• Rising fraud rates due to expansion into new SEA markets with diverse user behaviors.
• Legacy rule-based system unable to detect sophisticated or emerging fraud patterns.
• Slow decision-making causing 1-2 minute delays during transaction verification.
• High chargeback losses and operational burdens on fraud teams.
• Need for real-time detection without disrupting transaction flow or user experience.

‍

• Built an AI-powered fraud detection engine using:
  • Anomaly detection models
  • User behavior scoring
  • Ensemble machine learning models
• Integrated the AI engine with a Kafka streaming pipeline to enable real-time decisioning at scale.
• Implemented dynamic threshold tuning and continuous feedback loops to improve model accuracy over time.
• Deployed the system on AWS with auto-scaling, supporting high-traffic events like promotions or peak transaction periods.
• Ensured seamless integration into the client’s digital wallet and transaction authorization flow.

Impact

• 45% reduction in fraudulent transactions within 90 days of deployment.
• 30% decrease in chargeback losses, significantly improving revenue protection.
• Decision time reduced from 2 minutes to 5 seconds, enhancing user experience and reducing abandonment.
• Real-time fraud detection strengthened overall platform security and customer trust.
• Built a scalable foundation for ongoing fraud prevention as the client enters more SEA markets.

‍