Hyperpersonalization: A Tailor-Made Online Choice Environment

ANALYSIS REPORT NOVEMBER 2025 Hyperpersonalization: A tailor-made online choice environment

In short Technological developments make it possible to personalize online choice environments - such as apps and websites. Content, tone, order, and design can be adjusted to the characteristics, preferences, and behavior of consumers. In its most advanced form, this is called hyperpersonalization. In this report, the AFM explores these possibilities, their current and future application in the financial sector, and what this means for consumer protection. Although application is currently limited compared to other sectors, financial enterprises indicate they want to do more with personalization of the choice environment in the future. Personalization can influence the attention and behavior of consumers more targeted and effectively. This offers opportunities and risks. The AFM finds it important that enterprises deal with these new possibilities in a careful and responsible manner.

ANALYSIS REPORT Summary 3

1. Introduction 5
2. (Hyper)personalization of the online choice environment 10 2.1 Possibilities for personalization of the online choice environment 10 2.2 Hyperpersonalization: a fully tailor-made choice environment 12 2.3 Personalization increases the influence of the choice environment 14
3. Application in the financial sector 15 3.1 The first steps towards a personalized online choice environment 15 3.2 Why (hyper)personalization is usually not yet the case 16 3.3 Financial enterprises want to do more with personalization in the future 16
4. Opportunities and risks 19 4.1 Personalization offers opportunities to support consumers in financial choices 19 4.2 Personalization can have unwanted consequences for consumers 19
5. Responsible use 22 5.1 Careful and responsible application by financial enterprises 22 5.2 Implications for supervision 24
6. References 25 Contents

Hyperpersonalization: A tailor-made online choice environment 3 ANALYSIS REPORT Summary The combination of large amounts of individual data and the use of artificial intelligence (AI) enables enterprises to refine customer profiles and personalize online choice environments with increasing accuracy. Content, tone, order, and design can be adjusted to the characteristics, preferences, and behavior of the consumer. It is possible that the various elements of the online choice environment are no longer determined by a human, but by the outcomes of an algorithm. Through the use of generative AI (GenAI), the design of content can even be fully automated. In its most advanced form, we speak of hyperpersonalization of the choice environment: real-time data-driven, dynamic, and fully tailored to the individual.

This exploration by the AFM shows that financial enterprises want to do more with personalization of the choice environment in the future, but that current application is still limited compared to other sectors. The application by banks, investment firms, and crypto service providers usually is limited to small adjustments based on broad customer segments, such as age groups or customers with and without a mortgage. The first steps towards more fine-meshed and advanced applications focus on personalizing the content of advertisements, notifications, email reminders, and in-app banners. The design of the choice environment is usually still manual work. Financial enterprises can be deterred by legal risks and uncertainties. But also by practical obstacles – such as existing systems that are not flexible enough and limited scalability of personalization. Many of the financial enterprises that the AFM spoke to for this exploration want to focus on (further) personalizing the choice environment in the future and see advantages of personalization, for example for commercial purposes, customer retention, and supporting healthy financial behavior.

By personalizing the choice environment, enterprises can influence the attention and behavior of users more targeted and effectively. This brings opportunities and risks. Personalization can be used to the advantage of consumers, for example by providing better support in financial choices, more insight into the personal financial situation, and increased usability of digital services. However, personalization also increases the existing risk of unwanted influence by the choice environment. It can also be accompanied by unintended side effects, such as a negative effect on outcomes outside the algorithm, the strengthening of (previously made) unwise choices, and discrimination.

It is expected that conversational AI will play an increasingly important role within online choice environments; in the future, this could completely replace the traditional user interface. Thanks to this technology, consumers can communicate with financial enterprises in a personal and low-threshold manner. It is expected that conversational AI will become increasingly advanced, with the function shifting from customer service chatbot to full-fledged financial assistant. Due to the automated and dynamic nature of this technology, new risks also arise, including the risk of even more effective unwanted influence. These risks deserve extra attention in the coming years.

