2.1. Data for addressing societal challenges

Government organizations become increasingly acquainted with adding data and analytics to the evidence base on which decisions are made in different steps of policymaking (van Veenstra et al., Reference van Veenstra, Grommé and Djafari2020). Examples of such data use can for instance be found in the domains of public health (Danemayer et al., Reference Danemayer, Young, Green, Ezenwa and Klein2023; Urbano et al., Reference Urbano, Bartolomucci and Azzone2024), mobility (Liu and Dijk, Reference Liu and Dijk2022; Martinazzo, Reference Martinazzo2023), and social policy (Kersing et al., Reference Kersing, van Zoonen, Putters and Oldenhof2022; Dietrich et al., Reference Dietrich, Malerba and Gassmann2024). It is argued that data can help to understand the issue at hand, estimate the effects of interventions, arrive at decisions that contain less uncertainty, and ultimately achieve better policy outcomes that can also be monitored and evaluated with the use of data (Giest, Reference Giest2017; Head, Reference Head2022). In the execution of policy, it is believed that making decisions based on data and algorithms can improve efficiency and remedy favoritism that might inadvertently be introduced by prejudiced individual human judgments (Kersing et al., Reference Kersing, van Zoonen, Putters and Oldenhof2022; Rij et al., Reference Rij, Dekker and Meijer2024). At the same time, the extent to which value can be derived from data is highly dependent on the capacity of decision makers to comprehend it and weigh it against other forms of information (Giest, Reference Giest2017).

In this article, we zoom-in on the use of data in the public sector in relation to decision making about societal challenges. The principles about data informed policymaking as outlined before still hold in this context, but the dynamic becomes more complex because addressing a societal challenge is not contained to work being done in one governmental department or organization. Klijn (Reference Klijn2002) outlines three characteristics of the complexity in decision making on societal problems.

The first concerns knowledge uncertainty, meaning that there is a lack of knowledge about both the nature of the problem as well as the right way to solve the problem (Klijn, Reference Klijn2002). Over the past 20 years, the data revolution has significantly increased opportunities for attaining knowledge about a problem through data, but even when these data or information are available, it is scattered among a large set of actors that each may hold relevant information (Head, Reference Head2022). When information is fragmented, it is desirable, or even necessary, to set up collaboration between actors to arrive at a proper information basis for decision making (Alford and Head, Reference Alford and Head2017).

This relates to the second characteristic, namely institutional uncertainty (Klijn, Reference Klijn2002). Large societal challenges do not fit well in existing organizational networks. Societal problems can span multiple policy arenas, and each arena in turn can contain actors of different networks. It takes time and effort to find out where certain data and information is present, and establish arrangements in which this is shared between actors (Alford and Head, Reference Alford and Head2017).

Setting up such arrangements can be complicated by the third characteristic of complexity, namely strategic uncertainty (Klijn, Reference Klijn2002). Different actors can have conflicting values and interests (Head, Reference Head2022), for example, because they have diverging mandates or responsibilities. Actors may even choose to strategically deploy control over to their data, in order to gain an advantage in protecting their interest (Alford and Head, Reference Alford and Head2017). Cumulatively, these three characteristics outline key complexities in decision making for addressing societal challenges, which have implications for the data ecosystems in which these actors interact and collaborate with each other.

2.2. Data ecosystems for policymaking

Oliveira et al. (Reference Oliveira, de Fátima Barros Lima and Farias Lóscio2019) define data ecosystems as “socio-technical complex networks in which actors interact and collaborate with each other to find, archive, publish, consume, or reuse data as well as to foster innovation, create value, and support new businesses” (p. 589). Propensities of data ecosystems typically include intentionality, value creation, dynamism, demand–supply relationships, and sustainability as well as evolution (Harrison et al., Reference Harrison, Pardo and Cook2012). According to Heimstädt et al. (Reference Heimstädt, Saunderson and Heath2014) actors may play the following roles in a data ecosystem: data suppliers, data intermediaries, and data consumers. Following Harrison et al. (Reference Harrison, Pardo and Cook2012) that mention an additional role of a central actor “taking the initiative […] to achieve specific goals” (p. 907), this initiating and coordinating actor is termed convener of the data ecosystem by Susha et al. (Reference Susha, van den Broek, van Veenstra and Linåker2023).

