Data governance in smart cities: Challenges and solution directions

2.1.Smart cities

Due to current rapid urbanization worldwide, cities have a huge impact on every aspect of our lives. In the last three decades, therefore, the concept of smart cities has emerged as a possible solution direction to address the problems stemming from this rapid urbanization [1,4,27]. The range of these problems spans from sustainability-oriented ones to efficiency ones [50]. Sustainability problems are related to social values (e.g., equity, community autonomy, citizen well-being, quality of life and gratification of fundamental human needs), economy vitality and diversity, and environment conservation (e.g., flora, fauna, and natural resources). Efficiency-oriented problems are related to efficient management of urban processes like transportation, education, and administration.

Consequently, the interest in smart cities has risen considerably among scientists, practitioners, and public policymakers in recent years. Despite this popularity, there is a lack of conceptual clarity around the term of smart cities due to the plethora of existing definitions [4,50]. According to [1], lack of a general agreement about the term smart cities may be attributed to the fact that the term has been applied to two different kinds of domains, namely hard and soft ones. In hard domains (such as, buildings, energy grids, natural resources, water management, waste management, mobility, and logistics), the Information and Communication Technology (ICT) plays a decisive role in the functions of the systems while in soft domains (such as education, culture, policy innovations, social inclusion, and government) the application of ICT is not usually decisive. The approach of hard domains aims at harnessing technology, particularly the ICT, to create a city that is, according to [29], instrumented (i.e., for capturing and integrating real-world data), interconnected (i.e., for communicating the resulting information among various city services) and intelligent (i.e., supporting operational decisions within city). The approach of soft domains sees smart city far from being limited to the application of technologies to cities as ICT eschews the actual knowledge about how cities function and disregards its complexity. This approach, instead, focuses on people and promotes nurturing citizens’ creativity and knowledge (via education, culture and art) as well as their participation in order to achieve a sustainable growth.

Despite lack of a unified definition of smart cities, ICT plays a major role in shaping the concept of smart cities according to most definitions of smart cities, for an overview refer to [1,4,27,30]. According to the findings of [2, Section 4], the standardization bodies such as ISO, the ITU, the UK Standards BSI, and the US NIST have defined smart cities as “innovation – not necessarily but mainly based on the ICT –, which aims to enhance urban life in terms of people, economy, government, mobility, living and environment”. For smart cities, ICT provides an infrastructure (and means) for collecting data from various data resources of a city, integrating the collected data into useful information, communicating the resulting information among city services, and supporting decision makers with collective intelligence that is based on the exchanged information [17].

As the collected data in a smart city impact its citizens on every scale, we believe that the ICT systems that collect and capitalize on them should not be perceived as pure technological systems. The way that data are collected, which are often blended with sensitive information about individuals, groups, and businesses, and the way that data driven ICT systems are devised, designed, implemented, deployed, and (mis)used, are going to impact us deeply at both individual and societal levels. As such, social transformations derived from data-driven systems must be attentive and respectful of the principles of social justice and ethics as postulated as the outer and inner boundaries of all human activity. Therefore, we are going to elaborate on our vision about how data in a smart city must be managed in a responsible way, as there are many data related blind spots like data quality issues and threats (e.g., “forms of dataveillance, social sorting and redlining, predictive profiling and anticipatory governance, nudge and [behavioral] change, control creep, and system security”) in smart cities rhetoric [31, pp. 12].

2.2.A reference model for smart cities

Given that ICT is a main enabler of the concept of smart cities, it is useful to sketch a unifying model or architecture of smart cities to capture an abstract view of the complex context where ICT systems, in general, and data governance, in particular, operate. One interesting reference model for smart cities is presented in [46], which we have adopted with some adaption below for our purpose. The reference model, sketched in Fig. 1, comprises three levels, namely: organizational level, informational level, and technical level.

Fig. 1.

A reference model for smart cities, adopted with adaption from [46].