The Authority for the Financial Markets (AFM) encourages financial enterprises to explore how they can use personalization of the choice environment to promote sound financial choices. It is important to handle these new possibilities carefully and responsibly. This means that enterprises carefully weigh the application of personalization and refrain from it when the customer interest is not served by it. When defining the goals pursued with personalization, the commercial interest should not stand above the customer interest. Furthermore, it is important to actively monitor the influence of personalization on (outcomes for) consumers, with special attention to unintended side effects.

Hyperpersonalization: A tailor-made online choice environment 4 ANALYSIS REPORT personalization is pursued, the commercial interest should not stand above the customer interest. Furthermore, it is important to actively monitor the influence of personalization on (outcomes for) consumers, with special attention to unintended side effects. The AFM sees that enterprises are rightly hesitant about the automated use of GenAI for designing the choice environment. This allows enterprises to maintain control over content quality and limit legal risks. Furthermore, it is important to give consumers actual control over the personalization of the choice environment. This means that they can determine at any time and for each application whether they want a personalized choice environment or not – supported by accessible settings and clear explanations.

It is important to respond in time to developments in the field of personalization of the choice environment. These developments create new possibilities that lie around the legal boundary between information provision and advice. The AFM finds it important that financial enterprises gain more clarity on when personalization falls within the boundaries of advising. The AFM will further delve into such ambiguities and take action where necessary. We also strive for accessible and targeted support for consumers in financial choices, paying attention to the opportunities that personalization offers.

Finally, the possibilities for personalization of the choice environment raise new questions about the manner of supervision by the AFM. Knowledge building is crucial for effective supervision. With this and previous explorations, we deepen and enrich our knowledge about the opportunities and risks of digitalization, AI, and personalization (AFM, 2024b, 2025b). The AFM will consider how it can maintain adequate visibility and supervision over increasingly personalized online choice environments and their influence on consumers.

Hyperpersonalization: A tailor-made online choice environment 5 ANALYSIS REPORT

1. Introduction Enterprises have more and more possibilities to personalize the online choice environment; in its most advanced form, we speak of hyperpersonalization. Consumers spend more and more time online and leave digital footprints continuously. These data enable enterprises to form an increasingly sharp image of the person behind the screen. Combined with developments in artificial intelligence (AI), it becomes possible to personalize online choice environments – such as apps and websites – with increasing accuracy. As a result, consumers no longer all see the same choice environment: content, tone, order, and design can be adapted to someone's characteristics, preferences, and behavior. It is possible that the various elements of the online choice environment are no longer determined by a human, but by the outcomes of an algorithm. Through the use of generative AI (GenAI), the design of content can even be fully automated. In its most advanced form, we speak of hyperpersonalization of the online choice environment: real-time data-driven, dynamic, and fully tailored to the individual.

Outside the financial sector, personalization of online choice environments is already widely applied. Think of the Netflix overview adapted to viewing preferences, the TikTok feed that shows exactly where someone 'hangs', Booking.com that makes travel recommendations based on previous choices, or a webshop that recommends products that other, similar consumers bought.

It is very likely that personalization – also in the financial context – will assume increasingly advanced forms. Consumers make their financial choices increasingly online. Investing, saving, borrowing, paying, insuring: it all happens via an app or website. Financial enterprises therefore have more and more data at their disposal. They can make predictions based on that about the effect of adjustments in the online choice environment on different groups. Think of the online portal of a pension fund that is adjusted based on the age of a participant, a bank app that sends specific notifications aimed at customers with an investment account, or an investment app that adjusts the order of displayed products based on a customer's previous trading behavior.

Transfer B.L. VAN DONGEN NL22 ABCB 3134 3772 44 €2,411.54 B.L. VAN DONGEN NL33 ABCB 4348 5625 32 €1,533.35 Save Pay Your salary has just been deposited. Don't forget to put some money aside for your savings goal? Welcome, B! Transfer A.R. SAHIN NL22 ABCB 4029 3432 32 €504.34 A.R. SAHIN NL33 ABCB 4532 4465 12 €21,143.08 Save Pay Tomorrow your mortgage will be deducted. Attention: your balance is now insufficient. Welcome, A! Figure 1.1. A fictional example of how personalization in the financial sector might look. In a banking app, customers see different messages.

