Navigating publicness in digital innovation: big data and AI adoption in European public sector organisations

The growing role of data-driven technologies in the public sector has brought big data (BD) and artificial intelligence (AI) to the forefront of public service innovation. Although closely connected, BD and AI represent distinct yet complementary functions within digital governance. While BD refers to the large-scale collection, processing and analysis of structured and unstructured datasets from various sources (Mergel et al., 2016; Klievink et al., 2017), AI involves the use of algorithms to learn from these datasets and perform tasks with some level of autonomy (European Commission, 2018). In public sector settings, BD serves as the foundational input, enabling richer insights and evidence-based decisions, while AI offers automation, responsiveness and predictive capabilities that can transform the delivery of public services (Pencheva et al., 2020).

In practice, BD and AI are applied in various ways across sectors. For instance, municipalities increasingly rely on real-time traffic data to optimise mobility and reduce congestion (Zheng et al., 2015), while national statistics offices collect and harmonise vast administrative datasets for evidence-based policymaking (Manzoni, 2018). AI, on the other hand, has been trained to detect fraud, predict crime hotspots and enhance citizen interaction through chatbots (Van Noordt and Misuraca, 2022). Nonetheless, these technological opportunities also bring significant challenges, particularly in the public domain. Issues of algorithmic bias, lack of transparency, ethical governance and lack of citizen trust complicate AI adoption (Wirtz et al., 2019; Aoki, 2020). Likewise, BD practices suffer from fragmented data infrastructures, legacy systems and varying degrees of data readiness across organisations (Guenduez et al., 2020).

What makes BD and AI uniquely complex in the public sector are the distinct characteristics that define public organisations – namely, publicness (Seepma et al., 2021; Senadheera et al., 2024). Unlike private entities, public institutions are subject to political oversight, public visibility and legal constraints, which significantly shape their technological choices and implementations (Bozeman, 1987, 2013). These dimensions of publicness can either facilitate or constrain the development and use of BD and AI. For example, while public organisations may be favoured in funding to experiment AI, political influence and control can restrict the independence needed to scale those technologies. Furthermore, publicness may create significant consequences for data collaboration between organisations (Klievink et al., 2017), often imposing formal procedures and bureaucratic red tapes. Therefore, the application of BD and AI in public service delivery cannot be understood without considering these governance-specific publicness factors.

Despite this, the growing research on BD and AI in public service delivery has often engaged in theory or through lenses more suited to private sector contexts (Fatima et al., 2020; Mikalef et al., 2019). In doing so, it has tended to overlook how the publicness of organisations influences the nature and impact of digital transformation (Sun and Medaglia, 2019; Kankanhalli et al., 2019). Recent research already addresses the calls for using a public value-based perspective when exploring the implementation and use of AI in public administration (Madan and Ashok, 2023). However, by foregrounding structural dimensions, a publicness perspective may enrich our understanding of how and why public agencies pursue certain technological strategies, and what constraints or enablers arise from their institutional identity.

In this context, the present study aims to investigate empirically how publicness characteristics influence BD and AI activities in public organisations. It looks beyond technical or infrastructural concerns and additionally investigates how governance structures, political mandates and institutional roles shape the collection, management and use of BD and AI technologies. Thus, this study aims to answer the following research questions:

BD and AI have emerged as transformative forces in public administration, promising to improve efficiency, responsiveness and evidence-based decision-making across service domains (Wirtz et al., 2019; Pencheva et al., 2020). Although often discussed jointly, they represent different yet complementary technological paradigms.

BD refers to the accumulation and processing of vast and varied datasets, often highlighting its three primary features, volume, variety and velocity (Desouza and Jacob, 2017). In this study, drawing on Mergel et al. (2016, p. 931) BD is defined as “high-volume data that frequently combines highly structured administrative data actively collected by public sector organisations with continuously and automatically collected structured and unstructured real-time data that are often passively created by public and private entities through their Internet interactions”. This includes traditional government records as well as sensor outputs, social media content and transaction logs. AI, in turn, refers to “digital systems capable of performing tasks that typically require human intelligence, such as learning, reasoning, or decision-making” (European Commission, 2018). It encompasses the computational techniques that extract value from the collected vast BD, through prediction, classification or decision support (Mergel et al., 2016; Boyd and Wilson, 2017).

These technologies are increasingly embedded within public organisations, applied across diverse functional areas and typically progressing through interconnected stages. Drawing on the literature, the application of BD follows a sequence of interconnected stages of collection, management, use and sharing (Klievink et al., 2017). Firstly, the data must be collected from internal (e.g. administrative records) and external (e.g. sensors, third-party platforms) sources, often without predefined purposes, in anticipation of future utility (Constantiou and Kallinikos, 2015). This is followed by data management, which involves cleaning, integration and preparation for use. Many public bodies struggle at this stage due to legacy systems, low data literacy and the lack of unified standards (Miller and Mork, 2013; Chen et al., 2014). Once managed, data can be used to analyse and generate actionable insights, supporting service design, operational decisions and policy evaluation. Finally, sharing these data results, whether across interested departments or with the public, is key to enhancing transparency and enabling holistic governance (Klievink et al., 2017). Each of these stages forms part of a continuous BD cycle that supports public value creation.

Similarly, AI activities are shaped by development and application stages. Development entails training models using BD inputs, often through collaborations with academic or private partners, given the lack of in-house technical capacity (Wirtz et al., 2019). These models are then applied to decision-support or automation settings, where their outputs may guide resource allocation, risk assessment or public communication. However, research reveals that the willingness to implement and use AI technologies in government is contingent upon a series of positive and negative perceptions about the new technologies as well as the long-term outlook on the role of AI technologies in society (Ahn and Chen, 2022).

Despite their promise, the actual operational use of BD and AI in the public sector remains fragmented and relatively limited (Guenduez et al., 2020; Yigitcanlar et al., 2023a). While public institutions are among the largest data collectors, their ability to leverage this data is often constrained by legal, technical and organisational challenges. Fragmented IT infrastructures, outdated systems and the absence of interoperable platforms impede data integration and analytic scalability (Desouza and Jacob, 2017). Low levels of data literacy, bureaucratic resistance and unclear roles and responsibilities further complicate adoption (Moody et al., 2019). At the same time, legal and ethical concerns, including data protection regulations and the need for transparency, introduce necessary, but often burdensome, compliance requirements that can delay or limit AI deployment (Yang and Maxwell, 2011; Guenduez et al., 2020).

To realise the full potential of BD and AI in the public domain, interoperability and cross-organisational coordination are essential. When public bodies can effectively share and integrate data, supported by common standards, secure data pipelines and harmonised governance protocols, they are better positioned to achieve comprehensive insights and coordinated action (Karlsson et al., 2017; Rotta et al., 2019). Such inter-organisational collaboration allows data to be reused across contexts, reducing duplication and increasing efficiency. However, interoperability is not only a technical issue but also involves aligning organisational goals, addressing trust concerns and overcoming bureaucratic inertia. The degree of publicness within an organisation can further influence what data can be shared, with whom and under what conditions.

