Creating value through data collaboratives

2.1Data collaboratives

The term ‘data collaboratives’ is relatively new, though the phenomenon is not necessarily novel. We use the definition of data collaboratives provided by Susha et al. (2017a, p. 2691): “cross-sector (and public-private) collaboration initiatives aimed at data collection, sharing, or processing for the purpose of addressing a societal challenge”. This definition puts data collaboratives at the interface between (a) data-driven innovations; (b) inter-organisational information exchange and collaboration between organisations (government and others) (Gil-garcia et al., 2007); and (c) collaborative forms of governance, via partnerships or collaborations (Klievink et al., 2016; Lowndes & Skelcher, 1998) or via other forms of networked governance (Ansell & Gash, 2007; Mcguire et al., 2010). As a mode of data-related innovation, the collaborative nature of these initiatives also links them to the paradigm of inclusive innovation approaches, which we will briefly touch on later in this section.

At the core, data collaboratives are about combining data and collaboration to enable creation of new value, beyond the immediate goals, capabilities and logics of each of the participating actors. In that sense, the interests of data collaboratives are much wider than merely data standards and interoperability, though those are clearly important. As a field of study, data collaboratives is relatively new, with most existing research largely conceptual or illustrative.

Following our definition, data collaboratives are initiatives related to the goal-directed collection, sharing or processing of data. The data come from somewhere, but can be of various types (e.g., on people or phenomena) and in different forms (e.g., in words or transactions); it can be specific or not, structured or unstructured (Susha et al., 2017a). Often, open government datasets are quite specific and structured, as they come from specific administrative departments or levels and originate from specific administrative processes, activities or sources. The less specific or structured a dataset is, the harder it is to provide the necessary metadata and to create a useful dataset (Zuiderwijk, 2015). As data collaboratives are initiatives in which data is collected, processed and shared by various organisations, even from different sectors, the data they use can be quite diverse. As Susha et al. (2017a) observed, this also implies a significant diversity in data providers, which collaboratives must take into account, given the inherent multiplicity of interests and incentives to join and share data within a data collaborative (Klievink et al., 2016). This diversity of data and actors is precisely what makes data collaboratives valuable. By combining their different capacities for data generation, analysis and processing, the data, goals and expertise of the various partners in the collaborative are brought together to create something none of the individual parties could create alone. Or, a problem might be addressed that none of the individual parties could tackle alone.

2.2Tensions that seem inherent to data collaboratives

Our above conceptualisation of data collaboratives shows that our view on data collaboratives is one of a collaboration and innovation phenomenon rather than a data phenomenon, at least for the purpose of this article. As such, the challenge of setting up such a collaborative is more of an organisational and incentivisation nature than it is about data standardisation or processing. In the literature, two ‘paradoxes’ can be found that illustrate the complexity of the interactions among partners in a data collaborative. These are the aforementioned goals paradox and paradox of simultaneous control and generativity.

First, the goals paradox refers to the fact that the involvement of diverse actors is a source of both value and conflict (Vangen & Huxham, 2011). Data collaboratives bring together a certain breadth of parties. Indeed, this serves to make the collaboration valuable, as each has different pieces of information, different perspectives, complementary expertise, involvement or targeting of different user groups and access to other sources. Yet, having a more diverse set of actors involved also increases the range of goals in relation to the collaboration, in turn increasing the potential for conflicts between those goals (Vangen & Huxham, 2011). Goals can be public or commercial and related to various products or services. Some parties may join a collaboration to get a seat at the table; others might seek to explore new business models or to cut costs by bringing others on board. Also, a collaborative approach to innovation can be a goal in itself, or it can represent an attempt to gain legitimacy for an endeavour by engaging other stakeholders. Whatever the goals or incentives, the more there are, the greater the risk of a clash.

Second, and related to this, is the control and generativity challenge. This comes down to the fact that the actors control assets that are used in a process that they don’t fully control and that leads to new uses, products or services which they don’t control and which may even be opposed to their interests. This is our version of a tension that is well documented in the literature on digital platforms and inter-organisational systems (Eaton, 2012; Markus & Bui, 2012). The collaboration (not necessarily the data) needs to be relatively open to novel, potentially unforeseen, value creation. That ‘generativity’ can lead to the innovations with the highest impact. However, information and information systems are both an object of and tool for control by organisations, and control is tightly linked to autonomy and competitive advantage (Gawer & Cusumano, 2013; Johnston & Vitale, 1988; Tilson et al., 2010; Tiwana et al., 2010). Data are a key organisational asset, and opening up data to new uses, potentially by others, means giving up some control over that asset and relinquishing some autonomy to the collaborative. Fear over what others might do with the data could be a disincentive to collaboration.

