ICE: LLMs drove 60% increase in ICE Chat trades

Intercontinental Exchange has seen a 60% increase in the number of transactions executed via its ICE Chat messaging platform as a result of refining proprietary large language models that help automate trading over the platform.

“We’re seeing good demand for adoption of our proprietary LLMs, and we are seeing the effect of that not just in our data services revenues, but also in other segments, such as energy trading,” said Lynn Martin, president of NYSE and chair of ICE’s Fixed Income & Data Services business, during the exchange’s Q2 results call, held earlier today.

Martin said the exchange has used LLMs in its ICE Chat offering for “quite some time,” starting with early steps around a decade ago, but added that these have been significantly refined over the past five years, and have become a contributor to growing transactions in its energy markets, including oil, gas, and utilities markets.

ICE Chat is a desktop and mobile messaging platform that includes a marketplace directory of market participants, and allows clients to message one another, participate in chat rooms, blast messages to multiple users for price discovery, and also uses quote and trade recognition logic and APIs to recognize and extract data from chats into actionable data for trading.

The LLMs enable greater automation so that the chat platform understands that participants are agreeing a trade, and facilitates a seamless integration with ICE’s trading platform and clearinghouse, while also providing analytics to ensure traders can execute at the best price available.

ICE has improved these LLMs over the past couple of years to allow traders “to automate more of their workflow,” Martin said, adding that the LLM development is still in its early stages. Nevertheless, it contributed to a 15% year-on-year increase in ICE Chat users, which now number almost 120,000.

But while it’s still only a “small contributor to volumes,” she said ICE is “very excited” about the potential of the LLMs—not just to drive transactions in the commodities markets, where revenues from energy trading and agricultural products rose by 34% and 27%, respectively, over Q2 last year to $355 million and $77 million, but also in other asset classes such as fixed income, which is “desperate for automation.”

This could include contributing to ICE’s automated evaluated pricing services, and other areas of the Fixed Income & Data Services business—which delivered revenues of $546 million overall, including $277 million from fixed income data and analytics, and $157 million from other data and network services—as well as seeing the effects of the LLMs “bleed into” its ICE Bonds market, where transaction revenues increased by 17% to $28 million.