TD Securities Leverages AI for Real-Time Equity Insights
Despite its highly regulated nature, the equity trading sector has consistently led technological innovation within financial services. Yet, when it comes to the broader adoption of AI applications and intelligent agents, many traditional banks have exercised considerable caution. Breaking from this trend, TD Securities, the dedicated equity and securities trading arm of TD Bank, launched its TD AI Virtual Assistant on July 8. This sophisticated tool is designed to empower the bank’s front-office institutional sales, trading, and research professionals by streamlining their workflows and enhancing access to critical insights.
According to Dan Bosman, CIO of TD Securities, the virtual assistant’s primary objective is to equip front-office equity sales and traders with deeper client insights and comprehensive research. Bosman explained that the initial version began as a pilot before being scaled for broader deployment. The core functionality revolves around making vast amounts of internal equity research data, generated by the bank’s analysts, readily accessible and “sales-friendly” for the client-facing teams. Operating within a fast-paced trading environment means the AI assistant must understand the unique lingo and context of user queries, delivering natural and intuitive responses derived from the latest market intelligence.
The genesis of the TD AI assistant was an innovative idea from a member of the equity sales team, brought to life through TD Invent, the bank’s internal platform where employees can propose projects for evaluation by the innovation leadership team. Bosman highlighted the organic adoption of the tool, noting that its inherent value meant it didn’t require extensive internal promotion. Instead, teams across the bank quickly recognized its potential, actively seeking to integrate it into their operations. This collaborative approach, integrating investments in data, cloud infrastructure, and human ingenuity, proved instrumental in the assistant’s successful development.
TD Securities constructed the TD AI virtual assistant by leveraging OpenAI’s advanced GPT models, collaborating closely with its internal technology teams, the Canadian AI firm Layer 6 (acquired by TD in 2018), and other strategic partners. The assistant seamlessly integrates with the bank’s existing cloud infrastructure, granting it access to a wealth of internal research documents and real-time market data, including regulatory filings like 13F reports and extensive historical equity data. Bosman characterizes the TD AI assistant as a sophisticated Knowledge Management System. It employs retrieval augmented generation (RAG) processes to efficiently retrieve, aggregate, and synthesize complex information into concise, context-aware summaries and actionable insights, enabling sales teams to respond to client inquiries with unparalleled speed and accuracy.
Furthermore, the TD AI virtual assistant provides users with access to TD Bank’s overarching foundation model, TD AI Prism. Launched in June, TD AI Prism is deployed across the entire bank, not just within TD Securities. During its unveiling, the bank stated that this model significantly improves the predictive performance of its applications by processing 100 times more data than previous systems, replacing single-architecture models while ensuring all customer data remains securely within the bank’s internal network. Developing the generative AI solution presented unique challenges, particularly concerning governance and controls, given the technology’s relative novelty within the organization at the project’s outset. Despite these hurdles, the initiative fostered significant cross-enterprise collaboration, culminating in a cutting-edge solution. A standout feature of the assistant is its text-to-SQL capability, which translates natural language prompts directly into SQL database queries, simplifying data access.
To optimize the assistant’s performance, TD Securities developed proprietary enhancements in its training methodology. Bosman noted that patent-pending optimizations in prompt engineering and dynamic few-shot examples retrieval allowed them to achieve the desired business performance through context learning alone. Crucially, these innovations eliminated the need for extensive fine-tuning of the underlying OpenAI model, enabling seamless interaction with both unstructured and tabular datasets.
TD Bank and TD Securities are not alone in their pursuit of advanced AI solutions. Across the financial industry, a growing number of institutions are transitioning from basic AI assistants to more sophisticated AI agents. BNY, for instance, has begun offering multi-agent solutions to its sales teams, assisting with complex queries such as foreign currency support. Wells Fargo has also observed a significant increase in the usage of its internal AI assistant, while Capital One developed an agent specifically to enhance auto sales for its customers. Many of these sophisticated AI applications, much like in other industries, emerge after months of rigorous pilot testing. However, financial institutions face the additional burden of stringent customer data privacy regulations and significant fiduciary responsibilities, which necessitate an even more meticulous approach to AI deployment.
Bosman observed that many employees, even on the business side of the bank, are increasingly familiar with consumer-grade AI tools like ChatGPT. The primary challenge in pilot testing these advanced assistants and agents lies not in educating users about the tools themselves, but rather in establishing best practices for their use, seamlessly integrating them into existing workflows, understanding their inherent limitations, and developing effective mechanisms for human feedback to mitigate potential inaccuracies or “hallucinations.” Ultimately, Bosman envisions the assistant evolving to a point where its value extends beyond internal teams, becoming an indispensable tool for external customers interacting with TD Securities. This future-forward perspective underscores AI’s potential to significantly enhance both client and colleague experiences, driving stronger engagement and operational efficiency.