An approach for detecting emerging operational risks from textual data

Operational risk (OpRisk) constitutes a significant non-financial threat with wide-ranging implications for financial institutions. While traditionally centered on regulatory compliance—encompassing data collection, capital requirement estimation, and the generation of managerial reports—OpRisk functions are increasingly shifting toward proactive strategies designed to anticipate and mitigate potential risk exposures. Artificial Intelligence techniques are playing an expanding role in this transformation by extracting actionable insights from unstructured data sources. In this study, we investigate the application of text analysis methods, a key component of natural language processing, to OpRisk event descriptions. The contributions of this work are threefold. First, we present a fully integrated workflow that applies text analysis techniques not only to internal event narratives but also to external data sources, including publicly available web content. Second, we introduce a tailored variant of latent Dirichlet allocation for clustering OpRisk events, aimed at uncovering latent patterns and identifying the underlying causes of risk. Third, recognizing the limitations of conventional models in mitigating the impact of future loss events, we explore the enhancement of traditional data sets with alternative sources that enable earlier detection of emerging risks. Specifically, we develop a real-time analysis of relevant content from X (formerly Twitter) to enable continuous monitoring of the evolving risk landscape and to detect early warning signals of novel risk types. This study demonstrates how these diverse methodologies can be cohesively integrated into a unified OpRisk management framework, fostering a more holistic, forward-looking, and adaptive approach to risk mitigation.