I need to reduce noise in my name screening process, where do I start?

Reducing Name Screening Noise

Given the recent heightened sanction against Russia, many institutions experienced a significant increase in hits, a vast majority of which, are false positives, which burden the investigation team or demand significantly more resources. This calls for immediate actions to reduce the noise, or, within an established model validation framework, ad-hoc tuning. The question is, where do you start and what are you aiming for? Here are some tips, talk to us for details if you need any further help:

1. Data Source:
   • conduct a quick data source review to identify extra or enhanced data sources that might help your name screening to perform better
   • conduct a quick data quality check to see if any obvious data glitch that could result in low-quality alerts
2. Threshold tuning:
   • use quick and dirty threshold mapping techniques to check if the current thresholds show obvious deviation
   • if your name screening solution uses fuzzy logic, check fuzzy logic thresholds too
   • use investigation feedback to do fine-tuning as soon as enough data is accumulated (see below)
   • and last but not least, if you use APIs, talk to your vendor and understand if and what are they doing to help you reduce the noise
3. Investigation feedback:
   • annotate hits with more granular feedback if possible, and create new labels as needed
   • feed annotation to tuning and also to triage processes (see below)
   • this step is known to the data scientists as “feature engineering”
4. Triage:
   • Apply simply rules
   • Apply unsupervised machine learning techniques to identify groups that can be triaged, especially those that can be immediately closed as “not interesting” and those that can be immediately escalated. The annotations from the feedback loop produce more features for data scientists that develop the UML algorithms, therefore it is recommended that the data scientists work closely with the investigators when annotating screening results.

Analytics is at the centre of People, Technology and Processes, following analytical good practice goes a long way when especially for fighting financial crimes.