The theory of publicness originated in public administration literature during the 1980s (Bozeman, 1987), challenging the simple dichotomy between “public” and “private” organisations by proposing that all organisations embody varying degrees of publicness. It is not simply a legal status; rather, it refers to “the extent to which an organisation is influenced by political authority and subject to public accountability, regardless of whether it is formally public or private” (Bozeman and Bretschneider, 1994). Bozeman conceptualised publicness as a continuum, along which organisations experience varying degrees of political, financial and institutional control. This was a shift from earlier public administration debates that focused narrowly on whether organisations were government owned. Organisations with a high degree of publicness tend to prioritise societal goals, such as transparency, equity and citizen engagement, unlike their private counterparts, which are more commonly guided by market efficiency and profit maximisation (Rainey et al., 1976; Bozeman, 2013).

Early applications of publicness theory examined cases where the boundary between public and private was blurred, such as healthcare, education and research. Rainey et al. (1976) observed that public organisations emphasise accountability and equity. Bozeman and Crow (1990) showed how funding shaped research agendas in hybrid institutes, while Bozeman and Bretschneider (1994) demonstrated that political influence constrained IT management in government agencies. These studies confirmed that organisational behaviour cannot be explained by ownership alone but is shaped by institutional arrangements, funding flows and political oversight.

While earlier research used publicness theory to explain governance contexts and organisational performance (Boyne, 2002), attention has since shifted towards values and outcomes. Bozeman’s later work connected publicness to broader societal concerns such as trust, transparency and equity (Bozeman, 2013). More recently, Anderson (2012) showed that healthcare providers operating across public–private boundaries must balance political authority, economic autonomy and public values to achieve service outcomes. Similarly, Seepma et al. (2021) applied publicness theory to criminal justice supply chains, finding that statutory control structures determined inter-agency governance and that integration mechanisms were used to manage tensions rather than emulate private-sector efficiency. Most recently, Senadheera et al. (2024) reviewed the Metaverse in the context of local governments through the publicness theory highlighting that despite the promise of greater inclusiveness, the Metaverse may, in reality, exacerbate the digital divide. Together, these developments demonstrate how the theory evolved from distinguishing organisations to explaining complex governance challenges.

Building on this evolution, applying publicness theory to service delivery highlights how publicness shapes not only organisational governance but also the everyday provision of services. Public services are defined as tasks that support the direct delivery of services to the public or facilitate communication with the public for regulatory or other purposes (Moulton, 2009). They require the voluntary or involuntary involvement of multiple actors in the design, management, delivery of services, including the end user: the citizens (Osborne et al., 2016). Unlike private organisations, where users are typically treated as customers, public sector institutions serve a broader collective who are not only recipients but also key stakeholders in government action. Therefore, the legitimacy and effectiveness of public organisations are deeply tied to public trust and democratic responsiveness, which are fundamentally shaped by their degree of publicness, a key factor influencing how services are designed and delivered.

According to Bozeman (1987), there are four key dimensions of publicness: (1) ownership, (2) funding, (3) goal setting and (4) control structure. Ownership refers to whether an organisation is publicly held and governed; funding denotes the proportion of its budget derived from public resources; goal setting captures the extent to which political or social mandates influence strategic direction; and control structure relates to the degree of oversight, visibility and compliance, the organisation is subjected to. These dimensions interact to produce both enabling conditions and constraints for digital innovation in the public realm. For instance, organisations primarily funded through taxation often benefit from political support to pursue data and AI projects that align with government priorities. However, such funding may also come with conditions tied to political cycles, changing agendas or procedural scrutiny, which can slow innovation or shift focus (Bozeman and Crow, 1990; Berman, 1998). Similarly, goal setting can determine whether BD and AI are used for service efficiency, equity-driven interventions or regulatory monitoring, based on the values embedded in public mandates.

The control structure dimension influences how organisations approach new technologies. Public agencies that are highly visible and accountable to political actors and the public face greater demands for transparency and explainability, especially in AI applications (Yigitcanlar et al., 2023b). In domains such as social welfare or public safety, where decisions carry significant ethical implications, AI tools must be auditable and align with public values, often necessitating human oversight in decision-making (Wirtz et al., 2019; Aoki, 2020). This reinforces the need for responsible and citizen-centric AI governance frameworks (Zouridis et al., 2019). Here, the communication dimension of publicness, although not always separately listed in Bozeman's original four, plays a vital role. Public sector organisations are not only expected to share vertically with government actors but also horizontally with other government agencies, civil society and the public. This requirement for cross-boundary transparency and coordination significantly shapes how information is shared (Seepma et al., 2021), and together with the core dimensions of publicness, influences how BD and AI are applied in practice.

Recognising that public organisations vary in the extent to which they exhibit these dimensions, it becomes evident that not all management approaches drawn from private-sector models are universally applicable (Boyne, 2002). Publicness imposes unique constraints and obligations, shaping how data are collected, used, shared and technologies are adopted. Accordingly, this study examines a range of public service settings to explore how different degrees of publicness shape the stages of BD and AI adoption. The research framework for this study is illustrated in Figure 1. This framework illustrates how data-driven and AI-enabled public service delivery emerges through iterative, while being shaped by governance dimensions of ownership, funding, goal setting and control. It highlights the interplay between technological practices and institutional structures in enabling public value creation.

This research adopts a multiple case-study approach to empirically investigate BD and AI efforts in public service delivery (Eisenhardt, 1989). Case study research is particularly suited to explore a real-life phenomenon as it allows for an in-depth description of specific practices such as information sharing (Eisenhardt, 1989). The inductive and flexible nature of the case method favours the emergence of new conclusions. Simultaneously, it enables this research to gain an understanding of holistic events such as organisational strategies. While the findings from this approach may not be easily generalisable, examining multiple cases offers a broader understanding of current challenges and creates room for generating hypotheses (Yin, 1994).

This study identified public organisations as a unit-of-analysis, selecting enough cases helping to define the limits for generalising the findings, thereby sharpening external validity (see Table 1). The cases operate in different public sectors, which has significant implications for their possibilities of collecting and using data, given the prevailing laws and predetermined roles. In addition, political control structures prescribe data sharing permitted between public service organisations which influences BD efforts. For this reason, it is expected that the scope of BD and AI as well as the level of inter-organisational data sharing are varying between the different sectors. For understanding how data sharing mechanisms affect the scope of BD and AI and whether this is influenced by the characteristics of the public sector, case study research is appropriate (Yin, 2009).

