The Challenge
The bank was processing $10T in monthly payments, but 2-3% of transactions required manual intervention due to sanctions hits, formatting errors, or liquidity checks. This manual repair queue created a massive operational bottleneck, costing 100+ FTE hours daily and introducing significant regulatory risk.
The Transformation Strategy
-
1
Process Mining
Deployed Celonis to map the exact "happy path" vs "exception path" workflows.
-
2
Intelligent Routing
Built a Python-based classification engine to auto-resolve "False Positive" sanctions hits.
-
3
Agentic Auto-Repair
Deployed an LLM agent to parse unstructured SWIFT messages and suggest repairs to operators.
The Outcome
Within 12 months, the system achieved a 60% reduction in manual touchpoints.
60%
Reduction in Manual Work
Zero
Audit Exceptions