AI Usage in the Dutch Asset Management Sector

ANALYSIS REPORT
AI in the Dutch Asset Management Sector: Usage is Growing, Risks are Growing Alongside It

In Brief - The use of AI applications by asset managers is increasing. AI is currently mainly used for information gathering and data analysis. The application of (generative) AI for, among other things, improving trading strategies is expected to increase. The use of AI brings benefits to asset managers, but the increased deployment and growing complexity of AI models also bring challenges, for example regarding explainability. Managing these risks remains extremely important for asset managers. Asset managers are also expected to be transparent with their clients about the extent of AI's role in their investment policy or portfolio composition.

MARCH | 2026

Management Summary
The deployment of advanced/self-learning algorithms and other Artificial Intelligence (AI) applications by asset managers is increasing. This has prompted the AFM to conduct research into the current use of AI, developments in the sector, and the associated risks.
The survey of 323 institutions shows that about half (53%) already use AI or plan to do so within a year. Large asset managers and proprietary traders (up to 75%) are leading the adoption of AI. AI is currently mainly used for data analysis and information processing; proprietary traders also use AI for price forecasting and optimizing trading strategies.
Although AI application is increasing, the level of investment remains limited. For example, 71% had no specific AI budget for 2024. Where investment does occur, it often involves small amounts (less than 1% of revenue), although a small group spends more than €1 million per year. Most asset managers (60%) expect to increase their investments in the coming period.
The expected impact of AI varies: a majority currently sees no direct cost savings, but does see potential revenue growth through more efficient processes, better data processing, and more accurate risk assessment. In the longer term, AI could contribute to improved portfolio allocations and market analyses, although it remains unclear whether AI will actually start making autonomous investment decisions. It remains crucial that asset managers are transparent about the role of AI in their investment policy. With opportunities come risks. Asset managers face algorithmic biases, data quality issues, limited explainability of complex models, and dependence on a small number of (mostly non-European) technology providers. The research shows that a part of the sector is still insufficiently prepared: a quarter has no policy for AI use, and for generative AI, this figure is even over two-thirds of participating institutions. There is also much room for improvement regarding ethically responsible AI use: more than half have no ethical handbook or code of conduct specifically addressing AI. Although AI deployment offers great potential, its increasing use requires careful integration into business operations, including clear policy frameworks, attention to explainability and data quality, and safeguards for ethical use. The AFM will therefore continue to supervise the controlled and proper application of AI within the sector.

Table of Contents

Introduction 4
Results 5
2.1 AI Usage in More Detail 5
2.2 Personnel 9
2.3 Policy and Controls 10
2.4 Benefits and Challenges 11
2.5 AI Agents 13
Next Steps 14

Introduction
Technological developments and digitalization are causing changes in the business operations of asset managers and the asset management sector. The overall availability and use of (generative) AI has increased rapidly in recent years. AI systems bring benefits, but from a supervisory perspective there are also points of attention, such as transparency, risk management, and outsourcing. The AFM supervises asset managers and ensures that they have their business operations in order. An important part of this is that these institutions manage their risks effectively. This also applies to the risks arising from the use of AI applications. Therefore, the AFM explicitly looks at whether asset managers carefully incorporate these AI risks into their supervision. The AFM gained insight into the use of AI1 by Dutch asset managers in the first half of 2025 via an exploratory questionnaire. The questionnaire was distributed among Dutch investment firms, managers of UCITS and AIFs (including MiFID top-ups), custodians, proprietary traders (PTs), trading platforms (MTFs and OTFs) (hereinafter collectively referred to as 'asset managers'), with the aim of obtaining a current picture of the current level of AI adoption in the Dutch asset management sector. The questionnaire contains 37 mostly closed questions covering six core themes: AI usage and details, personnel, policy and controls, AI finances, benefits and challenges, and AI agents2. In total, 323 asset managers participated in the questionnaire. The number of respondents per license category is shown in Figure 1.
Figure 1. Number of respondents per license category
1 In this report, we use the definition of AI by IOSCO (2021, page 5, FR06/2021 The use of artificial intelligence and machine learning by market intermediaries and asset managers) and the AI definition of the OECD (OECD Legal Instruments) 2 Definition of AI agents: AI systems that can autonomously perform actions on behalf of a user with minimal or no human intervention (CR/01/2025 Artificial Intelligence in Capital Markets: Use Cases, Risks, and Challenges).

