Every trading firm we meet wants to transform their trading business with AI. Some firms streamline their organizations, using automated tools to enrich their human workflows. Others use the latest predictive tools to improve trading profitability. And many firms are curious about how they can utilize AI, but haven’t dipped a toe in the water yet.
Note: There are common terms used in this space: Artificial intelligence (AI) and Machine Learning (ML). AI in trading typically refers to replacing or augmenting human workflows, while ML refers to the mathematical modeling techniques on large data sets. In business conversations, people use AI to refer to all of these concepts, so we’ll do the same here, using AI for that larger, generic context.
Successes
We see a number of ways to succeed, whether it’s a focus on increasing profitability, human efficiency, or enabling non-technical employees with data insights.
1. Development
Using agents to speed up development is in the news, for good reason. These tools are uncannily amazing, especially for quickly prototyping new ideas. That said, as our CTO points out: building something is easy – maintaining systems is the hard part. Agents are like well-caffeinated, eager interns writing a lot of code that can result in tech debt. For new projects, teams can quickly generate ideas for trading strategies,
2. Predictions: Medium-Term
Many hedge funds hold risk on the order of days-weeks. AI enables complex modeling techniques on large data sets, including cross-asset market data, economic events, news, and more. That complexity can create models that outperform normal linear techniques, but there are risks, as outlined below.
3. Predictions: Short-Term
High-frequency trading operations use AI techniques on rich order book data to make predictions on the order of seconds. These predictions feed execution algorithms and tweak quoted prices for market makers. The edge for these signals may be less than bid-ask spread, but even a 0.1 pip improvement in an FX prediction can have an impressive impact on P&L, when coupled with low-latency execution and passive market-making.
4. Relationship Management
In FX, we’re all familiar with the role of relationship managers. Those roles may be a dedicated team at an ECN who determines the participants in liquidity pools or sell-side traders that tweak the streams shown to different client types. Teams can spend hours a week monitoring client market impact curves or volume dips to adjust spreads. With automated processes, those teams can get the same decisions in minutes.
5. Research
Buy-side firms’ research groups need to find nuggets of key data points in large data sets. For example, in corporate bonds, teams extract structured data from 10-K and 10-Q, or a macro firm might determine significant items from various news sources. Using AI tools to extract detailed data frees up time from research teams.
6. Summary Trends
AI tools are great at summarizing data. In our daily lives, search engines provide pithy sentences aggregating text across many websites. Likewise, using these tools on order data can give great insights. However, you must think through data privacy!
Risks
1. Data Privacy
For all of finance, data privacy is an existential topic. A mistake in mishandling client data clearly bears reputational and regulatory risks. As a result, trading firms cannot simply use ChatGPT or Claude to summarize trade data.
Ideal creates private Model Context Protocol (MCP) servers to enable analysis on privately held data. This architecture gives the benefits to the users without uploading sensitive data to Microsoft, OpenAI, etc.
2. Model degradation in regime changes
The downside is that if the data fundamentally changes, which is common in financial markets, the complex model can break without the users understanding the root cause. Traders who experienced market disruptions in 1998, 2001, 2008, and 2020 understand all too well that market shifts can be sudden and dramatic.
In statistical terms, financial data is clearly not independent, and identically distributed (IID). Firms need to measure and manage the trade off of predictive accuracy vs model complexity.
Get started
Each trading business has their own competitive advantage in the market, which contributes to the richness and efficiency of financial markets. A credit fund with long-term positions will get more business impact from improving their research process vs execution. Once a team finds the starting point in AI, they can build solutions in-house or by partnering with external firms.
1. Benchmark
When using complex modeling techniques, benchmark the results to simple models and assess if the outperformance is worth the increased model risk.
2. Healthy skepticism
AI does not solve every problem. Large advertising budgets can make the latest tool sound exciting. A small amount of upfront due diligence can save hours of time.
3. Learn
Here are some resources our team found useful:
www.deeplearning.ai
https://openlearning.mit.edu/news/explore-world-artificial-intelligence-online-courses-mit
4. Practical start
Small wins build confidence for the organization to keep investing in efficiency and performance. These projects also educate teams about pitfalls.
When Ideal started using Claude and Cursor to accelerate code development, we also noticed that these models can generate excessively verbose code, so we adjusted when we use these tools and increased our code reviews of their output.
Success
Each business is different, so there’s no single “right” path to incorporate AI tools. You know your clients and business risks, so have confidence in your own journey. After some initial success adopting AI, your organization will expand its ambition to achieve increasingly grander goals.
