Reducing Operational Risks and Trading Frauds with Generative AI

Background:

The Client is a part of one of the top 5 Financial Services conglomerate in India, offering retail broking and investment services. The client is a full-service broker offering execution along with research, advice, financial planning, etc.

Challenges Faced:

Our client, offering a Call & Trade facility for stock market transactions, faced significant challenges in reconciling trades made through this service. The facility, designed to help customers trade during system downtimes or when they lack internet access, became a source of compliance issues, leading to reputational damage and financial fraud amounting to millions of rupees.

The core issue lay in the manual process of recording and verifying customer orders placed via phone, which led to disputes such as unauthorized trades, incorrect order quantities, and misuse of funds. To address these challenges, the objective of this case study is to showcase how our solution enhanced the client’s ability to maintain accurate records, resolve disputes effectively, and restore trust in their Call & Trade service.

Objectives:

• Automated Trade Verification: Implement AI-driven verification of trades to ensure accuracy in order execution, minimizing errors such as incorrect quantities, prices, or stock selections.

• Dispute Resolution Optimization: Leverage AI to analyze recorded calls and automatically flag potential disputes, providing faster and more accurate resolutions by identifying inconsistencies in trade instructions.

• Fraud Detection and Prevention: Utilize AI to detect and prevent fraudulent activities by monitoring trade patterns, identifying suspicious behaviors, and flagging unauthorized transactions.

• Compliance Assurance: Ensure adherence to regulatory requirements by automating the documentation and retrieval of trade records, including call recordings, thereby reducing the risk of non-compliance and associated penalties.

• Enhanced Customer Trust: Increase customer confidence by providing transparent and reliable trade processing, backed by AI-generated audit trails and real-time feedback on trade execution status.

Solution Summary:

The proposed solution focuses on accurately reconciling stock trades by identifying stock, quantity, and price per trader, per order, and matching this data at the end of the trade process. Any discrepancies are promptly notified to the client’s designated office with detailed information.

Traditional tools like Speech-to-Text, translation, RPA, or Hyper Automation offer limited capabilities in analyzing audio files and building meaningful relationships. To address this gap, we utilized Large Language Models (LLMs), which excel at recognizing speech, processing it, and identifying keywords and relationships with high accuracy.

Key components of the solution include:

1. Speech Recognition: Whisper AI, an offline LLM, is employed to transcribe and translate audio data, ensuring accurate capture of trade instructions.
2. Custom Keyword Identification: The LLM is trained on custom data, enabling it to recognize specific keywords relevant to trade reconciliation and to establish relationships between them. This approach significantly outperforms traditional NLP models, which require substantially more data to achieve similar accuracy.

By leveraging these advanced AI technologies, the solution enhances the ability to detect and resolve trade discrepancies, improving compliance and reducing the risk of fraud.

Tech Stack:

• Development : Python

• Transcribing & Translating : LLMs – Whisper AI, Bloom

• Named Entity Extraction and Relationship Extraction : Approach 1 (NLP) – Labelstudio, BERT | Approach 2 (LLM) – Falcon, Red Pajama, SalesForce X Gen

• AWS Services : EC2, Sagemaker