Hyperpersonalization: A tailor-made online choice environment 6 ANALYSIS REPORT The influence of a personalized choice environment on consumer behavior is easily overlooked. Public and policy attention often goes to the influence of AI applications in the financial sector that make direct decisions for or about people, such as in credit assessment or algorithmic trading (AFM, 2023b, 2025b, DNB, 2024). Less attention is paid to the use of data and AI for the personalization of the choice environment. This application does not directly lead to different outcomes for consumers, but can indirectly influence their choices in a subtle and often unnoticed way (Fisher & Oppenheimer, 2025; Susser, 2019). Precisely because this form of influence is easily overlooked but may have a large impact on consumers, it deserves explicit attention. In recent years, the AFM has paid much attention to the influence of the, non-personalized, online choice environment. For example, we have conducted observational research into the design of the online choice environment in investment apps and crypto apps (AFM, 2023a, 2024a). Furthermore, we have called on the sector to ensure a design of the choice environment that promotes sound financial choices (AFM, 2021a). With the introduction of the choice guidance standard for pension providers, the AFM has also explicitly paid attention to the choice environment for pension participants (AFM, n.d.-a).

In this exploration, we look at the new possibilities that arise for financial enterprises to (hyper)personalize choice environments, and at the opportunities and risks associated with this for consumers. We look at a broad spectrum of applications: from simple segmentation to advanced forms where an app adapts in real-time to the behavior of the individual user. In chapter 2, we discuss the broad range of possibilities for (hyper)personalization of the choice environment and the influence on consumer behavior. In chapter 3, we discuss the current and future application in the Dutch financial sector. Then, in chapter 4, we discuss the opportunities and risks for consumers that (extensive) personalization in the financial sector may bring. We end the report with recommendations for responsible use (chapter 5).

The content of this report is based on scientific publications, sector and policy reports, and a series of conversations with experts, software parties, and financial enterprises. The findings on the current and future application of personalization are based on conversations with a selected group of banks, investment firms, and crypto service providers. They therefore do not necessarily provide a complete picture of the Dutch financial sector. Although the insights are broadly applicable, the focus in this exploration is mainly on banks, investment firms, and crypto service providers. It is expected that these parties are the most advanced with personalization of the choice environment. This is because they have relatively much customer data, due to the high frequency of online customer interactions and transactions. The potential of personalization is possibly large for these enterprises: due to the broad and diverse product offering, there is much variation in customer behavior and much room to guide customers to suitable products. This does not mean that other financial enterprises – such as mortgage lenders, pension providers, and insurers – also have possibilities to personalize the choice environment.

Hyperpersonalization: A tailor-made online choice environment 7 ANALYSIS REPORT Trending Share A Share B Share C Share D Share A Share B Share C Share D Buy now Increase 24.77% €0.2808 €0.2204 €0.2024 Crypto X Crypto X €0.2204 24.77% €0.2808 €0.2024 Figure 1.2 Two fictional examples of how personalization in the financial sector might look. (1) In an investment app, the same products are presented and highlighted in different ways, and the ease with which a certain action can be taken varies. (2) In a crypto app, the same information is displayed in different ways.

Hyperpersonalization: A tailor-made online choice environment 8 ANALYSIS REPORT Box 1. The building blocks of online personalization In a broad sense, three building blocks are needed for personalization: the data available at the individual level, the interpretation of this data, and the application of these insights for personalization. See the figure below for an overview of the building blocks (based on Strycharz & Duivenvoorde, 2021).

Building block 1: Data – the digital footprint People leave a lot of data online. This is also called the digital footprint. On the one hand, this concerns data that consumers explicitly share with companies and institutions, such as demographic data (e.g., age, gender, place of residence), account information (e.g., preferences and settings), financial data (e.g., income, transactions, purchased products), or feedback (e.g., reviews, customer satisfaction surveys). On the other hand, there is much data that is shared continuously – in a more implicit way – such as data on online click, search, and browsing behavior (e.g., previously viewed pages or products, viewing duration, search terms), usage patterns (e.g., time, frequency), location data (e.g., IP address, GPS location), or technical characteristics (such as device type or browser). The increase in the use of AI-driven chatbots creates a new source of much data. Furthermore, data can be enriched with other (contextual) data, such as the weather at a location or socio-demographic information of a postal code.

