Zopa Enhances Fraud Detection with Amazon SageMaker Clarify
Zopa, a UK-based digital bank and P2P lender, addresses identity fraud in loan applications through advanced ML models. The bank uses Amazon SageMaker Clarify to provide explainability for ML fraud detection models, improving compliance, transparency, and underwriting efficiency. Models are trained using a combination of XGBoost and TensorFlow on Amazon SageMaker, deployed via Jenkins CI/CD pipelines as microservices. SHAP values generated by SageMaker Clarify enable detailed insights into individual predictions and feature impacts for operational and compliance needs.
- Organization
- Zopa
- Industry
- Finance
- Location
- United Kingdom
- Published
- February 2021
Reported outcomes
Strategic outcomes
Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks
- Customer
- Zopa
- Provider
- AWS
- Maturity
- Production
- Linked source
- AWS Machine Learning Blog
Models are trained using a combination of XGBoost and TensorFlow on Amazon SageMaker, deployed via Jenkins CI/CD pipelines as microservices
Primary read
Use case focus
Showing 2 of 2
- 1Fraud Detection
- 2Explainable AI
- Combat identity fraud in loan applications while ensuring explainability of ML models for regulatory compliance and operational validation.
- Improve underwriters' manual review process with trustworthy model explanations.
- Need for model transparency to demonstrate fairness and support regulatory obligations.
- Implemented Amazon SageMaker Clarify to generate SHAP-based explainability for individual ML model predictions.
- Trained fraud detection models on Amazon SageMaker using XGBoost and TensorFlow frameworks.
- Deployed models as microservices via Jenkins CI/CD pipelines for real-time fraud detection.
- Provided aggregated and instance-level model explanation reports to underwriters to focus reviews and reduce customer friction.
- Improved model transparency and regulatory compliance with detailed explainability.
- Enhanced operational efficiency by enabling focused manual reviews based on explainability insights.
- Increased confidence in ML-driven fraud detection results among data scientists and underwriters.
Architecture
Zopa uses a fraud detection system where models are trained on Amazon SageMaker, deployed via Jenkins CI/CD pipeline as microservices. Amazon SageMaker Clarify produces SHAP explanations both during training for validation and after deployment for monitoring. Explanations are computed using synthetic contrastive data and stored in Amazon S3. This architecture supports real-time fraud detection with operational explainability for manual underwriter review.
Sources & evidence1
- Customer explicitly identified
- Deployment status explicitly supported
- Technical implementation details available
AI-generated summary. Verify important details with the linked sources before relying on this case.
Explore related AI use cases
Was this useful?
Community
Comments
Loading comments...