How are Machine Learning Models used in Banking?

How are Machine Learning Models used in Banking?

By Jacob Kosoff, Head of Model Risk Management & Validation, Regions Bank


Where Can Machine Learning Models Be Used In Banking?

Machine Learning models have already started to be used widely in banking.  On Day One of Risk Americas 2019, Jacob Kosoff, the head of Model Risk Management and Validation at Regions Bank, will discuss how to validate these common uses of machine learning models in banking.

Data Security, Fraud Detection and Anti-money Laundering:
ML models can be flexible and adaptable so that they can pick up emerging cyber risk, fraud or money laundering patterns and more efficiently screen thousands of transactions or data points.

Automated Suspicious Activity Report (SAR) Narratives:
Through the use of ML techniques, data analytics, and metadata tools, the unstructured and structured data related to BSA/AML cases can be automatically converted into SAR Narratives that consistently comply with FinCen guidance.

Fraud Monitoring:
Many fraud models, including cyberfraud, have historically utilized judgmental rules, but fraud modeling lends itself naturally as an application of ML that has been adopted at many financial institutions. As long as the model risks are mitigated, ML can enhance banks’ surveillance technologies and materially reduce fraud losses.  However, the fraud model should not be used just as a black box.

With sufficient transparency and rigor, ML can improve the understanding of the drivers of fraud activity and emerging risks as well as aid in fraud monitoring and investigations. Specifically, an ML model may identify new patterns previously unknown to fraud investigators, e.g., slight increases in activity similar to a customer’s previous patterns that indicate fraud, or new combinations of factors with regard to the transaction/product/customer that can be better used by the fraud team in focusing their monitoring and investigation efforts.

However, the increased monitoring power and automation of ML do not remove the need for human intervention, as a new fraud scheme that produces novel data patterns may not be correctly labeled as fraud without human intervention, or the ML model could pick up false patterns that are not fraud (which could be identified by a subject matter expert) that then increase false positives and overwhelm investigators.

Cyber Risk Monitoring:
Malware codes can have similar properties or follow similar patterns as they try to penetrate or move around in a bank’s systems. ML can provide cyber risk professionals with tools to analyze patterns of attack and develop stronger systems and their own set of protections that are more resilient to hacking.

Credit Risk Forecasts or Credit Scoring:
ML can be more flexible than traditional econometric or logistic regression models in selecting and recognizing the information value in multiple drivers and risk segments.  However, these models often need to have intuitive structures and transparency to the management using them.  Additionally, a lack of transparency for underwriting models can lead to difficulty determining whether they meet the requirements of consumer compliance regulations.

Financial Trading:
High speed, high volume trading based on ML identification of patterns of valuations can take advantage of market opportunities.  However, these models do have the risks of sell-offs or other mistakes that have led to large losses.

There is the potential to utilize ML to better assess customer propensity in a highly saturated consumer market.  These models can be updated quickly and tailored to specific marketing campaigns. Marketing is an area for growth in ML applications to better target marketing campaigns and online adds to customers’ wants.

Escalation Management:
ML combined with the appropriate IT tools can be used to analyze customer complaints, internal escalations, curate case files, automatically escalate to the appropriate case manager, and track escalation resolution. The ML approach can create consistency, trend analysis, and reduce administrative costs.

The opinions expressed in the article are statements of the author and are intended only for informational purposes, and are not opinions of any financial institution and any representation to the contrary is expressly disclaimed.