As data ecosystems are sociotechnical systems (Bartolomucci and Bartalucci, Reference Bartolomucci and Bartalucci2024), both technical and social elements are required for their functioning. Technical elements that are needed to facilitate data sharing within the ecosystem include data sources, the presence of a (data) infrastructure, (digital) resources, services and applications, software, and standards (Shin and Choi, Reference Shin and Choi2015; Oliveira et al., Reference Oliveira, Oliveira, Batista and Lóscio2018, Reference Oliveira, de Fátima Barros Lima and Farias Lóscio2019). Social and organizational elements are the social, cultural, legal, and ethical context of a data ecosystem. Moreover, participants that share data and broader (societal) stakeholders play an important role within the data ecosystem. Additionally, the skills and expertise of participants and stakeholders are of importance (Zeleti and Ojo, Reference Zeleti and Ojo2017; Oliveira et al., Reference Oliveira, de Fátima Barros Lima and Farias Lóscio2019). Between all these elements, interactions and dependencies can be identified.

Data ecosystems are often embedded in a specific context (Harrison et al., Reference Harrison, Pardo and Cook2012). Depending on the purpose, scope, and level of maturity of the data ecosystem, the way organizations work together can vary (Lis and Otto, Reference Lis and Otto2021). It is becoming more common that data ecosystems are set up with the aim to create, store, and share data to support policy or decision making more structurally (Lanza, Reference Lanza2021). Data ecosystems are considered to be more sustainable than interorganizational information sharing (Susha and Gil-Garcia, Reference Susha and Gil-Garcia2019) and may evolve intentionally from data collaborations (Susha et al., Reference Susha, van den Broek, van Veenstra and Linåker2023).

Literature has identified success factors and barriers for data ecosystems. These include technical barriers such as the lack of technical knowledge and resources to maintain the ecosystem (Dawes et al., Reference Dawes, Vidiasova and Parkhimovich2016; Oliveira et al., Reference Oliveira, de Fátima Barros Lima and Farias Lóscio2019; Diran and van Veenstra, Reference Diran and van Veenstra2020). Moreover, there are also a number of organizational and network barriers that are of influence to the success of the data ecosystem, such as the high complexity of tasks, lack of actor participation and interaction, a lack of organizational structure, unclear division of roles and responsibilities, data literacy of stakeholders, data ownership, fragmentation of data, and data privacy (Oliveira et al., Reference Oliveira, de Fátima Barros Lima and Farias Lóscio2019; Acar et al., Reference Acar, Raes, Rosseau and Satta2021). Also, “lack of clarity in terms of roles and responsibilities and turf issues (non-cooperation between organizations with seemingly common interests) have been identified as important challenges” (Susha and Gil-Garcia, Reference Susha and Gil-Garcia2019, p. 2899). It is yet unknown whether primarily technical or organizational factors have a more significant role in the success of data ecosystems that address societal challenges, and how these play out in practice.

4.1. Use the policy question as guide to articulate data demands

Both cases experienced that it is important to have a central demand or policy question at the heart of the data ecosystem that is derived from the societal issue. This policy question can serve as the glue that holds the data ecosystem together and guides the next steps that need to be taken. As a respondent notes: “data must be at the service of the social task” [V3], and having a clear research goal helps to define what data (sources) are necessary to gain the required insights that can bring the social task forward. Working with this approach requires a reversal from a supply driven to a demand driven mindset. It was stated that when the demand is not clearly articulated, collaborating parties will commonly ask for as much data as they can possibly receive, in which case, the availability of the data becomes leading in the development of new knowledge. However, when you spend more time to figure out which question people wish to answer, it becomes evident that certain data are not needed. A demand driven approach serves as a filter for assessing which of the data sources are relevant to use and provides an incentive to go the extra mile to try and unlock data that are not readily available but would provide valuable observations.

