Need smarter insights in your inbox? Sign up for our weekly newsletter to get only the things that matter to enterprise AI, data and security leaders. Subscribe now
Despite being a highly regulated industry, stock trading has been consistently at the forefront of innovation in the financial services sector. However, when it comes to agents and AI applications, many banks are I took a more careful approach Adopted.
TD Securities, TD Bank’s equity and securities trading division deployed TD AI virtual assistant on July 8thaimed to become a front office sales, trading and research expert, helping to manage workflows.
TD Securities CIO Dan Bosman told VentureBeat that the virtual assistant’s main goal is to help front office stock sales and traders gain client insights and research.
“The first version of this started as a pilot and then expanded,” Bosman said. “It really is about accessing the stock survey data that analysts put out and bringing it to the sales team’s hands so that it’s within the hands of the sales team.”
AI scaling reaches its limit
Power caps, rising token costs, and inference delays are rebuilding Enterprise AI. Join exclusive salons and discover what your top team looks like.
- Turning energy into a strategic advantage
- Architects efficient inference for real throughput gain
- Unlock competitive ROI with a sustainable AI system
Make sure you have your place to stay first: https://bit.ly/4mwgngo
Bossman feels that being around the trading floor means being exposed to many jargon, and the context in which users ask some questions is very unique. Therefore, AI assistants should be able to hear natural, intuitive and access insights generated by traders.
Building TD AI
Bossman said the idea for the AI assistant came from members of the stock sales team. Fortunately, the bank has a platform called TD Invent, allowing employees to bring ideas and the innovation leadership team to responsibly evaluate projects.
“Someone from our Equity Research Sales Desk came over and I had this idea and brought it to TD Invent,” Bossman said. “The thing I love most about this is when you don’t have to make something very magical and go out and sell or show up. People come and say, ‘I want this, I need this, or I have an idea.’
TD Security has built the TD AI Virtual Assistant by leveraging OpenAI’s GPT model. Bossman said TD collaborated with the technology team and the Canadian AI company Layer 6, and worked with other strategic partnerships acquired by the bank in 2018. The assistant is integrated with the bank’s cloud infrastructure, providing access to internal research documents and market data, such as 13F filing and historical equity data.
Bosman calls TDS AI a knowledge management system. This aggregates terms and information, including the ability to search through the Search Augmented Generation (RAG) process, and combines them into “concise context-conscious summary and insights,” allowing sales teams to answer client questions.
TD AI Virtual Assistant allows users to access TD AI Prism, the underlying model of TD Bank.
The model, released in June, is used not only by TD Securities, but also by the entire bank. During the launch, the bank said that TD AI PRISM will process 100x more data, replace a single architectural model, and improve the predictive performance of TD Bank’s applications by keeping customer data inside.
“Development poses unique challenges as development was relatively new to the organization when the initiative began and requires careful navigation of governance and control,” Bosman said. “Nevertheless, the project brought together a diverse team across the enterprise, fostering collaboration and providing cutting-edge solutions.”
He added that one of the outstanding features is the ability of SQL from text to convert natural language prompts into SQL queries.
To train assistants, Bossman said TD Securities developed optimizations to facilitate the process.
“Due to rapid engineering and patent-pending optimizations in searching dynamic few shot examples, we have successfully achieved the desired business performance through context learning,” Bossman said. “As a result, fine-tuning of the underlying OpenAI model was not required to interact with both unstructured and tabular datasets.”
Banks slowly enter the age of agents
Of course, TD Bank and TD Securities are not just banks interested in expanding from assistants to AI agents.
BNY told VentureBeat it has begun offering multi-agent solutions to its sales teams to help answer customer questions like those related to foreign currency support. Wells Fargo also saw an increase in the use of internal AI assistants. For car sales customers, Capital One has built an agent that will help them sell more cars.
Many of these use cases, like in all other industries, emerged after months of pilot testing. However, financial institutions have an additional burden of strict customer data privacy and fiduciary responsibility.
The bossman of TD Securities noted that many employees are increasingly familiar with tools like ChatGPT, even in the business side of the bank. The challenges of pilot test assistants and agents are less likely to teach about tools, but they will establish best practices for using assistants, integrate them into existing workflows, understand their limitations, and understand how humans can provide feedback to alleviate hallucinations.
Ultimately, Bossman said that even the things assistants want to use when communicating with TD Securities will evolve.
“My vision is that we can not only see AI as something that can add value to us, but we can also see it in external customers at the bank. Nowadays, driving a stronger client experience and providing a better peer experience is a huge opportunity,” Bossman said.
Source link
