Business Objective
Our client is a leading European financial institution, offering banking, wealth management, and payment solutions to their retail and business clientele.
The financial institution’s internal audit team responsible for money laundering wanted to:
- Develop an independent data-driven system to leverage client transaction information
- Intercept high-risk transactions and flag clients not yet reported as suspicious transactions
- Increase the effectiveness of internal control currently based on random samples or selected by experiential triggers
Challenges
- Limited number of tagged cases corresponding to various money laundering scenarios
- Disparate data sources within the bank
- Limited visibility on suspicious operations among non-customers
- No benchmark of model performance: Not all cases flagged by the model were investigated, making an evaluation of false/true positive cases complex
Solution Methodology
- Stitched together customer 360 data from a variety of data sources within the bank
- Prepared training dataset capturing 1000+ behavioral features across different banking products and channels
- Created velocity features for transactions across different channels
- Ran supervised and unsupervised learning approaches on past investigations to generate a SAR ‘signal.’
- Fine-tuned model performance to reduce false alerts, without compromising on effectiveness in capturing maximum suspicious events
- Validated model performance in terms of AUC, precision and recall at optimized thresholds across train, test and out-of-time validation set
Business Impact
- High precision of the model optimized investigation activities for high-risk customers
- The model was also used to reduce audit effort by 80% for very low-risk customers