Within the technical level there are various data sources, telecommunication infrastructures, and data processors. Data sources, which provide raw data as well as (semi)processed data, include Internet of Things (IoT) sensors, social media, directories and registers within public and private organizations and enterprises, open data initiatives, and Internet websites. These data are offered commercially, openly (i.e., being public, free of charge and for any usage as in case of open data) or conditionally (e.g., for a given purpose by specific data consumer groups). Telecommunication infrastructures encompasses data communication networks of various types (like sensory networks, cellular networks, wireless and wired local access networks, and Telecom/Internet core networks) as well as telecommunication protocols to interconnect various devices in the smart city ecosystem. Data processors adapt and enrich heterogeneous data (by means of advanced data analytics, statistics and artificial intelligence) to produce information to the components of the informational level.

In the informational level there are various applications and services, marketplaces, and usage support services. Via applications and services, the processed data in the technical level are further enriched to support the semantic relations and understandings, needed for applying the resulting information in a smart city application setting. These applications range from those for critical systems (e.g., traffic light regulation and energy distribution systems) to those for infomercial services (e.g., about city traffic peak hours and citizen energy consumption). Marketplaces are online platforms enabling citizens and organizations to discover, rank, purchase, and deploy integrated smart city applications and services in different city domains. Usage support services help the end-users of smart city applications and services to download the apps to their devices (like smartphones, tablets, personal computers, and notebooks), to maintain these services, and/or to monitor the way that these services are used. Note that the end-users can be citizens, journalists, companies, or even other automated systems (like actuators to control physical systems in the smart city).

In the organizational level there are procedures, objectives and policies, and city governance. The procedures include the business procedures for commercial enterprises (like billing and payment) and administration procedures for (local) governments (like smart social services), which are partially enabled with (advanced) ICT and data-driven applications. The objectives and policies refer to the business and city governance objectives and policies that are based on business and city administration models, missions and visions and that are used to define, design and shape the above mentioned (business and city administration) procedures. The city governance refers to the values, social norms, principles, laws, rules, regulations and activities that guide the behaviors of the stakeholders (e.g., civil servants and citizens) and organizations towards defining and realizing the above mentioned procedures, objectives and policies. As such, city governance includes all organizational aspects of city governance.

In the next section we discuss the role of ICT resources in the reference model for smart cities.

2.3.Data governance

To achieve the sustainability and efficiency objectives of smart cities, we need to manage the ICT resources embedded in the informational and technical levels of the reference model (see Fig. 1) within and across the various organizations of a smart city ecosystem. For managing these ICT resources, as indicated in Fig. 2, we distinguish between governance and management, where the latter can be seen as the technical implementation of the former [23]. Two main categories of ICT resources are Information Technology (IT) resources and data resources. IT resources are the hardware and software components needed for collecting, storing and sharing smart city data. Management of these IT resources involves, among others, their installation, configuration, monitoring and maintenance and is mainly done within every organization in the ecosystem of smart cities with minor cross organizational coordination. Management of data resources, however, demands much more cross organizational collaboration and coordination relatively, due to the inherent property of data in traversing organizational boundaries. The difference in cross organizational natures of IT resource management and data management is indicated by the heights of the corresponding building blocks in the smart city reference model of Fig. 2.

Fig. 2.

The reference model of smart cities, extended with governance and management components.

In this contribution, we are going to describe our vision on data governance and, as such, our scope is beyond specific organization, excludes the IT resource management and IT resource governance aspects, and does not cover the implementational aspects (i.e., it excludes the data management aspect).11 In [12, pp. 432], data governance is defined as “the exercise of authority and control (planning, monitoring and enforcement) over the management of data assets”. Data governance aims at defining the organizational structures, data owners, policies, rules, processes, business terms, and metrics for the whole lifecycle of data (i.e., collection, storage, use, protection, archiving, and deletion [23]. Cuno et al. [20] relate data governance in urban settings to identifying the entities who are involved in the management, provisioning and utilization of urban data (like data owners, data stewards, IT system administrators, data policy coordinators and data policy officers), and the required communication and interactions among these entities. These aspects (i.e., which entities and their interactions) are embedded in the decision-making process for data quality management, data access management, general data lifecycle management, and metadata management for the datasets in an urban environment.