In selecting specific public institutions, this study purposefully sought out organisations from different public sectors to investigate the various BD and AI applications for delivering services to citizens. The selection was guided by three criteria: (1) sectoral diversity, to capture variation in mandates and service contexts; (2) organisational hierarchy, including both operational and strategic roles, to reflect differences in emphasis and perspectives on BD and AI; (3) interdependence on third-party information providers, as these organisations typically engage in extensive data sharing to fulfil their obligations. These criteria ensured that the cases would generate insights into both intra-organisational dynamics and cross-organisational integration challenges. Moreover, the different hierarchical levels provided varying accounts of technical, organisational and legal barriers to BD and AI adoption.

To enhance the robustness of the study, attention was given to established quality criteria for qualitative research. Credibility was pursued through triangulation of perspectives across multiple organisations and hierarchical levels, ensuring that findings reflect a balanced understanding rather than single viewpoints. Transferability was supported by selecting cases from diverse public sectors, allowing insights to be relevant across a broader range of government contexts.

The main source of data in this study were 17 semi-structured interviews collected between September and November 2022 (see Table 1). This study mostly focused on cases operating in the Netherlands, a digital frontrunner within Europe (ranked third in the Digital Economy and Society Index), providing a profound basis for more advanced analytics techniques and thus, state-of-the-art insights. Besides, three cases from Germany, one case from Belgium and one from the UK were interviewed.

For each case, individual online interviews with a representative of the public organisation were organised. As this research is particularly interested in studying the development and use of BD and AI, interviewees were chosen based on their knowledge of innovation and data science (Table 1). Grounded in the initial literature review, an interview protocol was developed (see Table A1 in  Appendix 1), allowing for comparability of answers, and improving the reliability of the study. Accordingly, the interviews followed a standard procedure organised under broadly defined topics in relation to current projects, with open-ended questions and probes to encourage detailed responses. Interviews took between 45–60 min each.

Each interview started with broad questions regarding the interviewee’s role and the mission of their organisation. The discussion then shifted to exploring specific use cases of BD and AI within their organisation. The primary focus was on how these technologies were leveraged to improve service delivery, and whether implementation occurred within a single organisation or involved collaboration across organisations. All interviews were conducted with consent, recorded and transcribed. To ensure data reliability and accuracy, the transcripts were shared with the participants for validation and were revised, when necessary, based on their feedback or clarifications.

Last, documents provided by the interviewees were collected to underpin and supplemented the information gathered through interviews (Table 2). Employing a variety of data collection methods offers deeper insights and reinforces the theoretical foundation through evidence triangulation, ultimately leading to more robust support for constructs and hypotheses.

The interviews were analysed using the three steps proposed by Miles and Huberman (1994): data reduction, data display and conclusion. The coding and analysis were performed using ATLAS.ti as it enables systematic data organisation, facilitates in-depth thematic exploration. The analysis began by condensing the interviews into relevant sentences or paragraphs that directly addressed the research questions (first-order codes). These first-order codes were then categorised into more descriptive second-order themes, such as “mandatory data provision” and “stringent rules governing data use”. This enabled the deduction of third-order themes in relation to BD and AI activities identified from the literature and complemented by emerging themes from the interviews. An excerpt of the coding is presented in Table A2 and Table A3 in  Appendix 2.

Subsequently, the focus of the analysis was on individual cases to understand the circumstances of each organisation and gain familiarity with data and preliminary theory generation (Eisenhardt, 1989; Yin, 2009). In that regard, the organisational characteristics and services were closely examined and summarised accordingly for each case ( Appendix 3). This paper then proceeded to a cross-case analysis, initially across cases in relation to each sector and then across all 17 cases to also see evidence through multiple lenses. In doing so, this study was seeking to identify patterns that could explain how data sharing mechanisms relate to the specific BD and AI activities and, ultimately, enhanced public service delivery. Moreover, this study aimed to identify sectoral differences regarding the scope of BD and AI based on the influence of publicness dimensions (Goodsell, 2017).

To ensure the trustworthiness of the qualitative data and its analysis, several measures were implemented. This study followed a recursive and iterative approach, first relating the data and findings to existing theoretical frameworks and literature (Eisenhardt and Graebner, 2007), and second, eliminating alternative interpretations (Yin, 2009). It is important to note that due to the qualitative depth of the data collected, which aligns with the research objectives, the focus of this study is on analytical generalisation to theoretical concepts rather than on statistical generalisation.

Our findings highlight an ongoing need to clarify the concept of BD, as respondents provided inconsistent definitions. These different understandings of BD are not inherently problematic. However, they could pose challenges in defining organisational and technical standards and in forming a common vision for BD use in public administration. Such misalignment may slow down the development of potentially valuable BD applications. While some recognise BD simply as a buzzword, for others it is “all the data that we gather today in the most unprocessed, raw format” (C2). One case described BD in terms of the capability to cheaply store and combine very large datasets:

Interestingly, the two statistical offices (A1, A2) use the term administrative data for government-owned information. In their view, BD is generated and owned mostly by private organisations; thus, “government owned data, we do not call that Big Data, but administrative data and […] from our perspective, that's out of scope of Big Data'” (A1). This divergence in definitions likely reflects role-specific incentives: statistical offices anchor BD in long-standing administrative data standards, whereas service providers meet BD through emerging, privately held streams where storage cost declines changed practice.

Our analysis indicates that public organisations rarely face obstacles in obtaining data from other public institutions, provided the data will be used for a legally prescribed purpose. Many local and regional government cases are obliged to fulfil public service duties and thus have free access to citizen data registers. Moreover, specific laws in statistics and law enforcement enable data requests on a legal basis. As one interviewee explained, “most processes are very strict and very much set in place already” (H1). Accordingly, most organisations follow a use-case–driven approach; in other words, they rarely collect data without a specific immediate purpose:

Public organisations collect data from various sources, and the private sector plays a key role since it owns the majority of BD (as opposed to government-held administrative data). Some examples in the public sector are satellite images used for agricultural statistics (A2) and highway planning (G2), road sensor data for traffic flow optimisation (B2) and social media data for citizen sentiment analysis (G3). Public organisations also generate some of their own data (e.g. via web scraping (A1) or text analysis (G1)). However, privately owned data is often higher in quality, possibly because public bodies lack awareness of data quality's importance (C1) or because data collected was never intended for secondary use (E1, K1).

Nevertheless, data collection can be performed in several ways. First, publicly available data can be retrieved online as governmental bodies increasingly publish statistics on dedicated websites (B1). Second, data may be acquired in bulk through purchases or licenses or streamed via sensors and APIs (N1). Purchasing or licensing data becomes necessary when the intended use falls outside an agency’s official mandate (G2). In such cases, there is no legal basis to access the data freely, so it must be obtained through agreements. Last, data collection may be handled by a central unit (B1, A1) or decentralised to the responsible department (C2, K1). Centralised models lower coordination costs but narrow scope, while decentralised collection increases responsiveness at the price of heterogeneity and later harmonisation work.

