A big data in banking example - Improved the alert mechanism to flag off fraud transactions by building a surveillance workbench

Case Studies:Mu Sigma
Published On: 19 December 2015
Views: 5566

What We Did: Developed a workbench to streamline transaction monitoring process of a leading financial services company and highlight fraudulent entries.


The Impact We Made: The operational efficiency of the transaction review process has shown a great improvement and the investigation time has reduced by 50%. 


Summary - Anti-money laundering system

The operational team of Anti-Money Laundering segment (AML) of the client’s organization is responsible for adequately managing and mitigating the risk embedded in banking activities. Most of their activities revolve around Transaction Monitoring. The team approached us with an objective to streamline their transaction monitoring process and develop an accurate alert mechanism to flag off fraudulent activities.


About The Client - Large financial services company

The client is one of the leading financial services company and also one of the largest credit card brand in the United States. The card division accounts for more than 20% of the total dollar volume of transactions in the US.


The Challenge - High volume of manual investigations

The existing alert generation process was inefficient and resulted in several thousands of alerts on a monthly basis. The bank’s AML analysts were reviewing these alerts, which was a gigantic task. Since the alert generating engine was separate from other data sources, the analysts spent 40% of their time gathering information to investigate an alert. The client learned that this entire process was time consuming, heuristic in nature and susceptible to manual errors.


The Approach - Automation and software based pre-processing to reduce manual effort

Mu Sigma regularized the analyst investigation process as well as optimized the alert generating engine:

  • The team started off by developing a detailed understanding of the analyst alert review process. This was done by shadowing the alert investigation process.

    • A lot of focus was around understanding the actual review process, different elements involved in the context of a transaction, and setting up SLAs against industry benchmarks

    • Key transaction elements were identified such as internal EDW information, location information, and transaction channels that were reviewed in the process

  • Data aggregation to create an integrated data source was set up as multiple data from different banks and third party sources were obtained (and used by the reviewers) in silos.

  • Post that, rules to identify a transaction as fraudulent were set based on the learning from Analyst Review Process and discussion with other key stakeholders.

  • Multiple models were run on past data to result in correct identification of fraudulent activities to derive the most relevant alert mechanism. After that, a rich KYC report and other reports were rolled out to share with business stakeholders. The alert mechanism and report generation module together formed AML surveillance workbench.

  • The above AML workbench was used as the major single-screen review framework for the entire transaction monitoring process. The visualization of this workbench captured every essence related to the analyst investigation process:

    • The user could interact with the tool by picking up a transaction, assessing historical flows, analyzing the entire context of the transaction.

    • The tool was hosted on a rich parallel bench which was designed to easily scale across users and enable very high usage by the business’ entire team


The Outcome - Increased analyst productivity

The client appreciated the AML surveillance workbench as it encapsulated all the information and activities related to an analyst investigation process through a series of appealing visuals. The rich interactive mode with holistic on-screen context delivery reduced the investigation time by over 40-50%; an analyst who was earlier performing 3-4 reviews in a day could now complete 6 reviews in a day.

This is an example of big data in banking.

Back to Top