In 2023, Broadridge became one of the first capital markets-focused companies to roll out a GPT-powered tool. Now it is looking across use cases and foundational models to determine where to plant its flag next.

No technology has captivated the capital markets in the last year more than generative AI and large language models (LLMs). Some have seized on the moment cautiously. Others, like fintech provider Broadridge, sought to plant their flags early.

Broadridge CEO Tim Gokey tells WatersTechnology how he set the company’s tone on AI: “What we have said is we intend to be a leader in AI in our space, and that means we’re not going to hire PhDs to create the next large language model. Within the pretty arcane space of capital markets and the things we do, it makes way more sense for us to invest in bringing AI to that than to have each of our clients make an investment in that area individually and have it all add up.”

Gokey says there are currently 27 different initiatives underway across the company, all of which aim to either bring AI into existing products or create new, AI-powered solutions.

Part of that approach started last June, when the company unveiled BondGPT, a generative AI-powered tool for its LTX bond trading platform, which aims to answer traders’ bond-related questions and assist them in the identification process of corporate bonds on the platform. Built on OpenAI’s GPT-4 LLM, Broadridge’s bet on the technology marked one of the first clear efforts by a trading platform to implement the burgeoning technology.

LTX CEO Jim Kwiatkowski told WatersTechnology last fall that the platform was the byproduct of internal conversations and requests from clients to create a dashboard that provided them with more data while taking up minimal screen space. He says the growing popularity of generative AI and LLMs that underpin genAI helped guide the team’s thinking beyond the traditional ways of digesting data—something they saw as creating a lot of work for an end-user.

“In a tiny chat box, BondGPT’s natural language capabilities allow users to ask questions that would typically require access to multiple different datasets, often through different user interfaces, all at once,” Kwiatkowski said. The answer they get in return might be a table, chart, or textual response, but ultimately the goal is to dedicate the screen real estate necessary for the answer instead of dedicating space to “a whole bunch of data they’re not using in the moment.”

In October, Broadridge launched BondGPT+, an enterprise version of the application that allows users to point the technology at their internal, proprietary data. Gokey sees a future where AI will be incorporated into most products. Currently, the vendor is utilizing OpenAI models, Meta’s Llama 2, and models from Anthropic. “We’re definitely agnostic right now, in terms of who’s going to be the ultimate winner,” he says. “Different people will be ahead at different times and different models are tuned for different purposes.” Broadridge must also consider which models run best locally on premises, which Gokey says at this point is Llama 2.

While utilizing different models for different use cases, Broadridge has identified the need to have workloads that stay within its walls. To address this, all developers, products, and associates flow through its enterprise-wide AI platform, so that everyone working with these different models follows the same security and compliance protocols.

“Each of our product teams is not separately going out to negotiate with OpenAI on their model,” Gokey says. “People can build their stuff, they can go to a central place, hook into multiple models, decide which one is working best, and leverage that with all the appropriate data checks.”

Joseph Lo, head of enterprise platforms at Broadridge, adds that this commonality also allows for centralized monitoring capabilities, such as ensuring that sensitive data does not leave the firm and controlling model performance.

Broadridge has also looked at genAI as a means not just to better serve its clients, but to serve its own interests. “These AI tools are great ways to attract talent,” Lo says. “The top engineers [and] top product managers expect the company they work for to have AI.”

In their efforts, the team created an AI assistant called BroadridgeGPT, or BroadGPT, that launched in September. “It was my dream that a Broadridge associate would start work at Broadridge with their laptop, with their email address, and their own personal AI assistant,” Lo says. Upon its launch, it was clear that employees liked the idea, Lo says, but they weren’t entirely sure how to use it. Since then, usage has increased five times and Lo estimates that about 30% of Broadridge employees are using the assistant.

BroadGPT pairs with another initiative centered around developer productivity. Developers are being onboarded onto Microsoft’s CoPilot, with around 400 active users at Broadridge. Gokey estimates that the number will settle around 1,000 by the end of the year.

Expanding use cases

In January, Broadridge rolled out its second GPT-powered tool: OpsGPT. The tool is aimed at the post-trade lifecycle to assist operations managers, analysts and teams. With the US, Canada and Mexico slated to move from a T+2 settlement cycle to T+1 next month, assisting post-trade operations has become a top priority for technology providers.

Lo describes the tool as an AI layer that sits on top of Broadridge’s core processing engines. It uses transactions, settlements and positions data to provide real-time visibility for faster fails resolution, researching next best actions and prioritizing key risk items in a single interface, according to the announcement issued at the beginning of the year. 

While it’s a new product, its data is not new. “Some of the things that it does, our clients have always wanted to get more out of,” Lo says. “How are my trades being settled, and who are my largest counterparties? It brings those things together.”

Gokey says solutions that seek to unite AI and capital markets should be built by existing financial technologists, rather than come from the broader tech world. “We think the best people to do that are our existing product teams, because they have the domain knowledge to say, ‘What’s the problem that would be most relevant in this area to apply?’” he says.

Broadridge holds this belief for its buy-side products as well. The vendor’s Distribution Insights tool, for example, allows asset managers to understand trends currently impacting ETFs and mutual fund flows globally. “How that works is that’s built on top of the data and the research and analysis we’ve already done—so new products, but within what we do already,” Lo says.

Not all AI applications that have come to market over the last year look the same, but vendors are flocking to the same delivery mechanisms. As the buzz grew around ChatGPT, many vendors have looked to chatbots or conversational interfaces to enhance their offerings.

S&P Global added a chatbot interface, ChatIQ, to its Capital IQ Pro workstation offering at the end of 2023, while FactSet rolled out its LLM-based knowledge agent, FactSet Mercury, as a part of its AI roadmap. 

Other vendors, like Nice Actimize, have outlined why they are opting to stay away from chatbots. Donna Weiss, senior director of product strategy and marketing at Nice Actimize, told WatersTechnology last month that the vendor is not looking to rely on conversational chatbots as much as other vendors have, mainly due to the challenges around prompt engineering. 

“You need to train users on how to be concise with the context of the question you’re asking,” she says. “It could spin in circles, and it can take a lot of time to go back and forth to finally get the answer that the individuals seek.” Lo recognizes the chatbots’ value in certain use-cases but overall doesn’t see a future where everything will be a chatbot. “Sometimes a button works way better,” he says. “I think in this current world we live in, chatbots are really good for when discovery is required.”