Curve Uses BigQuery Graph to Sniff Out Fraud Rings

Curve, the UK-based financial super-app, is fighting fraud with a new weapon: BigQuery Graph. Instead of just looking at individual transactions, they’re now mapping the hidden networks of fraudsters—and it’s saved them a cool $12 million in 2025 alone.

The Multi-Hop Problem

Fraudsters rarely work alone. They share devices, cards, or contact info across multiple accounts. In a standard database, finding these links means complex multi-hop joins—think massive self-joins that bog down performance. At Curve’s scale (millions of users, billions of connections), that’s a swamp of computational complexity.

Graph Analytics to the Rescue

Curve moved their network analysis into BigQuery Graph, using its native Graph Query Language (GQL). No more data migration—they kept everything in their existing data warehouse. By modeling users as nodes and shared identifiers as edges, they simplified queries from dozens of lines of JOINs to intuitive GQL patterns.

Key Benefits

• Traverse billions of connections with high performance
• Unify data workflows: combine graph traversals with SQL, search, and ML in one query
• Accessible to more analysts without needing a new language

Results That Speak Volumes

• $12M saved in transaction losses from automated blocks
• 72% accuracy in identifying fraudulent users, reducing false positives
• Faster rule refreshes: from one-hop queries to multi-hop in hourly rules
• ML-ready: moving toward streaming traversals for real-time fraud detection

Curve is now incorporating IP address connections into real-time loops and exploring graph visualization. By treating data as a living network, they’re keeping security as smooth as their customer experience. For more on how cloud platforms handle such scale, check out our Google Cloud review.

Original announcement published on Google Cloud.