This all leads to perceived vulnerabilities, especially if it is unclear whether data may be used for purposes, or by users or groups, other than those identified beforehand. Working with real time and raw data is more challenging than with data that was pre-processed, altered or specifically selected (Tan et al., 2011). Government can act as a facilitator and create a neutral or ‘safe’ environment to get a collaboration going. Yet, next to the competitive concerns of private partners, government partners also face challenges in sharing, for example, related to public values or statutory roles. A government role can, furthermore, bring additional challenges, such as fear among private partners that data may be ‘misused’ by government. For example, real-time data on arrival times of trains and buses is great for customers and operations, yet may reflect negatively on the provider when the time comes to renew a concession (Klievink et al., 2018).

It is therefore key to determine potential and novel uses of data and to instil trust in others so that all can feel comfortable opening up data as part of a collaboration (Thomson & Perry, 2006). To arrive at our analytical model for studying what factors contribute to value creation by data collaboratives, we explore the literature on how to mitigate these challenges. We start with a focus on trust relations within a collaborative. We then move to the role of government.

2.3Trust relations within data collaboratives

The topic of trust has been well-researched in several institutional contexts. Trust is said to facilitate interaction among actors by lowering transaction costs (Ring & Van de Ven, 1994). Without trust, more formal arrangements would be needed to anticipate and cover risks. This idea suggests that risk and trust are intertwined. Trust implies vulnerability (Newell & Swan, 2000; Seppänen et al., 2007). In data collaboratives, this vulnerability – or risk – is that others will opportunistically use the available data, to the detriment of one’s own goals within or beyond the collaboration. Much of the literature on trust relates to hierarchical relations in an organisation (Nooteboom, 1996; Norman, 2002; Six, 2007). Relationships within data collaboratives are typically mixed, ranging from horizontal to, indeed, control relations. Still, within this setting harm can be done, for example, in market relations if sensitive data are made available to competitors, or in power relationships if data are used to reinforce uneven power positions.

The role of trust in fuelling interaction seems conditional, however, as it is only wise to accept vulnerability if the other actors are trustworthy. Yet, collaborators’ assessments of ‘trustworthiness’ are inherently subjective and beyond calculation (Möllering, 2014). Trust represents a ‘leap of faith’ in regard to others’ commitment to shared values, good conduct and the sought-after outcomes (Nooteboom, 1999; Ring & Van de Ven, 1994; Van der Voort, 2017). The more complex relationships are, the more difficult it will be to assess these elements and the causal associations between them.

These assessments are easier if actors already know each other through a previous interaction. Interaction provides actors feedback on each other’s behaviour. If feedback is positive, trust increases (Nooteboom, 2002; Six, 2007). Thus, collaboration and trust are mutually reinforcing. They feed each other in processes of interaction. However, this makes it hard to just start with one and hope the other will appear. Without a history with the other actors, there are just too many unknowns to take that ‘leap of faith’. For example, if a data collaborative or a specific use of data is part of a continuing collaboration, then parties will typically be more willing to provide their data; or at least, there will be existing channels of communication and coordination to discuss and resolve any potential hurdles (Hart & Saunders, 1997). If the collaboration is ad hoc or it is a first attempt to set up a collaborative, these experiences will need to be acquired first, for example, through piloting (Klievink & Lucassen, 2013). Bachman and Inkpen (2011) suggested the importance of institutions to ease this first step. They identified three roles of institutions in a trust relationship. Institutions such as laws, contracts and informal ‘rules of the game’ provide meaning to the context of a relationship, they structure an interaction process among actors, and they serve as a trustworthy intermediary themselves.

Applied to data collaboratives, the literature informs us that trust and collaboration are mutually dependent. The positive feedbacks between trust and collaboration suggest the possibility of a catalyst effect, with collaboration feeding trust, feeding collaboration, et cetera. This process is bounded and facilitated by institutions (Van der Voort, 2017). In other words, antecedents of collaboration have a volatile part – trust and cooperation fuelling each other – and a more stable, facilitative part – institutions shaping this volatile process.