Results
Half of asset managers use AI or will start using it in the near future. Of the 323 asset managers who participated in the research, 170 entities (53%) use AI or have plans to do so in the next 12 months. Among different license types, proprietary traders (72%) and OTF operators (75%) showed the highest AI adoption percentages. The study also shows that larger asset managers (in terms of Assets Under Management) make more use of AI applications than smaller managers. Table 1 provides an overview of AI usage across different license types.
Table 1. Number of respondents currently using AI
License Type/Activity | Yes | No | Yes, %
Investment Firms | 74 | 99 | 43%
Fund Managers | 59 | 36 | 62%
PTs | 26 | 10 | 72%
MTFs | 5 | 4 | 56%
OTFs | 3 | 1 | 75%
Custodians | 3 | 3 | 50%
Total | 170 | 153 | 53%

The majority of asset managers participating in the research had no AI budget for 2024 (230 asset managers, 71%). For those parties that did have an AI budget in 2024, this budget was less than one percent of their revenue. Possible explanations are that investments in the use of AI applications fall under a general budget and are not reported separately, or that publicly available free services such as ChatGPT and Microsoft Copilot are mainly used. There is also a small group of asset managers that invests significantly in AI applications, with budgets exceeding EUR 1 million. Most asset managers (60%) expect to increase their investments in AI applications over the next two years. Slightly more than half of the respondents (172 asset managers, 53%) do not expect cost savings or efficiency gains from AI adoption in the next 2-3 years. They do have higher expectations regarding what AI adoption can do for their expected revenue. 223 asset managers (69%) expect a low to moderate revenue increase. 9 asset managers expect a very high revenue increase.

2.1 AI Usage in More Detail
In the research, we looked at what asset managers use AI for, what types of AI applications and models are used, and where these AI solutions are hosted. The presented information is based on the answers of the 170 asset managers who indicated in the research that they use AI or plan to do so in the short term.

© AFM 2026 | AI in the Dutch Asset Management Sector: Usage is Growing, Risks are Growing Alongside It 6 At the time of answering the questionnaire, slightly less than half (82 asset managers; 48%) of the 170 asset managers indicated using AI applications for obtaining information to support analysis and decision-making. In this process, information from various sources is identified, collected, and verified to support analysis and decision-making. In addition to information sources, AI is often used for analyzing unstructured alternative 'big data' (48%) and writing research reports (37 asset managers; 45%). Proprietary traders stand out for using AI in more complex activities. In addition to information sources (73%), they mainly use AI to optimize trading algorithm parameters (65%), improve trading strategies (62%), and forecast price movements or market rates (62%). For a detailed breakdown of daily AI usage by activity type, see Figure 2.
Figure 2. Number of asset managers already using AI tools daily in specified activities
Among the different types of AI technologies, asset managers mainly use Natural Language Processing (NLP) applications. In the questionnaire responses, 136 respondents (80%) stated they use NLP applications. This type of AI application processes text and spoken words and understands their meaning, so this information can be used as input for a (statistical) model. NLP can or cannot be based on Machine Learning (ML) – a form of AI where a system learns/adapts itself without having received explicit instructions for this. Deep Learning is a specialized form of ML, as it can also convert unstructured data (such as images and sound) into data from which learning occurs. Parametric models, on the other hand, use fixed parameters, in contrast to the flexibility of Deep Learning.3

Given the number of asset managers deploying AI applications for obtaining information, it is not surprising that NLP tools are used much more than more complex AI techniques such as Deep Learning.

