How Cognitive Investigator is taking a bite out of financial crime

Financial crime accounts for US$1t to US$3t in annual activity globally, yet less than 1% is detected. While much investigative work goes into cracking sophisticated crime schemes, the results are often plagued by costly errors and low-quality outcomes.

One large global financial institution recently turned to EY for guidance. Their compliance lifecycle was burdened with inefficient, highly manual processes. On any given day, investigators worked across many disparate systems and tools, manually collecting data, reviewing information and ultimately entering it into a case management system. Investigators struggled to review large volumes of customers and alerts, with a lack of risk prioritization often resulting in full population reviews and many unproductive alerts. Once the reviews were complete, analysts had to manually source key data points and write a disposition narrative, which suffered from inconsistencies due to the subjective nature of the process or simple clerical errors.

These challenges were extremely concerning for the bank. As their compliance costs and quality issues continued to grow, the institution looked for a sustainable technology solution to alleviate these pain points. While they were familiar with technologies such as artificial intelligence (AI) and machine learning, they were facing resource constraints, did not fully understand the complex nuances for building, testing and ultimately deploying these technologies into production, and also were concerned about regulatory adoption and acceptance of these technologies.

The bank engaged EY to assess their current processes and identify areas for improvement through the introduction of technology. Of utmost importance was the need to reduce the overall manual effort of the review process, prioritize high-risk customers and alerts, and standardize the narrative writing component of the investigation.

To combat challenges like these, EY has developed Cognitive Investigator, a technology platform that enables the rapid deployment of advanced technologies, such as AI, machine learning and intelligent automation. Cognitive Investigator has been designed to attack inefficiencies and error-prone processes in the compliance lifecycle. Unlike comparable solutions, Cognitive Investigator integrates directly with existing systems and tools, as opposed to replacing them, and was built with flexibility and reusability in mind.

After an initial assessment, EY identified an approach to deploy the integrated technologies from Cognitive Investigator’s toolkit, including opportunities for smart decisioning and intelligent process automation that addressed the client’s primary objectives. EY’s ultimate intention was to enhance the decision-making process by shifting investigators’ focus from manual, time-intensive processes to high value-add activities.

When the bank had utilized AI and machine learning in the past, there was significant lead time before they could deploy their models into production, often outweighing the projected benefit of the solution. With Cognitive Investigator, EY accelerated the realization of these benefits, paving the path toward future technology innovation for the financial institution.

Deploying the technology

After reviewing the financial institution’s processes, EY recommended three potential opportunities for technology enablers: implementing an advanced customer and alert risk prioritization methodology, automating the negative news search and creating an automated “smart template” for the narrative generation process. These enhancements would help the client’s investigators to focus their efforts on more complex activities, such as high-risk customers and alerts, while reducing time spent on heavily manual processes such as sourcing information and writing repetitive narratives.

EY’s smart-decisioning risk-scoring framework was implemented to provide an initial understanding of risk that was used to prioritize high-risk customers for investigation. The AI-enabled framework was designed using a machine-learning model that identified key customer attributes indicative of risk and utilized a sustainable feedback loop, where historical due diligence and investigations data allowed for future training and refinement of the scoring model.

Additionally, EY developed a predictive modeling methodology to evaluate alert risk, enabling the process with machine learning technology that utilized feedback from historical investigations. Risk tolerance levels were configured based on the bank’s risk appetite, employing data from alert generation, transaction monitoring and other systems. The tool’s smart-decisioning model pinpointed suspicious activity and automated the process of sifting through and reviewing unproductive alerts. This reduced the investigators’ workload significantly by eliminating false positives from the review population.

One of the key attributes identified for risk scores was relevant negative news associated with a customer. A higher negative news score drives a higher risk score and could raise a red flag in the relationship with that customer.

EY’s negative news solution automated and enhanced the typical negative news search process. The tool’s technology automatically scraped negative news information from open web and third-party data providers for relevant customers and counterparties, evaluating and summarizing each article, and ultimately computing a comprehensive negative news score for the customer. The summarized negative news report was generated in a fraction of the time required to manually perform the search (from hours to seconds), enabling quicker analyst decisions and more focus on value-add efforts.

For the alerts that were automatically identified for closure, EY generated a disposition narrative summarizing key information and rationale for closure. The narrative generation technology was integrated with existing client systems and EY’s other technology enablers to pull key information from upstream systems and identify areas requiring manual input within the narrative. This enhancement standardized the disposition narratives for alerts and shifted analyst focus to higher-risk alerts, providing significant time savings and reduced costs for the client.