Authored by Ian Stirling, CEO, Streets Consulting
At TradeTech in Amsterdam, one thing became very clear: AI on the trading desk has moved well beyond the pilot phase. Under the Chatham House Rule, senior industry leaders skipped past the hype and straight into discussing where AI is already being embedded into real workflows, from pre-trade research and execution support through to post-trade analytics and governance.
That matters because trading desks have always been information-processing businesses. The difference now is the sheer volume and messiness of the information being consumed, from voice notes and chat messages through to broker commentary, market data, positioning, risk limits and compliance obligations. AI is increasingly becoming the layer helping firms process that complexity faster and more intelligently across the trading lifecycle.
Pre-Trade: Structuring the Unstructured
One thing that particularly resonated with me was how closely this mirrors what parts of the digital asset and DeFi world have already been experimenting with for years. Crypto markets have long looked beyond pure pricing data, drawing signals from social sentiment, influencer activity, community momentum and behavioural patterns across platforms like X and Reddit. Some of that has admittedly been noisy, and occasionally outright chaotic, but the underlying premise was directionally right: markets move on behaviour as much as numbers.
Traditional finance is now building a far more institutional-grade version of that same concept, combining alternative intelligence with cleaner datasets, stronger controls and clearer accountability.
That is especially evident in the pre-trade environment, where desks remain flooded with unstructured information. Voice transcripts, Bloomberg chats, analyst commentary and internal discussions all contain valuable signals but have historically been difficult to process at scale.
AI is increasingly being used to pull structure and intent from that noise, pairing communications with market data, positioning, historical trading behaviour and counterparty context to create more actionable decision support for traders and portfolio managers.
Another major shift discussed was the rise of natural-language querying across institutional datasets. Rather than navigating multiple systems and dashboards, traders can increasingly interrogate data conversationally, dramatically reducing the time needed to surface insights.
Credit markets were repeatedly highlighted as an area where this could become genuinely transformative. Compared with equities, credit trading has historically remained more relationship-driven and operationally fragmented. Structuring messy datasets and surfacing intelligence faster has the potential to make credit workflows materially more scalable and systematic than many legacy processes allow today.
Execution: The Smart Trading Assistant
The execution layer was where the discussions became particularly practical. This was not about replacing traders, but reducing friction around decision-making in fast-moving environments where liquidity, timing and risk tolerance can shift quickly.
AI is increasingly being positioned as a real-time execution assistant sitting alongside the trader. By ingesting order data, market conditions, risk parameters, algorithmic options and historical trading behaviour simultaneously, systems can help desks make faster and more informed execution decisions under pressure.
In practical terms, that may mean flagging orders at risk of missing completion targets, identifying trades requiring additional compliance scrutiny or recommending alternative execution strategies as conditions change.
Transaction Cost Analysis (TCA) also featured heavily in discussions. Rather than reviewing performance after the fact, AI-driven TCA can continuously evaluate historical outcomes, liquidity conditions, broker behaviour and market impact to help guide broker selection, algorithmic strategy and order placement dynamically.
Importantly, most firms are still keeping humans firmly in the loop. Fully autonomous routing may come over time, but the immediate value today is augmenting trader judgement rather than replacing it. Traders remain accountable for the decision-making process, while AI increasingly acts as the analytical layer helping surface the best course of action faster than traditional workflows realistically allow.
I would think that continued emphasis on human oversight will likely be welcomed not just by traders, but by the allocators, risk teams and stakeholders ultimately accountable for where capital is deployed.
At-Trade: Real-Time Supervision and Delegation
Speaking of accountability, supervision was another area where the conversations became far more practical than theoretical.
As orders actively hit the market, AI is increasingly acting as a real-time monitoring layer across execution activity. Systems can instantly flag deteriorating liquidity conditions, unusual execution behaviour or trades drifting outside acceptable tolerances, allowing traders to react far earlier than traditional workflows would typically allow.
At the desk level, this also subtly shifts the role of the trader from pure operator towards supervisor and decision-maker. Traders can increasingly delegate repetitive and monitoring-heavy tasks such as meeting summaries, workflow recaps, routine information gathering and operational checks.
None of these are especially headline-grabbing use cases, but collectively they free up cognitive bandwidth for the higher-value judgement calls where, from everything discussed in Amsterdam, human experience and market intuition still matter most.
Post-trade: Analytics and the Feedback Loop
Post-trade may be one of the least glamorous parts of the trading lifecycle, but it is also where AI can deliver some of the fastest operational value. Several firms discussed how AI is already being used to rapidly process broker TCA reports and post-trade execution data, acting as a first-pass reviewer that can quickly identify anomalies, execution outliers or areas requiring further scrutiny across venues, brokers or algorithms.
The operational benefits are equally significant. Reconciliation, record-keeping and exception management have historically consumed huge amounts of manual effort across operations teams. AI is increasingly helping accelerate those workflows by matching records faster, identifying breaks earlier and reducing operational friction.
Perhaps more importantly, post-trade data is increasingly feeding directly back into future pre-trade analytics, execution strategies and routing decisions. Over time, the system starts becoming continuously self-improving, with each trading cycle refining the next.
Research and Analytics: Democratising Quant Capabilities
In the research realm, AI acts as a powerful institutional copilot (no pun intended). Several leading firms have developed firm-wide quant bots that utilise a unified API to seamlessly access analytics libraries, market data, and risk systems that allow users to ask complex questions in plain English rather than relying solely on specialist quantitative teams.
Tasks that may previously have taken weeks of manual analysis can increasingly be completed in hours, while internal AI agents are also making vast archives of documentation, policies and compliance manuals instantly searchable across organisations.
Governance: The Bedrock of AI Adoption
None of this, however, scales safely without governance. Firms are putting increasingly strict guardrails around AI deployment, ensuring legal and fiduciary accountability remains firmly with human operators. Many are now adopting dedicated governance frameworks defining exactly what AI systems are permitted to do, what data they can access and how outputs are monitored.
Ultimately, integrating AI into trading workflows feels less like a technology upgrade and more like a cultural transformation. The firms likely to benefit most will not necessarily be those chasing full automation, but those learning how to combine human judgement with AI-driven speed, scale and analytical capability most effectively.
