Overview
Financial leakage through procurement inefficiencies, duplicate invoices, and off-contract spending costs large organisations millions annually. This system builds an automated detection layer that flags anomalies before they become audit findings.
Detection Mechanisms
Rule-Based Layer
- Duplicate invoice detection (fuzzy matching on vendor + amount + date)
- Off-contract spend flagging (PO validation against approved vendor list)
- Budget overage alerts by cost centre
ML Layer
- Isolation Forest for transaction-level anomaly detection
- Temporal pattern analysis — identifying unusual spend cadences
- Vendor clustering to detect fraudulent entity splitting
Data Pipeline
The pipeline ingests SAP transaction data daily, applies both rule and ML layers, and produces a prioritised exception queue for the finance team. Each exception is scored and annotated with the reason for flagging.
Impact
In active testing. Early false-positive rate below 8%, with the model recovering 3x its development cost in the first quarter of operation (projected).