The ability to articulate a clear question is a capability that both data ecosystems cultivated over time. In ZoO, it was first experienced that “it was not possible to conduct the data analysis in one or two weeks. […] As a result it became clear that a big question first needs to be narrowed down and further articulated before you know which data source you can use to answer the question” [Z1]. They found that there are often several questions that arise at the same time, which must be brought together. Because of this experience, ZoO established a way of working to do so, where they gather input from all partners to formulate initial research questions, and then hold thematic meetings to refine and prioritize the questions. Once this is completed, it is explored how feasible it is to do the data analysis and afterward data preparation is started. ZoO found that this process helps to make it easier to translate a question toward data analysis.

VIVET also found that it was a challenge to formulate good questions. One of the complicating factors was that, at the onset of VIVET, the energy transition field was very new and dynamic and many things were yet to be explored and defined. A tool that they experienced to be helpful is a vision document. Such a document paints a picture of how societal partners, in this case decentralized governmental bodies such as municipalities, envision the future. From there, it is possible to distill data demands and work backward. In general, VIVET experienced that reasoning along the lines of a societal challenge helped to get things moving and made it easier to establish data sharing.

4.2. Invest in a common language and shared definitions

The complexity of data projects and societal challenges requires that policy and domain expertise, data expertise, and legal expertise—across several institutions—come together in the initiative. This indispensable collaboration between multidisciplinary actors in a data ecosystem makes it a necessity to invest time and energy in making sure that all participants understand each other well. Interviewees remark that people with data expertise approach things differently than policymakers. It is therefore not always easy to establish a common language: “Dedicate a lot of attention to what are we exactly talking about? How do we define crime? When you zoom in it quickly becomes more complicated” [Z2]. It is recommended to make room to ask and answer such detailed questions, because, as an interviewee warns, otherwise there is a chance that information is produced that people do not agree on and that everyone interprets in their own way.

Investing in shared definitions is not only necessary so that participants have the same understanding about what a concept entails, but it also impacts whether data can be connected and compared. One interviewee presented a detailed account of how the process for establishing shared definitions took place in the VIVET case, to harmonize data about solar energy and wind energy on land. Several parties publish relevant information about these concepts, but their data are not always the same. This causes a problem because it leads to discussions about the validity of the different numbers, rather than about the actions that should be taken on the basis of these numbers. VIVET chose to appoint an independent project manager to facilitate the harmonization of these data, because all other parties had a particular interest in the outcome. The process was started by taking stock of the various definitions that were used by the different parties and for what reason. In some cases, there was a good argument for using a particular definition, for example, because this was a requirement based on European legislation. This inventory was used for a kind of negotiation between the different parties. A proposal was made for which definitions would be best to use, based on how well the definitions matched with policy documents. Then there was a discussion about whether or not it would be possible to use these definitions and what advantages or disadvantages this choice entailed. From that, a conceptual framework was created, which was validated over the course of a year. This way of working was appreciated and they want to return to it in the future when they need to harmonize new concepts.

4.3. Secure capacity through managerial support

To be able to bring together all the necessary expertise in a data ecosystem, sufficient capacity, that is, the availability of experts with required skills, must be secured. Managerial support is highlighted as a crucial factor that impacts whether this capacity is available. As an interviewee states: “It is important to manage expectations and secure managerial support, and therefore also the required resources. Freeing up people in the organization who can participate. [..] Without this managerial support, no extra capacity will be created. Without created capacity, you miss out on quality and expertise and you are in fact dependent on yourself, which is too weak for such a major development” [Z1]. When sufficient capacity and budget are available, it not only makes collaboration possible, but it also becomes easier for a data initiative to solve technical challenges, such as transforming data, improving data quality, and keeping data up-to-date.

Data ecosystems addressing societal challenges are vulnerable to fluctuations in managerial support, because whether societal issues remain high on administrative agendas is dependent on various factors, such as the political climate, notable events, and whether new urgent issues emerge. ZoO experienced that when the initiative was started, the topic of subversive crime was high on the agenda. However, as time passed, it slowly faded away and other issues were started to be seen as more important. This made it hard to keep people involved who initially said yes or signed an agreement, and had a direct impact on capacity in the data ecosystem: “maintaining administrative commitment is a very difficult one and it has had an impact on project leaders, who were overshadowed by their department heads with more work. Many project leaders [indicate that they] would like to be able to continue contributing, but just don’t have the time for it anymore, and do it more on the side. [… It is] very sad because most project leaders have good will, are really interested in this subject, […] but often do not have the time to seriously devote time to this” [Z1].

