Data Science in Financial Services – Improving Transaction Monitoring with Artificial Intelligence
How Capco Used Machine Learning to Transform a Tier 1 Bank’s Transaction Screening and Monitoring Process
In recent years, interest in data science and machine learning has increased dramatically. The tech industry was the first to adopt these technology-driven approaches, and now that they’ve entered the mainstream, companies in all industries, including financial services, want to understand how they can best use their data. However, it is important to distinguish potential use cases from real applications. In our experience, data science and machine learning are underutilized in large financial institutions and to deliver tangible results. At Capco, we strongly believe that data science adds significant value to financial services across multiple functions and can deliver a host of benefits.
Transaction screening and monitoring is a critical aspect of anti-money laundering (AML) practices within banks and financial institutions. It has been estimated that the amount of money laundered worldwide is over $3 trillion, much of which is used to finance illegal activities. Financial institutions are strongly encouraged to identify fraudulent transactions. Fines imposed by regulators for inadequate controls can be substantial, not only financially, but also reputationally. Financial crime (FC) units in financial institutions often have multiple layers and functions to monitor suspicious activity, ranging from Know Your Client (KYC) or Customer Due Diligence (CDD) to Name Screening (NS) and Transaction screening.
For most financial institutions, the typical AML workflow is usually a linear pipeline that connects customer transaction data sources to a simple model or rules-based system. If the transaction is deemed suspicious by the system, it is then flagged and passed through several levels of filtering involving human analysts who decide whether the transaction is indeed fraudulent. To ensure that financial institutions have strong anti-money laundering controls, regulators require companies to implement very strict screening and reporting standards. These checks are often very expensive and require several teams and considerable man-hours, especially if the account base is very large.
High volumes of reported transactions can often lead to the following issues:
- Bad customer service: Most reported transactions are usually false positives (genuine transactions). Approval of these transactions is often delayed, resulting in poor customer experience. Financial institutions need to manage service quality. Customers potentially waiting for a transaction to be processed could lead to poor quality of service and potentially breach their SLA.
- Escalating costs: Costs associated with increasing the manpower required to review an increasing volume of alerts (often seasonal). The need for human validators will be immense for large financial institutions. On top of that, validation staff typically have two or more levels of reviewers and therefore require a significant number of man-hours to maintain a high level of service quality. This problem is often exacerbated when there are seasonal peaks (eg towards the end of tax season).
- Escalating Errors: Human error of forwarding a transaction that should be forwarded to regulators. These false negatives lead to the undetected processing of fraudulent transactions and can be a source of substantial regulatory risk for any business.
Approach: addressing duplication and complexity –
Capco was engaged by a Tier 1 investment bank to design, analyze, test and implement a machine learning solution to optimize the transaction screening workflow. This solution would process more than 100,000 screening alerts per month. Partly driven by regulatory requirements, the client had a complex process for approving these alerts, with multiple levels of human review. Each additional filtering requirement increased the volume of alerts for the teams performing the reviews. Most of the alerts were false positives, and many of them were duplicate alerts that required the same type of investigation. Growing volumes of alerts and time spent on reviews strained the bank’s resources. To address these challenges, the Capco team worked with the Operations and Technology functions to define the following approach that would allow the bank to optimize resources and achieve significant cost savings.
Model Workflow: Automate reviews and escalations-
After the initial review is complete and an alert is created, it is fed into the machine learning model. This provides a level of confidence for expected results based on historical data from the selection workflow. The decision to systematically perform the necessary action is then based on the level of confidence calculated by the model as below:
- Suggested Action: Provides suggestions for the initial review outcome (excludes approval, escalation, and/or required feedback) based on historical data; serves as an intermediate phase before automatic escalation and approvals.
- Automatic escalation: Automatically escalates transactions to secondary review (bypassing initial review) if the confidence level is above the pre-determined threshold.
- Auto Approve: Automatically approves and adds comments to close false positives
Methodology and results: advantages of a dual approach
One of the main achievements of this model was the automation of more than 60% of the control workflow. This is done in two phases: self-escalation and self-approval.
The goal of auto-escalation is to automatically escalate screening alerts from the initial exam to the secondary exam, only if the outcome can be predicted with a high level of confidence. The type of model used to achieve this is a remote stacked assembly model. This model breaks down into:
- Historical data: Screened telegraph transactions for the last 8 weeks.
- Features: Includes transactional information from the company’s web-based workflow tool used to present potential sanction matches and keywords.
- Distance calculation: Each transaction (with filter details) is “flattened” to a line and measured against the number of differences between the current transaction and previous alerts and auto-escalations.
During the pilot phase, machine learning was used to suggest actions and feedback to reviewers. After five months of monitoring the accuracy of the forecasts, the proposal was reviewed and approved by the company’s Global Financial Crimes Unit and Operational Controls Department. Using the trial period parameters, the projected benefits and savings were presented to stakeholders, and an appropriate level of confidence was agreed upon.
The purpose of the automatic approval phase is to automatically approve and close false positives during the initial review, only if the outcome is predicted with a high level of confidence.
Since auto-approval has more associated risks, alerts must pass through two separate models to be auto-approved. The models used for this phase were a Distributed Random Forest Distance Classifier (DRF) model and a Natural Language Processing (NLP) model. If the alert passes through the first DRF model with high confidence of approval, it is then passed to the second NLP model for the final decision.
- Distributed Random Forest Distance Classifier: This is a decision tree-based classifier, which uses historical filtered transaction inputs from the previous 8 weeks, including information about the reason for the filtering.
- Natural Language Processing Classifier: This is the main classifier because it is the final decider. It converts the details of each transaction into human-readable text. This mimics the actual process of a human which mimics the process a human would perform.
There are several applications of machine learning in financial services that can prove extremely beneficial to businesses. Transaction monitoring models, like the one described in this post, are one such use case. By deploying innovative machine learning techniques, the company in question was able to significantly reduce costs, improve the accuracy of fraud detection and, most importantly, was able to improve the overall customer experience. Our Data Science in Financial Services series aims to highlight the solutions Capco has provided to its customers and further demonstrate how these solutions can apply to your organization. Capco combines cutting-edge machine learning techniques with financial services expertise to help you achieve your goals
By- Jacqueline Gheraldi, senior consultant at Capco