Artificial intelligence in finance: Insights from Luxembourg’s financial sector

Key findings from the CSSF/BCL AI survey

Based on a broad and representative sample of over 460 financial institutions, the 2024 survey highlights the evolution of AI adoption since the last edition in 2021.

1. Strong participation and maturing practices

The strong response rate underlines the growing strategic relevance of AI, with a clear trend towards operational maturity, especially among larger banks and group-level structures.

2. AI adoption is accelerating

AI is increasingly being adopted across entities, with a notable increase in live or near-production use cases. Banks in particular are leveraging AI in development or experimentation, often drawing on centralised group expertise.

3. Generative AI surpassing traditional ML

Generative AI (GenAI) has overtaken traditional machine learning (ML) in terms of reported use cases and institutional attention. While GenAI dominates in experimentation and internal use, ML remains the more mature approach for risk, compliance, and regulatory applications.

4. Strong investment momentum

2024 saw a rise in investments in AI and distributed ledger technologies (DLT), with AI clearly leading the trend — especially at group level. Investment plans for 2025–2026 suggest continued growth, particularly in GenAI applications.

5. Use cases focused on internal efficiency

The vast majority of AI use cases are for internal optimisation rather than client-facing tools. Top categories include:

• Search and summarisation
• Process automation
• Virtual assistants and chatbots
• Text content generation
• Translation

6. Limited GenAI governance

Despite 64% of institutions allowing access to public GenAI tools, only 40% have a formal policy in place. Credit institutions tend to be more restrictive, with over half denying access to such tools entirely.

7. AI organisation and training on the rise

Many institutions, especially larger banks, have now established dedicated data science teams, mainly at group level. 84% of respondents either offer or plan to implement AI-related training, though smaller entities may still face challenges in upskilling.

8. Progress on trustworthiness

The survey shows improvements in bias detection, human oversight, and cybersecurity. Auditability and explainability, however, remain key areas to address — particularly for complex GenAI models.

9. Data remains the primary challenge

As in 2021, data remains the most significant hurdle. Respondents cited:

• Data quality
• Data protection
• Data governance
  as the top three barriers to successful AI deployment.

10. Awareness of regulation is growing

The AI Act is broadly acknowledged, but there is still a need for greater regulatory clarity, especially among non-bank institutions that may lack the resources to fully engage with evolving compliance requirements.

Insights from the panel discussion

A distinguished panel of industry experts shared practical insights on the realities of AI implementation. Discussions focused on:

Regulation, governance and culture

How to balance innovation with compliance, and how to embed the right skills and mindset to scale AI responsibly.

AI literacy was also highlighted as a critical element. While the vast majority of survey respondents have either already implemented or plan to implement a range of AI training programmes for their employees—particularly in view of the AI literacy obligations introduced by the AI Act—it remains unclear what specific expectations exist around the scope and depth of such training. Nevertheless, a general culture of AI awareness within institutions was seen as essential to support the safe and effective development and internal use of AI tools.

Data, technology and use cases

From legacy integration to cybersecurity, the panellists emphasised the need for:

• Cross-functional collaboration
• A robust data governance framework
• Trust in AI across teams
• Cleaning and structuring data early in the project lifecycle
• Exploring the future impact of agentic models
• Model exchange between institutions for training purposes

Future outlook and strategic priorities

To be AI-ready, institutions must:

• Continue investing in data quality and infrastructure
• Strengthen governance and oversight
• Collaborate across the ecosystem — including regulators and vendors
• Embrace a culture of experimentation and responsible innovation