Empowering Regulatory Compliance Teams Through Generative AI

In the rapidly changing financial technology environment, staying ahead of compliance challenges is vital. Financial institutions must continuously seek innovative approaches to increase effectiveness, maintain competitive advantage, improve operations, and reduce costs.

This is where the integration of Generative Artificial Intelligence (GenAI) can come into play. GenAI at a high level refers to models which can generate text, images, video, and/or various combinations of these in response to prompts. One form of GenAI tools are Large Language Models (LLMs), which are models trained on large historical data sets which can comprehend and generate human language. Use cases for these models can cover a wide range of areas that include knowledge-based answering, text classification, generating original written content, or summarizing documents; and are limited only by imagination.

The Financial Crime Compliance area is well known for necessary but repetitious tasks, often carried out in accordance with defined policies and procedures. This area can substantially benefit from incorporation of task specific LLMs. Separating out the tasks of gathering and organizing information from various sources in a logical, consistent, and structured presentation from the judgment-based activities that need to be performed by a human investigator is a perfect opportunity for financial crime compliance to leverage the power of LLMs. The time intensive activity of data gathering can be done in an automated way such that it increases consistency and quality, thus allowing investigators to have all the information they need at their fingertips almost instantly.

For example, anti-money laundering transaction monitoring programs can generate thousands of alerts requiring investigation. These investigations can take anywhere from a few minutes to a few hours of review time by a human investigator. In addition to the time required for review, the output of these investigations can vary in quality and completeness.

This is where LLMs can provide significant assistance. In this example, one or more LLMs or can be leveraged to:

1. Summarize alerting and non-alerting transaction activity
2. Identify anti-money laundering or fraud red flags and incorporate transaction analysis
3. Organize and draft a list of aggravating and/or mitigating factors; and/or
4. Prepare a draft narrative in a standard template.

Applying LLMs to one or more of these activities would be a strategic reinvention with the possibility to not only reduce investigation time, but also standardize investigative output to improve quality and control (POC’s with our clients are showing time savings of 30-50%).

One of the keys to successful implementation is to remain focused on processes that can be rigorously tested and easily reproduced, allowing for oversight by internal and external reviewers. This can be done through a combination of targeted model implementation, leveraging financial crime compliance subject matter experts, identifying the key areas where investigators remain the decision makers, and ensuring flexibility. Examples include:

• Targeted Implementation — Selecting a single transaction monitoring rule (or a handful of similar transaction monitoring rules) for the model to handle; applying the model to summarize only transaction activity; targeting products with lower behavioral variation (gift card vs. cash).
• Leveraging Financial Crime Compliance Subject Matter Experts — Informing the design for the LLM output template to ensure alignment with procedures; sharing investigative input on the scope of red flags to be evaluated for each transaction monitoring rule; providing feedback on the quality and accuracy of the LLM narrative summaries.
• Human Driven Decisions — Ensuring investigators make all determinations regarding suspicion; keeping underlying transaction data visible and accessible to reviewers; using an LLM only at the triage level.
• Flexibility — Designing modular prompts to allow for adaptation to emerging risks, typologies, and behaviors; incorporate investigator feedback.

Other potential use cases in financial crime compliance include negative news searches, KYC/EDD reviews, and sanctions screening, among others. In an industry where ensuring process effectiveness is as important as being innovative, integrating LLMs into financial crime compliance operations is giant leap forward. Leveraging GenAI while navigating complex financial regulations will be a key differentiator in driving growth. Guidehouse is eminently well-positioned to help financial institutions, payment processors, and fintech companies with responsible design and implementation of GenAI frameworks, infrastructure, models, and operational integration while navigating the nuances of the regulatory landscape.