Broadridge’s LTX looks to GenAI as it competes for market share

Breaking into a market where giants already loom large isn’t for the faint of heart, nor the shallow of pockets. In fixed-income trading platforms, there are the Big Three: Bloomberg, MarketAxess, and Tradeweb. But two years ago, Broadridge ambitiously threw its hat into the bond trading ring with its LTX platform.

LTX’s value proposition is that broker-dealers stand to gain better execution and price improvement by aggregating liquidity across multiple buyers. It offers an AI-powered trading platform that targets “natural” buyers and sellers for each security and gives respondents the opportunity to bid or offer their desired amounts through its proprietary, patented trading protocol, RFX, as an alternative to the widely used request-for-quote (RFQ) protocol.

Innovation has been seeping through fixed income for decades, and that evolution is rapidly expanding. This year, LTX expanded its AI-driven efforts to capture clients with the roll-out of BondGPT, built on OpenAI’s GPT-4 large language model, in June. The generative AI-powered tool aims to answer traders’ bond-related questions and assist them in the identification process of corporate bonds on the platform. Broadridge’s bet on the technology marks one of the first clear efforts by a trading platform to implement generative AI, though it’s still unclear how much it will help LTX stand out in an LLM hype frenzy that sent firms of all stripes scrambling to develop use-cases.

Jim Kwiatkowski, chief executive officer of LTX, says 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 real estate. He says the growing popularity of generative AI and the large language models 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 tells WatersTechnology.

The answer they get in return might be a table, chart, or textual response, he says. 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.” Kwiatkowski says that in the 95^th percentile, the application is delivering answers in less than 20 seconds.

“People are saying that those same answers might require them to go to multiple data sources and combine an interim answer from each and do some work on a scratchpad or in their head,” he says. “When people are busy, those seconds matter.”

After taking in more client feedback following the initial BondGPT rollout, LTX looked to build on what they had started. Last month, they introduced BondGPT+, the enterprise version of the application.

The new version allows a user’s proprietary data to be integrated, as well as third-party datasets to sit alongside the critical information needed for trading. Building on the natural language interface of the application, users can now “favorite” queries and questions and schedule them for certain times of the day or when a market event occurs. The initial offering of BondGPT included sample questions and client feedback indicated that the ability to insert their own questions would be helpful. BondGPT+ can also now be integrated into a trading workflow and the amount of screen real estate used can be determined by the user.

Status report

When LTX launched in 2021, there were skeptics. WatersTechnology spoke to several people who expressed doubts about the platform’s chances of attracting users and about co-founder and former LTX chief executive, Jim Toffey. Kwiatkowski took over the company a year ago, while Toffey became chairman of LTX’s board of directors.

Doubts also exist as to how useful generative AI can be across multiple use-cases in fixed-income trading. Vuk Magdelinic, CEO of Canadian fixed-income analytics company Overbond, is no stranger to artificial intelligence and its applications in the capital markets. His company has been using AI since “the dawn of time,” or, more realistically, since the company was founded in 2014. He believes that while the wave of expectation around generative AI is spurring a lot of interest, it is not something that naturally lends itself to the problem-solving associated with fixed-income trading.

“Not every capital markets problem lends itself to a type of problem where generative AI can move the needle,” Magdelinic says. He believes that workflow assistance—similar to BondGPT’s objective—is the way forward for generative AI in fixed-income trading, and he does not think the technology can be effectively applied to more common problems around liquidity and pricing due to error margins and potential hallucinations that arise from the software.

“I recently wrote an essay on fixed income using ChatGPT. The essay was written in a matter of two minutes, I put in as many parameters as I could, and it was 80% usable content, 20% erroneous content,” he said. “It cut maybe 5 hours of my time. If we applied the same error margin to a bond trade, I would have been smoked. My P&L would have had $10 million lost today, and it wouldn’t matter how much [time] it saved me. The problem and solution need to meet the risk profile we’re talking about.”

Not every capital markets problem lends itself to a type of problem where generative AI can move the needle

Vuk Magdelinic, Overbond

LTX’s Kwiatkowski acknowledges that the company faced an uphill climb when breaking into the market, but he emphasized the need to innovate to compete with other established players for screen real estate.

“There are other trading platforms, and we entered a space looking to differentiate ourselves in a couple of ways as a trading platform, but there’s no denying that, as a new trading platform, you need to earn desktop screen real estate. BondGPT has helped us do that,” Kwiatkowski says. “Now we are being asked to build on the foundation that we came up with [for] BondGPT, so getting us a permanent place on our customers’ desktops helps us to earn trading volume.”

LTX shared that the most recent user data shows more than 200 firms and 700 users overall after four months of BondGPT in operation.

A fixed-income specialist who spoke to WatersTechnology in 2021 said that a platform like LTX is great in theory, with the caveat that what’s great in theory doesn’t always translate into what gets used and liked. Traders, especially in the insular fixed-income community, are creatures of habit, and LTX—and other platforms that purport to upend well-established workflows—are likely to encounter resistance.

That issue is still a consideration today. “For people doing RFQs, I don’t know how motivated they are to change if that’s the bulk of their trading,” the consultant says, adding that RFQs are becoming more commoditized for investment-grade investing. “If you’re going out to five or 10 dealers and you’re getting the same type of responses, no matter where [you choose], you’re going to use the cheapest platform.”

The willingness to take on a new platform comes from its ability to do something that nobody else can or because it is significantly additive to an arduous process, they say. Adoption also takes time and other platforms—such as fellow electronic trading platform Trumid, founded in 2014—are a testament to how long it can take to permeate a space.