2.4Government and collaborative governance

Data collaboratives are here presented as a tool of modern public governance. The focus on the active involvement of non-government actors puts these initiatives in the area of collaborative governance and co-creation (Ansell & Gash, 2007, 2017; Mcguire et al., 2010; Voorberg et al., 2015). One of the conceptual models discussed in the literature, and the one we use here, is that provided by Ansell and Gash (2007). It is well-cited and offers useful concepts for the present study, though these authors emphasised that their model concerns general organisational logic, and is not confined to information technology (Ansell & Gash, 2017).

The core of the model is a collaborative process that can be likened to our notion of trust in that it is an iterative process in which trust facilitates commitment, which helps in developing a shared understanding (of the problem or goal), leading to joint outcomes and dialogue, which again reinforce trust. This iterative, collaborative process is affected by three main factors: starting conditions, institutional design and facilitative leadership (Ansell & Gash, 2007). These factors provide key components relevant to our understanding of how data collaboratives may be effective. We discuss each of them in turn.

2.5Analytical lens

In the introduction, we asked what factors contribute to value creation through data sharing in a data collaborative. This section has explored the literature on data collaboratives, trust and collaborative governance. This literature provides us a lens to identify contributive factors and to understand the collaborative process as a setting in which value creation, control and trust are balanced and iteratively lead to a result.

Based on the tensions discussed, it is not hard to fathom why data collaboratives are so hard to set up, why their initial coverage seems quite limited and why so much effort goes into discussions on their governance, covering questions like who should have access to a data platform, how, when and who decides (Klievink et al., 2012; Susha et al., 2017a).

The literature discussed suggests that trust can counterbalance the dynamics and challenges raised in this section and catalyse successful collaboration. A positive feedback loop between trust and collaboration can foster a collaborative process with shared, intermediate gains that become self-reinforcing, gradually strengthening the collaborative’s capacity to cope with greater potential threats, like opportunistic behaviour, and to generate new value creation. Over time, parties will also have invested in the collaborative and face costs (financial, reputational and operational) if they withdraw from it.

Summarising this literature, we build our lens on four components to understand the data collaborative process. Each of these four components are domains of the aforementioned struggle between the capacity to produce new value in innovative and unforeseen ways, and individual actors’ need for control over their data and data-related activities, goals and interests. Figure 1 presents the analytical model (inspired by Ansell & Gash, 2007).

The main components, based on this section, can be summarised as follows:

5.1Starting conditions

The starting situation was potentially challenging, as much of the data were in the hands of private or semi-public transport operators and there was no clear party in charge with the power, be it regulatory, financial or otherwise, to enforce ‘data collaboration’.

Yet, the starting conditions were apparently conducive to the innovation. Even though there was no meaningful prior collaboration, the situation was generally non-competitive, not just in terms of the use of data but also in terms of the parties’ core processes. There was also enthusiasm to do more with the data that the operators increasingly had. Most parties had similar incentives; that is, the idea that sharing would benefit public transport in general. This made it relatively easy to start a conversation on what they, as public transport operators, could collaboratively do. This could be considered a form of peer pressure, in which the collaborative incentive prevailed over the lack of individual incentives among some of the partners. As the platform started to work and parties found themselves in a situation without the materialisation of risks to their core services, trust developed, which mitigated perceived vulnerabilities. The actors did not feel a strong need for control over their activities and data. Within a decade, this led to real-time sharing of even raw data, not just with their customers, but also with each other.

5.2Trust and collaboration

The literature presented in section two on the dynamics between trust and collaboration suggests the existence of a positive feedback loop between them. In our case study, some reciprocity in the data-sharing relationship might have existed but was limited because the parties differed significantly in size and stakes. The passenger rail service company operated nation-wide, whereas most of the other transport operators served a smaller region only. A comparison with similar platforms in other countries covered in a project of which this case study was part (Veeneman et al., 2018) shows that it is not uncommon for a similar set of parties to fail to share such data. This is even the case when a transport authority forcefully pushes for data exchange. Our analytical lens suggests that 9292 succeeded here because of trust, based on prior collaboration. The collaborative process reinforced trust, leading to expansion of the platform. The starting conditions were conducive to a situation in which trust could develop between the parties bringing data to the collaborative platform.

When advanced traveller information systems entered the scene, a need arose to share real-time information, which helped to further institutionalise 9292. The incentive here was initially primarily of the ‘nice to have’ variety; undoubtedly, it was nice to have the electronic signs announcing the up to the minute arrival times of the various transport options, and no real competitive costs seemed to be involved. This happened at about the same time as the first competitive tendering procedures began. But at that time the transport authorities did not yet have the desire or know-how to use the 9292 data for contract supervision, and such use of the real-time data was not yet included in the public transport contracts – though this changed later.