In [19], the governance of smart city data is viewed more broadly than just focusing on the substance of rules around the collection, use, sharing, retention, and disposal of data. Their suggestion is to focus on how these rules are made, disputed, and changed (i.e., on who is making the rules and what processes are used in rule making). To this end, the authors adapt four principles of Ostrom’s 8 principles that are applicable to the context of smart city data. These procedural principles [19, pp. 7] are:

Eke and Ebohon [22, Section 3.2] consider a data governance that “incorporates data from all relevant sources, focuses on diverse values for all stakeholders and aligns with strategic objectives for and by residents”. As such, this view pays special attention to the impacts of the inferential decisions made with the data (due to, for example, exclusion of some groups in the population). This view extends the scope of data governance from “ensuring maximum value creation from the data while adhering to ethical and legal requirements” to “the establishment of processes, policies, roles and responsibilities that foster the effective management of data for the benefit of relevant stakeholders that will be affected by the decisions derived from the data” [22, Section 3.1]. To this end, smart cities should set up a data governance framework with the right data (i.e., data without representation bias, measurement bias, evaluation bias, aggregation bias, population bias, sampling bias or data linking bias), the right algorithms (i.e., in being accountable, explainable and fair), the right people (i.e., the right mix of people and expertise to adequately address inclusion issues), and the right policies/standards to build trust, transparency, accountability and responsibility) [22]. This is an interesting viewpoint as it asks for monitoring the impacts of data when the results of data-driven applications are applied into practice. In other words, managing the use of smart city data belongs to the domain of data governance in addition to managing the quality and diverse nature of the data that inform data driven applications in a smart city eco-system.

In Section 3, we discuss several functionalities for managing the quality and diverse nature of data. Section 4 is devoted to a set of building blocks for managing the use of smart city data.

3.1.Addressing data quality issues

An important factor that should be considered is to establish an acceptable level of data quality. A bad quality of data often results in a low utility of the data since the data will be considered as unsuitable for use. Therefore, it is necessary that the data meet a standard quality level. However, to predefine an acceptable data quality level is a tough task. As will be argued, this is because the notion of data quality is hard to capture and an acceptable data quality level depends on the type of the application and its deployment context.

The broadness of the notion of data quality makes it hard to capture in a single definition. Therefore, there is no widely accepted definition for data quality. Various definitions of data quality can be found in literature, ranging from defining some data quality dimensions (such as accuracy, completeness, timeliness, usability, relevancy and reliability) to more comprehensive (i.e., generic) definitions [26,47,51,52]. Some of these dimensions, e.g., accuracy, may be classified as objective while the other dimensions can be seen as subjective, e.g., relevance. This holds also for the comprehensive definitions of data quality. For example, an objective and comprehensive definition is that data should adequately represent (parts of) the real world, while a subjective definition is that data should fit the usage purpose of an application. Although the definitions of data quality may look different, the comprehensive definitions subsume the core elements of the more specific definitions.

Assessing the degree to which a data set represents the real-world is a predominantly objective definition of data quality. As elaborated in [51], this implies that there should be a valid and unambiguous mapping from the data set to a complete description of a real-world phenomenon. Although this definition raises a number of questions, such as how to determine the completeness and which real-world phenomenon should be represented (e.g., those of today or the past), it provides sufficient guidelines for operationalization. Based on this definition, scholars have specified several data quality dimensions such as data completeness and accuracy. Assessing the degree to which a data set fits its usage is a predominantly subjective definition of data quality. Also this definition raises several questions such as to which usage the data should fit and how this fitness should be defined. Elaborating upon this definition, scholars have specified several data quality dimensions such as data relevance, usability and understandability. These dimensions, which depend on the type of application at hand, provide more criteria for operationalization of the data quality concept.

Thus, from data quality definitions a wide range of subjective and objective dimensions can be derived. In Fig. 4, we depict some of these dimensions, see [11] for a detailed overview.

Fig. 4.

Some data quality dimensions.

Alternatively, in [42] a semiotic view is taken on data quality, which results in an analytic framework for data quality. Three analysis levels have been distinguished in this framework, namely: the syntactical level, the semantical level and the pragmatic level. The syntactical level focuses on the degree of adherence of the data to predefined meta data characteristics. As such, it measures the consistency of the data within a data set. The semantic level focuses on the meaning of data, for example, whether it is complete, and up to date. The pragmatic level is concerned with the utility of data in terms of, for example, usefulness and usability. On this level the assessment of data quality depends on the involved professionals, the application(s) at hand, and the organizational context of the use.

