Scientific Beta

While artificial intelligence (AID) and machine learning (ML) increasingly are used in investment management, many AI models operate as “black boxes”, assuming a fictional, frictionless world.

While artificial intelligence (AI) and machine learning (ML) are increasingly used in investment management, many AI models operate as "black boxes", assuming a fictional, frictionless world.

This issue exercised institutional investors at Scientific Beta's recent Days Europe flagship conference in Paris, with many asking how they could responsibly adopt AI techniques while maintaining transparency and robust investment processes.

In its presentation, Scientific Beta highlighted machine learning's potential to exploit expanded data sets, but emphasised that to turn AI into real returns, transaction cost control had to be embedded in the machine learning engine itself.

"Quant investing has always been criticised for working with a narrow information set," Research Director Felix Goltz said. "Past returns and accounting metrics are useful. But they hardly provide a complete picture of what drives a firm's risks and cash flows. Systematic strategies have long ignored hard-to-measure aspects like management quality, product fit, market power, technological change and industry dynamics."

Machine learning models, by contrast, can improve the quality of the signal and can, for example, help investors make more accurate and robust return and risk forecasts. This is achieved by incorporating a 'high-dimensionality' set of firm characteristics, while neural networks can capture non-linear relationships and interactions.

"With currently available AI tools, we can analyse massive amounts of alternative data, such as company filings, patents and job ads, to look beyond the usual fundamental ratios and past returns," Goltz said. "The hard part isn't the technology. It's doing it rigorously, with a clear economic rationale and robust implementation discipline. I am confident the industry will figure this out over the next few years," he added.

"AI does not change the rules of portfolio construction – it raises the bar. At BNP Paribas Asset Management, machine learning signals are subject to the same stringent validation as any factor: robustness over time, controlled risk exposures and disciplined turnover. The real breakthrough is not model complexity, but integrating AI into portfolios that remain explainable, trusted by clients and fully investable across market cycles," explained Robinson Rouchie, CIO for Systematic and Quantitative Investments at BNP Paribas Asset Management at the event.

Indeed, the conference heard that the industry is slowly becoming more comfortable in the use of machine learning. A survey of participants found risk and data management are currently seen as key roles for AI, with some potential seen in portfolio construction, trade execution, alpha generation and back-office efficiency.

But questions remain, such as the explainability of ML methods for governance and fiduciary oversight. Others wondered whether, as ML models become more powerful and the industry adopts them, the signals they identify will persist.

Looking further out, some asked if AI systems will operate largely autonomously, or whether human oversight will remain central to investment decision-making.

Responding to what was a lively conference discussion, Maxime Ricomes – Director of Allocations and Selection at Mercer – praised Scientific Beta's work "in shedding light on the "undeniable trend" of integrating AI into asset management."

That process is already underway, though tentatively. A recent study by French market regulator AMF showed that about 90% of French market participants who responded to its recent survey said they already used AI or planned to do so within the next year.¹

So far, in France at least, market participants are mainly using AI for internal purposes, such as enhancing productivity, improving data extraction, analysis or monitoring, summarising information or generating content.

However, Scientific Beta's Research Director Felix Goltz said these uses inevitably would increase as AI tools provided the means for managers to incorporate much wider and deeper data sets into investment approaches in such a way that makes systematic investment the default.

"AI will not just lead to an evolution of quant investing," he said. "Instead, it will fundamentally reshape systematic investing and massively expand its adoption."

In an industry long worried about black box thinking, Scientific Beta is helping participants to find a way of thinking outside those former confines.


This article was written as part of the Scientific Beta Spring 2026 Spotlight Edition. Access all articles here.