Further institutionalisation of the platform and formalisation of 9292 as a collaborative, combined with the popularity of the joint value-added services and absence of trouble between the partners, led to a positive feedback loop as described in the literature.

5.3Institutions and context

Institutions facilitate and constrain the interaction between trust and collaboration by making the context more stable. In our case, we could argue that the collaboration itself, and its formalisation, became an institution that mitigated the risk of opportunistic behaviour, persisting even when the threat of data use beyond the envisaged goals substantially increased over time.

The collaboration’s perseverance can be explained, first, by the absence of disincentives and absence of noteworthy transaction costs, due to the coincidental use of the same systems. However, the introduction of competitive tendering did provide for disincentives, at least for some of the parties. Still, this system change occurred almost ten years after the collaboration started – an aeon in ICT development terms. By the time it became clear that authorities could use the data for supervision purposes, the parties were so invested in the collaboration and in the platform that backing out could come only at great cost. In other words, the trust and collaboration process began an institutionalisation process that was hard to stop. The entity 9292 was too well-established in the routines of travellers and transport operators to be stymied by changes in context.

This might also explain why in other countries a push by authorities has been much less effective than the collaborative process was in this case. The collaboration started relatively small and quickly grew into a national system. A look at other countries shows us that elsewhere government involvement has been less of an enabler than the push to exchange data with Google’s Transit (as the latter forced operators to adopt GTFS as a data standard). Compare that to the case at hand, in which 9292 became the source of the data and Google a user.

5.4Government and leadership

Consciously or not, government played a vital role in establishment of the platform. Government was responsible for the initial incentive structure in which public transport providers were not direct competitors. This created fertile grounds for collaboration. At this meta-level, government was a main facilitator of 9292. Once government introduced tendering, the collaboration process was already too institutionalised to be harmed. Also, the potential for the transport authorities to use the data for control produced little conflict. As an institution with a clear governance structure and strong leadership, 9292 had over time become strong enough to keep parties on board in the collaborative, even though the contracts eventually stated that they must open up the data to government organisations for supervision. The data collaborative had thus gained sufficient independence from its sponsors and members to start playing a role not in the interests of those same parties. The platform evolved into an important instrument for satisfying travellers’ information demands, and the transport operators would not want that real-time information to disappear, for instance, from the electronic information signs announcing the arrival times of public transportation virtually everywhere.

The governance of the mobility data platform was captured in the governance of the association, composed of key data providers and data users. It helped, too, that open data sharing had become so ‘common’ over the years. Indeed, today a widely-supported idea in the Netherlands is that data generated directly or indirectly by publicly funded activities should also be public. This ‘open data culture’ has become an important incentive for the parties to sustain the collaborative in the near future.

6.1Conclusions

We started out with two paradoxes that illustrate the complexity of the interactions among partners in a data collaborative: the goals paradox and the paradox of simultaneous control and generativity. Theoretically these represent major challenges. Data collaboratives rely on parties opening up their data in or through the collaborative. Yet, the participating parties may find it difficult to disclose, for example, real-time information gained through channels that they own themselves. They may additionally fear that competitors could gain early or advanced access to sensitive information. And it does not help that the government could potentially use the data to assess the performance of transport operators.

We explored the literature on how these tensions could be mitigated. Academic work on trust and on collaborative governance formed the basis of our analytical lens, featuring trust and collaboration as a mutually reinforcing process, with starting conditions, institutions and leadership as antecedents. All these features were present in some form in the case.

We found four types of antecedents of collaboration and collaborative innovation that both confirm and complement those found in the literature:

Next to these antecedents, we found the context of a data collaborative makes a difference. Context factors include, for example, institutional and governance frameworks, cultures of inclusiveness or transparency, the interest demonstrated by government in data collaboratives, and the means by which collaboration is legitimised as a strategy for governance or innovation. Changes in the context may significantly affect the incentives and goals of the actors and threaten the collaborative process. However, in our case such changes did not harm the collaboration.

In our case, the reinforcement process between collaboration and trust, as found in the literature, resulted in institutionalisation of the collaboration itself. The collaboration was recognised by the public, by clients and by authorities. This recognition provided positive incentives to pursue collaboration. We therefore propose that the collaboration process and the institutionalisation process are also potentially mutually reinforcing. Our case suggests that these positive feedbacks can result in such strong collaboration that it continues even if the original incentives to collaborate change or disappear.