In order to make the concept of data quality tangible, we combine the semiotic view with the data quality dimensions, resulting in a data quality framework. This framework can be represented by a matrix as depicted in Table 1. Each dimension, which constitutes a column of the matrix, may be decomposed into a set of indicators. An indicator pertains to one of the levels, which constitute the rows of the matrix.

The following example illustrates how each dimension is decomposed into a set of measurable indicators and how the appropriate level is determined for each indicator.

As stated before and in line with [32,35], we believe that the assessment of data quality is eventually dependent on the applications at hand. Consequently, knowledge about the applications and their situational context, referred to as domain knowledge, is of vital importance to assess and improve the quality of data. In turn, the domain knowledge determines the data quality dimensions that should be selected and the proper levels on which the data quality can be assessed. From a practical point of view, we foresee that a crucial task of a data ecosystem is to come up with typical sets of data intensive applications for an application domain. For each set of applications, the proper data quality dimensions, the corresponding data quality indicators and the proper level(s) for each data quality indicator should be determined. This will result into a filled data quality matrix as depicted in Table 1, and for each element in the matrix an (acceptable) value should be specified. We refer to such a filled matrix as an instantiation of the data quality matrix. The values assigned to the elements in the data quality matrix may represent the minimum or desired values for an indicator at a certain level that should be met for a desired quality of the data. Actually, an instantiation represents a typical scenario within a domain, i.e., represents a marketing scenario. In a marketing scenario, city marketers need data to promote the city. Therefore, the granularity/accuracy of data (i.e., an exact spelling of the hotspots in a city) will be less important for the quality of the data.

The challenge for every application domain is to come up with a number of useful data quality instantiations.

3.2.Addressing trust issues

In addition to addressing data quality issues, the success of smart cities depends on addressing the trust issues that may arise due to security, privacy and ethics related threats. Collecting, communicating, storing, processing and using (privacy) sensitive data should respect the required and expected level of security, privacy and ethics in a (loosely coupled) data ecosystem. Otherwise, the threats and risks introduced to privacy, human rights, and ethical or democratic values may not only hamper the success of smart cities but also inflict harmful and irreparable impacts on citizens and society. In this section, we elaborate on privacy governance (Section 3.2.1) and security governance (Section 3.2.2), where the latter is done in a less elaborative way than the former. Due to space limitations, we leave out the governance of ethical aspects here, noting that some of its characteristics are common with those of privacy and security governance.

3.3.Recapturing the concept of data governance

Like [12], we consider data governance as the exercise of authority and control (planning, monitoring and enforcement) over the management of data assets. As such, data governance aims at defining the organizational structures, identifying the actors (like data owners, data stewards, IT system administrators, data policy coordinators and data policy officers), defining the policies and rules, designing the processes and business terms, and specifying the metrics that are relevant for the whole lifecycle of data (i.e., collection, communication, storage and processing, dissemination and use). Note that

We elaborated on two main aspects of data governance: Improving the data quality (Section 3.1) and enhancing the trust of stakeholders (Section 3.2). This elaboration and the data governance model shown in Fig. 3 are inspired by the study of [12].

4.1.Data domain

A data domain refers to the set of organizations (including their human capitals and technical infrastructures) that serve a specific (set of) data-driven application(s). Every data domain may have a data ecosystem among the participating organizations for managing its data. To determine which organizations should be part of a data domain (or its data ecosystem), it is worthwhile to identify the organizations that perform the tasks related to the data of the data domain. These data related tasks can be broad but roughly we distinguish the following main data related tasks (see also Fig. 3):

The organizations that are involved in performing at least one of these tasks are eligible for participation in the data ecosystem and are members of the corresponding data domain.

The organizational structuring of a data domain (i.e., the structure of the relationships of the organizations in a data ecosystem) depends on the stakeholders involved and the generic domain in which the data ecosystem is embedded. At a high abstraction level, a generic domain may be considered as a set of organizations that serve a specific aspect of our society such as healthcare, science, justice, public safety and education. The organizations in a generic domain are coupled strongly or loosely. In other words, these organizations have already an existing formal or informal structure for collaboration, like a chain structure, peer to peer structure or a hierarchical structure. The hierarchical structure is mainly found within an organization (i.e., among its various departments), while the other structures are often found across organizations. For example, the supply chain within the logistics domain has a chain structure typically, while collaboration in the science domain has a peer-to-peer structure generally. In a local government (like that of a city) or a central government several variants of the chain structure can be recognized. The organizational structuring of a data domain, generally, inherits that of the corresponding generic domain.

