The European Securities and Markets Authority (ESMA) has published comprehensive research examining how artificial intelligence is transforming EU investment funds, revealing significant insights for finance and wealth management professionals. This analysis and key findings, published in February 2025, provides crucial guidance on both the operational adoption of AI tools and the investment flows into AI-related companies. 

In this article, we break down the key findings of the research by ESMA as well as sharing Infront’s own findings relating to AI and details of how we are already using these technologies across our solutions.

1. Limited operational adoption

• AI use is still marginal: out of 44,000 EU investment funds, only 145 funds explicitly stated that they use AI or machine learning (ML) in their investment process.
• Minimal market share: these AI-promoting funds represent less than 0.1% of total assets under management (about €1 billion as of Q2 2024).
• Peak and plateau: the number of funds promoting AI use peaked in 2023 and has not grown since.

2. How AI is used

• Augmenting, not replacing: most asset managers use AI, especially generative AI and large language models (LLMs), to support human-driven investment decisions and enhance productivity in research, risk management, compliance and administrative tasks.
• Few fully AI-driven funds: only a minority of funds use AI as the primary driver of investment decisions. Most use AI to inform or augment rather than determine investment choices.

3. Performance and investor response

• No clear performance edge: funds that promote AI use have not delivered significantly higher or lower returns compared to traditional funds.
• Mixed investor flows: these funds have seen alternating periods of inflows and outflows, indicating no consistent investor preference for AI-branded strategies.

4. Industry structure and challenges

• Large firms lead: larger asset managers are more likely to experiment with and deploy AI, while smaller firms face barriers such as limited budgets and expertise.
• Third-party reliance: smaller firms often depend on external AI vendors, which introduces operational and concentration risks if many firms use the same providers.

5. Market trends

• Growing interest, slow adoption: while surveys show most asset managers view AI as transformative and plan to increase adoption, actual integration into investment processes remains rare.
• AI for productivity: the most common use cases are for data analysis, risk management and compliance rather than core investment decision-making.

ESMA research in summary

Looking at the findings of the research, it is clear AI adoption by EU fund managers is still at an early stage, with only a tiny fraction of funds explicitly using AI in their investment process.

Most use cases involve AI supporting human work, rather than being fully automated and the impact on fund performance is neutral so far. 

And while larger firms are leading the way, industry-wide adoption is hampered by cost, expertise and reliance on third-party providers.

Infront’s investments and research into AI and AGI 

Infront has done research into and implemented a diverse set of AI technologies within our product portfolio, global data catalogue and news.  

At the moment, AI or AGI is generally seen as the usage of large language models (LLMs) –OpenAI, Gemini, Llama, DeepSeek, Claude and others. Our research shows that although LLMs excel at learning from documents, text and data patterns due to their architecture and training methods, they face significant limitations when performing complex real-time calculations for financial markets. 

Here is why 

LLMs do have strengths in learning from documents and text. 

• Advanced contextual understanding: LLMs are trained on vast amounts of text, enabling them to understand context, jargon and nuanced expressions in financial documents, news and reports.
• Transfer learning flexibility: their broad pre-training allows them to be fine-tuned for specific tasks with relatively little additional data, making them adaptable to new topics and languages.
• Scalability and real-time analysis: LLMs can process large volumes of text rapidly, making them useful for sentiment analysis, summarising reports and providing real-time insights from news and market information.
• Explainability and customisation: they can generate human-like explanations and be tailored for specific financial instruments or market conditions.

But LLMs have limitations in complex real-time financial calculations. 

• Numeric and mathematical accuracy: LLMs are primarily trained on language and not (yet) on structured numerical data or complex mathematical operations. They often struggle with exact calculations, especially in multi-step financial problems and can produce errors or "hallucinate" incorrect results. Would you feel comfortable basing your short or long-term investments on potentially “hallucinated” numbers?
• Domain-specific expertise: general-purpose LLMs lack deep domain-specific knowledge. While finance-focused models like Bloomberg GPT are trained on financial data, they still may not perform well on highly specialised tasks such as regulatory compliance or complex asset class analysis without further customisation.
• Complex reasoning and context integration: financial market calculations often require integrating diverse data sources, understanding nuanced market conditions and applying judgment – skills where humans still outperform LLMs 
• Computational constraints: LLMs have fixed token limits and are not optimised for real-time, high-frequency numerical computations, which are critical in trading and risk management.

LLMs are highly effective at processing and understanding language, making them valuable for text-based tasks in finance such as sentiment analysis, summarisation and document review. However, their limitations in numeric accuracy, complex reasoning and real-time computation mean they are not yet strong enough to autonomously perform complex, real-time calculations required for financial markets. 

Specialised models and hybrid approaches that combine LLMs with numerical and machine learning techniques are often needed for these advanced financial tasks.

Infront's products and AI

In the latest releases of Infront Professional Terminal and Investment Manager web-based terminal, we are using natural language processing and Gemini’s LLM models to retrieve global financial news and summarise and translate this in real time. In addition, we are using advanced machine learning and quantitative algorithms into our market data catalogue and data streaming for trading and portfolio management solutions.  

Our newly announced Infront Quant IQ Risk solution uses a lot of Quantative Risk models that are based upon algorithms and deep data analysis – bringing that analysis into existing Infront tools. Also, our Infront Analytics product with our patented GPRV models for Growth-Profitability-Risk-Value is an analytical framework to assess the relative attractiveness of listed companies through fundamental analysis. Analysing over 45,000 companies with deep fundamental data comparisons is not easy without using machine learning and algorithms under the hood to provide number crunching, calculations and analysis. 

As a next step into AI, we have developed a Gemini chatbot with Google based on the specialised hybrid approach that is required to perform complex real-time calculations, combining the Gemini LLM with vector database techniques, machine learning and our Core Data APIs. As well as enabling performance of complex real-time calculations, for example in fixed income instruments, this project is also promising in the potential it offers for an AGI agentic framework in future. 

If you are interested to learn more about our research and want to exchange experiences, reach out to Infront via your account contact or by using the form below. 

Read the original article here.