Data has long been described as the new oil. At the Fixed Income Leaders Summit, the conversation had moved on: the question is no longer whether data is a sore of value, but how firms can actually refine that value, and who owns the output.
Multiple panels noted the problem in fixed income is not a shortage of data. Trading volumes are at record highs. Pre-trade analytics have improved markedly. Real-time axe data, post-trade reporting, pricing feeds and trace information are more widely available than at any point in the market’s history.
The challenge, as speaker after speaker emphasised, is synthesis. “It’s not that we don’t have the information or the data,” said one senior figure from a major trading venue. “It’s so much that — what’s the quality of it, and can you trust the data that’s actually prompting you to execute?”
This quality problem can be grouped into two areas. First, is axe data. Real-time dealer inventory information has proliferated, and the connectivity infrastructure to pipe it into EMS platforms now exists. But for larger block trades, the data is often stale or indicative rather than firm, leading to negative interactions that discourage further use. Second, unstructured data, which captures a vast body of credit research, market commentary, voice communications and internal analysis sits outside any systematic workflow, generating no signal and adding no value.
To help overcome the first problem, AI tools are already being used to parse dealer axes, overlaying broader information sets to assess if axes are executable, and to measure patterns of dealer behaviour.
On the second problem speakers reported that unstructured data is increasingly be captured by natural language tools in order to facilitate pattern tracking and modelling in manual workflows, which would potentially allow the mapping of ‘natural’ trading activity, which could then be systematised.
New tools — retrieval augmented generation, natural language interfaces, AI-powered chatbots — are beginning to address the unstructured data problem. Tradeweb’s TARA, a newly launched AI research assistant built on proprietary transaction data, was cited as an early example of what is possible when clean, structured data is combined with large language model capability. The system allows clients to query trading history, transaction costs and dealer behaviour in plain language, removing the need for SQL skills or dedicated data science support. Its architects were clear, however, that the hardest part of the project was not the model – it was curating a clean, validated set of questions and answers to train it on. Data readiness precedes AI readiness.
A question that is beginning to receive serious attention is who controls data once it leaves a firm’s own systems. Several panellists noted that master service agreements with data vendors, signed years ago, contain no provisions governing whether that data can be used to train third-party AI models. Reviewing those contracts and renegotiating where necessary is becoming a compliance priority. In a market where proprietary data is the primary source of competitive advantage, the stakes could hardly be higher.
©Markets Media Europe 2025
