Multi-National Bank Transforms Fraud Detection, Customer Analytics, and Live Monitoring with Real-Time Operational Insights

A large, multi-national bank with over 100 million customers needed to gain immediate, actionable insights from its data streams to enhance fraud detection, improve customer experience analytics, and optimize platform observability. Operating in multiple regions and processing hundreds of millions of transactions annually, the bank needed a platform that could support its growth while ensuring security and performance. After evaluating several options, the bank chose StarTree to address its business operations and analytical challenges.

Challenge: Keeping pace with growing data demands

The bank was experiencing significant issues as its customer base expanded and the cost to ingest, store, and query first-party data exploded. Meanwhile, the existing infrastructure was struggling to keep pace with growing demands from a performance and management standpoint.

In fraud detection, the bank relied on risk models that only processed data in daily aggregation batches. This approach created vulnerability windows where fraudulent transactions could remain undetected for up to 24 hours. 

Cost management was another concern. The bank had initially deployed commercial solutions including Amplitude for behavior analytics, Redis for in-memory caching, and Splunk for log analysis. As data volumes grew, licensing costs increased proportionally, creating an unsustainable cost structure and limited areas to improve the user experience. Amplitude and Redis struggled to handle increasing user loads without significant cost increases. 

The bank also struggled with limited visibility into its technical operations. Engineering teams lacked real-time insights into system performance and user behavior, making it difficult to diagnose and resolve issues quickly. Observability data typically had a 30-minute to 24-hour delay, further complicating troubleshooting.

Their Apache Spark environment, which powered critical data processing workloads, was particularly problematic. When Spark jobs failed, teams needed between two and four days to properly diagnose root causes due to limited visibility into task-level metrics and log data. This extended troubleshooting time directly impacted business operations.

These challenges culminated in operational complexity, with their analytics architecture requiring the movement of data between multiple systems, creating a fragile and expensive environment.

Transforming fraud detection with StarTree

The bank implemented StarTree, powered by Apache Pinot, as a unified platform to deliver real-time insights across several key operational areas. With StarTree, the bank was able to  address these challenges, targeting high-value use cases first while building a foundation for future expansion.

Fraud detection platform

The bank deployed Pinot to replace their existing NoSQL implementations, enabling a shift from latent to real-time, event-level risk models. The new system analyzes transactions as they occur, applying machine learning models to streaming data, identifying potential fraud signals within milliseconds.

StarTree’s innovative indexing enables the bank to aggregate high-cardinality data with processing times under 200 milliseconds for fraud detection queries, even during peak loads of over 100 requests per second. The system consolidated multiple data sources to provide a comprehensive view of customer activity with real-time visibility.

Clickstream analytics

The bank addressed its analytics cost challenges by replacing Amplitude with an in-house solution built on Apache Pinot. This platform collects and analyzes user interactions across web and mobile applications, providing product managers, analysts, and customer support services with live insights into customer behavior.

The platform was later extended to replace Redis for customer support operations, saving 7-figures of annual budget by consolidating systems. This expansion utilized the existing Pinot infrastructure, not only improving data freshness but maintaining costs as use cases grow.

Using StarTree alongside Spark observability platform for internal live monitoring

The bank created an observability solution for their Apache Spark environment that evolved the user experience for Spark across the bank. This platform consolidated logs previously scattered across Splunk and BigQuery, creating a unified view of Spark operations in Pinot in real-time.

The stack leverages StarTree’s ability to send logs in real-time from Spark to Kafka, Flink and Pinot—integrating with visualization tools like Superset and Grafana. This immediate visibility has enabled the team’s ability to detect issues as they happen rather than hours or days later.

The observability platform was designed to handle substantial data volumes—up to 49 billion events per day and one terabyte of log data daily.

Results

Real-time fraud prevention

The new fraud detection platform proved effective, with projected fraud prevention savings of $10 million annually through the primary detection system. Secondary detection systems contributed an additional $1.6 million in projected annual fraud reduction.

The real-time nature of the new system allows the bank to identify and block suspicious transactions before they are completed, rather than detecting them after the fact, with query latencies for fraud signals dropping from 300 milliseconds to 50 milliseconds (P99) and average query response times (P50) as low as 7 milliseconds.

Cost reduction

By replacing Amplitude with Pinot, the bank reduced its annual analytics licensing expenses by $800,000. Similarly, the replacement of Redis for customer support analytics saved an additional $1.2 million annually in infrastructure costs.

The observability platform prevents an estimated $730,000 in annual costs associated with incident management and resolution by identifying and addressing potential issues before they escalate.

In total, the bank was able to save $2.8 million in annual technology infrastructure costs.

Operational improvements

The new analytics infrastructure improved operational efficiency. Customer support agents reduced resolution time by 30% to triage issues, resulting in improved customer satisfaction. 

The impact on Spark job troubleshooting was significant. Prior to the implementation, identifying and resolving issues with failed Spark jobs typically required between two and four days of engineering time. With the new observability platform, most issues can be diagnosed and addressed within minutes.

The bank also recorded improved dataset stability. Analysis showed that 13 out of 21 performance-related dataset crashes could have been prevented through the proactive monitoring capabilities now available.

Perhaps most importantly, the new analytics infrastructure empowered technical teams to optimize their own systems without extensive support from specialized data engineers.

Future plans

Building on the initial implementations, the bank continues to expand its use of StarTree and Pinot across the organization:

• The analytics team is implementing anomaly detection across both business metrics and technical performance indicators to automatically identify unusual patterns that might indicate fraud, performance issues, or business opportunities.
• The product team is developing a real-time experimentation platform to support A/B testing of new features and experiences, allowing for faster iteration and more data-driven product decisions.
• The infrastructure team is implementing StarTree’s observability support for metrics and logs data, projected to save costs and improve query performance by >90%.  
• The credit risk team is enhancing their feature store capabilities to support more sophisticated lending analytics, aiming to improve credit decision accuracy, loan performance, and speed.

By implementing StarTree as their platform for operational insights, the bank has transformed their ability to understand what’s happening across their business. This real-time visibility has delivered substantial value through improved fraud detection, enhanced customer experience, and optimized platform operations—all while significantly reducing costs. StarTree provides both the performance and flexibility needed to support diverse use cases across the business, enabling them to act on fresh data with sub-second response times at scale.

Build with StarTree on AWS

The company was deployed on AWS as its cloud provider. StarTree is an AWS ISV Accelerate partner. Discover how to build your business on AWS, provisioning StarTree Cloud directly from the AWS Marketplace.

Get started with StarTree Cloud today

Interested in how your organization can improve queries and reduce costs with StarTree Cloud? Contact us for a demo. We’d love to listen to your needs, understand your use case, and answer any questions you may have on whether StarTree Cloud is the best database for your real-time analytics needs. You can also get started immediately in your own fully-managed serverless environment with StarTree Cloud Free Tier.