Public-sector organisations face considerable challenges in data management, particularly in data storage, preparation and combining datasets. One major issue is the lengthy process of data preparation, which requires significant time and resources. As one data specialist noted, “[…] we have a lot of different organisations that provide data, it's never just in one format. The pre-processing cleaning steps are always very annoying” (H1). Moreover, combining BD from multiple sources remains difficult for public organisations. This difficulty also affects higher levels of government that rely on aggregated data from lower-level bodies.

This can be traced back to the predominant existence of legacy systems that hinder the integration of multiple data sets. A data scientist (K1) described this legacy issue as a form of technical debt, where past technical compromises require ongoing maintenance efforts (Metaxiotis et al., 2010). Thus, insufficient investment in IT infrastructure, coupled with data collection processes not designed for reuse, hinders the public sector's ability to fully exploit BD. In addition, there is a lack of transparency about what data exist across the public-sector supply chain. This lack of visibility hampers data sharing and use, and it complicates accountability. One interviewee (B2) describes the challenges as follows:

Importantly, this challenge is not only local, but recognised across multiple public-sector institutions. This awareness has led to new projects aimed at improving data transparency and interoperability. One key step is investing in data transparency and quality: “make transparent who has what, then improve quality” (C1). For example, a Dutch province created a data strategist role to identify “what kind of data we have within our organisation, what data is essential and how can we use it for the better of the province” (G1).

BD use is on the rise across various public-sector domains and government levels, although the scope of applications varies considerably by context. First, data is used for numerous services by governmental bodies. At the municipal level, BD is used for policymaking (B1), traffic flow optimisation and crowd management (B2). Regional government bodies use BD for spatial planning (G1), optimising waterway flows (G2) and monitoring grant subsidies (G3). Our national government case did not provide specific examples of BD use. However, another case (C1) described building a BD model to support a national government's control function. Beyond these examples, BD use in other public organisations varies, largely depending on sector-specific factors that can enable or hinder data initiatives.

In the statistics and law enforcement sectors, BD use is facilitated by legal provisions that grant access to a wide range of administrative data sources. For example, police authorities have legal powers to compel private organisations to share data (H1). By contrast, statistical offices cannot compel sharing and instead must persuade companies that providing data serves mutual interests (A2). These broad data access rights enable wider use of data for statistical analyses and criminal investigations. However, laws also mandate that data can only be used for its originally collected purpose (A1, H1), and both the Statistics Act and Police Act outline strict rules governing data use. One interviewee from law enforcement (H2) illustrated these legal limits, saying:

Outside of statistics and law enforcement, legal frameworks are generally less conducive to BD use. Nevertheless, our healthcare case (K1) already employs BD at strategic, tactical and operational levels. Transportation and infrastructure providers have fully operational BD use cases (e.g. energy demand forecasting and asset failure prediction in C1 and C2). This extensive use of BD stems from the vast data volumes generated by sensors and the relatively less sensitive nature of the data involved (C1). Hence, early BD use concentrates in sectors with (a) lawful access to multi-source data; (b) sensor-rich operations, because these jointly raise data availability while keeping privacy risk relatively lower.

Last, BD use in social benefit processes is minimal, even though some believe that “the biggest benefits are in the social, in the benefits, in the subsidies” (G1). On the other hand, it was noted that “the biggest risks are within customs, benefits and so on” (G1). Multiple cases (B1, E1, G1, G2, H2) mentioned earlier government missteps in applying BD to social welfare. Those failures have increased public and political scrutiny (G2) and created internal resistance within departments (B2). Additionally, many existing processes function without BD, so new BD use-cases are rarely explored. One interviewee (B2) attributed this to a lack of incentives for innovation: “If they're not asked, if they're not cheered for by their bosses for trying something new, why would they try something new? There's no reward” (B2).

Data sharing between public institutions and private organisations is widespread. The exchange is managed in varying frequencies, ranging from ad-hoc requests (G1) to monthly bulk transfers (A1) and to continuous live data sharing (B1). Data are shared in various formats: as aggregated datasets (A2), compiled reports (H2) or even visual dashboards (B1). Notably, organisations often pseudonymise or anonymise data before sharing to protect confidentiality and privacy. Last, data sharing responsibilities can be centralised and handled by a dedicated unit (B1, G2) or decentralised to the responsible department (C1, K1).

Public organisations deploy several exchange structures to facilitate the sharing of data. One simple method is to grant stakeholders temporary access to an internal database (B1, D1). This approach avoids physical data transfers, thereby sidestepping potential data ownership disputes and technical challenges of moving large datasets. (D1) noted that this approach “has a lot of advantages given this size of the data. It's getting more difficult to physically move the data.” Another approach that avoids data transfer is data virtualisation. Unlike the traditional extract-transform-load (ETL) process, virtualisation leaves data at its source and provides real-time access to it (D1). This method reduces the risk of errors that can occur during data movement.

Public service organisations that require extensive BD exchange tend to use more sophisticated data-sharing structures. For example, law enforcement authorities use secure shared platforms for data access and exchange. Two other cases (C2, N1) used APIs linked to external databases to automatically pull in data. These approaches are often chosen for large or jointly owned datasets because they avoid the need to move data, reduce errors and keep clear audit trails. Last, government institutions are increasingly publishing open data to boost transparency and public trust (B1). They typically release this data through official websites or dedicated open-data platforms accessible to citizens and businesses. The key points of BD collection, management, use and sharing from all cases are summarised in Table 3.

Like BD, AI also lacks a universally agreed definition among respondents. In practice, most cases equated AI with machine-learning (ML) algorithms. This narrow view reflects organisational risk preferences: explainable, well-scoped models align more easily with audit requirements than broad, open-ended AI notions. One interviewee (N1) defined AI as “algorithms or techniques that enable machines to process data, learn from it and make big decisions with the human in the middle or moving towards more autonomous scenarios.” Proposed AI regulations also provide only broad descriptions rather than precise definitions. As a result, “in every discussion and interaction we come to understand each other's definition” (D2).

High-quality, sufficient-volume data is a prerequisite for effective AI development (C2, F1). These conditions are necessary to train reliable AI models and to prevent biased data from yielding undesirable outcomes, a particularly important concern in the public sector. Accordingly, responsible AI emerged as a major theme in our interviews. Several public organisations are taking measures to ensure their AI systems are socially and organisationally acceptable (D2, G2, F1, N1). For example, several government organisations have jointly developed AI registers to catalogue where and how AI is implemented, to monitor its effects on citizens (G1). N1 further provides an example of stepping up organisational enablers for responsible AI:

Furthermore, explainability is crucial for AI development and application. Public-sector AI algorithms must be explainable, because if results cannot be independently verified, they will not be accepted. “We have very high-quality standards, so it means when we use these methods, we have to be very sure about the quality of the results and that's what our users expect” (A1). Especially in the criminal justice sector, judges must trust the truthfulness of AI-generated results. If an algorithm's workings are not transparent, its outcomes are not admissible for use in legal proceedings (H2, 2021). The need for explainability holds true in statistical offices as well, since their outputs inform important policy decisions (A2). Consequently, most organisations have so far limited themselves to relatively simple ML algorithms. Only recently have some begun experimenting with more complex AI techniques – such as deep learning (for image recognition) (G2, H2) and natural language processing (for text and speech recognition) (G3).