© AFM 2026 | AI in the Dutch Asset Management Sector: Usage is Growing, Risks are Growing Alongside It 7
Figure 3. Most asset managers use Natural Language Processing tools
There are also differences between license types. For example, many proprietary traders prefer to use Machine Learning (ML) in their daily activities.
Regarding the choice for Machine Learning techniques, 'supervised learning' is the most applied technique, used by 36 of the respondents using ML (80%). Other ML techniques, such as 'unsupervised learning' (20 asset managers; 44%) and 'reinforcement learning' (13 asset managers, 29%), are used less frequently. These results align with observations in the AFM's 2023 report on the use of machine learning in trading algorithms4. Almost all asset managers (94%) use General Purpose models – artificial intelligence models designed to perform a wide range of tasks across different domains, rather than being specialized for one specific function. Custom models developed internally (18%), custom models from external providers (14%), and open-source models (12%) are used much less frequently. The figure below, Figure 4, shows the number of asset managers per used AI model type.
Figure 4. Most asset managers use General Purpose models

In the research, 23 asset managers indicated using AI models developed by an external provider that specifically align with the characteristics and objectives of the asset manager's own business model. The advantage of these custom models is that they know the organization better and may even be trained on the asset manager's specific data. The research has shown that custom models are developed by a wide range of different financial technology companies. There is no single external developer used by many asset managers. The majority (114 asset managers, 68%) use commercial cloud solutions to host their AI solutions. The advantages of commercial cloud solutions may relate to scale and ease of use. However, commercial cloud solutions require extra attention for, among other things, data handling and data privacy, and operational dependence/vendor lock-in. A smaller group (15 asset managers, 9%) uses only private or specialized infrastructure, including some proprietary traders. The remainder (34 asset managers, 20%) use a hybrid environment, which combines cloud and private hosting—this group consists mainly of proprietary traders.
Figure 5. Most asset managers host their AI solutions on commercial cloud solutions

2.2 Personnel
As of February 2025, organizations that develop or use AI must promote that their personnel possess sufficient knowledge and skills to deploy AI responsibly. This expectation regarding 'AI literacy' is included in the European AI Act.5 With AI literacy, it is meant that everyone who works within or on behalf of an organization with AI systems must have skills, knowledge, and understanding not only of the technical operation of the AI systems but also of the social and ethical aspects.6 This paragraph addresses the survey results regarding the promotion of this knowledge among employees. The results are based on the answers of the 170 asset managers who previously indicated using AI tools. The majority of asset managers (95 asset managers, 56%) provide training for their employees responsible for AI applications (including the use and operation of AI systems) or are in the process of developing a training program that will be rolled out shortly. Almost half of the asset managers (80 asset managers, 47%) provide general awareness training for all employees on AI (including the use and operation of AI systems), while 15 asset managers state they have set up specific advanced training for AI developers/data scientists, including four proprietary traders. There is also some overlap, as nine asset managers provide both general AI awareness training and specific advanced training.
Figure 6. Most asset managers provide training to employees responsible for AI or will start doing so shortly

© AFM 2026 | AI in the Dutch Asset Management Sector: Usage is Growing, Risks are Growing Alongside It 10 5 EU Regulation 2024/1689 6 See also the explanation on the website of the Dutch Data Protection Authority regarding AI literacy

