About the customer – Customer is one of the top 10 largest banks in the world.
Customer Challenge
- Bank lost billions of dollars in penalties and trading losses
- Challenge posed was to detect the intent to commit fraud proactively by monitoring communications in order to avoid penalties and damage to reputation
- Large team of ~30 people were dedicated to ensure compliance using keyword search but face too many false positives
NucleusTeq Solution Approach
- Delivered a semi-supervised learning-based NLP solution to aid in detecting fraudulent activity
- Coordinated multiple rounds of reviews to improve solution accuracy across 10 categories
- Attained expected precision in SLA within 2 months
- Implemented solution on Stanford NLP, Spark and Tachyon
Outcomes Delivered
- Delivered expected accuracy to detect email messages with scores for fraud categories
- Elimination of a large part of the data via identification of non-relevant messages using learning algorithms
- Defined templates unique to each type of fraud based on presence of different entities (persons, places etc)
- Created an approach to profile all the users in the data by means of graph analytics algorithms