Sigma’s AI: Revolutionising risk management and compliance

Sigma's AI Revolutionising risk management and compliance

Sigma, recently ranked in the AIFinTech100 for 2023, is driving innovation with AI to offer robust and scalable risk management solutions. Their unique use of AI technologies makes the platform indispensable for current and prospective clients.

The AI of Sigma combs through more than 4.5 million articles every month from over 200,000 publishers. Its objective is to filter out negative news involving companies or individuals. By employing event tags, it identifies and categorises risks. Additionally, it highlights entities referred to in the articles. To streamline the due diligence review process, articles pertaining to the same real-world event are grouped together.

Sigma’s cutting-edge performance in news and adverse news screening can be attributed to the utilization of large language models (LLMs). These models are particularly proficient at handling various natural language processing (NLP) tasks.

The Adverse News model by Sigma selectively extracts news relevant to risks as deemed significant by the end user from over 200,000 news sources. This model benefits from transfer learning from the DistilBERT LLM. The model also uses event tags determined by a supervised deep learning model built on the RoBERTa LLM.

The Sigma model is trained on a proprietary and growing corpus of expertly tagged news articles. It covers a range of events including terrorism, financial crime, environmental crime, legal risk, arms trafficking, and human rights violation.

Risk managers can fine-tune the efficiency of their teams by choosing only the event tags that interest them. Notably, Sigma’s model also considers news events that may not be adverse but could still hold significance for risk managers.

Sigma’s Named Entity Recognition (NER) technology spotlights and extracts specific entities such as organisations, individuals, and locations from news articles. This technology significantly enhances the relevance of articles linked to the entity of interest, compared to outdated keyword matching systems. Sigma’s NER model capitalises on BERT (another LLM), powered by self-attention among other methodologies.

Sigma is looking to the future with plans to only show articles where the risky actors match the entities the user is interested in. By extracting pertinent entity relationships from news articles, Sigma aims to establish robust links between entities, providing risk and compliance teams with an additional advantage in identifying network-based risks.

Sigma’s matching system has emerged as an industry leader, offering a ten-fold increase in speed due to the implementation of Golang. Furthermore, they are set to incorporate AI and deep learning models into matching algorithms to enhance matching accuracy.

Traditional strategies focusing solely on individual transactions or client screening no longer suffice in the face of increasingly sophisticated global crime networks. Sigma is set to launch Sigma360 Platform to address these concerns. This platform provides users with the ability to assess connections between entities in a swift and scalable manner. The Sigma360 platform’s network graph feature allows users to explore the connections of an entity to the broader world, providing a single stream of related risk intelligence.

Sigma’s AI platform is favoured by leading financial institutions and high-risk businesses for its capability to identify, screen, monitor, and review clients and their relationships. The platform offers real-time data updates on news, corporate registries, sanctions, and enforcement actions, all in one dashboard. It also boasts the best-in-class entity resolution and a network-based knowledge graph, ensuring a comprehensive view of risk is never missed.

Read the full report here.

Keep up with all the latest FinTech news here

Copyright © 2023 FinTech Global

Enjoying the stories?

Subscribe to our daily FinTech newsletter and get the latest industry news & research


The following investor(s) were tagged in this article.