2.3 Policy and Controls
To use AI applications responsibly and manage risks effectively, a clear governance framework is necessary, as well as well-defined processes to guide employees. This means clearly assigned responsibilities regarding the development and use of AI applications, and established policies and controls that match the level of AI adoption by the organization.
This paragraph covers governance arrangements regarding the use of AI applications. The scope of this paragraph includes all 323 questionnaire respondents.
At most asset managers, multiple organizational functions are involved in supervising the use of AI applications. These include compliance (254 asset managers, 79%), information security (209 asset managers, 65%), risk management (190 asset managers, 59%), and DPO/data protection officer (94 asset managers, 29%). It is important to note that ultimately the board of directors and/or senior management is responsible for establishing policies and controls and supervising the management of risks related to AI use by employees within the organization. However, a quarter of the participating asset managers (85 asset managers, 26%) have not established a policy (and have not even started developing one) to regulate or limit the use of AI applications by employees. In line with this, over a third of asset managers (112 asset managers, 35%) have not established technical and/or procedural controls regarding the use of AI tools by employees. This could include techniques to detect, address, and/or prevent AI-specific threats such as data or model poisoning7, adversarial and evasion attacks8. The controlled use of AI applications is part of the general obligation under the Financial Supervision Act (Wft) for asset managers to ensure controlled and ethical business operations9. Strong governance, policies, and controls that align with the complexity of the AI models used and the intensity of AI model usage within the organization will support the effective understanding and management of AI (model) risks. Risk awareness and corresponding behavior among employees is also crucial when using Generative AI applications. Generative AI offers the ability to make predictions, outlining future developments based on data. Generative AI can thus create its own content that may not be entirely based on actual facts. This greater complexity of Generative AI implies greater uncertainty and unexpected behavior. At 237 asset managers (73%), all employees have free access to public GenAI applications available on the internet. However, the number of asset managers with targeted policy for this is much smaller. Only 92 asset managers (28%) state that they have implemented specific policy for the use of Generative AI applications or that they have general policy in which this topic is addressed.

7 Where data is manipulated before being used to train the model 8 Techniques that manipulate the model, for example by adding extra data to a model or manipulating new data 9 Article 4:14 of the Financial Supervision Act, Wft)

© AFM 2026 | AI in the Dutch Asset Management Sector: Usage is Growing, Risks are Growing Alongside It 11 The use of AI applications brings ethical risks alongside operational benefits, such as biases and privacy breaches. An ethical handbook or code of conduct supports employees in dealing responsibly with AI systems. However, the survey shows that more than half of the respondents (175 asset managers, 54%) have not implemented a special ethical handbook or code of conduct addressing the use of AI in the organization.

2.4 Benefits and Challenges
Of the 323 asset managers who participated in the survey, 226 asset managers (70%) experience efficiency gains as a major benefit of using AI applications. Additionally, asset managers also consider the ability to analyze data (144 asset managers; 45%) and the improvement of internal processes (123 asset managers; 38%) as important benefits, as shown in Figure 7.
Figure 7. Efficiency is the most frequently mentioned benefit10

Data quality appears to be the biggest challenge for asset managers (160 asset managers, 50%). Another frequently mentioned challenge is data protection (137 asset managers, 42%). Challenges that few asset managers face are production implementation (28 asset managers, 9%) and model drifting (33 asset managers, 10%). This suggests that most asset managers are not yet at an advanced stage of AI development and therefore do not yet encounter the challenges associated with, for example, developing their own AI model.
Figure 8. Data quality is the biggest challenge asset managers face11

Although AI offers many opportunities for the financial sector, there is also considerable attention being paid to its possible consequences. Surprisingly, most respondents expect AI to have little impact on specific parts of the asset management sector in the coming years. They expect AI to make little difference regarding contagion risk, liquidity, market concentration, net market return, and product assortment. However, several asset managers expect that, alongside the increase in legal requirements, AI will mainly impact internal processes, market efficiency, regulatory requirements, and skilled labor, as shown in Figure 9.
Figure 9. What will be the impact of AI on the following areas in the next three years, according to asset managers? (number of asset managers)?

2.5 AI Agents
AI agents have the ability to improve efficiency, decision-making, and scalability across a wide range of tasks and sectors. At the same time, they have the advantage of being highly accessible to entities. This paragraph covers the extent to which asset managers plan to implement AI (multi)agents, and whether the use of AI agents is already embedded in AI policy. The scope of this paragraph includes all 323 questionnaire respondents. A significant majority of asset managers (253 asset managers, 78%) state that they do not plan to implement AI agents or multi-agent systems within the next 12 months. When respondents are broken down by license type, it appears that OTFs (2 asset managers; 50%) and MTFs (4 asset managers; 44%) are more positive about this. Currently, 287 asset managers (89%) lack a formal AI policy that regulates the use of AI agents within their organizations. Only

Similar documents from other regulators