Fixed income markets have undergone substantial electronification over the past two decades, but parts of the primary issuance process remain surprisingly manual. On busy new-issue days, salespeople can still spend significant amounts of time reading investor orders and amendments from chat conversations and manually entering them into bookbuilding systems.
Sense Street, the capital markets AI company, is targeting that workflow through a new integration with S&P Global Market Intelligence’s IssueBook platform, using AI to turn unstructured conversations between investors and salespeople into structured order information that can be fed into the bookbuilding process.
The integration brings together Sense Street’s language models, which have been developed specifically to interpret capital markets conversations, with IssueBook, S&P Global Market Intelligence’s primary market bookbuilding platform.
For Sense Street CEO Haroon Hassan, who spent much of his career on credit trading desks before founding the company, the opportunity reflects a longstanding inefficiency in primary markets.
“If you walk down a trading desk when the new issue calendar is busy and you’ve got five or six new issues coming, it can be a complete mess,” he tells TradingTech Insight. “You’ve got some very highly paid salespeople taking orders from clients in conversation and physically transcribing them into bookbuilding software.”
From Conversation to Order
The problem isn’t that primary market communications haven’t been digitised. Investor conversations already take place extensively over electronic messaging platforms. Rather, much of the information contained within those conversations remains unstructured.
During a bond issue, an investor might place an order and subsequently increase it, reduce it or withdraw altogether as market conditions change. Each instruction has traditionally required somebody on the sales desk to interpret the message and update the book accordingly.
Sense Street’s models are designed to identify that order intent within the conversation and convert it into structured information that can be passed into IssueBook.
For this particular integration, the source is Bloomberg IB Chat, although Sense Street already processes conversations from Bloomberg, ICE and Refinitiv across other fixed income and commodities workflows.
The technology also needs to understand the changing state of an order rather than simply extract individual messages in isolation. Once the relevant information has been identified and passed into IssueBook, the salesperson retains a final validation step.
That human oversight is deliberate.
“As we think about AI and automation and all these amazing things that are happening, it’s very important that you cannot take a human out of the loop, especially when essentially what you’re doing is executing a trade,” says Hassan.
Freeing Up the Sales Desk
Automating order entry potentially changes the role salespeople play during a busy bookbuild. Rather than repeatedly switching between client conversations and the bookbuilding platform to update orders, they can spend more time discussing the issue and broader market conditions with investors.
“The real win here is for sales teams because they’re getting valuable time back on a day when things are busy,” says Hassan. “Currently, much of their contribution to the new issue process is just entering orders. Now they can engage clients in meaningful conversations about the quality of the book, the nature of the order and what’s going on in the market.”
Sense Street has already run initial pilots of the workflow with a bank, according to Hassan, with the experience suggesting that removing some of the administrative burden can change the nature of client interaction during an issue.
That distinction is particularly relevant in fixed income primary markets, where relationships between investors, sales desks and syndicate teams remain important. Although technology could potentially automate more of the interaction between the different parties, Hassan argues that doing so isn’t necessarily desirable.
Information flowing back from IssueBook to investors during the bookbuild, for example, continues to be communicated manually by salespeople. Maintaining those points of contact gives sales teams opportunities to discuss the market and the order with clients rather than reducing the entire process to a self-service electronic workflow.
For Hassan, the aim is to remove rote work from trading and sales desks rather than remove people from the process.
A Practical Use Case for AI
The integration also reflects a broader shift in the way financial institutions are approaching AI.
Much of the early discussion around AI in capital markets focused on its ability to generate new insights from large datasets, improve decision-making or ultimately identify new sources of revenue. Those possibilities remain, but Hassan sees banks increasingly focusing on narrower applications where the benefits can be more readily identified.
“There’s a yearning for tangible things to come out of all of this AI discussion, and that means people are looking at workflows,” he says. “Where people are feeling tangible benefits is when you focus on workflows. I think the more prosaic workflow conversations are what’s happening.”
Primary bond issuance provides a useful example. The underlying client conversation remains largely intact, as does the salesperson’s responsibility for checking the resulting order. AI instead acts as an intermediary between an unstructured communications channel and a structured workflow.
For trading firms assessing where AI can deliver practical benefits in the front office, that model may prove increasingly relevant: identifying the repetitive work surrounding trading and client interaction, automating those tasks where appropriate, and leaving traders and salespeople to concentrate on the parts of the process where human judgement and relationships still matter.
Subscribe to our newsletter
