Customer Story

Machine learning at scale to fight credit card fraud

About the company



Barclays, a British universal bank based in London, moves, lends, invests, and protects money for businesses and clients worldwide. Older than the UK itself, Barclays has a long history of offering innovative financial products, including personal and corporate banking, cards, insurance, mortgages, payments, ecommerce, wealth management, and more.


Fragmented platforms caused inaccurate decisions and missed fraud activity

With the ever-growing number of devices, channels, and amount of available data online, opportunities for fraud are also on the rise. Fraud techniques and tools are becoming more sophisticated and accessible with easy access to low-cost computing. Customer needs and behaviors have become more complex, with demands for real-time seamless transactions, declines in brand loyalty, and increased comfort in sharing personal information over social media.

Barclays experienced an 83% increase in data from 2015 to 2020 and was projected to generate 463 exabytes daily by 2023. With over 20 million customers making over 30 million payment transactions a day, Barclays business units found it difficult to accurately analyze user profile information to detect and prevent credit card fraud.

Problems included:

Inefficiencies and inconsistencies

With multiples of one product, business units couldn’t share user profile data or fraud rules, resulting in false positives or false negatives.

Ballooning costs

Costs to manage and run multiple custom platforms were uncontrollable — hardware resourcing, technical staffing, operations.

Missed fraudulent activity

Vertical and horizontal scale-out problems led to an inability to store or manage the user profile data required to capture fraudulent activity.

Inability to meet SLAs

Performance at peak loads and unpredictable latencies caused delays in detecting fraud, resulting in failures to meet payment cycle SLAs

Limited opportunities to grow and scale

Barclays found it increasingly difficult to evolve complex, bespoke engineering solutions to achieve goals and meet requirements.

A single, shared machine learning platform

The payment fraud team at Barclays implemented a machine learning fraud detection platform, powered by Aerospike, as a single platform across all business units. The platform accesses and analyzes user profile data for credit card fraud at sub-millisecond speeds. Characteristics of the solution include:

High throughput with low latencies

Scalability to process millions of transactions daily and handle 6x data growth

Strong consistency and security

Barclays has eliminated false positives and negatives, loss of user profile data, and is able to keep up with security requirements as they evolve

Real-time data handling

Aerospike’s Hybrid Memory Architecture allows Barclays to stores indexes in RAM and data on disk

Single platform

Simplified architecture and reduced number of platforms compared to the original structure

Aerospike delivers performance and reliability while handling 6x data growth

With Aerospike, Barclays was able to fight financial fraud in real time with shared rules, user profile data, and information across all its business units. The machine learning platform is able to meet current and future projected performance requirements with a simplified architecture and unmatched reliability.

80% reduction in latency

Vastly improved latencies compared to the original system

4x greater throughput

Handles over 10M transactions per day

<100ms response times

Less than 100ms response time for 99.99 percentile of transactions

6x data growth

Scale to handle growth from 3TB to 30TB+ over 3 years

Surpass limits with Aerospike