Artificial Intelligence (AI) should not replace but enhance investment decision making

Artificial intelligence (AI) biases and bad sampling stand out as core risks for data-driven fund managers operating at the intersection of data and emerging technologies.

The challenge when embracing AI and machine learning (ML) is not a lack of data, but the frequent misinterpretation of it. This is the view of Matis Mrazik, Investment Manager Systematic Equities at London-based Jupiter Asset Management, and Portfolio Manager of the Old Mutual Global Equity Fund. The fund won the 2025 INN8 Invest Diamond Award in the category “Best Global Equity General”.

“At Jupiter Asset Management, we are building on our long-standing leadership in systematic investing by exploring how AI and machine learning can enhance our already robust research framework. This means resisting the temptation to adopt untested algorithms and automations in favour of data interpretation, model design, and statistical learning – all areas in which we excel. Statistical learning means using emerging technologies and applying data-driven techniques, testing hypotheses, and refining models under real-world constraints,” Mrazik says.

AI has become a catchphrase to signal innovation and future readiness. However, simply tagging a fund as ‘AI-powered’ and its decision-making as ‘autonomous’ does not guarantee enhanced performance. Similarly, using AI to process reams of unstructured data does not necessarily yield better forecasting in dynamic, noisy financial markets.

“Our preference remains for statistical learning and the continued, natural evolution of our research process, already 20 years in the making,” Mrazik adds.

Much of the work the Jupiter Systematic Equities team engages in is technical, occurring behind the scenes. One example is Natural Language Processing (NLP), a form of AI that enables algorithmic models to interpret natural language. Jupiter uses NLP to handle large volumes of unstructured data, processing analyst reports and company commentary, and converting that qualitative text into structured inputs to complement traditional datasets. This allows the team to consider investor behaviour alongside underlying macroeconomic factors in its signal weightings.

To build on its advantage, Jupiter collaborates with academics in fields such as AI, behavioural finance, economics, econometrics, and econophysics. Results from these collaborations are only incorporated into Jupiter’s core model framework where they prove to add alpha or risk adjusted investment returns.

“Machine learning cannot be seen as delegating decision-making to black boxes or trusting back-tests run on historical data. It is far more complex, involving the framing of economically sound hypotheses, validating patterns on data not used in model training, and testing these observations across markets and environments over time,” Mrazik explains.

In high-noise domains such as finance, collecting more data often translates to collecting more noise unless one understands the structure, and mechanisms that generated it.

Mrazik continues: “We certainly see ourselves as practitioners of machine learning, with the caveat of approaching the field with discipline, as any scientist would. This discipline applies to the data inputs we rely on and the subsequent modelling, in acknowledgement that market regimes change, structures evolve, and the environments our models are trained on may cease to exist.”

The Systematic Equities team at Jupiter continues to build on its foundation of rigorous science rather than AI sloganeering. “The focus remains on testing and refining the investment philosophy in pursuit of clarity and consistency. This means balancing technology and statistics with rigorous testing, and intellectual clarity,” concludes Mrazik.

Click here to read more...