Managerial caution and risk aversion significantly hinder AI application development in the public sector. To date, public-sector AI is predominantly applied in services that do not directly affect social welfare. This cautious approach stems from previous scandals that eroded citizen trust and heightened public administrators' awareness of AI's risks. For example, one municipality (B1) piloted an AI system to predict which citizens were likely to fall into debt and proactively reach out to them. Despite the pilot’s technical success, the project was ultimately stopped because “people at the department said, we don't really want that. Let the people just come to us.” Hence, where services are tightly coupled to individual rights (education, social benefits), goal setting and control structures raise the bar for fairness and contestability, delaying AI development. Contrary, in domains where AI's impact on citizens is minimal, concerns about public acceptance are less pronounced, making it easier to implement AI solutions.

Our findings indicate that AI application in public organisations is still in its early stages, with considerable variation between sectors. While many public institutions are still just experimenting with AI application-cases (A1, B2, D2, G1), some organisations already have fully operational machine-learning algorithms in their workflows (C1, C2, H2). These differences stem from several factors, particularly how much a given public service impacts citizens, which in turn drives the need for transparent and fair AI outcomes. Additional factors include the availability of BD, the organisation’s technical capabilities and management’s willingness to adopt AI.

For example, a criminal justice authority (H2) explained that they “don't have the choice because of the amount of data”, implying they must use AI to handle the volume. By contrast, a national healthcare provider (K1) is “using AI to forecast, to understand, to help the support of management decisions”, highlighting AI’s role in decision support. Other use cases mentioned were as follows: traffic flow prediction (G2), optimised asset management (C2) and automatic text detection to prevent website abuse (K1). However, up to now AI has been deployed only in a handful of processes that involve very large data volumes (G1, G3, H1, H2). Most local government services remain manageable with traditional (non-AI) methods (B1, B2).

Apart from a lack of immediate need for AI in some services, certain public-sector characteristics also limit AI application. First, because public institutions are required to deliver services transparently, they are constrained to use algorithms that are explainable and accountable. Goal setting in citizen-facing services further moderates AI adoption. Take education as an example: due to privacy concerns and the high stakes of AI-driven decisions on individuals’ futures, AI application in education remains exploratory. Moreover, proposed legislation will classify AI applications by risk level, with high-risk systems facing additional requirements:

As explained in our theoretical framework, the degree of publicness describes the extent to which an organisation is subject to economic and political influence, and whether it is constrained or enabled by political authority (Bozeman, 1987). To understand the advancement of technology-enabled service delivery, it is relevant to examine how political influence and other publicness dimensions facilitate or hinder data collection, management, use and sharing in these cases. Therefore, expert statements were thematically coded according to the four dimensions (ownership, funding, goal setting and control structure) and grouped under the main BD and AI activities. The influence level (low, medium, high) was then assigned through comparative assessment of severity and frequency. To strengthen reliability, the coding was cross-checked with insights from relevant literature (e.g. Seepma et al., 2021), ensuring that the classifications reflect both expert judgement and wider sectoral evidence. Table 4 provides an overview of how each publicness dimension influences BD and AI activities.

First, this study found that funding considerably influences public organisations' BD efforts. Most funding comes from state contracts or government grants, which often carry specific mandates or expectations on how the money is used. Furthermore, funding is tied to each organisation’s policy agenda, which in turn shapes what data is collected and how it can be used. Cases noted that current technology infrastructures often do not fully support BD initiatives without additional investment. However, “usually IT is not particularly hip. So, there's not really much money going into it” (B1). In addition, legal and regulatory control structures heavily influence BD efforts. For example, in law enforcement, specific laws empower police to request data from other organisations. Conversely, the same regulations also impose strict limits: authorities cannot collect data without a prior suspicion, cannot reuse data beyond its original purpose (H2), cannot combine certain individual datasets (H1) and in some cases must share certain data or results with other bodies.

Moreover, unlike private firms, public organisations must deliver services equitably, in addition to efficiently and effectively (Bozeman, 1987). Therefore, public managers need to ensure any AI applications produce equitable, bias-free outcomes. However, ensuring fairness and absence of bias is challenging, given the complexity and opacity of many algorithms (E1). Additionally, new mechanisms like AI registers (i.e. public listings of an agency's AI systems) are increasing accountability to political overseers. While this transparency is beneficial, it may also reinforce the already strong culture of risk aversion among public managers. Past AI missteps in social services have drawn intense media scrutiny, heightening ethical concerns and public distrust. In response, public organisations are now engaging ethics advisors (G3) and collaborating on joint frameworks (G1) to guide responsible AI implementation.

Across cases, the interaction of publicness dimensions channels the BD and AI pipeline in predictable ways. Funding and ownership shape where data capacity sits (central versus decentral), thereby setting the feasible scale of collection and management. Control structures (law, audit) determine how data may flow, and which AI models are acceptable, privileging explainable approaches. Goal setting calibrates risk tolerance by proximity to citizen rights: the closer the service is to individual entitlements, the higher the evidentiary and fairness thresholds. Together, these influences steer innovation towards infrastructure- and operations-facing use cases first, with citizen-facing AI lagging until compliance routines and safeguards reduce perceived risk.

The use of BD in the public sector has been described as limited (Guenduez et al., 2020) despite its promise for service delivery and policymaking (Kuziemski and Misuraca, 2020; Pencheva et al., 2020). This study finds that public organisations are committed to collecting, managing, using and sharing BD, though activities vary considerably across sectors as mandates, regulations and data environments shape both enablers and constraints. Transport applies BD for optimisation, statistical agencies use standardised datasets for precise calculations and law enforcement integrates sensitive data under strict oversight (H2). These variations show that BD maturity is not uniform but tied to functional needs, data availability and institutional capacity (Van Veentrsa et al., 2021).

Contradicting Constantiou and Kallinikos (2015) and Müller et al. (2016), this study finds collection largely use-case driven, enabled by legal authority (H2), public data (B2) and administrative expertise (G3). The scope and structure of exchanges differ by project or service. Yet organisations face major challenges from incompatible IT systems, technical debt and unstructured datasets (Moody et al., 2019). These system-level issues (Pencheva et al., 2020), linked to the “networked” nature of BD (Desouza and Jacob, 2017), hinder sharing across boundaries. As one interviewee noted: “We need more standardisation. […]. Because the data now is a cowboy world. It's not being standardised at all. Especially within government” (H2).

