This insight was originally published as a guest article in TabbFORUM. Click here to view.
As markets evolve at an unprecedented pace, investment firms are increasingly turning to AI and machine learning not as novel tools, but as foundational elements of their operating model, writes Sandra Mostacci, Global Head of Delivery at Lab49. Increasingly, AI is not just augmenting traditional methods – it’s reshaping the very structure of investing, Ms. Mostacci explains.
Alpha in the Age of AI: New Signals, Old Mandates
Despite rapid technological innovation, the underlying mandate in investing remains unchanged: generate alpha, manage risk, and protect capital.
Shortened sentiment cycles, alternative data, and narrative-based alpha now dominate the landscape. Portfolio managers are increasingly searching for “narrative signatures” in real-time. These signals—such as geopolitical sentiment or CEO tone—can impact securities rapidly and briefly. Identifying, quantifying, and acting on them at the right moment is the new battleground for alpha.
Despite rapid technological innovation, the underlying mandate in investing remains unchanged: generate alpha, manage risk, and protect capital.
Generative AI, particularly large language models (LLMs), now supports this effort by synthesizing massive volumes of unstructured data. However, while AI models can scan entire universes like the Russell 1000 in minutes, human-in-the-loop oversight remains critical. Ultimately, hallucinations are easy to catch at the extremes – it’s the subtle ones that are dangerous.
While high-frequency trading (HFT) continues to reward speed, the industry is pivoting towards examining lower-frequency signals—developed over hours, days, or even weeks – and strategies that blend pattern recognition with reasoning and domain context.
This is where agentic AI comes in: autonomous systems trained to act, adapt, and learn in evolving environments. AI factories (clusters of specialized agents that collaborate in dynamic workflows) may soon simulate entire economies or financial systems – digital twins running complex “what-if” scenarios in real time.
Machine Learning That Matters: Explainable, Accurate, Resilient
That said, to really be effective, and most importantly, trusted by investors, machine learning must go beyond accuracy. It’s important that models can capture the right balance between speed, accuracy and resilience. Resilience in this context means the ability to cope with unexpected market shocks, as well as regulatory compliance. Most importantly, AI must also be explainable. Investment firms can’t afford to deploy black-box models that fail silently – quite rightly, their stakeholders and regulators wouldn’t permit it.
Most importantly, AI must also be explainable. Investment firms can’t afford to deploy black-box models that fail silently – quite rightly, their stakeholders and regulators wouldn’t permit it.
Models are already moving in the right direction. Bayesian estimation is one way, for example, to regularize predictions and push models toward known reference behaviors, helping stabilize model behavior in the face of uncertainty or limited data.
Data Quality: The Unshakeable Foundation
Ultimately, as sophisticated as AI models have become, they are only as good as the data that fuels them. And in capital markets, that data must be not only clean, but contextually correct, regulatory-compliant, and ready for scale.
Too often, firms try to jump to model innovation without solving their foundational data issues. The most effective organizations treat data quality as a product in its own right – governed, observable, and continuously improving.
The best way to achieve this is through a layered approach to data management including the following:
- Foundation Layer: Automated systems ensure that data follows the correct format, is complete, tracks its origin, and meets delivery standards
- Domain Intelligence Layer: Combines statistical models with human oversight to check for context and anomalies, especially for sensitive data like client orders, liquidity, or regulatory information.
- AI-Ready Layer: Structuring unstructured data and making sure it is time-aligned, versioned, and explainable for use in machine learning and AI applications.
Ultimately, as sophisticated as AI models have become, they are only as good as the data that fuels them.
Data quality is a critical enabler of safe, effective, and scalable AI. Without it, even the most elegant models collapse under real-world complexity.
The next generation of investment leaders
From risk modeling to alpha generation, from crypto markets to macroeconomic forecasting, the role of AI is growing. To be successful, however, adoption must be intentional. The firms that win will be those that build resilient, explainable machine learning models, while investing scalable, clean data infrastructure. They will balance agentic intelligence with domain-specific oversight, and embrace new market structures while anticipating their unintended consequences.
It’s too early to tell who will succeed in this new era, but the next dominant strategy may not come from a trading desk. It might emerge instead from an AI-native engineering pod, co-located with portfolio managers. Against the backdrop of large firms’ dominance, which have the capital to invest in people and machine power, a quiet revolution is underway. Smaller, specialized teams could find themselves well positioned to outperform their better-resourced peers – leveraging AI, synthetic data, and flexible infrastructure to move fast and smart.
The fusion of deep technical capability, financial domain expertise, and human creativity will define the next generation of leaders in capital markets – and it might not be the ones you expect.