Nubank Transforms Fraud Detection Across 65M Accounts with Real-Time Graph Intelligence
The Challenge
Nubank faced 9,000+ scam reports and approximately $1.8M in monthly losses. Legacy infrastructure produced 60%+ false positives and struggled to deliver broader coverage, higher precision, and real-time intelligence across 65M+ accounts.
The Solution
Implemented TigerGraph Cloud on AWS to connect 65M accounts and inject 30 graph features into Nubank’s machine learning pipeline, surfacing high-risk relationships through dynamic, real-time relational insights.
The Results
Enabled real-time detection and action across 65M accounts, significantly boosted model performance without increasing infrastructure complexity, and integrated graph-powered investigation and insights directly into Nubank’s operational workflows.
The Challenge
Nubank’s AML and fraud teams faced an overwhelming increase in scam-related activity. More than 9,000 reports were being generated each month, accompanied by approximately $1.8 million in monthly losses. At the same time, the bank’s existing detection environment was producing more than 60% false positives, consuming investigator time and limiting the effectiveness of fraud operations.
The underlying infrastructure also struggled with the scale and complexity of the problem. Detection models needed to operate across more than 65 million accounts, yet Nubank faced limited transaction coverage, rigid rules-based monitoring, high operational costs, and SLA windows exceeding 45 days. Critically, the existing approach could not dynamically incorporate network-based risk factors.
Nubank needed higher precision, broader coverage, and real-time intelligence capable of identifying risk through relationships across accounts, transactions, and profile data.
The underlying infrastructure also struggled with the scale and complexity of the problem. Detection models needed to operate across more than 65 million accounts, yet Nubank faced limited transaction coverage, rigid rules-based monitoring, high operational costs, and SLA windows exceeding 45 days. Critically, the existing approach could not dynamically incorporate network-based risk factors.
Nubank needed higher precision, broader coverage, and real-time intelligence capable of identifying risk through relationships across accounts, transactions, and profile data.
The Solution
Using TigerGraph Cloud running on AWS, Nubank implemented a graph-powered solution that enhanced its machine learning capabilities through entity resolution and feature generation. TigerGraph enabled Nubank to create a graph connecting 65 million accounts through transactions and other profile data, providing the infrastructure to surface high-risk relationships and support real-time detection and action.
TigerGraph then enabled Nubank to inject graph-native intelligence directly into its machine learning pipeline, including SageMaker and related services. Nubank calculated 30 graph features, including shortest paths and total path counts to known mule accounts, enriching its detection vectors with dynamic, real-time relational insights. The solution calculated proximity to labeled fraud accounts across the 65-million-account graph and integrated directly into Nubank’s existing machine learning workflows.
Nubank also implemented TigerGraph Insights, TigerGraph’s no/low-code visualization and analytics tool, providing a front-end solution tailored to its operational workflows and supporting case investigation and insights.
TigerGraph then enabled Nubank to inject graph-native intelligence directly into its machine learning pipeline, including SageMaker and related services. Nubank calculated 30 graph features, including shortest paths and total path counts to known mule accounts, enriching its detection vectors with dynamic, real-time relational insights. The solution calculated proximity to labeled fraud accounts across the 65-million-account graph and integrated directly into Nubank’s existing machine learning workflows.
Nubank also implemented TigerGraph Insights, TigerGraph’s no/low-code visualization and analytics tool, providing a front-end solution tailored to its operational workflows and supporting case investigation and insights.
The Results
TigerGraph enabled Nubank to apply graph-powered fraud intelligence across 65 million accounts, surfacing high-risk relationships and supporting real-time detection and action. By injecting graph-native intelligence directly into its machine learning pipeline, Nubank significantly boosted model performance without increasing infrastructure complexity.
The implementation extended graph intelligence into Nubank’s fraud operations through TigerGraph Insights. The no/low-code visualization and analytics tool provided a front-end solution tailored to Nubank’s operational workflows, enabling investigator UI integration for case investigation and insights.
Together, TigerGraph Cloud, graph-based feature generation, machine learning integration, and TigerGraph Insights gave Nubank the infrastructure to surface high-risk relationships, dynamically incorporate network-based risk factors, and enable real-time fraud detection and action across its 65-million-account environment.
The implementation extended graph intelligence into Nubank’s fraud operations through TigerGraph Insights. The no/low-code visualization and analytics tool provided a front-end solution tailored to Nubank’s operational workflows, enabling investigator UI integration for case investigation and insights.
Together, TigerGraph Cloud, graph-based feature generation, machine learning integration, and TigerGraph Insights gave Nubank the infrastructure to surface high-risk relationships, dynamically incorporate network-based risk factors, and enable real-time fraud detection and action across its 65-million-account environment.
More stories
Fintech Customers
FinTell. Improving Risk Management for Banking and Fintech with Graph and Machine Learning
Fintech Customers
Intuit. Leveraging Knowledge Graphs to Accelerate Intuit’s AI Journey
Fintech Customers
Global Bank Saves $50M By Transforming Fraud Detection with Real-Time Graph Analytics