Building block 2: Interpretation – insight into the human behind the data The digital footprint can reveal a lot about who someone is. Innovations in the field of AI – including machine learning, natural language processing, and GenAI – enable companies to recognize patterns in data and predict preferences. Companies can make increasingly accurate estimates, also about aspects that people are not always aware of.

Initially, data was mainly used for predictions of customer behavior and preferences; for example, the chance that someone would buy a product or cancel it (Ascarza et al., 2018; Chaudhuri et al., 2021; Haleem et al., 2022). But due to the increased availability of individual data, it is now also possible to predict deep psychological characteristics (Matz & Netzer, 2017). Think of deriving properties such as extraversion, openness, risk attitude, materialism, need for interaction, and quality and price awareness based on various data sources, such as social media and smartphone usage, conversations with chatbots, and financial transactions (Azucar et al., 2018; Blömker & Albrecht, 2025; Gladstone et al., 2019; Marengo & Montag, 2020; Peters et al., 2024a; Ramon et al., 2021; Shumanov & Johnson, 2021; Stachl et al., 2020; Tovanich et al., 2021).

GenAI also makes it possible to use data types that were previously difficult to analyze – such as written text and photos – for estimating psychological and personal characteristics (Guenzel et al., 2025; Peters et al., 2024a). Even moods, emotions, and sensitive information such as income, sexual orientation, and political orientation appear to be derivable to some extent from online behavior (LiKamWa et al., 2013; Matz et al., 2019; Wang & Kosinski, 2018; Kosinski, 2021).

The accuracy of these estimates is certainly not perfect. It varies depending on the nature and quality of the data, the way models are trained, and the extent to which data sources are combined (Azucar et al., 2018; Hinds & Joinson, 2024; Peters et al., 2024b). Some properties are also more difficult to perceive and therefore harder to predict (Hinds & Joinson, 2024). There are also questions about the validity of certain estimates. For example, the Dutch Data Protection Authority (AP) recently expressed concerns about emotion recognition models (Autoriteit Persoonsgegevens, 2025). Despite these caveats, it is expected that estimates will become more complete and accurate in the coming ten years. This thanks to developments in (Gen)AI, the availability of even finer-meshed data, and the integration of data sources (Matz, 2025).

Building block 3: Application – from insight to personalization Enterprises can use data and the insights derived from it to distinguish customer groups with similar characteristics. At the most basic level, this means dividing customers into a limited number (often manually selected) segments or personas, for example based on age, satisfaction, or shared preferences. Thanks to the increased availability of data and computing power, companies can combine data sources and analyze patterns. This allows them to move from static segmentation to more detailed data-driven and even dynamic segmentation (Alves Gomes & Meisen, 2023). In the most advanced case, it is no longer about customer segments, but about a profile at customer level (a 'segment of one', Olayinka, 2021).

Enterprises can use these increasingly fine-meshed segmentations for personalizing the online choice environment. But other elements of the financial service process can also be personalized. For example, the AFM recently published on the personalization of insurance premiums (AFM, 2024b, 2025a). We also published a publication on the possibility that the credit process and offer will be increasingly personalized in the coming years (AFM, 2025b).

Hyperpersonalization: A tailor-made online choice environment 9 ANALYSIS REPORT Building block 1: Data – the digital footprint People leave a lot of data online. This is also called the digital footprint. On the one hand, this concerns data that consumers explicitly share with companies and institutions, such as demographic data (e.g., age, gender, place of residence), account information (e.g., preferences and settings), financial data (e.g., income, transactions, purchased products), or feedback (e.g., reviews, customer satisfaction surveys). On the other hand, there is much data that is shared continuously – in a more implicit way – such as data on online click, search, and browsing behavior (e.g., previously viewed pages or products, viewing duration, search terms), usage patterns (e.g., time, frequency), location data (e.g., IP address, GPS location), or technical characteristics (such as device type or browser). The increase in the use of AI-driven chatbots creates a new source of much data. Furthermore, data can be enriched with other (contextual) data, such as the weather at a location or socio-demographic information of a postal code.