To realize data governance within a data domain, several measures should be taken. Data governance realization may call for introducing some (new) roles within a data ecosystem such as data steward and (chief) data officer roles. Although their importance is being recognized and they have been introduced in practice slowly, these roles and their corresponding tasks are not well defined and stablished generally. An elaboration of these roles and their corresponding tasks should be on the future research agenda of smart cities. Moreover, for realization of data governance, we need, among others, to educate the participating organizations about the data governance and implement the key functional features of data governance. In Sections 4.2 and 4.3 we are going to elaborate on these data governance knowledge distribution and the key data governance functional features, respectively.

4.2.Knowledge distribution

Realization of data governance (and the data management) in practice asks for deploying data quality management mechanisms and data protection mechanisms (like data security and personal data anonymization mechanisms) within and across smart cities data domains. Often, these mechanisms are complex and therefore a successful deployment of smart cities asks for disseminating data governance knowledge among the involved parties so that they (i.e., citizens and organizations) who reside closely to their data can play their roles in maintaining the quality of their data as well as protecting their data.

The knowledge and expertise about (new) data governance (and data management) can be deployed within the data domains of a smart city in various ways. To start with, we investigate two structures that specify the two ends of the deployment spectrum, namely: fully-distributed deployment and centralized deployment. In the fully-distributed deployment, every organization in the data domain hosts the knowledge about data governance/management. This is the most desired format because the domain knowledge (i.e., the knowledge of the organization) can fully be exploited to adapt data quality to a desired level while protecting the raw data at the source (e.g., without letting the data leave the premises and boundary of the organization in charge of personal data collection). However, it is a serious challenge for every organization to master data governance (and data management) knowledge and carry out it independently and uniformly.

The second option centralizes the data governance (and data management) knowledge at one organization (i.e., at the central party). However, on the downside, the central party in this model must be trusted to receive the raw data of all local parties (i.e., it must be a Trusted Third Party, TTP). This option is challenging as, for example, the collection of privacy-sensitive data at a central party is a classical threat for privacy protection. Furthermore, the scalability of the data governance (and data management) functionality becomes an issue as the central party becomes a bottleneck with the current fast growth of data and data sharing. Table 3 summarizes the pros and cons of the fully distributed and fully centralized options.

Considering the pros and cons of the first two options mentioned above, our envisioned framework for disseminating data governance (and data management) knowledge in smart city settings is based on a third option that offers a partially distributed data governance (and management) to benefit from both distribution and centralization while avoiding their disadvantages as much as possible. In this way, member organizations do not become overwhelmed with demanding data governance (and management) tasks. For distribution of data governance (and management) knowledge between the central organization and the member organizations, we envision gradually transferring the data governance (and management) knowledge to the member organizations as much as possible. In this way, creating data governance (and management) expertise at member organizations occurs gradually, without imposing immediate burden on them to learn and apply complex and demanding data governance (and management) tasks. Initially, we foresee that the member organizations learn the fundamentals of data governance (and management) and learn an initial set of guidelines and apply them into practice.

There are several benefits of learning about data governance (and management) as much as possible. Firstly, data stewards at local parties can better understand the data quality issues and the (privacy, security and ethics related) risks associated with the data they collect, process and share. This understanding can be helpful to project own domain knowledge into data quality management and data protection. Secondly, the data stewards are often processors/consumers of external data sets, which are adapted (e.g., anonymized) by other parties. Knowing about the technical details of data governance (and data management), they can avoid misunderstandings when processing such data. Learning the initial set of guidelines can enable member organizations to carry out routine data governance (and management) tasks as much as possible. The central party, which comprises several data governance (and management) experts, can gradually educate member organizations and offer consultancy to them.

