With a background building and leading data science teams across several global banks, Nikhil Aggarwal is intimately familiar with the challenges financial institutions face in leveraging the information to increase revenue, streamline operations and manage credit and financial crime risk.
Aggarwal, a director at IBM's Promontory Financial Group, now works on the Financial Crimes Insight for Watson platform and also provides consulting to firms.
IBM has touted Watson's
Aggarwal said that the next step for many banks is to build a uniform system to better identify and manage the links between different compliance risks.
The following interview has been edited for length and clarity.
Based on your experience working at and consulting for banks, what do you find lacking in the ways that banks think about compliance in their overall business model?
NIKHIL AGGARWAL: Banks have made steady progress in building out their compliance programs. Traditionally, these programs often have had stand-alone components that remain siloed. There is opportunity to build a more unified, integrated compliance program that allows us to better identify and manage linkages between different compliance risks such as money laundering, sanctions, fraud, and bribery and corruption.
IBM’s regtech solutions, including components of Watson, seek to address thematic linkages in governance, risk and compliance, money laundering and financial crimes, surveillance and broader financial risk. The goal is to empower banks to manage risk by building out a more comprehensive mitigating control framework.
What kind of vision do you think banks should have for the kinds of customers they should be targeting?
Banks have defined risk appetite policies that state the types of customers that they would like to onboard and maintain an ongoing relationship with. A bank would, at an overall portfolio level and at an individual customer level, focus on revenue targets and delivering a set of products and services that result in a superior customer experience. It is imperative to forge a balance by coupling the risk-revenue dynamics with a positive customer-centric experience. At the same time, a bank must also maintain its risk posture and adhere to all regulations.
What’s the best way for banks to find low-risk, high-revenue customers?
A robust metrics/reporting program will help to carve out current revenue performance and risk levels, both at a customer level and at an overall portfolio level. Predictive analytics models such as a revenue propensity model and dynamic customer risk rating model can help determine future revenue streams and changes in behavior, including an uptick in risk. Banks can rank order both these dimensions to identify customer clusters and segments, and develop segment management programs.
Do banks currently have the manpower and technology to be able to identify these customers?
Banks have increasingly been building analytics teams and investing in cognitive technology in recent years. While revenue/front-office analytics teams and solutions are more mature, compliance risk analytics as a domain presents an opportunity. There is a critical need for individuals and solutions that can factor in both domain and data science nuances to better identify and mitigate compliance risk. It is especially important to understand inherent risk drivers and interweave these perspectives into analytical models.
How does IBM+Promontory aim to help banks identify risky customers?
Promontory’s deep understanding of the regulatory context and customer risk profiles coupled with IBM’s cognitive technology has allowed us to build out a dynamic customer risk rating model that can rank customers on their current and future risk. Furthermore, we are developing novel features to measure risk and continuing to add incremental data to the risk engine to generate up-to-date risk ratings.
What are your top three recommendations for banks trying to weather the amount of false positives they are encountering while also trying to identify bad actors and avoid financial crime?
No. 1, develop ensemble models (models that have multiple analytics layers) that zero in risk vectors and minimize the noise.
No. 2, bolster the investigative feedback loop, as qualitative insights result in new learnings that can help in making necessary adaptations to future models.
No. 3, focus on building a robust data layer and architecture. Data normalization, data lineage and data extract, transform, load are necessary hygiene steps in measuring the effectiveness and efficiency of the program.
