ShieldAI™ - Intelligent Fraud Prevention

Real-time fraud detection combining anomaly detection, network analysis, and behavioral biometrics.

The Challenge

First-party fraud, synthetic identities, and organized fraud rings cost lenders billions annually. Traditional rule-based systems generate 90%+ false positive rates.

What It Does

Real-time fraud detection combining anomaly detection, network analysis, and behavioral biometrics to catch fraud while minimizing customer friction.

Model Metadata

Model Architecture: Ensemble of Isolation Forest (anomaly detection), Graph Neural Networks (ring detection), and XGBoost (classification)
Training Data Size: 50M+ applications, 500K+ confirmed fraud cases
Features: 400+ identity, behavioral, and network features
Update Frequency: Daily model updates (online learning)
Inference Speed: <80ms per transaction
False Positive Rate: <5% (vs. 90%+ for rule-based systems)

Business Outcomes

68% reduction in fraud losses
94% reduction in false positives vs. rules-based systems
$10M+ annual fraud prevention (typical $1B lending volume)
<0.5% good customer decline rate
Detection of organized fraud rings (10+ linked applications)

Training Approach

Anomaly DetectionUnsupervised Isolation Forest identifies unusual patterns in applications without requiring labeled fraud examples. Particularly effective for novel fraud schemes.

Semi-Supervised LearningCombines labeled fraud cases with much larger unlabeled dataset using techniques like positive-unlabeled learning.

Graph Neural NetworksLearns representations of applicants based on their connections (shared devices, addresses, phone numbers, IP addresses) to identify fraud rings.

Online LearningModel continuously updates as new fraud patterns emerge, adapting within hours rather than waiting for quarterly retraining.

Active LearningStrategically selects uncertain cases for manual review to improve model most efficiently.

Data Sources

Identity Verification

Socure (synthetic identity detection, document verification)
Jumio (biometric verification, liveness detection)
Persona (identity verification API)
GIACT (phone, email, address intelligence)

Device & Behavioral Data

iovation (device fingerprinting, velocity checks)
Threatmetrix (digital identity risk scoring)
BioCatch (behavioral biometrics during application)
Application session data (time spent, field changes, navigation patterns)

Network Analysis

Proprietary fraud ring database (shared attributes across applications)
IP geolocation and proxy/VPN detection
Email and phone number reputation scores
Social network analysis from connected applications

Credit & Identity Data

Early Warning Services (account validation, ChexSystems)
Experian Precise ID (identity verification)
Credit bureau fraud alerts and security freezes
SSN validation and deceased file checks

Consortium Data

Shared fraud databases across financial institutions
CFPB fraud reports
FTC identity theft database
Law enforcement fraud intelligence (when available)

Fraud Types Detected

Synthetic identity fraud
First-party fraud (bust-out schemes, misrepresentation)
Third-party fraud (stolen identities)
Account takeover
Organized fraud rings
Straw borrower schemes