To overcome the barriers of adopting complex data governance (and data management) in organizations, we envision an evolutionary process to expand the initial data governance (and management) knowledge, conveyed from the central party to member organizations. The distributed knowledge can be of awareness, principle and how-to types as defined in [33]. These are the different types of innovation knowledge used during the technology adoption process for mobilizing potential adopters. The awareness knowledge is about the existence and key properties of an innovation, the how-to knowledge is about how to use an innovation properly at individual and organizational levels, and the principle knowledge is about the functioning principles underlying the innovation.

4.3.FAIR principle

To promote re-usability aspect of the FAIR principle, it is necessary that data are findable and accessible. Furthermore, many applications demand that various data sets are integrated. Therefore, these data sets should be interoperable. Thus, re-usability requires the compliance of the findable, accessible and interoperable aspects of the FAIR principle. Good data governance should take care of the embedding of the FAIR principle among the organizations within a data domain (or data ecosystem).

There are many efforts and best practices reported in the literature to make data findable and accessible. Choices should be made within a data ecosystem about the best ways to make data findable and accessible within a data domain. Furthermore, it should be made clear which organizations in the data ecosystem are responsible for these aspects.

Interoperability is a more challenging topic and requires more coordination to address it properly. In general, the organizations involved in a data domain collect and store data sets to optimize their core business processes. Whether these data sets are interoperable with the data sets stored at other organizations is not their primary concern. Since interoperability is a necessary condition for data integration, it should be on the agenda of data governance. Standards to exchange data should be defined. Attributes that every data set should include for interoperability purposes, e.g., primary/foreign keys, should be defined as well. Finally, the quality standard that a data set should meet should be defined. As discussed in Section 3.1, data quality may be decomposed in different dimensions and on several levels.

For re-usability of data sets, the findability, accessibility and interoperability of data sets are necessary conditions, but they are not sufficient. To promote re-usability for every data set, it is necessary to clarify the semantics of the data, the purposes for which the data may be used, and the conditions of the data usage. There are several ways to realize this, for example, by specifying the metadata that provide insights in the data quality, and providing an unambiguous description of every attribute in the data set, giving insights about the type of analysis that the data is suitable for. All these mechanisms may be tailored for different types of users. Good data governance includes well-considered choices about these mechanisms and defines the responsibilities and tasks among the entities in the data ecosystem.

An important emerging side effect of re-usability is function creep. Function creep refers to the situation where data to the corresponding applications are gradually used differently than they originally were intended to be used. Facebook was once intended to connect people, but it is now also a tool for organizations to collect as much data as possible. It often starts with collecting microdata to provide services to users. Microdata concern the data of individuals, which consist of some attributes about them. However, these individuals and/or data providers have not full control on the operations that are performed on the data over time [6]. For example, the data can be combined with other data, then aggregated and used for (direct) marketing purposes. Aggregated data can also be used to formulate policies. The microdata can also be used for surveillance purposes in combination with other data. Function creep often takes place gradually and out of the sight of stakeholders [15]. As a result, it takes a while before stakeholders become aware that the data are used differently than that was initially agreed upon. If function creep becomes a frequent occurrence, especially when the stakeholders are harmfully affected, then the support for maintaining a data ecosystem will decrease. The impression may arise that function creep is a strategy to present stakeholders with a fait accompli. Therefore, it is necessary to monitor function creep closely to prevent its occurrence. In particular, one should prevent using data for tracking and tracing (of any group) of individuals as a result of function creep. In the Netherlands, we are already observing an increasing resistance of citizens in using an app that is intended to control the spread of COVID-19 due to the fear of function creep. Finally, we note that function creep may lead to a violation of the privacy of citizens and to stigmatization of groups of individuals. We believe that an appropriate data governance should take care of function creep.

As illustrated in this section, the embedding and elaboration of the FAIR principle come with a number of challenges. Since the FAIR principle is considered as a cornerstone for data ecosystems, these challenges need to be systematically addressed in the future research.