As a conclusion, Fig. 2 presents a model (loosely based on the models of Ansell & Gash, 2007; Robey et al., 2008) of the factors found in the literature and their relationships as found in our case. It is a further specification of the analytical model summarised in Fig. 1.

Figure 2.

Conceptual model for data collaboratives.

Our findings suggest that next to the constellation of actors involved, timing is key. The model implies at least two time-related factors that contribute to collaboratives’ ability to create value. First, the collaborative is more likely to take off if the antecedents are favourable. In our case, the rules of the game (the low competition level) and the information systems (actors used the same systems) were especially favourable to collaboration. Second, once collaboration has commenced, the favourable conditions need to remain until a degree of institutionalisation has taken place. In the Netherlands, 9292 was strongly institutionalised, partly because for many years it was the only multi-modal infomobility data service. If the collaboration had not been institutionalised, for instance, because there were multiple infomobility data services, it would have been more vulnerable to changing conditions.

Summarising the findings and conclusions, there are at least three key ideas to take away from our study:

6.2Discussion, limitations and future research

Our case spanned 25 years of development and evolution. Hence, data collaboratives may be seen here as ‘merely’ a new lens on a long-established phenomenon. As pointed out, cases labelled data collaboratives in the literature might be relatively ‘easy’ cases built on datasets already being shared in some form, and just being put to a new use. Yet, we think it is worthwhile to attempt to identify established data collaboratives as well. Established ones may already have had significant impact on policy or public governance but may not have been studied from the perspective of how the data enabled those practices and their impacts. New data collaboratives can tell us more about what is possible today, as incredibly large sets of data are becoming available, with unprecedented degrees of interconnectedness and strong personal, public and commercial motivations to make use of the data. Studying new and emerging collaboratives may help to inform existing practises, to gain more from the institutions and trust already in place to share and re-use data for new value creation.

Still, to really enable the use of data collaboratives as a tool for public governance, there are many challenges that need to be solved in situations that are rendered much more difficult because of differences in goals, fear of opportunistic behaviour and lack of incentives to join the collaborative. Our case suggests the importance of organising the process and context in such a way that perceived vulnerabilities and fear of misuse are dealt with.

The right choices in the design of the technology and governance of the data collaborative depend highly on the perspective and position of the actors. Which parties should be involved? In what capacity? By whom? These seem to be the million-dollar questions for data collaboratives. Regarding infomobility platforms, regional views will likely differ from local views, and public and private views will differ as well. Service provider and service consumer perspectives will not be the same, and data contributors and data users will have different perspectives. Further, we must consider all the others, ranging from technology providers through to local politicians, visitors of a metropolitan area and many more. To make a mobility platform work, some or most of those actors will have to be connected to the data collaborative in some way, though perhaps not be active in it. Such a platform will be highly dependent on the contributions of these actors. If data providers don’t provide data, if no one funds the platform, if travellers don’t use the services on offer, the platform can create no value. If one of those actors have blocking power and an incentive to use it, it may not fly at all.

We aim and hope to have offered insights relevant to the emerging practice of data collaboratives, yet our study is limited in several ways. This article covered just one case. We cannot make any broader inferences based on this research, and the model in Fig. 2 should be considered an exploratory and relatively broad attempt at identifying what makes data collaboratives work. The emerging nature of the phenomenon and the explorative character of this study severely limit the contribution of this research at the current stage. The case featured here was studied inductively, and the lack of research rigour further adds to the validity problems this research presents. The contribution of this article is therefore largely conceptual, and the proposed conceptual model is proposed as a subject for further testing and verification. Further research is required employing more rigorous designs in both quantitative and qualitative follow-up. Qualitative designs will aid in further refining the model and in gaining a more in-depth understanding of the questions and paradoxes raised here, across multiple cases and domains. Quantitative research could test and verify those findings. This research is a first step towards describing and understanding this phenomenon by looking at potential tensions and paradoxes. The analytical lens presented, bringing together insights from collaborative governance, trust and data collaboratives, could be further expanded with other literatures, such as that on inter-organisational information systems and innovation systems – both of which we only touched upon here. We aim to further develop and refine the analytical model and commit a comparative and in-depth study of other data collaboratives, including those with different degrees of complexity, generativity and success.