Tech • IA • Crypto
Trading Signals That Trade Themselves
8/10TL;DR
Man Group has deployed AI-driven systematic trading signals using governed internal workflows, highlighting that strong data and process governance is key to scaling AI in finance.
KEY POINTS
AI enters high-stakes asset management
Man Group, which manages over $200 billion for pension funds and sovereign investors, is integrating AI into investment processes where errors directly affect real-world savings. The firm emphasizes that deploying AI in this context requires extreme reliability, as mistakes could translate into tangible financial losses for individuals and institutions.
Systematic trading as a prime AI use case
A major focus is systematic trading, where algorithms scan thousands of securities across global markets to generate investment signals. These signals rank assets for buying or shorting, similar to selecting players in a fantasy sports lineup based on performance expectations and timing.
Backtesting remains central to validation
Investment strategies are evaluated through historical backtesting, often spanning 15 years or more. Key metrics include annualized returns, drawdowns, and Sharpe ratios, which measure risk-adjusted performance. Despite advanced modeling, uncertainty remains inherent, as past performance cannot guarantee future outcomes.
End-to-end AI-generated trading signals
The firm has deployed signals in production where AI handled the full pipeline: idea generation, data sourcing, backtesting, proposal writing, and implementation. Human oversight remains in place, but AI operates as the central engine in the workflow.
The hidden complexity beneath signals
While signal creation appears straightforward, most effort lies in underlying processes such as data cleaning, infrastructure management, and backtesting consistency. Discrepancies across teams can produce conflicting results, making standardized workflows critical.
Early failures in AI adoption
Initial rollout efforts prioritized rapid adoption, with employees building custom tools or “skills.” However, many were created by power users rather than process owners, leading to fragmented solutions optimized for individuals rather than the organization.
A cautionary example of poor governance
One internally shared automation tool for expense reporting unintentionally routed approvals to the wrong department due to a hardcoded cost center. The incident illustrated how unchecked AI tools can create operational inefficiencies when not governed properly.
Shift to centralized skill governance
To address fragmentation, the firm introduced a centralized skill marketplace where tools are owned, reviewed, tested, and lifecycle-managed. This ensures consistency, transparency, and reuse across departments, transforming ad hoc tools into enterprise-grade infrastructure.
AI powered by institutional context
Rather than retraining models, the firm focused on giving AI access to proprietary data, workflows, and institutional knowledge. This “context layer” enables AI systems to operate effectively within the firm’s unique environment.
Demonstrated signal using alternative data
In one example, AI analyzed credit card transaction data to predict stock performance, identifying correlations between consumer spending and equity returns. A backtest suggested that a $1,000 investment in 2021 could grow to about $2,500, outperforming a basic buy-and-hold approach, though broader validation was required.
Scaling usage across the organization
The firm reports around 750 active users of its AI coding platform across departments, supported by over 100 governed skills. Adoption extends beyond technical teams to functions such as finance and human resources.
Foundation for autonomous agent systems
With governed workflows in place, the firm is preparing for agent-based AI systems capable of autonomously exploring investment opportunities. These systems rely on standardized, trusted building blocks to operate at scale.
CONCLUSION
Man Group’s experience shows that the real challenge in enterprise AI is not model capability but organizing and governing internal knowledge, which ultimately determines whether AI can deliver consistent and scalable results in complex domains like finance.