FILS Europe is one of the few events where most of the room trades on the buy side, which makes it a good place to listen. I spent it moving between panels and in one-to-one conversations with heads of trading on both sides of the market. Before the event, I recorded podcasts with three FILS speakers: Gareth Coltman of MarketAxess, David Rickard of LSEG and Matthew Coupe of Susquehanna. The same themes came up on stage, around the halls and in those recordings.
For all the talk of automation, a lot of fixed income still gets done in chat. On the chat communications panel, Crédit Agricole CIB's Guillaume Carreno was direct about it: "Emails, chat, they're here to stay for sure." LGIM's Darren Moore said much of his desk's communication is verbal or in chat rooms, often before an order reaches the OMS (The DESK). Matthew Coupe put it more bluntly: "I can't remember the last time I took an order over the phone", in the Capital Markets Fintech Forum podcast.
David Rickard explained why chat persists. How electronic a market gets depends on how standardised and how liquid its instruments are. US municipal bonds have 55,000 issuers and over a million securities, and some don’t trade for two or three months at a time. Bonds like that trade through relationships, by voice and chat.
The change is in what desks can do with chat now.
Matthew described buy-side traders searching their chat rooms for an ISIN to see who has said what. ‘There’s a massive lake of information, and people are starting to use search tools to pinpoint what they want when they want it.’ In my own conversations, programmatic access to chat came up repeatedly. Some desks are building their own tools on top of it. Trade capture and booking STP from chat was the use case I heard most often.
That is the problem ipushpull works on. Quote Hub captures axes, runs, RFQs, orders and trades from all chat platforms, Excel, APIs and other services and structures them, so what arrives in chat can be used as data instead of being read once and scrolled past.
The sharpest frustration of the week was the volume of axes.
Allianz Global Investors' Michał Pondel said: "We don't rely on axes much anymore." His desk looks at hit ratios and how much of the street price is firm instead. BNP Paribas Asset Management's Benjamin Mahe said dealers who fake axes "will be removed from the RFQ", and MarketAxess's Aria Pattinson-Speirs agreed the market is flooded with them (The DESK).
It doesn't help that the buy side doesn't agree on what an axe is. A Susquehanna survey published the same week found only 10% of buy-side traders define an axe as a firm price, with no consensus at all on what makes an axe a voice axe (The DESK). If one dealer’s axe is a firm price and another’s is a loose indication of interest, a trader can’t compare them. Desks building tools to pull quotes out of chat run into the same problem, and so does AI. Matthew Coupe co-chairs the FIX Trading Community in EMEA, and he made the point about data generally: ‘You don’t want interpretation.’ When the FIX AI working group first met, it agreed to start with data standards, ‘because otherwise AI will be useless.’
Gareth Coltman described where dealer selection is heading. Desks used to ask who they usually call. Now they ask, ‘who is most likely to be able to price risk for me right now?’ In his words, ‘that’s a data question, not a relationship question.’ A desk that can see from its own history who responded, who won and whose prices held up needs fewer axes to make that call.
That is what the analytics capability in Quote Hub is built to show, ranking dealers by win rate, spread and response time. We cover it in more detail in our latest blog.
On the RFQ panel, Aurel Partners' Adam Ferencz said: "RFQ will be obsolete in the next few years in its current form." State Street Investment Management's Maile Robichaud pointed to pre-trade data and interoperability between systems as the next step. ODDO BHF AM's Sven Rudolf argued for one-click RFQ on platforms with reliable direct connectivity, so only the chosen counterparty sees the order, as long as the pricing can be relied on. Pattinson-Speirs summed up the mood: "Having a static decision-making process doesn't work." (The DESK)
Gareth described the alternative as a toolkit. It includes block protocols built around negotiation, portfolio trading for large baskets, and pre-trade TCA so a desk can choose the protocol, counterparty and timing before it goes to market.
Whichever version of the protocol wins, it needs pre-trade data that the desk trusts. In Quote Hub, a trader can run a what-if against RFQ history before an RFQ goes out and rank the dealers they are about to send to.
The cross-desk collaboration panel was about how traders and portfolio managers work together, and information leakage ran through all of it. Traders described joining PM huddles early in the decision, with a lot of dialogue happening before an order is in the OMS. Excel is still used around the OMS, and nobody wants overcrowded chat rooms. One trader on a hybrid desk put the priority simply: you can't spook the market. A poll of the room showed many firms now run central dealing desks.
Gareth’s view from the platform side differs: ‘clients’ fear of information leakage tends to be ahead of the reality.’ In liquid credit, he said, desks now go to the market in size and to a broad set of participants. They keep targeted protocols for large trades and illiquid bonds. Both views can be right. Which one applies depends on the trade, and only the desk’s own dealer data can show it.
Overcrowded rooms cost dealers too. Matthew Coupe keeps his team selective about what goes into client chat rooms because a trade can get lost in the noise. ‘Let’s face it, we never want to lose a trade.’
Knowing when to send an RFQ to a narrower, better-informed set of dealers means fewer counterparties see the position being worked. That is one of the most practical things better dealer data gives a desk.
Nobody I spoke to is handing execution to GenAI. BNP Paribas Asset Management's Inés de Trémiolles described a desk of 37 traders with no AI use cases in execution and only a few in pre-trade. S&P Global's Olivier Dermaux said of buy-side traders: "They want to keep a manual hold on the order." (The DESK) The Susquehanna survey found 72% of buy-side traders don't use AI to evaluate their data.
Gareth sees agents doing pre- and post-trade work, with a human pressing the button. In his words, the ‘human-in-the-loop is still critical to everybody on an institutional trading basis.’ David Rickard said most firms he works with are somewhere between proving AI’s value and embedding it into a workflow. Compliance and data teams are being asked to sign off on data sets faster than ever.
The one-to-one conversations I had added detail. In new issues, I was told automation is still below 90%, with a lot of human checking. Where errors are corrected, it is often developers fixing them, with traders' corrections not yet feeding back into the system. One platform's advice was not to let AI make any decision that can't be reversed. Several asset managers talked about long periods of data clean-up before new systems go live, and about needing visibility of all the data behind a decision. The survey backs this up: 39% of buy-side traders don't systematically measure slippage, and two-thirds of those cite lack of data.
That matches how we build. Structuring and enriching data is where AI earns its place today. Traders stay in control; they can correct what has been captured, and every change sits under Control Hub permissioning with a full audit trail.
One sceptic put it well to me: chat is only as good as its network, and it needs data and analytics on top to be useful. Matthew Coupe gave the simplest example of what goes wrong without them: ‘Who’s buying and who’s selling? If that’s not communicated clearly, it can flip very quickly.’
Chat is staying; the buy side wants dealer performance over axe volume, and people want to stay in charge of the order. The firms that get value from all of this will be the ones that turn what is said in chat into data they can trust and act on.
If that sounds like your desk, we'd be glad to talk. Discover more with ipushpull