Today, the discussion around AI in support of banking operations can seem synonymous with generative AI. At CTFSI, AI Operationalized always starts with an understanding of operational workflows before we even begin to consider what AI capabilities to bring to bear. But at the point where the problem or opportunity has been identified, we tap into two specialized “AI Quivers” – one for our PEP and UDP platforms respectively – each tailored to meet distinct challenges. Having a selection of “AI Arrows,” or AI technologies, to choose from enables us to tackle a broad range of operational challenges and avoid the proverbial attempt to fit round pegs into square holes and deploy the right AI technologies and tools for the job at hand – be it gen AI or otherwise.
Two AI Arrow Quivers: Precision Tools for Distinct Challenges
CTFSI’s capabilities are built around two key product platforms, each represented by its own “AI Quiver”:
The first we call our Prediction Exception Processing (PEP) platform, which generally deals with structured, numerically-rich tabular data. The “PEP AI Quiver” includes “AI Arrows” designed to address complex exception processing problems such as predicting securities trade fails. Accurately predicting when and if a securities trade is going to fail involves analyzing extensive datasets and identifying patterns that may indicate a likely issue. We utilize approaches and techniques such as learning contextualized embeddings for categorical feature values, multinomial naive Bayes and deep learning classification, and ensemble modeling. By combining deep industry domain expertise, rich data enrichment, and feature engineering, we orchestrate an end-to-end process that generates accurate forecasts and enables proactive exception management in advance of the actual settlement date.
“This strategic use of diverse AI tools ensures that each solution is not only innovative but also practical and tailored to each solution’s needs”
The second AI Quiver supports our Unstructured Data Processing (UDP) platform, which is focused on managing operational risks and addressing inefficiencies related to processing unstructured data. “UDP AI Arrows” are capable of parsing, grouping, extracting, reasoning, and summarizing information from PDF documents and other data sources. For example, producing financial statements requires cross-checking numerical content in rendered PDFs. The use of Retrieval Augmented Generation (RAG) and LLMs to extract the table content from the PDFs can empower automated workflows that greatly reduce the time for human quality assurance of these documents. Our solution includes the use of agentic workflows with vendor or hosted open LLMs, including multi-modal LLMs and VLMs. Depending on the use case, hosted LLMs can be fine-tuned in a customer-specific, secure VPN. In addition to LLMs, our solution for unstructured content involves other AI/ML components such as clustering and NER. Like our PEP-driven solutions, we bring our full industry knowledge to bear to accelerate practical deployments of AI that are employed to process and synthesize large volumes of text, automating complex content management tasks and reducing manual effort.
More Than Just Gen AI: A Comprehensive Approach to AI Operationalized
While generative AI is a valuable component of CTFSI’s toolkit, the concept of “AI Operationalized” involves much more. CTFSI’s approach encompasses a broad spectrum of AI techniques and tools, carefully selected to address specific challenges married to our deep domain expertise and a workflow-first approach. The PEP Quiver focuses on utilizing traditional and advanced ML algorithms to process structured data, while the UDP Quiver leverages state-of-the-art LLMs and RAG techniques for unstructured data.
This strategic use of diverse AI tools ensures that each solution is not only innovative but also practical and tailored to each solution’s needs. By integrating both traditional AI methods and cutting-edge Gen AI technologies, CTFSI provides solutions that go beyond mere technological adoption, delivering real, actionable value.
Conclusion
CTFSI’s dual Quivers—PEP for structured data and UDP for unstructured data—represent a comprehensive approach to using AI to solve for operations challenges in institutional banking. By deploying targeted AI solutions through these specialized platforms, CTFSI ensures that AI is operationalized effectively to address the sector’s most pressing challenges. This methodology provides institutional banks with precise, impactful tools that enhance efficiency, manage risk, and streamline processes, all while leveraging a mix of advanced AI technologies.