Organisational infrastructure is not the only barrier. Informed consent for public sector data integration is increasingly difficult, and linking datasets amplifies re-identification risks. The “computational turn” in public administration reflects the shift towards algorithmic methods (Felzmann et al., 2020). These approaches promise efficiency and predictive capacity but also bring opacity and ethical concerns. A key challenge is that systems often rely on complex rules that confound human comprehension and constrain action (Smith et al., 2023). Transparency must therefore be built into AI design from the outset to foster trust and enable responsible algorithm sharing.

Given these challenges, most organisations report AI applications remain experimental, with few fully operational systems. Still, awareness of AI's benefits is widespread, supported by knowledge exchange and collaboration through governmental coalitions (N1, D2). While evidence of sharing applications is limited, organisations acknowledge that “the models we build are multiply useable” (H2). Publicness plays a distinctive role, as government actors emphasise responsible and explainable algorithms to ensure accountability and citizen acceptance (Felzmann et al., 2020). Governance instruments such as AI registers and FAIR principles (findable, accessible, interoperable, reusable) are seen as pathways to visibility and trust (F1).

At the same time, organisations adopt a cautious stance given risk of bias and the high transparency demands of the public sector. Findings suggest explainable AI fosters trust among decision-makers, while human-centric AI strengthens citizen acceptance. Current practice therefore emphasises human–AI collaboration, where AI supports but does not replace professional judgement. As one case put it: “We never use AI to actually decide stuff. They're always supporting a professional … and that's kind of a basic law” (H2). This reflects both technical limitations and reluctance to delegate decisions (B1, H2, N1). Hence, AI is more often applied in technical or administrative settings with minimal direct citizen impact, rather than in socially sensitive services.

Ownership and control structures exert the strongest influence on BD and AI activity stages in public organisations. Ownership defines who has authority over data and technological assets, which in turn determines how collection and management are organised. Control structures add a further layer of influence by specifying the tasks, responsibilities and procedures for each organisation, determining with whom and to what extent they can collaborate (Seepma et al., 2021). Together, these dimensions ensure that public agencies cannot simply pursue technological opportunities for efficiency but must embed BD and AI within legal and ethical frameworks. This influence is particularly pronounced in the use and application of AI, where explainability, fairness and oversight are critical.

Funding also shapes how BD and AI are developed and applied, though its influence is more uneven across stages. Access to financial resources determines whether public organisations can invest in modern infrastructures, build analytical capacity and sustain innovation. Funding, however, is rarely neutral; it is often tied to political cycles and shifting agendas, which may restrict continuity and prioritisation of certain projects over others (Seepma et al., 2021). As such, funding can be both an enabler, by providing resources for experimentation and scaling, and a constraint, when tied to short-term political goals that disrupt long-term strategies.

Goal setting acts as the strategic dimension of publicness, ensuring that BD and AI initiatives align with wider societal objectives. Political and organisational mandates determine which data should be collected, how it should be used, and for what purposes AI applications are developed. This influence extends beyond strategy to values, embedding concerns for equity, inclusion and public trust into the design and deployment of technologies (Senadheera et al., 2024). In practice, goal setting directs resources, making it a key driver of whether BD and AI are used for efficiency-oriented reforms, equity-focused interventions or regulatory enforcement. Figure 2 illustrates these influences of Publicness on BD and AI activity stages.

The dimensions of publicness interact in ways that jointly shape organisational behaviour. Ownership influences who provide resources and how funding is allocated, which in turn constrains or enables the types of goals an organisation can pursue. For instance, publicly owned entities financed by tax revenues tend to emphasise equity, accessibility and long-term public value, while private or mixed-ownership organisations often prioritise efficiency and financial sustainability. Funding streams, whether public, private or hybrid, also feed back into goal setting by determining the scope of activities and the level of risk organisations are willing to assume.

Control structures tie these relationships together by ensuring that ownership and funding are channelled towards legitimate goals. Regulatory and accountability mechanisms enforce transparency and public oversight, guiding how organisations balance competing priorities. Where control is strong, it aligns ownership and funding arrangements with socially desirable outcomes, whereas weaker or fragmented control can create inconsistencies across sectors or jurisdictions. Taken together, these interactions show that the publicness dimensions form a dynamic system in which ownership, funding, goals and control are mutually reinforcing and collectively determine the degree to which organisations act in the public interest (Figure 3).

Although the interviews were conducted in 2022 and BD volumes have since grown, the findings of this study remain highly relevant and resonate with recent literature. First, the creation and maintenance of an appropriate technological architecture and ICT infrastructure, as well as an increase in the scale of its storage, analysis and transmission is emphasised in literature (Yukhno, 2024). This study identifies similar challenges, particularly legacy IT systems and incompatible databases. Yet, unlike the broader literature, it emphasises that today's data infrastructure remains constrained mainly by limited funding. At the same time, the fact of having BD does not automatically imply that it can be used to solve real problems. Bringing together technical experts, operational professionals and management with ethical and legal expertise is essential (Latupeirissa et al., 2024). Achieving this in practice is complex and requires new organisational structures, processes and routines, underscoring the importance of both technological and organisational developments (Selten and Klievink, 2024).

In addition to the increase in BD, AI systems experienced unparalleled advancements in recent years. Nevertheless, Berman et al. (2024) confirm this study's finding that efficiency gains in AI adoption are still marred by transparency, interpretability and stakeholder engagement challenges. This demonstrates that data-driven automated service delivery continues to face both technological and organisational barriers in public settings. In addition, research found that the importance of technological and organisational factors in AI adoption shifts across stages: in early stages, technical readiness (e.g. infrastructure, data availability) is most critical, while in later stages, organisational aspects like leadership, staff skills and governance become decisive (Neumann et al., 2024).

In regard to the latter, this research supports the findings by Ahn and Chen (2022) who revealed that the willingness to implement and use AI technologies in government was contingent upon a series of positive and negative perceptions about the new technologies as well as the long-term outlook on the role of AI technologies in society. Here, especially the scandal by the Dutch tax authority which used an algorithm to spot suspected benefits fraud overshadowed the potential benefits and lead to great risk aversion among public organisations.

Moreover, economic pressures for efficiency often overshadow the need for ethical considerations and stakeholder involvement, highlighting the strong influence of funding (Berman et al., 2024). This is surprising given the government's role in producing outcomes that matter to the public (e.g. trust, equity, well-being), not just efficiency (Madan and Ashok, 2023). Nevertheless, the findings are in line with this study and highlight how funding influences the adoption of AI systems but also, how critically adequate funding is for developing responsible AI systems that meet the needs of internal stakeholders and citizens. In contrast to this study's emphasis on funding, Neumann et al. (2024) attribute AI adoption success mainly to internal capabilities and leadership, a finding shaped by their Swiss context outside the scope of the EU AI Act, classifying AI systems by risk and imposing strict requirements on high-risk applications, including data quality and human oversight.