Building block 2: Interpretation – insight into the human behind the data The digital footprint can reveal a lot about who someone is. Innovations in the field of AI – including machine learning, natural language processing, and GenAI – enable companies to recognize patterns in data and predict preferences. Companies can make increasingly accurate estimates, also about aspects that people are not always aware of.

Initially, data was mainly used for predictions of customer behavior and preferences; for example, the chance that someone would buy a product or cancel it (Ascarza et al., 2018; Chaudhuri et al., 2021; Haleem et al., 2022). But due to the increased availability of individual data, it is now also possible to predict deep psychological characteristics (Matz & Netzer, 2017). Think of deriving properties such as extraversion, openness, risk attitude, materialism, need for interaction, and quality and price awareness based on various data sources, such as social media and smartphone usage, conversations with chatbots, and financial transactions (Azucar et al., 2018; Blömker & Albrecht, 2025; Gladstone et al., 2019; Marengo & Montag, 2020; Peters et al., 2024a; Ramon et al., 2021; Shumanov & Johnson, 2021; Stachl et al., 2020; Tovanich et al., 2021).

GenAI also makes it possible to use data types that were previously difficult to analyze – such as written text and photos – for estimating psychological and personal characteristics (Guenzel et al., 2025; Peters et al., 2024a). Even moods, emotions, and sensitive information such as income, sexual orientation, and political orientation appear to be derivable to some extent from online behavior (LiKamWa et al., 2013; Matz et al., 2019; Wang & Kosinski, 2018; Kosinski, 2021).

The accuracy of these estimates is certainly not perfect. It varies depending on the nature and quality of the data, the way models are trained, and the extent to which data sources are combined (Azucar et al., 2018; Hinds & Joinson, 2024; Peters et al., 2024b). Some properties are also more difficult to perceive and therefore harder to predict (Hinds & Joinson, 2024). There are also questions about the validity of certain estimates. For example, the Dutch Data Protection Authority (AP) recently expressed concerns about emotion recognition models (Autoriteit Persoonsgegevens, 2025). Despite these caveats, it is expected that estimates will become more complete and accurate in the coming ten years. This thanks to developments in (Gen)AI, the availability of even finer-meshed data, and the integration of data sources (Matz, 2025).

Building block 3: Application – from insight to personalization Enterprises can use data and the insights derived from it to distinguish customer groups with similar characteristics. At the most basic level, this means dividing customers into a limited number (often manually selected) segments or personas, for example based on age, satisfaction, or shared preferences. Thanks to the increased availability of data and computing power, companies can combine data sources and analyze patterns. This allows them to move from static segmentation to more detailed data-driven and even dynamic segmentation (Alves Gomes & Meisen, 2023). In the most advanced case, it is no longer about customer segments, but about a profile at customer level (a 'segment of one', Olayinka, 2021).

Enterprises can use these increasingly fine-meshed segmentations for personalizing the online choice environment. But other elements of the financial service process can also be personalized. For example, the AFM recently published on the personalization of insurance premiums (AFM, 2024b, 2025a). We also published a publication on the possibility that the credit process and offer will be increasingly personalized in the coming years (AFM, 2025b).

Hyperpersonalization: A tailor-made online choice environment 10 ANALYSIS REPORT 2. (Hyper)personalization of the online choice environment 2.1 Possibilities for personalization of the online choice environment The design of the choice environment influences how people consider their options and which choices they ultimately make (AFM, 2021a). Designers can subtly steer the attention and behavior of people. This can be done, for example, by presenting certain options or considerations more prominently. Or by making follow-up actions easier or more difficult. Here we discuss the many aspects on which the online choice environment can be adjusted to align with insights about the user. Think of personalizing text and images, structure and design, and support in choices (based on Münscher et al., 2016). We do not focus exclusively on possibilities within the financial se...