AI Meets
Value Investing

No, because the design possibilities are limitless, shaping every step of the process.

Consider investment style and philosophy. At ACATIS, we gravitate toward value companies, while other investors chase high-growth opportunities. Some intentionally target small companies or specific regions, while others aim for short-term trades or long-term investments. Some prefer equities, others bonds, or a blend of both in balanced funds. These diverse approaches, with varying goals, lead to a vast array of distinctions. Then there’s the ocean of data sources—ranging from key balance sheet figures to satellite images, patent data, social media, and economic indicators. Finally, AI itself offers countless architectural possibilities, ensuring no two approaches yield the same result. In the world of AI-driven asset management, we take a niche path:

Our strategy is rooted in the principle of value: focusing on solid business fundamentals, a long-term holding period, and a long-only stance.

Explainability is still a puzzle, and while researchers are tirelessly working on it, we need to shift our expectations. If the decision isn’t being made by a human, why should we expect a human-like explanation? Take image recognition, for instance—methods like SHAP values break down how much each pixel influences a prediction. It’s not a narrative, but rather a precise number tied to each pixel. Yet, the insight holds value. Now, apply this to stock picking: the question becomes, which key metric does the model use to shape its prediction? In the end, you won’t get a neat story as a human would offer, but a clear analysis of the factors guiding the decision. It’s not about crafting a tale—it’s about understanding the building blocks that drive the choices.

This isn’t the dawn of AI taking over; the above-average fund manager won’t be out of a job because AI isn’t a silver bullet. In many areas, the true power lies in the partnership between humans and AI. AI isn’t a universal solution. In emerging markets, for instance, the data often falls short—it’s incomplete, inconsistent, or shaped by varying standards that make comparison a maze. That’s why we trust our AI only in developed, tightly regulated markets. Also, for companies fresh out of the gate, there’s usually too little data to forecast their growth, or they don’t even have a product yet, rendering any fundamental figures meaningless. Our AI won’t touch companies with less than three years of history. And when it comes to the financial sector, where key figures like sales and debt carry a different meaning, we step back entirely.

In these scenarios, AI is not the answer; it’s the human portfolio manager who steps in. The one who can navigate the unknowns and gauge the potential of something like an IPO—an expert who can assess where the future might lead.

Domain expertise is the compass guiding an AI developer. To teach a machine to understand and analyze companies, the developer must have a keen sense of company valuation. The key point is that during the model’s creation, the human developer is making hundreds of subtle choices—each one shaping the model’s direction.

That’s why, as developers, we step into the shoes of active investors. We visit companies, seek out the insights AI can’t yet access—things like touring production facilities, speaking with management, experiencing a product firsthand, and grasping the long-term vision and business model. We dive into the human side of the equation to uncover what the model can’t see. By understanding the gaps in our approach, we can build the right workarounds and make the model more robust.

Ultimately, humans are the architects of the model’s foundation. But once the model is complete and starts suggesting companies, the human hand can be removed from the decision-making process, allowing the machine to navigate on its own.