Research by Hjaltalin and Sigurdarson (2024) found that while efficiency and service delivery dominate the discourse, citizen engagement remains underemphasised. This finding is equally prevalent in this study, highlighting that public organisations focus on AI applications with less proximity to citizens which might be attributed to less structured data availability or concerns over public backlash. Most common use cases for AI applications in local government citizen engagement are chatbots which can enhance outreach but also pose ethical concerns including accuracy, accountability and exclusionary assumptions (Senadheera et al., 2025).

Hence, the AI-driven transformation and smart data management for the public sector are primordially characterised by an operational change, which includes human resources and know-how as the spearhead (Valle-Cruz and García-Contreras, 2023). Therefore, recent literature highlights the need for improved AI literacy among both management and personnel to improve their understanding and perception about the new technologies as well as their potentials in government that will foster a culture of innovation towards sustainable and impactful AI integration into public decision-making processes (Ahn and Chen, 2022; Berman et al., 2024). Moreover, integrating data science teams into already existing operational departments improves the alignment between AI and primary processes (Selten and Klievink, 2024). This studies' recommendations are outlined in the following chapter.

The study generated several actionable insights for policymakers, public administrators and practitioners seeking to harness the benefits of BD and AI while navigating the constraints of publicness. In practice, organisations may consider a phased approach: first, secure leadership support and strengthen internal data governance foundations; second, foster a transparent and responsible AI culture; and finally, advance digital education and public engagement to build long-term capacity.

First, it is essential for public sector executives to develop a nuanced understanding of the time, technical complexity and resource demands involved in BD and AI implementation. This includes recognising the challenges faced by data science teams, such as data cleaning, algorithm development and system interoperability, and taking proactive steps to enable their success. Executive support should prioritise cross-functional collaboration and iterative evaluation mechanisms to ensure the responsible and effective adoption of these technologies. As Mergel et al. (2016) emphasised, the successful integration of BD in public administration depends not only on technical readiness but also on leadership capacity and inter-organisational coordination, the same also applies to AI.

Second, enhancing organisational data visibility and governance is essential. A prevalent issue identified in this study is the limited awareness and coordination regarding data assets within public institutions. Many agencies lack knowledge of the data they possess or the formats in which it exists, leading to inefficiencies in data reuse and delays in sharing. To address this, public organisations should implement robust metadata management systems, develop organisation-wide data inventories, and appoint dedicated roles such as data stewards or strategists to oversee data quality and facilitate cross-departmental coordination (Van Donge et al., 2022). Encouraging a networked understanding of data and AI, where systems are viewed as interconnected across government levels and agencies, helps break down silos and promotes more integrated, equitable outcomes. Prioritising standardisation and interoperability are crucial, supported by IT architectures that enable seamless and secure data exchange.

Third, fostering a responsible and transparent AI culture is essential in public settings, where the stakes for accountability and fairness are high. Training and awareness initiatives should target all organisational levels, including legal, operational, managerial and technical staff, to build a shared understanding of ethical and governance imperatives. Institutions should adopt tools such as algorithm registers, explainability assessment frameworks and AI risk classification schemes to promote transparency and align AI applications with public values. While transparency alone is an inadequate ideal for achieving accountability, it is advised to look across a systems relational aspects, rather than simply inside a black box, for developing more effective algorithmic accountability (Ananny and Crawford, 2018).

Fourth, public engagement is vital for building long-term capacity and public sector readiness for data-driven innovation. Effective strategies should inform citizens about data collection and use, provide clear consent mechanisms and enable feedback or redress. In an era of growing state datafication, this requires a visible, accessible and ongoing institutional channel for consultation and oversight (Smith et al., 2023). Such mechanisms deepen accountability, build trust and ensure that public sector data use reflects the futures desired by diverse communities. As Van Noordt and Misuraca (2022) highlight, inclusive governance and digital skill-building are essential to realise AI's transformative potential while safeguarding trust and accountability. Citizen participation in the design and oversight of AI systems further strengthens legitimacy, while transparency, explainability and human oversight remain critical for responsible implementation (Wirtz et al., 2019). Ultimately, fostering a responsible AI culture is key to protecting digital rights and preventing algorithmic harm.

These implications highlight the need to align BD and AI with the institutional context and societal responsibilities of the public sector. Rather than viewing them as stand-alone technical tools, they must be embedded in governance systems shaped by leadership, ethical safeguards and inter-agency collaboration. Their value will depend not only on technical capacity but also on trust, accountability and citizen participation. As governments continue to navigate the complexities of digital transformation, adopting a citizen-centred approach to BD and AI will be essential to ensure that technological advancements contribute meaningfully to public service improvement and democratic accountability (Kankanhalli et al., 2019). Ultimately, governments that adopt a rights-based, citizen-centred vision can turn BD and AI into drivers of fairness, empowerment and democratic renewal, rather than risks to them.

While the study offers useful insights, some limitations should be noted. First, the research focuses on a diverse sample of public sector organisations, each with different degrees of publicness and legal context. This variation, while enriching the analysis, also limits the direct comparability and generalisability of findings. Nonetheless, the diversity of cases enabled a more comprehensive exploration of how publicness manifests across sectors and governance settings.

Second, the expert interviews were conducted just before the rapid rise in AI applications, following OpenAI's public chatbot release in late 2022, a key turning point in mainstream adoption. Despite this early examination of AI adoption stages, the study provides valuable insights into foundational opportunities, concerns and organisational thinking that continue to shape current developments. The core insights and recommendations remain highly relevant, and future research should continue to monitor evolving AI applications in public service delivery.

Third, the study does not include direct performance measures to evaluate the outcomes of BD and AI initiatives. Although interview data revealed perceived benefits in terms of cost-efficiency, accuracy and decision support, these remain anecdotal. Future research should aim to empirically assess the relationship between BD–AI integration and public sector performance, using measurable indicators such as service timeliness, resource utilisation or citizen satisfaction. Such work could also explore the trade-offs between quantitative metrics and qualitative values like equity, justice and procedural fairness.

Lastly, future studies could adopt comparative approaches to deepen understanding of the institutional factors that influence digital innovation. Longitudinal research tracking the evolution of BD and AI across time, or cross-national studies examining policy variation, could provide further insight into the systemic and cultural dimensions shaping public technology adoption.

This study examined the development and use of BD and AI in the public sector, with a focus on how the distinctive characteristics of publicness – namely ownership, funding, goal setting and control structure – influence the adoption, implementation and integration of these technologies. Drawing from extensive empirical evidence across 17 European public organisations, the study found that BD is increasingly embedded in public service delivery, albeit at varying levels of maturity, while AI remains largely exploratory and constrained by institutional and societal factors.