5.1.Design thinking

Design thinking is a methodology to express concerns, needs and wishes of stakeholders in an explicit manner and to create support and consensus among them. Design-thinking has shown to be useful in settings where user needs and concerns are insufficiently formulated and are hidden in tacit knowledge. In such settings, there are often different and poorly communicating stakeholders [5]. A typical design-thinking process comprises the following stages: Empathize (to understand the real concerns of stakeholders), define (to find out the deeper roots of the needs), ideate (to explore and generate solutions), prototype (to make tangible objects for some ideated solutions) and test (to evaluate the prototypes with the end-users and learn from them). The design-thinking stages may occur concurrently and are rapidly iterated, where the prototyped artifacts (e.g., products, services, tools or processes) are tested per iteration. The practical experiences gained with the prototyped artifacts in every design round inform the following round about how to improve the artifacts. Via improving the most viable concepts in more detail, the designer aims at attaining a viable product eventually.

The starting point in design-thinking is a good understanding of the practice field. Further, the design process is informed by the practice and is highly collaborative, where all stakeholders are involved in all phases of developing the artifacts. Involving all stakeholders, especially early on in the design process, prevents disappointments in that the artifacts do not cater stakeholders’ real needs. This early involvement of stakeholders enables discarding suboptimal solutions as soon as possible. This so-called fail fast approach gears the design process towards producing viable products with high chance of user adoption. Giving users the opportunity to give their inputs, ideas and viewpoints during the design process, is important in developing social services where citizens’ participation and acceptance are of outmost importance (like in case of smart cities).

5.2.Need for trade-offs

There are many trade-offs needed to be made in designing data governance instruments in a given smart city context. Every design should determine which combination of data quality measures and data privacy, security and ethics related measures should be in place. Often, these measures are adversely related. For example, improving the anonymity of personal data requires reducing the data quality (like the exact age of 24 years old is transformed to age range of 20-29 years old). Adversely, a high data quality may result in personal data breach, see for example the attack described in [37]. As the design space is complex and multidimensional, many designs can be created. For example, for the privacy-by-policy measures, every design determines how far data subjects can be put in control of their data. This control can be at varying levels, ranging from full control to no control. Each choice, alone and in combination with other choices, has implications that should carefully be weighted and made.

Making tradeoffs among competing values is studied, for example, by [13] between secrecy and transparency and by [24] between privacy and public good. These works focus only on data collection technologies or on the control of the collected data. When the number of dimensions in the design space increases, as explained above, it might be difficult to come up with some design options and choose the most viable design in a deterministic way (i.e., only by, e.g., engineers and with engineering rules). Using design-thinking, we envision, one can integrate the true knowledge (i.e., the models and theories from science) with the how knowledge (e.g., the technological opportunities demonstrated by engineers) through an active process of ideating, iterating, and critiquing the potential solutions to make the right thing [54].

In Fig. 5, the idea of creating a series of artifacts, denoted by solutions S₁, … and S₆, is illustrated in a simple two-dimensional design space of data trustfulness versus data quality. In Section 5.1, we argued how design-thinking can help to derive these six viable design options by making trade-offs among various criteria and also how it can help the stakeholders to achieve consensus on a most viable one, for example, solution S₂ shown in Fig. 5.

Fig. 5.

An illustration of creating multiple design options, adopted with adaption from [7].

5.3.Need for collaborations

So far, we have identified the importance of several building blocks for an appropriate data governance in a data ecosystem. These building blocks pertain to functionalities (Section 3) as well as organizational aspects (Section 4). Each of these building blocks entails several challenges and needs further elaboration for a specific data domain. The elaboration may evolve in several directions which are not necessarily divergent. As we expect that the involved parties (i.e., individuals and organizations) in a data ecosystem will be organized as a peer-to-peer organization and as argued there will be a shared responsibility among these parties, there should be full consensus among the member organizations (and the corresponding user groups such as citizens) about the direction along which the challenges should be elaborated and about the selected strategies to meet these challenges. This requires a sound collaboration among the entities within a data ecosystem. There are several means that might be deployed for an effective and efficient collaboration, such as proper communication, mutual understanding regarding the tasks and responsibilities, transparency, clear definition of tasks assignment, and exchange of best practices. The entities in a data ecosystem should come up with a set of mechanisms that they perceive as useful for their collaboration and should prioritize these mechanisms. Furthermore, they should agree upon the deployment of each of the prioritized mechanisms.

With many stakeholders in a data ecosystem and different interests it is not always easy to come up with a set of mechanisms that earn the approval of all stakeholders. Design thinking, we suspect, can be exploited to obtain a prioritized list of the mechanisms approved by the stakeholders.