At FILS Amsterdam, AIs and LLMs were in most panels but two different AI mindsets were on display: a subset of buyside portfolio managers want control and guardrails, while vendors focus on measurable accuracy and scale focusing on what they can build with the tech.
Orlando Gémes, CIO at Fourier Asset Management, said the biggest benefits of AI for PMs comes from putting senior investors, not the latest model, in the driver’s seat.His team records weekly investment meetings, and uses transcripts to surface bias amongst the team. It then uses these inputs to send curated and relevant meetings insights to the relevant PMS.
“Model choice matters far less than the user. The biggest edge comes from getting the right PM behind the wheel, not swapping between foundation models,” Gémes said. He insisted on caution and guardrails around his new assistant.
He said: “We treat AI like a new intern—never client-facing, never anything with price impact—and we always check its work.”
That caution is very contrasted with the enthusiastic product-builder stance from Ivan Mihov of Boltzbit.
He described systems designed to extract pricing signal from messy market data. Boltzbit claims it can prove numerically the superiority of its AI powered pricing. <
Mihov said: “We collect billions of data points a day from unstructured chats and attachments and turn them into price-discovery data.”
On top of that data layer, Boltzbit trains size-aware pricing predictive models, think bid/mid/offer for a specified trade size, and also delivers confidence metrics to desks. Many venues / platforms offer predictive pricing, sometimes augmented with AI or machine learning techniques.
Read more: Bond traders predict data science/AI use explosion in 2026
But Mihov specified that: “Purpose-built models with clear KPIs beat generic GenAI hype, always ask what data it’s trained on and how accuracy is measured,” he added.
In practical terms, for portfolio construction, some PMs like Gémes are using the tool to curate the data decision-makers will have to sieve through in their investment processes and also use LLMs to question the narratives and factors associated with their investment thesis. For Fourier, the aim isn’t to outsource trading: investment decision but to better it.
For fixed income execution, Transaction cost analysis, and pricing, vendors are industrialising the data plumbing and data science, AI layers looking to better predict pricing at scale, while scoring the new tools against specific KPIs.
We take away that at the very least, the new AI tools are at minima like ever eager interns, helpful but whose work need to be carefully checked.
Where will the actual human interns start in trading?
©Markets Media Europe 2025