A core finding is that effective BD use depends heavily on structured, inter-organisational data sharing. Public organisations that successfully integrate BD into their operations often benefit from clear legal mandates, established governance protocols and standardised data management practices. Nevertheless, many still face persistent challenges, including fragmented legacy systems, lack of interoperability, technical debt and data invisibility across the public service supply chain. These issues hinder the timely and efficient sharing of data within and between government levels, and between public and private entities. Promising efforts, such as the introduction of data strategists, open data platforms and virtualised access solutions, illustrate emerging approaches to overcoming these barriers and enhancing visibility and reuse of public data assets.

Unlike BD, the development and use of AI in public institutions is constrained by higher levels of caution and risk aversion. This caution is especially evident in domains where algorithms interact directly with citizens or make decisions with ethical or social implications. In such contexts, demands for explainability, transparency and human oversight are non-negotiable. The study found that public organisations are actively pursuing responsible AI principles, often implementing AI registers, algorithm governance boards and internal guidelines to ensure that emerging systems meet public standards for accountability and fairness.

Nevertheless, AI projects are largely confined to technical domains, such as infrastructure monitoring, energy forecasting and traffic management, where the impact on individual citizens is indirect. In contrast, attempts to introduce AI in socially sensitive domains like welfare, education or healthcare are often abandoned or delayed due to concerns about fairness, bias and social backlash. This is compounded by past scandals involving algorithmic discrimination, which have left lasting impressions on public trust and institutional reputations.

These findings highlight that the concept of publicness is central in understanding both the opportunities and limitations of BD and AI in government contexts. Funding structures influence the capacity of organisations to invest in the infrastructure required for data-intensive technologies, while political goal setting and control structures shape how data are collected, shared and governed. Where public value mandates are strong, organisations tend to prioritise transparency, accountability and social outcomes over speed and efficiency. This creates both safeguards and constraints on innovation. By demonstrating how these dynamics condition the adoption of emerging technologies, this study extends publicness theory into the domain of digital governance and underscores its relevance for contemporary public administration knowledge.

In addition, the study points to a possible reciprocal dynamic. While publicness appears to condition the adoption and governance of BD and AI, the use and management of these technologies may also begin to reshape the degree of publicness within organisations. For instance, by potentially shifting ownership arrangements, funding dependencies or accountability structures. This suggests that publicness could be understood not only as a contextual condition but also as an evolving outcome of digital innovation. Future research may examine this reciprocal relationship more explicitly.

To conclude, this study affirms that BD has gained meaningful traction in the public sector, particularly in statistical analysis, law enforcement and infrastructure management. AI, by contrast, is still in its relative infancy, and its trajectory will likely remain shaped by evolving legal frameworks, social norms and political priorities. As such, efforts to adopt BD and AI in public service delivery must be grounded in the realities of publicness, ensuring alignment with democratic values, institutional responsibilities and the lived experiences of citizens.

A1 is a Dutch statistics provider that delivers quantitative information for public debate, policy development and democracy. It operates as an autonomous administrative authority financed by the state. The organisation has existed for over 100 years and has a long tradition of innovation and new technology adoption. A2 is the Belgian statistics office responsible for collecting and disseminating reliable data about the country, the economy and society. It operates under the Belgian Statistics Act, emphasising simplification, privacy protection and utility for academics. A2 has a modern mandate to improve production processes and embrace Big Data opportunities. Of the roughly 100 statistics it produces, around 75% are legally mandated by the EU.

B1 is a mid-sized Dutch municipality serving more than 200,000 inhabitants. It is responsible for 30–40 data-related processes such as waste, housing, welfare and taxation. Known for its cycling culture, the city is actively involved in monitoring and improving the flow of pedestrian and bicycle traffic through a dashboard that feeds into internal workflows. B2 is a larger Dutch municipality with more than 650,000 inhabitants. It operates across both physical and social domains and is obligated under the Open Government Act to provide open and accessible government information. The city collaborates with private, academic and civil society actors within a quadruple helix model to deliver services and improve operations.

C1 is a German railway energy provider responsible for the 8,000-km electric network of the national railway infrastructure. The organisation focuses on energy supply, transition to renewable sources and electrification of the railway system. It is involved in the digitalisation of energy infrastructure and energy monitoring systems. C2 is a Norwegian railway infrastructure company tasked with maintaining and developing the national railway infrastructure. It oversees track and station development projects, traffic control and investment strategies. The organisation embraces digital technology to plan and optimise infrastructure and collaborates with public agencies and ministries.

D1 is a Dutch vocational and professional education provider offering courses in technology, healthcare, business and the creative sector. The institution educates about 13,000 students and works with industry partners to ensure relevance. It has implemented digital learning environments and learning analytics. D2 is a German university with a strong focus on applied sciences. It operates in an industrial region and offers programs tailored to regional economic demands. It has developed collaborations with surrounding municipalities and regional industry to improve data science education and municipal innovation.

E1 is a Dutch public–private consulting firm focused on digital government. It delivers research and consultancy on digitalisation, ethics and innovation in the public sector. The organisation works with municipalities, national agencies and academic institutions. E2 is an international development organisation based in Brussels, focused on improving public administration in Europe. It organises capacity-building programs, policy exchanges and training on e-governance and digital transformation. It works with EU institutions and national governments, offering frameworks and benchmarks for modernisation.

F1 is a Dutch provincial government responsible for spatial planning, regional economy, nature and culture. It coordinates with municipalities and the national government on regional policy and public services. The province manages open data platforms and regional dashboards. It supports data-driven policymaking, especially in sustainable mobility and spatial development.

G1 is a Dutch police organisation that supports intelligence-led policing. It develops data systems and coordinates innovation strategies across national and regional police units. The organisation is committed to ethical frameworks for AI and analytics and engages with public stakeholders on technology use. G2 is a Belgian fire and emergency services agency covering a region of several municipalities. It uses data systems to manage resources, monitor performance and optimise operations. The agency collaborates with regional governments and is involved in international knowledge exchange.

H1 is a Norwegian public health organisation responsible for coordinating regional health services, hospitals and digital health infrastructure. It is involved in implementing electronic health records, patient portals and telemedicine. The organisation works with national authorities to standardise digital solutions and ensure data security. H2 is a German university hospital operating as a centre for medical treatment, research and education. It implements clinical data systems, research platforms and smart hospital technologies. It partners with technology providers and academic institutions to advance healthcare innovation.

I1 is a Nordic national defence organisation responsible for cyber defence and digital infrastructure security. It develops strategic capabilities for cybersecurity and cooperates with NATO and EU agencies. The organisation is involved in civilian-military cooperation on critical infrastructure protection. I2 is a political innovation unit within a Dutch ministry that focuses on digital transformation and public innovation. It supports experimentation, new policy approaches and human-centred design in government. The unit operates semi-independently and connects different levels of government to foster systemic innovation.