Curve Saves $12M in Fraud Losses Using Google Cloud’s BigQuery Graph for Network Analysis

Quick Summary

Curve utilizes BigQuery Graph to identify complex fraud rings by analyzing billions of data connections natively within Google Cloud. This network analysis approach allows the fintech to detect hidden relationships between accounts, resulting in $12 million in savings and a 72% accuracy rate in fraud detection.

How Does BigQuery Graph Solve the Multi-Hop Fraud Problem?

Network analysis is essential for modern fintechs because fraudsters rarely operate in isolation; they typically share devices, cards, or contact details across multiple accounts. Traditional relational databases struggle with multi-hop analysis, which requires computationally expensive self-joins that often exhaust system resources when processing millions of users.

• Native GQL Support: By using Graph Query Language, Curve eliminates complex JOIN logic, replacing it with intuitive pattern matching.
• Zero Data Movement: Keeping data within the BigQuery environment saves significant time and costs compared to migrating to external graph databases.
• Massive Scalability: The system comfortably traverses billions of connections, including user-level, device-level, and card-level data points.

What Results Has Network Analysis Delivered for Curve?

The transition to network analysis has transformed Curve’s fraud mitigation from a reactive process to a proactive, high-precision operation. By treating data as a living network of relationships, the team can identify organized crime patterns that traditional models miss, significantly reducing manual review workloads.

• $12M Saved: Automated blocks based on graph insights prevented massive transaction losses in 2025.
• 72% Detection Accuracy: High precision ensures that fraud agents focus on high-certainty cases, reducing false positives.
• Faster Rule Refreshes: Optimized GQL scripts allow for hourly fraud rule updates, keeping the platform ahead of evolving threats.

How is Curve Integrating Graph Data into Machine Learning?

The ultimate goal for network analysis at Curve is moving from static rules to real-time machine learning features. While daily graph rebuilding is sufficient for model training, the team is now moving toward micro-batch and streaming traversals to serve fresh data during the sub-second window of transaction authorization.

• Billions of IP Signals: Curve is currently incorporating high-volume IP connections into real-time detection loops.
• Native Visualization: Future updates will include graph visualization tools to help analysts intuitively "see" fraud webs as they form.
• Unified Data Experience: Combining graph traversals with standard SQL analysis allows the entire data team to contribute without specialized upskilling.

FF NEWS TAKE:

This partnership between Curve and Google Cloud moves the needle because it democratizes network analysis for the broader fintech sector. By embedding graph capabilities directly into BigQuery, Google has removed the "data movement tax" that previously made deep link analysis too expensive for many. Curve’s $12M saving is a powerful proof of concept that network analysis is no longer a luxury - it is a fundamental requirement for any high-volume digital wallet.