Overview
Capital market organizations, including Broker-Dealers, Custodians, and Investment Banks, operate in a landscape fraught with various risks, spanning from credit and operational risks to market uncertainties. Navigating these challenges requires sophisticated risk management techniques, especially in today's BANI (Brittle', 'Anxious,' 'Nonlinear' and 'Incomprehensible') world.
Our client, a US-based investment banking firm, was grappling with conventional rule-based risk management processes. The inherent challenge lay in the inability of the existing architecture to effectively address unknown risks and anomalies, leading to a delayed time-to-market for rule changes and, subsequently, low responsiveness.
In response to these challenges, the client sought Birlasoft's expertise to develop an innovative risk management solution leveraging Generative AI to complement their existing rule-based approaches.
Birlasoft crafted a solution integrating LLM models from OpenAI with LangChain and conventional AI, specifically designed to identify anomalies in trading patterns. The models underwent progressive training on simple and sophisticated anomaly scenarios, fostering context awareness.
The Challenge
1. Identifying Unknown Risks/Anomalies
1.1 Ineffective Trading and Margin Anomaly Detection in Existing Risk Management Processes
Insufficient insight into anomalous trading activities and margin-related events in capital markets can lead to losses and non-compliance. The existing solution often needs to catch up, such as risky positions in high beta stocks or rising activity in volatile environments..
1.2 Revenue Loss due to Higher Margin Requirements
The absence of an effective margin check function exposed the client to credit risks. The existing margin check function did not efficiently capture anomalous behavior, resulting in higher margin requirements deployed across the board, reducing potential revenue.
2. Suboptimal Prediction of Risk Sentiment of Traders Around New Investment/Financial Products
Financial institutions must gauge changing risk sentiment around financial products for key asset management decisions. Our client sought insights into shifting risk sentiment around financial instruments and a better understanding of increased activity in risky scenarios.
Birlasoft Solution
The Birlasoft Gen AI Center of Excellence, drawing on its deep expertise in capital market risk scenarios, utilized a library of use cases to prototype and devise a high-performing insights-based risk management solution with Generative AI.
Key Techniques Employed:
- GUI-based Risk Management Solution with Generative AI Running on Azure OpenAI
- In conjunction with Prompt Engineering, Birlasoft's AI experts leveraged an OpenAI LLM on Azure's OpenAI service to develop a GUI-based risk management solution. The LangChain framework minimized development time, and the model was trained with synthetic trading data comprising 50,000+ records, reducing bias and false positives.
- Anomaly Scenarios and Synthetic Data
- Birlasoft developed a library of anomaly scenarios and synthetic data based on deep insights into trading patterns. The scenarios were scaled using GenAI to make the LLMs context-aware, breaking down anomaly scenarios into chains of thought.
- Mixed Approach to Reduce Cost of Running GenAI Models
- By making the LLMs context-aware, LangChain simplified data queries and allowed the team to leverage traditional AI in conjunction with GenAI, reducing costs and developing a cost-effective solution.
Key Business Benefits/Impact
The generative AI-based risk management solution delivered by Birlasoft provided the client with profound insights into complex market activity, facilitating better corrective actions for risk mitigation. Key benefits include:
- Deep Insights into Trading and Margin Anomalies
- The solution enabled the client to predict changing sentiments around risky financial products and margin accounts where debits or equity changes exceeded 25% over varying time periods, empowering risk managers to make better decisions and employ effective risk mitigation strategies.
- High-Precision Anomaly Detection
- The solution achieved a 40% reduction in false positives and a 30% reduction in false negatives, significantly enhancing the confidence of risk managers and improving overall risk management outcomes.
- Enhanced Risk Mitigation and Risk Segmentation
- The client could now better segment accounts based on risk, enabling targeted risk mitigation measures based on risk segments.
- Insights on Potential Clients Getting Margin Call Notice
- The solution provided valuable insights on potential clients receiving margin calls with an accuracy of 20-40%, empowering risk managers with crucial data.
- Enhanced Development Efficiency and Time to Market
- Training the Large Language Foundation Model using contextualized synthetic data resulted in an optimized Generative AI LLM. This approach accelerated the time-to-market for the solution by 10-20% and improved development efficiency by 15-25%.
