ipushpull

When will AI agents execute trades in fixed income? ‘The desire is there, but there are significant barriers to that.’ That’s David Rickard, Global Head of Fixed Income at LSEG, on the latest Capital Markets FinTech Forum podcast. He thinks adoption will come faster than we expect, but the barriers are real: models still struggle with real-time data, handing execution to a model faces heavy scrutiny and trading systems on many desks are ‘deeply entrenched’. For now, his view is grounded. Most firms are experimenting with the tools, proving value and embedding them into real workflows. Agentic, multi-step automation is still the frontier. I also liked his explanation of why chat and voice persist in fixed income. How electronic a market becomes depends on two things: how standardised the instrument is and how liquid it is. Where both are low, relationships matter. The question becomes ‘who’s the most likely to show you a price in times of stress’. That makes the unstructured data in chat a huge opportunity, but only if desks can trust where the data feeding their models comes from. Three takeaways: 1. AI on the desk is moving from experiments into real workflows. Execution will come later. 2. Chat persists where liquidity and standardisation are low, and that won’t change quickly. 3. Data provenance is becoming the gating factor for AI, and compliance teams are being asked to move faster than ever. David is hosting a buy side only workshop on 'designing fixed income workflows for the AI era' at FILS on Wed 14th at 1605 CEST. If you’re going, see you there. Full episode: https://lnkd.in/efV8eVVb #FILSEU2026

If chat becomes machine-readable, what stops everyone pouring even more into it? I put that question to Matthew C., Head of Fixed Income Sales and Product EMEA at Susquehanna International Group and EMEA co-chair of the FIX Trading Community. The general opinion he hears is that theres probably information overload and traders search their chat rooms by keyword or ISIN just to find what matters. The technology to change that has only really arrived in the last 12 to 18 months. AI and advancements in voice-to-text make it possible to extract information from chat that desks couldn't do before. But the same shift could make the noise worse... Matthew's answer "I would like to think as an industry that actually we could evolve to be smarter than that". For Matthew, two things make the difference: 1. His team at Susquehanna asks what a client will actually find useful, rather than sending a 'me too' message. Fill a chat room with too much colour and a clients order can get lost in it - nobody wants to miss a trade. 2. Context. Chat is one source among many channels. The value of comes from pulling chat, voice and electronic flow together, and structuring the unstructured information so it can be analysed alongside everything else. He also pointed to the FIX AI Working Group, whose first meeting agreed to start with data standards, otherwise AI will be useless. My takeaway was that machine-readable chat only helps if the people writing it stay selective, and the data lands somewhere it can be used in context. Matthew Coupe is on the 'View on Liquidity Panel' at FILS on Wed 14th 0910 CEST. Full interview https://lnkd.in/exmwxnuy Ill be there all week, so if you're going, lets hook up.. #FILSEU26

Human in the loop is still critical for every firm trading on an institutional basis. I spoke with Gareth Coltman, head of products for EMEA and APAC at MarketAxess, on the Capital Markets Fintech Forum podcast ahead of FILS Europe next week . Gareth describes traders moving from the dealers they usually call to whoever is most likely to price the risk right now. He sees AI and machine learning as the way to answer that. On data, the hard part is cleaning it, aggregating it and getting the result into the traders workflow in a form they can use. On AI adoption, Gareth sees agents being used to simulate the likely outcome of a trade. When it comes to pressing the button, he says human in the loop is still critical for everyone trading on an institutional basis. We build ipushpull around four pillars: chat, data, AI and governance. All four run through this interview. 1. Chat is where counterparty relationships have traditionally lived, so much of the trading history behind these models starts there. 2. Data has to be clean before a model can use it. 3. AI earns its place when the output lands inside the workflow. 4. Governance follows from Gareth's last point: if the human stays in the loop, firms need a record of what the agent suggested and what the trader decided. Full episode https://lnkd.in/e4nZu37W (Yes, that's scaffolding outside Gareth's window. A fitting backdrop for someone who spends his days building market structure.) I'm at FILS Europe in Juan-les-Pins from 13 to 15 October. #FILSEU26

Venezuela has the world's largest oil reserves but only produces 1% of the world's oil. Having the resource and being able to use it are completely different problems... Front office chat data has had exactly the same problem. In a recent interview with David Greely on the SmarterMarkets™ podcast we spoke about how valuable information is left on the table. Valuable information in a trader/sales/broker chat comprises of: a. Quantitative data - Axes, prices, RFQs, confirms b. Qualitative data - News, rumour, commentary, research c. Conversations - Chit chat amongst people in a room Billions of messages are sent every day across capital, commodities and crypto markets but almost no one is extracting any of the data, insights and signal from their own unstructured conversational data. Every firm is sitting on those enormous 'reserves' but for decades there wasnt a feasible way to extract anything out of them. The arrival of LLMs has meant a core piece of the refinery is available however a refinery on its own is not the whole process. You need all of the parts of the extraction and refinery process to create something useable, we frame it as the four pillars: 1. Chat - the interface that's already on traders desktops with no adoption battle to fight 2 Data - crude conversation extracted, cleaned and refined into structured data that other systems can consume 3. AI - The crucial tool that unlocks this data 4. Governance - permissioning, access control, identity, audit, logging Governance is the one people underrate, and it is precisely why some large firms have to move more slowly than smaller ones. It shouldn't be seen as a blocker or something slower you down. Instead the better governance in place the faster you can move with AI - think of an F1 car, the better the safety and control systems, the faster you're allowed to drive. Which pillar is your firm weakest on right now? Drop me a line if you want to discuss. Full episode: https://lnkd.in/d_wR_4KK Four Pillars of the Agent Era exec report: https://lnkd.in/ehHWtpX9

Most people never think about where their energy or food comes from, let alone how it gets traded. For those working in the markets, they know it doesn't just happen. Every trade relies on a complex network of systems, people and touch points. What you've seen us demonstrating for US Natural Gas isn't gas specific. It's the core capability of ipushpull's Chat Data Mining platform: capturing conversational data, understanding what's being agreed, and transforming it into structured, actionable records. Whether it's Oil, Power, Agriculture, Metals or Financials, the asset may change - but the gaps between the conversation and a booked trade has remained largely manual. Until now. https://lnkd.in/e3cwCYkv https://lnkd.in/diYjbkDW

No matter the size of your business, too much time is still being invested in manual tasks that no longer need to exist. OTC & Physical trade capture is just one of them. "We’ve seen strong results, with 25 traders booking over 50,000 trades on the platform. This has reduced trade input time by 85%, lowered errors, and improved daily operations.’ That's Darnell Bortz, Director of Natural Gas Trading at Koch Energy Services, describing what changed once trade capture & booking went live on their desk. It's a no-code setup, which is why a proof of concept can be running within days rather than months, connected to the desk's existing chat and ETRM without a development project behind it. Learn more here: https://lnkd.in/diYjbkDW