Customer Story

Protecting banks from fraud with unified, real-time behavioral intelligence

About BioCatch

BioCatch prevents financial crime by recognizing patterns in human behavior, continuously collecting more than 3,000 anonymized data points (keystroke and mouse activity, touchscreen behavior, physical device attributes, and more) as people interact with their digital banking platforms. With these inputs, BioCatch's machine learning models reveal patterns in user behavior and provide device intelligence that, together, distinguish the criminal from the legitimate. The company’s Customer Innovation Board, an industry-led initiative in partnership with American Express, Barclays, Citi Ventures, HSBC, Macquarie Bank, National Australia Bank, and others, collaborates to pioneer innovative ways of leveraging customer relationships for improved fraud detection. Today, more than 30 of the world's largest 100 banks and 311 total financial institutions deploy BioCatch solutions, analyzing 16 billion user sessions per month and protecting over 555 million people on more than 1.6 billion devices worldwide from fraud and financial crime.

Challenge

Building behavioral fraud detection at global scale

BioCatch’s behavioral analytics operate in real time, analyzing thousands of data points per session to produce instant risk scores. As the company scaled, its data volume increased substantially, while its architecture grew increasingly complex.

problem-icon

Fragmented architecture with dual systems

Online profiles were stored in distributed in-memory caches, while offline profiles lived in blob storage. The split increased complexity and made data synchronization difficult across regions.

problem-icon

High infrastructure costs for in-memory caching

The in-memory cache was costly to operate and maintain, requiring premium resources to meet performance SLAs for constantly updated profiles.

problem-icon

Inconsistent performance and race conditions

Cache warm-ups and delayed initialization often caused unpredictable response times, putting the 500ms SLA at risk.

problem-icon

Limited scalability for new fraud models

As more banks and behavioral signals were added, the existing system struggled to scale efficiently without adding more memory-intensive cache nodes.

Solution

Consolidating two data layers into Aerospike

BioCatch migrated both online and offline behavioral profiles into Aerospike, unifying its architecture under a single high-performance data platform built for predictable, near-in-memory performance at a fraction of the cost.

check-mark-icon

Unified behavioral profiles on one platform

Aerospike’s Hybrid Memory Architecture stores indexes in RAM and data on SSD, combining cache-like speed with durable persistence. This allowed BioCatch to merge previously separate cache and storage tiers into one high-performance database.

check-mark-icon

Predictable latency and bounded tail performance

Aerospike’s deterministic I/O path and bounded tail latency ensure that even the slowest responses stay within tight SLA limits, keeping fraud scoring consistent across billions of behavioral lookups.

check-mark-icon

Near-in-memory speed without cache dependence

With predictable latency regardless of cache hit rate, BioCatch achieved the same user-perceived speed as its prior in-memory cache, but with dramatically lower cost and risk.

check-mark-icon

Simplified operations across eight global regions

Running in eight Azure regions, Aerospike’s operational simplicity and self-healing design let BioCatch scale capacity horizontally while maintaining high availability and uniform performance across all regions.

Results

Real-time scale with lower cost and higher reliability

Migrating to Aerospike unified BioCatch’s data architecture, strengthened performance consistency, and cut operational risk and cost.

icon-integrity-white

50% reduction in operational risk

Eliminating cache synchronization and blob storage dependencies reduced failure points and race conditions, improving system stability by half.

tps-icon-white

Predictable performance under all conditions

Aerospike’s cache-independent design guarantees consistent sub-millisecond latency even during traffic spikes, warm-up cycles, or rebalancing events.

icon-cost-white

Lower total cost with near-in-memory performance

Consolidating from two data systems into one Hybrid Memory Architecture reduced infrastructure and maintenance costs while maintaining in-memory speed for real-time scoring.

icon-scale-up-white

Global scalability for AI-driven analytics

Aerospike now supports more than 5 billion profiles and 15 billion sessions per month, ensuring every behavioral insight and fraud signal is processed within 500 milliseconds.

Testimonials