99.99 percentile. Maybe the credit card fraud database is still relevant?
