As AI and large language models (LLMs) begin to play an increasingly vital role in financial markets, Matt Cheung, chief executive of ipushpull, explores how conversational data is becoming an important source of early market signals.
Financial markets have always depended on one fundamental ingredient: the ability of buyers and sellers to communicate. 400 years ago, that happened face-to-face in coffee houses or on the floors of the earliest exchanges. Today, aside from a few who resist the march of progress, communication, and therefore trading, has shifted to electronic venues, chat, and, to a much lesser extent, voice. Over recent decades, electronification has transformed how markets operate, with trading now driven largely by structured data.
These markets generate oceans of information, yet one of the most revealing datasets remains largely ignored: the daily flow of conversation between clients, sales and traders. Hundreds of millions of words and prices move through chat every day. In a market saturated with data, the next major source of insight will come from this conversational layer; price queries, questions, commentary and informal interactions that shape real-time decision making. These exchanges reveal intent, uncertainty and tone shifts, often well before such signals appear in traditional market data. Unstructured conversational flow is becoming one of the most important yet least recognised sources of intelligence in modern markets.
Historically, this data was almost impossible to capture or interpret at scale. Conversations are fluid, shorthand-heavy and context-dependent, meaning only humans could make sense of them. As a result, chat data remained an untapped reserve for two reasons: limited electronic access and no scalable way to process unstructured language. That changed in November 2022. Just as the browser unlocked the internet and mobile unlocked social media, ChatGPT became the interface moment for AI. Firms realised for the first time the need to adopt AI at scale. Interest in chatbots – and more recently, agents – accelerated across financial markets. Due to demand, Bloomberg opened APIs to its IB chat, giving firms programmatic access to conversational flow.
LLMs, advancing rapidly since Google’s 2017 transformer architecture paper, now offer a practical solution to the messiness and ambiguity of chat. Natural-language AI can interpret conversational flow with the same rigour historically applied to prices, quotes and orders. Combined with retrieval-augmented generation (RAG), LLMs can accurately extract and transform the information embedded in chat; prices, quotes, orders, confirmations and more.
Turning conversational data into actionable insight is powerful but must be governed with the same discipline applied to traditional data and compliance. The chat-to-data pipeline is complex: real-time capture, extraction, mapping, semantic layers, embeddings, metadata, lineage, integration into trading and booking systems, plus audit, logging and permissioning. Firms must consider far more than which model to use. Unsurprisingly, according to a recent MIT study, 95% of generative-AI pilots fail when built internally.
As regulation evolves, humans-in-the-loop will remain essential, providing judgement and oversight. In chat workflows, this can be integrated directly into processes, improving efficiency while maintaining control.
Handled responsibly, conversational intelligence enables salespeople and traders to streamline workflows, capture real-time signals on liquidity and interest, and detect patterns emerging across thousands of interactions. As firms invest in analysing this new dataset, markets will shift from reactive pricing to more predictive, context-aware decision making.
Early adopters across sell-side, buy-side and inter-dealer brokers are already digitising pre-trade workflows, booking trades directly from chat, summarising sentiment and generating analytics that link conversations to market behaviour.
Looking forward to 2026, use cases will expand rapidly: agentic workflows, sentiment-shift detection, early-warning signals, predictive liquidity awareness, self-service bots and a blurring of the lines between human judgement and AI-assisted insight. These developments will improve responsiveness and risk alignment across clients, sales and traders.
Market structure will also evolve as more markets become digitised and, later, electronified, enabling earlier discovery of trading intent and improving liquidity formation.
In a world where structured data is abundant, the next competitive advantage will come from understanding the conversations that precede every trade. Those who harness conversational intelligence will gain an edge, not through reacting fastest, but through understanding earliest.