Designing AML Pipelines for Data Quality, Lineage, and Drift Detection
A few months ago, I was investigating an unexpected spike in AML false positives. Every dashboard looked healthy. Every ingestion job had completed successfully. Infrastructure monitoring showed no issues. Initially, I assumed something was wrong with the pipeline. After digging deeper, I realized the real issue was the data itself. An upstream application had unintentionally changed how one field was populated. Nothing crashed or failed validation. But one missing attribute was enough to throw customer profiles out of sync and undermine the detection logic we trusted.
After tracing the problem through the pipeline, I discovered that an upstream application had stopped reporting closed-account status correctly. My AML models continued treating those accounts as active, which distorted customer profiles, inflated transaction histories, and generated investigations for accounts that should no longer have been in scope.
I only found the root cause after working backward from the spike in false positives, validating...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE
