Client profile

A European payment processor, monitoring over 1 billion transactions annually and facilitating large financial institutions and retail chains, was looking to enhance its transaction monitoring capabilities in the face of ongoing regulatory pressure. To meet the requirements and ensure compliance, they sought to enhance their existing transaction monitoring engine.

Key results

Quantitative

• 83% fewer false positives  

• True positives nearly doubled   
• Fewer rules deployed in existing monitoring engine (15 instead of 100+)  

Qualitative

• Better model explainability towards auditor and regulator 

• Decreased employee / analyst turnover

• More effective investigating 

Challenge – too many rules can create blind spots and inefficiencies

While it’s true that rules hold a system together, rule-based transaction monitoring can also create:  

• Blind spots. Rule-based transaction monitoring is typically based on formerly known fraud patterns that may no longer be accurate or relevant. After all, financial criminals tend to be agile and dynamic, updating their tactics them to make sure they stay undetected.  

• Harmful inefficiencies. Layering rules upon rules increases the complexity of rule-based monitoring over time. This leads to more alerts, most of them false positives. Nevertheless, they must all be checked by your analysts, which can result in a rather rigid system.  

The payment processor had 100+ individual Fraud and AML detection rules in place, each with several levels of depth. Such layered complexity and static character made auditability of the rule set low and maintenance time-consuming.

Over time this led to increasing amounts of false positives; a heavy operational burden and regulatory risk. What’s more, analysts were not able to review all alerts, resulting in a backlog. And going through high numbers of false positives all day eventually led to analyst fatigue and high employee turnover.  

Eventually, management was urged to undertake action: decrease the number of false positives or hire more analysts to review all alerts. 

Solution – automated model generation for Fraud and AML

Sygno Analytics automatically generated a combined Fraud and AML model with a transparent report based on 18 months of the processor’s raw historical data. Its automated machine learning surfaced granular patterns of legitimate behavior (know good, catch bad) in the processor’s raw transaction data and matched these against the processor’s regulatory requirements to generate a holistic detection model. 

After setting up our Sygno platform and loading the client’s historical data from their existing monitoring system, we generated new model(s) based on that data. Next, following internal model acceptance, we integrated the newly generated model into the client’s existing monitoring system. Sygno Analytic’s open architecture facilitates this integration smoothly. In the ten following weeks our generated models ran parallel to the client’s existing models to compare performance and finally, we took Sygno’s models into production.  

In the end, the client reported 83% fewer false positives, while the effective detection rate (true positives) nearly doubled. On top of the quantitative results, the client also emphasized how they were experiencing better model explainability towards auditors and regulator. Furthermore, rules deployed in their existing monitor engine went down to 15 from 100+. Lastly, by surfacing unknown-unknowns (making the blind spots smaller) and decreasing the clear unnecessary false positives, they could improve the quality of their investigations which resulted in a significantly lower analyst turnover. 

Eliminate false positives by optimizing your Transaction Monitoring through modeling good client behavior, allowing your analysts to focus on actual critical cases. Would you like to know more about how to achieve that? Feel free to reach out!