Performance outcomes from initial implementation
This reference shows how the platform delivers measurable results in a standardized AML use case. Real outcomes vary based on data quality, historical patterns, and operational integration.
IBM HI-Small transactions processed
Classifier parameters
Validation ROC-AUC
Streaming scoring interval
Measured Outcomes
Earlier detection and analyst efficiency
SAR turnaround time
False-positive review load
Pattern classes detected
Model Quality
Validation ROC-AUC improvement over rollout
Quality stabilizes as analyst feedback loops and edge-case tuning converge. Threshold tuning typically achieves production ROC-AUC within 4–6 weeks.
Month 1
0.84
Month 2
0.87
Month 3
0.89
Month 4
0.91
Implementation
Four-week rollout sequence
Data ingest and baseline benchmark
Connected transaction exports, profiled class imbalance, and validated baseline rule precision/recall envelope.
Model fine-tuning and thresholding
Fine-tuned the scoring model and calibrated alert thresholds against validation cohorts and review capacity.
Graph detector integration
Integrated topology detection and analyst pattern context into queue-level triage workflows.
Production rollout and governance handoff
Activated governed model version, enabled audit-ready dossier export, and scheduled retraining cadence.
System Stages
Four-stage processing pipeline
Data Pipeline
Cleans IBM HI-Small source rows, normalizes timestamps and currency values, and yields reproducible fixtures for model training and replay.
Classifier Scoring
Scores transaction descriptors with class weighting and negative undersampling to keep the extreme class imbalance trainable.
Pattern Engine
Detects structuring, fan-out, fan-in, and cycle motifs using graph windows and emits explainable evidence bundles for analysts.
Review & Retrain
Persists analyst dispositions and promotes nightly retrained models into the registry with threshold and metric lineage attached.