4.4. Foster collaboration through shared interests and expectation management

This narrative emphasizes the importance of managing the interpersonal relations within the collaboration. It is crucial for fruitful collaboration that people develop respect with and for each other. Especially when working together with different parties and different types of experts, it is necessary that people understand each other and respect each other’s varying positions. It helps when it is made evident what people’s expectations and interests are of participating in a data-sharing program, so that all participants know why certain parties are involved and what their priorities are.

The ZoO case illustrates how expectation management is part of this process. As soon as it becomes clear that parties are involved in the data ecosystem with expectations that deviate from what is previously been decided or what is in scope, this should be addressed. When the positions of the partners are clear, it can become evident that work needs to be done to align the interests. In ZoO, there was some frustration at the beginning of the collaboration because some parties had made inaccurate assumptions about the way of working of another party. An interim review and separate work session were organized to resolve this issue, with the goal of providing clarity about the interests of the various collaborating parties. Spending time on this subject created stability and mutual respect.

A successful way in which ZoO managed to create shared interests is that when a question was articulated and selected for further analysis, they took the approach of answering the question in such a way that more stakeholders could use the data. To illustrate this, if one municipality posed a question, the analysis was done for every municipality in the Netherlands. This led to new insights for the entire data ecosystem working on tackling subversive crime and not just for one specific stakeholder. Interviewees see this as one of the main success factors of the ZoO data ecosystem. Other factors that made collaboration a success and were briefly mentioned by interviewees are that working groups were kept small and that it was decided not to keep pushing when the collaboration met resistance.

Because the interest of the collaboration sometimes needs to be prioritized over the interest of the participants’ institution, flexibility of participants is required. This is emphasized by V2: “An important lesson learned is the flexibility that’s necessary […] How you can prioritize the interests of the collaboration over the priorities of your organization” [V2]. VIVET observed that parties were sometimes reluctant to share their data, but the urgency of the energy transition helped participants to take on a more flexible attitude than they might otherwise. Putting the societal challenge at the center helps people to see sufficient need for a particular activity so that they are willing to invest in it. “It is mainly about trust and that parties reason from the perspective of the social task and no longer from their organization. Then things start moving” [V1]. This sense of urgency can best be created in small work groups and then flow upward, as is elaborated further in Section 4.5. Moreover, an interviewee noted that flexibility is about allowing yourself space for learning together and keeping ego’s at bay. “You have to allow yourself to say ‘ok, that didn’t go so well, but now we will go in this direction’” [V2].

4.5. Organically create structured ways of working

In order to create more structure, both data ecosystems established a system to formalize the ways of working and governance of the data ecosystem. For example, an interviewee from VIVET mentioned that “we do not share data on a physical platform. We work on the organizational conditions that make it possible to share data” [V1]. Moreover, interviewees underlined that clarity about organizational structure is necessary. However, one interviewee also stated that “the organizational structure should emerge in an organic way that appeals to all participants” [V2]. This is how VIVET emerged from a bottom up manner. They realized that improving the information provision for the Energy Transition initiative would not start top-down. Therefore, the parties involved in VIVET saw the need to do this themselves.

One aspect that can contribute to achieving structure is that someone takes up a convener role and coordinates between different participants of the ecosystem. This role was present both within VIVET and ZoO. In VIVET, there was a program manager appointed that coordinated between different participants of the ecosystem. In ZoO, a ministry took on a facilitating role and there were two project leaders. Moreover, both data ecosystems established a formal meeting structure with different moments for meetings on different management levels, such as an interdepartmental meeting and an intergovernmental meeting. In ZoO, the core team worked in an agile manner.

Furthermore, as part of structuring processes in the data ecosystem of ZoO, a process description was developed. The process description consists of the 13 steps that detail in each step which type of roles should be involved. Important steps in this process are: (1) demand articulation, such as the prioritization and sharpening of questions and examining the feasibility of answering questions with required and available data; (2) data preparation and analysis; (3) validation of the analysis and output; (4) publication of the results and underlying data. This process description helped to create clarity on when and which stakeholders need to be involved in the process and made the established way of working explicit by outlining when certain steps need to be undertaken.