We asked Viston to modernize our fraud detection and Anti-Money Laundering (AML) monitoring using explainable AI and graph analytics while ensuring compliance with regulatory requirements.
Business Challenge
High false positives in card fraud detection, increasing mule account activity, and time-consuming manual KYC refresh processes.
Fragmented rule engines and limited case investigation capabilities reduced the effectiveness of fraud and AML operations.
Our Approach and Solution
Built a real-time streaming fraud detection pipeline using graph-based features and explainable machine learning models.
Enhanced AML monitoring with entity resolution, risk scoring, and automated Know Your Customer (KYC) refresh workflows.
AI Applications and Benefits Delivered
Real-Time Fraud Detection
Real-time fraud scoring powered by LightGBM models and graph-derived features to detect suspicious transactions more accurately.
Graph Analytics for AML
Neo4j-based network risk scoring to identify mule account networks and uncover hidden financial relationships.
Explainable AI
SHAP-based model explanations integrated into case management to improve investigator confidence and regulatory transparency.
AI Assistant for Investigations
LLM-powered assistant for investigator summarization and Suspicious Activity Report (SAR) drafting with strict compliance guardrails.
Cost of Implementation
USD 175,000, including data platform modernization and model validation.
Time to Implement
7 months to full production deployment across card payments and retail banking payment systems.
Tools and Technologies Used
- Azure Databricks
- Apache Spark Structured Streaming
- Apache Kafka
- Snowflake
- Neo4j
- LightGBM
- SHAP
- Elastic Stack
- Azure Kubernetes Service (AKS)
- Azure Key Vault
- MLflow
- Great Expectations
- Microsoft Power BI
Quantitative Outcomes
- 48% reduction in fraud losses across targeted products.
- 37% reduction in false positives.
- 60% faster KYC refresh turnaround.
- 22% increase in investigator productivity.
Key Performance Indicators (KPIs) Tracked
- Fraud loss rate (bps)
- False positive rate
- Model precision and recall
- Time-to-detect
- AML alert productivity
- Suspicious Activity Report (SAR) timeliness
- Model drift
Pre- and Post-Implementation Metrics
- Fraud Loss Rate: 7.8 bps → 4.1 bps
- False Positive Rate: 92% → 58%
- Average KYC Turnaround Time: 10 days → 4 days
- Alerts per Investigator per Day: 18 → 22
Stakeholder Quotes
Head of Financial Crime
“Explainability made model sign-off straightforward with internal audit and the regulator.”
Fraud Strategy Lead
“Graph features exposed mule networks we couldn’t see with rules only.”
Regulatory and Compliance Considerations
- UAE Central Bank AML/CFT guidance
- FATF-aligned controls
- PCI DSS scope containment
- Model risk policy adherence
- Challenger model testing
- Comprehensive audit logs
- Privacy-by-design principles
