AI in Trading: From Buzz to Practical Workflow at FMLS 2025
At the Finance Magnates London Summit 2025, a panel titled "Secret Agent Deploying AI for Traders at Scale 1" cut through the noise. The message was clear: AI is useful, but only where it improves workflows, data digestion, and decision quality-with humans staying firmly in control.
Moderated by Joe Craven of TipRanks, the discussion brought together David Dyke (CMC Markets), Guy Hopkins (FairXchange), Ihar Marozau (Capital.com), and Rebecca Healey (MindfulMarkets.AI). Their collective take: apply AI where it adds clarity and speed, not ambiguity or risk.
AI for Retail and the "Headless EMS" Idea
Craven framed AI as a way to make complex market data usable for retail investors. That's already happening through language tech and machine learning baked into platforms.
Hopkins called AI "a solution looking for a problem," noting the hype around large language models (LLMs) despite long-standing quant methods. Healey pointed to a shift toward a "headless EMS," where access to execution and insight is modular, personal, and stripped of clutter so users see exactly what they need.
Compliance First: Auditability, Explainability, Accountability
Marozau was blunt: "AI output must be auditable and explainable." Dyke warned against the illusion of progress when models produce answers you can't justify to a regulator or a client. If you can't explain it, you can't use it in production.
Hopkins flagged a common risk-over-trusting systems that feel smart. Healey pressed the basics: get clean, accurate data and be precise about AI types. Institutional teams often spend months standardizing broker data before training anything, and AI has surfaced many old data issues by making them more visible.
How Firms Are Actually Deploying AI
LLMs aren't doing the math. Hopkins explained they're used to parse intent and route tasks to deterministic agents that produce consistent results. That's how you avoid unpredictability in pricing or execution logic.
Dyke shared a similar model at CMC Markets: AI monitors customer communications to triage and assist, while humans supervise outcomes. The pattern is becoming common-LLM for understanding, deterministic systems for action, with controls wrapped around the whole stack.
Workforce Impact: Faster Onboarding, New Gaps to Watch
Dyke sees AI lowering friction for new hires by making knowledge accessible on day one. That's good for speed and support.
Hopkins cautioned that automating junior tasks removes key learning reps for future seniors. Healey countered that the scope of learning is changing; AI lets analysts build experience across multiple asset classes faster than any manual path. Marozau added that as knowledge becomes commodity, advantage shifts to conceptual thinking and flexible user interfaces.
Practical Concerns: Maturity, Mistakes, and Overconfidence
Audience questions circled model maturity and error risk. Panelists pointed to VIP coding tools and controlled experimentation frameworks that let teams test safely and iterate with guardrails.
Healey noted smaller firms and retail traders can experiment more freely versus heavily regulated desks. Across the board, personalization and efficiency came up as key gains-faster data digestion, better optionality, and judgment-free learning environments.
Market Effects: Speed Can Amplify Momentum
Faster workflows can push volatility and momentum. The panel agreed: keep humans in the loop. Use AI to assist, not replace, oversight.
Healey cited the shift from T+2 to T+1 settlement as an example of how infrastructure changes can alter engagement models, not just speed them up. For context, see the SEC's rule on shortening the settlement cycle to T+1: SEC T+1 announcement.
A Simple Playbook for Finance Teams
- Start with use cases that reduce time-to-insight: research summarization, broker data standardization, client comms triage.
- Separate LLM tasks (intent, summarization, explanation) from deterministic tasks (risk, pricing, execution).
- Enforce explainability: log prompts, model versions, data sources, and decisions for audit.
- Human-in-the-loop by default for anything that impacts capital, clients, or compliance.
- Stand up a sandbox with clear red/green lines and pre-approved datasets; promote fast, safe experiments.
- Invest in data quality and lineage. Bad inputs make every model worse.
- Upskill your team on AI risk and controls. The NIST AI Risk Management Framework is a practical reference.
End-User Benefits Without the Hype
Personalization came up again and again-tools that fit how each trader or analyst works. Healey framed it as moving from reaction to response: better optionality across assets and regions, fewer missed nuances.
Hopkins stressed faster assimilation of complex data. Dyke highlighted AI's role as a judgment-free coach-useful for new joiners and seasoned pros testing ideas.
Bottom Line
AI moves the needle when it clarifies decisions, automates repeatable workflows, and keeps a clean audit trail. It stalls when teams over-trust black boxes, skip data work, or cut humans out of the loop.
As Hopkins put it, getting AI to production "faces significant headwinds." Progress comes from pairing machine intelligence with human judgment-precision over hype, governance over guesswork.
Resources
- Curated tools for finance teams exploring AI: AI Tools for Finance
Your membership also unlocks:
