Visual Analytics: Transferring, Translating and Transforming Knowledge from Analytics Experts to Non-technical Domain Experts in Multidisciplinary Teams

Over the past decade, the Business Analytics (BA) field, also known as Business Intelligence & Analytics (BI&A) (Chen et al., 2012), has undergone an important shift. While in the past the main users of analytics technology were data scientists and IT professionals, these days there is a greater emphasis placed on self-service analytics for non-technical decision makers (Stodder, 2020). This is because domain experts, also known as subject matter experts, have the necessary domain knowledge and expertise to make sense of contextual data and turn them into value-adding insights. This shift, in turn, accelerated non-technical decision makers’ use of data visualization and visual analytics, as the new generation of analytics that does not require data science or IT backgrounds.

Thus, rather than just ‘consuming’ reports prepared by data scientists as they did in the past (Stodder, 2015), non-technical domain experts now have access to easy-to-use data visualisation tools that enable them to analyse data in a more intuitive way, driven by the problem-at-hand. Indeed, for today’s managers operating in data intensive environments, “the only way to make sense of large data sets is through data visualization” (Berinato, 2016, p.1). Through an ongoing process of interactive visual data exploration, they also keep discovering new questions along the way (Swoyer, 2013).

However, the term ‘data visualization’ (DV) is frequently reduced to mean static visual representations of numerical data (i.e. visual outputs) in the form of graphs or charts (Berinato, 2013a; Stodder, 2013). As Berinato (2016) explains, ‘[i]n some ways, ‘data visualization’ is a terrible term. It seems to reduce the construction of good charts to a mechanical procedure. It evokes the tools and methodology required to create rather than the creation itself.’ (p.16). Consequently, DV tends to emphasize data and DV tools and applications.

Similarly, the related term ‘visual analytics’ (VA) also interpreted in a variety of ways by different communities. On one side, data scientists use the term VA to describe specialized applications of advanced data analytics, such as those in machine learning and neural networks, used to visualize model execution (cf. Hohman et al., 2019; Yuan et al., 2021). On the other hand, when used by non-technical business decision makers VA is understood to combines visualization with new practices for analytical reasoning and test-and-learn inquiry through interactive data exploration and discovery (LaValle et al., 2010; Stodder, 2013).

While acknowledging different interpretations, in this paper we focus on VA for non-technical users and consider it to include interactive DV. Thus, we define VA as a process of interactive visual exploration by non-technical decision-makers with domain expertise and contextual knowledge necessary to make sense of data, gain valuable insights and take data-informed actions. This interpretation corresponds to Stodder’s (2013) notion of ‘visual analytics’ and Berinato’s (2016) “discovery-focused exploratory visualization” (Berinato, 2016, p.5). Used as a verb (i.e. a process of visualization) rather than a noun (i.e. visual outcome) as in DV, visualisation is thus interpreted here as intuitive and open-ended. Understood in this way, VA shifts the focus from tools, visual outcomes (e.g. graphs and charts) and even data themselves, to the process of sense-making through visual exploration of data, in a given context and by domain experts from any organizational level (not just executives).

While the industry demand for DV and VA tools is rapidly growing (Berinato, 2013b), industry practices are still limited by the lack of skills among non-technical users (Stodder, 2020). Thus, “[t]he greatest concern centers on whether employees will have adequate knowledge and skills to make effective use of the tools” (Stodder, 2013; p.5). As Berinato (2013a, b) notes businesses need insights “not just pretty pictures” (p.1).

Moreover, to deal with increasingly complex problems, organizations are now building multidisciplinary analytics teams, which include analytics experts and non-technical subject matter experts (Davenport, 2020; Talagala, 2019; Zhang, 2019). As Hindle et al. (2020) note “analytics is a multidisciplinary endeavour” (p.489). Teams are also created in response to the previously-held unrealistic expectation about the all-encompassing knowledge and skills of data scientists and other analytics experts, which were expected to span data science, IT and business (Baskarada & Koronios, 2017; Baumeister et al., 2020; Zhang, 2019).

However, organizations building teams around VA are now faced with a new kind of knowledge gap. On one side there is a VA expert with an in-depth understanding of analytical tools and techniques but lacking domain and contextual knowledge necessary for visual data exploration and interpretation. On the other side, there is a decision-maker, with a domain knowledge and contextual understanding of data, but lacking VA expertise, which involves more than data visualisation. At the same time, the tacit knowledge (Polanyi, 1966), which is involved in visual data exploration process, makes it very difficult, if possible at all, for a domain expert to articulate, let alone delegate this process to a VA expert. Although not in the same domain and thus outside of the scope of this paper, a similar scenario would be a medical expert exploring and making sense of visual data in a particular context, versus VA or DV expert looking at the same data without any medical expertise.

The need for further research on transfer of knowledge from analytical experts (i.e. data scientists and IT experts) to domain experts has been recognized by other researchers such as (Holzinger, 2016; Benbya et al. 2021). We aim to contribute to closing this research gap by focusing on a method of VA-related knowledge sharing from VA experts to domain experts in multidisciplinary teams. Guided by the stated differences between VA and DV, we recognise the role of a VA expert to be different from that of a DV expert. While a DV expert predominately focuses on data visualisation, the skills and competencies of a VA expert are much broader and include other aspects of visual analytics, such as data modelling, data quality process as well as design and implementation of a data exploration environment.

In framing of our research, we also recognize that the research problem of knowledge sharing in general, including knowledge transfer, has been investigated by the knowledge management (KM) field for decades – see for example Nonaka (1994), Alavi and Leidner (2001), Malhotra (2004), Newell (2014). However, analytics opens new challenges for KM (Pauleen & Wang, 2017; Tian, 2017), including sharing of analytical insights (Marjanovic, 2021). We extend this line of thinking to visual analytics, in particular to the challenge of VA-related knowledge sharing between VA experts and domain experts, which is yet to be investigated by the analytics, VA, KM and Information Systems (IS) communities.

The research described in this paper was conducted in the context of a long-term industry-wide research project called the ‘Visual Historical Atlas of the Australian Co-operatives’ (VHAAC) project (Patmore et al., 2020, 2021, 2022) . This national, large-scale project, funded by the Australian Research Council, was implemented over four years (2017–2020) by a multidisciplinary team comprising of VA experts with expertise in design and implementation of interactive VA environments, including data collection and modelling and domain experts with expertise in the history of the Australian Co-operatives, labour history and historical research methods. The project included design and implementation of the VHAAC environment – an industry-wide, online interactive visual data exploration environment of the Australian cooperatives, ranging from the 1820s to today. Now in its sixth year, the VHAAC environment is the most authoritative source of data on the Australian cooperatives to date.

Within this context, VA-related knowledge sharing between the VA experts and the domain experts was critical for the success of the VHAAC project. However, this knowledge sharing went beyond the common practice of co-design of various visualisations. Instead, its main objective was to empower the domain experts to use the VHAAC as an interactive, visual “thinking tool” (Thorp, 2013), guided by their knowing-in-practice (Orlikowski, 2002). The VHAAC environment thus provided an opportunity as well as created a necessity for a practice-inspired research project described in this paper.

To answer the stated research question, we conducted an action design research (ADR) project which resulted in an innovative method for knowledge sharing based on VA boundary objects. Informed by prior research on secondary design (Germonprez et al., 2011; Lakew & Aryal, 2015), we recognized VA experts as the ‘primary designers’ and domain expert users of VA as ‘secondary designers’ engaged in design-in-use. To emphasize and honour their principal role in visual data exploration, we renamed secondary into principal designers.

The paper thus proposes a method of VA-related knowledge sharing between primary and principal VA designers. The method is theorized as a as a three-phase process of knowledge transfer, translation and transformation, which each phase using a different type of VA-specific boundary objects (BOs). Phase 1 includes syntactic BOs in the form of static data visualisations (i.e. visual representations) and data-driven stories. Phase 2 includes semantic BOs in the form of co-created problem-driven visual stories and an exemplary external visual exploration environment used for collaborative future imagining. Phase 3 resulted in the visual data exploration environment (i.e. the VHAAC) itself becoming a pragmatic BO, which in turn enabled decision-makers to shift the focus from data to seeing and thinking ‘through data’. In addition to the proposed method of knowledge transfer, another type of ADR design knowledge includes a set of design principles related to VA boundary objects and their use.

This research offers several theoretical and practical contributions to the field of Information Systems (IS), at the intersect of analytics or more precisely VA, and knowledge management (KM). First, our research addresses the previously observed research gap related to the transfer of knowledge from analytics to non-technical experts (Holzinger, 2016; Benbya et al. 2021), which we investigate in the context of VA for non-technical decision makers. We also demonstrate that knowledge transfer is not sufficient and needs to be expanded with knowledge translation and transformation. Second, we also advance the the current research on BA by focusing on still under-researched VA for non-technical domain experts. We highlight the critical role of their domain knowledge for data exploration and sense-making in VA. In doing so, we extend the previous finding by Ghasemaghaei et al. (2018) about the importance of domain knowledge in analytics. Third, our research expands the well-established KM research on organizational knowledge sharing processes (Alavi & Leidner, 2001; Nonaka, 1994) to the emerging stream of research on VA. In particular, we explore the process of VA-related knowledge sharing in multidisciplinary teams at a very granular level and through the use of different VA-related boundary objects.

In terms of practical contribution, we propose a method that we developed, refined and evaluated-through-use in our own large-scale industry-wide VHAAC project, now for six years. We also explain how the same method could be adopted and further refined by other multidisciplinary teams in different contexts, aiming to better leverage tacit knowledge of their non-technical decision makers.

The paper is organized as follows. Section 2 describes the related work and builds the necessary foundations. Section 3 introduces our research context while Section 4 outlines our adopted Action Design Research Method (ADR). Section 5 describes the combined theoretical lens of boundary objects and secondary design, which we used as ADR kernel theories. Section 6 presents three groups of VA-related boundary objects (syntactic, semantic and pragmatic) and explains how there were used to facilitate knowledge sharing in our multidisciplinary team. Section 7 answers the research question by theorizing a three-stage method for knowledge transfer, translation and transformation from VA experts to domain expect. The proposed method is accompanied by a set of ADR Design Principles articulated in Section 8. Section 9 summarizes the main theoretical and practical contributions while Section 10 concludes the work by discussing study limitations and future opportunities for research.

In this project we used the Action Design Research (ADR) method (Sein et al., 2011). This method recognizes and conceptualizes “the research process as containing the inseparable and inherently interwoven activities of building the IT artefact, intervening in the organization and evaluating it concurrently” (Sein et al, 2011, p.1). ADR was originally proposed by Cole et al. (2005) with the two key objectives: (i) to use valid scientific research methods to solve (a) practical problem(s) experienced by the researcher/practitioner; and (ii) to contribute to the existing body of knowledge in relevant areas by creating new design artefacts.

ADR draws its origins from two research methods: Action Research (AR) and Design Research (DR), also known as Design Science Research (DSR). By proposing to combine AR and DR, Cole et al. (2005) argued that an integrated approach is required to emphasize the research relevance, problem solving through intervention in the real-life setting, reflection and learning (Baskerville & Wood-Harper, 1998). as well as knowledge created through design (Hevner et al., 2004).. It is important to note that ADR is more than a combination of the original AR and DR methods (Sein et al., 2011). In ADR the phases of building, intervention and evaluation (BIE) of a design artifact are not separated as in DR, but highly intertwined. A combined learning and reflection phase is an important phase of ADR, that is not separate but constantly emerging through BIE. Finally, ADR recognizes different kinds of design artefacts, including IT systems as well as organizational interventions.

Both DSR and ADR have been used for design and implementation of DV and VA tools – see for example Toreini et al. (2021). However, in this project we shift the focus from design of VA tools to design of an innovative method for knowledge sharing from VA experts to domain experts, who are making sense of data through interactive visual data exploration thus using VA on their own and as a thinking tool. Therefore, the main intended ADR design artefact in our research is not a VA technology, but a knowledge sharing method across disciplinary boundaries in a multidisciplinary team.

Our research project followed the ADR phases (Sein et al., 2011) depicted by Fig. 1 and implemented them as follows. The project was initiated to address a problem identified in a real-life setting experienced by the co-authors who were the leaders of the multidisciplinary team of VA experts and domain (cooperatives) experts.

Therefore, our research was practice-inspired (Principle 1) as the need for this project came from the practice of designing and implementing the VHAAC environment by a multidisciplinary analytics team. The main ADR design artefact in this project – a knowledge sharing method – was ‘theory-ingrained’ (Principle 2), because its design was guided by the theory of BOs (Carlile, 2004) and informed by the theoretical concept of secondary design (i.e. design-in-use) (Germonprez et al., 2011), as described in the next section. The Build, Intervene and Evaluate (BIE) phases of our ADR project were fully intertwined, with the emerging method shaping and being shaped through its use (Principle 3 of ADR), as described in the subsequent sections. The roles of VA experts and domain experts were mutually influencing (Principle 4 of ADR). We were also building the method while evaluating it in the authentic practice (Principle 5 of ADR). Throughout to the project, ADR-related research data were collected using two methods. The first method included an ongoing collective reflection-in-action (Levina, 2005) by the ADR team members, during each ADR cycle. The second method of data collection included observations of the effects of the actual intervention (Principle 5 of ADR), assessed by the domain experts’ improved ability to transform their practice (Carlile, 2004) through VA. The main design artefact gradually emerged and was shaped by its use in the real-life context (Principle 6). The proposed method was then observed as an instance of a class of problem that was sharing of knowledge from VA experts and domain experts in multidisciplinary VA teams (Principle 7). Formalization of learning included design knowledge in the form of the proposed knowledge sharing method and an initial set of design principles. The following section introduces the theoretical foundations, also known in ADR as kernel theories (Sein et al. 2011), which informed our design. Further details of our ADR project implementation are described in the subsequent sections.

This section describes different types of VA-related BOs we used to enable and facilitate VA-related knowledge sharing from the VA experts to domain experts in our project. Drawing from the ADR kernel theories, we recognize that any multidisciplinary collaboration involves a number of different boundary objects. In this research we focus only on VA-related boundary objects and their role in enabling the research team to work across- different kinds of knowledge boundaries. Also informed by the ADR kernel theories, we recognize VA experts as primary VA designers, and domain experts as secondary, here named “principal VA designers”. This renaming was done to emphasise their principal value-adding role in using their domain expertise to make sense data through visual data exploration. The three main phases of knowledge sharing and the associated BOs are described as follows:

In the first phase we used two types of syntactic VA-related BOs, with the overall purpose to raise the awareness of the domain experts of the new opportunities created by VA tools, when applied to our growing data set. The first type of syntactic BOs consisted of a number of data visualizations (DVs), created by the primary designers using the VHAAC data exploration environment, which at the time was in early stages. These DVs were used to inform the domain experts about the initial findings from the data.

The second type of BOs included data-driven visual stories that were created by VA experts out of static DVs, which were combined to illustrate possible decision-making scenarios. These stories presented to the domain experts, in a step-by-step manner, using the provided story telling features of the software application (Tableau). Both data visualization and visual stories were based on the VA experts’ understanding of decision-making needs of different intended users. In order to elicit these needs, the VA experts collected data through interviews with domain experts and design thinking activities.

Our use of static DVs as static boundary objects did result in new insights gained by the domain experts, both within the multidisciplinary team and the wider advisory group of industry practitioners. Presented in the visual form, the VA experts’ findings from the VHAAC were immediately augmented with new insights by the domain experts who could place these findings in the relevant historical context and understand their importance.

However, very soon these syntactic BOs were found to be very limited. Thus, asking users (with no previous VA experience) to communicate their data-related decision-making and data exploration needs so that they could be “transferred” into DVs and visual stories was not found to be effective. This is because of the well-known problem of “not-knowing what you don’t know”, especially when experiencing something for the first time.

At the same time the VA experts had very limited understanding of the cooperatives’ data. Apart from creating static DVs to answer experts’ initial questions, their ability to create insightful DVs and data-driven visual stories was equally limited. After evaluating our shared experience, collectively, we reached the limit of knowledge transfer across the disciplinary boundary thus creating the need for new types of VA-related BOs.

In this phase the VA experts decided to look for another type of VA-related BOs to enable translation of knowledge across disciplinary boundaries, not just its transfer, as in Phase 1. First of all, we looked for other examples of public visual data exploration environments developed in other industry sectors, similar to what we intended the VHAAC to become. Guided by the theory of BOs, we needed to make sure that any chosen example came from the context that was understandable to both VA and domain experts. We found an example of a nation-wide, visual data interactive environment using open data on senior citizens in Australia. Although very simple, this exemplary environment was very effective. First used as a syntactic BO, this environment helped the VA experts to illustrate how static data visualization (i.e. visulisation as output) differed from interactive visual exploration (visualisaion as a process) in order to facilitate the shift in their mindset from DV to VA. The exemplary environment was also used to demonstrate basic features of a DV tool, as much as it allowed us.

Informed by the previous insights from the ADR kernel theory again, in particular a possibility of using storytelling and stories for knowledge translation, we turned our attention back to visual stories. However, rather than using a data-driven method whereby static DVs are sequenced in a visual story (as in Phase 1), we shifted our collective focus from data to interesting industry problems and opportunities related to the cooperatives. In other words, our previous approach of looking at data in order to see what could be discovered there, turned out to be limited for two reasons. First, it constrained our exploration of possibilities, because it was limited by ‘what was there in data’ as perceived by VA experts. Second, our disciplinary knowledge barrier remained as data-driven stories still focused on knowledge transfer. It is important to note that this data-driven approach (i.e. looking at data in order to find and visualise interesting insights) is often practiced in the DV projects, especially if they are led by analytics experts. We discovered these limitations through our ongoing collective reflection.

Consequently, we decided to ‘avert our collective gaze from data’ and instead focus on problems and questions posed by the domain experts. For example, domain experts wanted to know: “Why are cooperatives demutualized over time”, “Why do cooperatives close down?” “What is the story of business cooperatives in Australia?” It is important to note that there were no pre-defined data attributes that could enable VA experts to search and visualize the existing data in order to answer these questions. Instead, these questions could be answered only through visual data exploration by domain experts, supported by VA experts. Consequently, we started using problem-driven visual stories as semantic BOs that in turn required very close collaboration between VA and domain experts. For example, domain experts would start to translate the problem into a number of scenarios and with VA experts proceeded to turn these into visual stories (as much as possible based on available data), while recording the limitations of the existing data sets. Whenever required, domain experts would engage in additional data collection, using for example, historical archives, databases of historical coops-related newspaper and other articles, further interviews and case studies. After collecting any new data, domain experts engaged in historical research, which was required to cross-reference, reconcile, validate and interpret the collected data in a particular historical context.

In this phase we also used a number of techniques that were designed to gradually turn domain experts, from VA users into principal designers. For example, primary designers (VA experts) started from a common practice of ‘walking with’ principal designers, first through visual representations towards interactive collaborative visual explorations, noting their comments. It is important to note that unlike the “traditional” user feedback sessions where the main objective would be to improve the main features of the user interface or application, our main objective here was not to improve what was essentially primary designers’ re-construction of domain experts’ stories. Similarly, the collected feedback was not used to capture and “reconstruct” subject matter experts’ way of thinking and visualise it. Looking from the knowledge management (KM) this would be an equivalent of trying to capture and document somebody’s tacit knowledge that according to prior KM research cannot be done effectively (Nonaka, 1994). Instead, the main objective was to facilitate the gradual process of ‘becoming’ principal designers by helping them to tap into their own knowing-in-practice more and more.

The shift from Phase 2 to Phase 3 occurred when domain experts started using visual data exploration guided by their knowing-in-practice, which resulted in different types of transformation of their own practices. The key enabler of this transformation was the VHAAC environment that we recognized as a pragmatic boundary object, which was constantly-in-making. Through the process of collective reflection-in-action we observed the following types of transformation. First, through the use of the VHAAC environment, domain experts – now in the role of principal designers (still supported by VA experts on the need basis)—started transforming and expanding the existing body of knowledge about Australian cooperatives, both in industry and research literature. For example, they started challenging the common myths about cooperatives, such as why they dissolve and de-mutualize (Patmore et al., 2021).

The second type of transformation occurred when domain experts stated providing insights gained through visual data exploration to the wider cooperatives industry sector to support decision-making by different stakeholders, especially during crisis. For example, during the Australian bushfire disasters in Nov 2019, domain experts were able to provide historical insights about cooperatives’ resilience and positive societal impact during similar natural disasters to the Australian Business Council of Cooperatives and Mutuals.

The third type of transformation occurred when the domain experts started to innovate on their own. For example, they invented a new practice of crowdsourcing of cooperatives-related data form other industry practitioners, using the form they designed for an easier upload into the VHAAC by the VA experts.

Our use of the pragmatic BO (the VHAAC) in Phase 3 also resulted in another shift of our collective VA-related mindset. While we started with the main focus was on VA technology features and data, in this phase we observed that principal designers started to explore and see their practice and domain knowledge ‘through data’. Consequently, both technology and data started to ‘fade into background’, thus becoming the means of VA-enabled transformation of their domain knowledge. This in turn impacted on the ongoing design-use of the pragmatic BO that is the VHAAC environment as well as the domain experts’ knowing-in-practice.

In response to the stated research question, we theorize a three-phase method for a gradual transfer of knowledge from primary to principal VA designers, using VA-related syntactic, semantic and pragmatic boundary objects. The method draws from our collective insights from the project. We recognize the problem of knowledge sharing in the VHAAC project as an instance of a class of problems, which is is knowledge sharing in multidisciplinary VA teams among VA experts and non-technical domain experts.

The proposed method is depicted by Fig. 2. As shown the first phase focuses on a knowledge transfer achieved by transfer of syntactic BOs across knowledge boundaries in the form of static DVs and their combinations. The second stage includes knowledge translation through co-design of semantic BOs as well as the use of exemplary BOs for collective future imagining. The third phase involves knowledge transformation, which is focused on empowerment of domain experts to ‘see through visual data’ in order to transform their own and wider industry practice. Further details of each phase are summarized in Table 1, including examples of different types of VA-related boundary objects. It is important to point out that depicted phases were not defined up-front. Instead, they were emerging throughout ADR cycles. Therefore, building, implementation and evaluation of the main ADR design artifact (i.e. the proposed method) were highly intertwined.

The following design principles capture design knowledge related to the proposed 3-phase method of VA-related knowledge sharing from primary to principal designers. They were derived through collective reflection-in-action (Levina, 2005) by the project team and refined throughout the ADR cycles.

While developing the principal designers’ ability to engage in a process of visual data exploration, we found that the actual sequence of boundary objects was important. For example, attempting to use visual stories as semantic or even pragmatic boundary objects immediately and without progressing through previous stages was not found to be very effective. Thus, creating awareness of data and technology through the use of syntactic BOs, followed by exploration of possibilities through ‘collective future imagining’ made possible by semantic BOs, enabled us to form the foundations of our shared knowledge. This in turn paved the way for co-creation of new semantic boundary objects, and later on our work at the pragmatic knowledge boundary.

When the initial static data visualizations were used as syntactic BOs, we observed that these BO were ‘transferred’ across knowledge boundaries. Using a metaphor, an ADM team member described this transfer as “tossing over the disciplinary wall”. Similar transfer of BO also occurred when primary designers used design thinking to elicit data and decision-making requirements from the domain experts, in order to turn insights into data-driven visual stories. In both cases these BOs remained unchanged through transfer. While initially effective (in Phase 1), these BOs did not result in the transfer of knowledge that would empower domain experts to effectively engage in visual data exploration on their own.

On the other hand, problem-driven visual stories used as semantic Bos in Phase 2, were effective because they were co-created through a collaborative data exploration process. More importantly, this exploration process resulted in what we termed ‘the critical shift’ in a gradual transfer of VA knowledge. The shift occurred when the principal designers) took over the co-creation of visual stories from the primary designers.

We found that BOs that engaged principal designers in ‘pattern thinking’ were very effective for knowledge translation and ultimately knowledge transformation. For example, the semantic BO we used in Phase 2 (a visual exploration environment from another industry sector), enabled secondary designers to ‘see’ new possibilities for the VHAAC environment by thinking in patterns. This way of thinking was stimulated by a deliberate practice of ‘collaborative future imagining’ implemented through brainstorming and guided by the questions designed to initiate pattern thinking. However, we found these BO to be effective only if the principal designers could relate to them either through their experience or common knowledge. For example, the VA environment related to aging was easy to use and understand due to our common knowledge. This in turn enabled more effective thinking in patterns by ‘translating’ the experience with this VA environment into imagined possibilities for VHAAC. We also observed the evidence of thinking in patterns in Phase 3, when the multidisciplinary ADR team came up with new ideas for the future VHAAC-like VA environments. An example is a future interactive shared VA environment for agricultural cooperatives to understand and negotiate ownership of their data.

We found that when the principal designers (i.e. domain experts) shared both semantic and pragmatic BOs with a wider group of industry practitioners, their learning to explore data became even more effective, and certainly more engaging. Even while these BOs were in the development stage. As these domain experts all have shared contextual knowledge, we found that the principal designers were able to share the experiential knowledge of “their own thinking process”. The same type of experiential knowledge could not be easily shared with the primary designers due to boundries in disciplinary knowledge.

We observed that syntactic BOs in the form of static data visualisations placed too much emphasis on ‘seeing data’ in a new visual form. While important, especially in Phase 1, we also found that this was limiting. We argue that the effective knowledge transfer occurs when the principal designers start to see their practice (i.e. problems-at-hand and future opportunities) ‘through data’. When this happens, the VA environment such as VHAAC becomes a thinking tool. The use of such a tool is not pre-defined as in DV. Instead, it is guided by the domain expert’s knowing-in-practice.

The proposed design principles are by no means exhaustive. As such, they are expected to be refined and extended through further research and use in other contexts.

This research offers several theoretical and practical contributions to the IS field, as follows. By focusing on VA for non-technical decision makers, we address an important research gap in the BA-related literature in IS, which remains focused on the mainstream analytics and to a much lesser extent, data visualization. The proposed method of knowledge transfer, translation and transformation from VA experts to domain experts also contributes to the growing body of IS research on organizational use of analytics (Abbasi et al., 2016; Vidgen et al., 2017). At the same time, our research opens an interesting research opportunity to investigate the proposed method as a potential mechanism for analytics value creation in an organization setting.

The proposed 3-phase method also addresses the previously observed research gap related to the transfer of knowledge from analytics to non-technical experts (Holzinger, 2016; Benbya, et al. 2021). Based on our research findings, we argue that knowledge sharing between VA experts and domain experts goes beyond VA-related knowledge transfer. As demonstrated in our research, this process also includes knowledge translation and ultimately, knowledge transformation, which empowers non-technical decision makers to transform their own practice, with VA becoming ‘a thinking tool’. The proposed method also confirms the previous finding by Ghasemaghaei et al. (2018) about the importance of domain knowledge in BA. We extend this research to VA and emphasize the role of domain experts who are best positioned to make sense of visual data through interactive exploration, guided by their knowing-in-practice (Orlikowski, 2002).

Our proposed method also contributes to the KM literature on knowledge sharing processes, exemplified by (Alavi and Leinder, 2001; Nonaka, 1994), which we advance by exploring a VA-related knowledge sharing process in multidisciplinary teams at a very granular level and using VA-related boundary objects. In doing so we also contribute to the emerging stream of research, which focuses on the role of KM in BA, as discussed by Hota et al. (2015), Pauleen and Wang (2017), Tian (2017) and Marjanovic (2021).

By situating our project in an industry-wide context, we also contribute to an important nstream of research on societal use of analytics (Gupta et al., 2018) as the new frontier of analytics research, in particular VA research. We argue that this is a important contribution as industry-wide projects of this complexity and duration, with the authors as co-leaders using the research outcomes to advance the practice of VA, are still rare in the IS literature. Also, given the societal importance of cooperatives (ICA, 2021a, 2021b), our project opens new opportunities for further IS research beyond VA in this important industry domain, which is yet to be explored by the IS community.

In terms of practical contributions, we offer a method that was developed, extensively evaluated and refined through use in our own ongoing large-scale VA project, now for more than six years. This is an important contribution, given the current industry trend of deploying multidisciplinary analytics teams.

The method goes beyond the common industry practice of developing visusal dashboards, whereby data visualization experts work together with the user to elicit user requirements and (co-)create various visualisual outcomes (i.e. visual representation). We used this approach in Phase 1. Then in Phase 2 we observed that the visualisation process of 1) getting to know data, 2) creating visual explanations of data and 3) framing and visualisation of stories, which is often used in practice, is entirely data driven. We then recognised the need to ‘avert our collective gaze from data’ which was an important shift in our VA mindets. This led us to Phase 3 in which we focused on developing non-technical domain expert’s ability to use VA as a thinking tool, guided by their knowing-in-practice (Orlikowski, 2002).

The proposed method could be adopted and refined by other multidisciplinary teams in different contexts beyond cooperatives. In particular, it is suitable for other industry-wide interactive visual data exploration environments, which are accessed by a diverse group of stakeholdersts with different decision-making needs that cannot be articulated in advance. Examples include future VHAAC-like environments in aged care or education sectors. We acknowledge that the nature of data in these environments would be different, as well as data collection methods that in our project included historical data. However, the challenge of knowledge sharing across disciplinary boundaries would be present, thus making our proposed method applicable to these settings.

Inspired by prior call for future IS research to address COVID-19 challenges by Chang et al. (2021), we could also envisage a future industry or society-wide environment for visual exploration of COVID-19-related data by different stakeholders with diverse domain experitise. Our proposed method for knowledge transfer, translation and transformation would be very relevant in the envisaged environments, given the stakeholders’ very diverse domain expertise coupled with the need for rapid knowing-in-practice, which is both used and developed through visual data exploration by the domain experts themselves.

Finally, our research calls for a different kind of VA training for domain experts. Instead of being a source of “user requirements” for design of visual dashborads, we argue for their knowing-in-practice to be recognised as the most important for VA value creation. We thus envisage the proposed method to be used as an ongoing method of organisational learning through VA, instead of a short-term VA-focused skill-based training.

This paper described an innovative Action Design Research (ADR) project focused on the practice-formed research challenge of knowledge sharing between VA experts and domain experts working in the same multidisciplinary VA team.

The proposed knowledge sharing method and the articulated design principles have important consequences for education and training of non-technical decision-makers in using VA for visual data exploration rather than just static data visualization. Therefore, workplace training and classroom learning, where the main emphasis is on skill-based training and technical features of a data visualization platform, is less likely to develop practitioner’s ability to use visual data exploration as a thinking tool. Rather than going through training manuals, which we observed to be the main mode of training in many workplaces, it is important to engage practitioners in a gradual process, first through the use of syntactic boundary objects (developed by VA experts), semantic boundary objects co-created through knowledge sharing and then gradually shaped into pragmatic boundary objects through design-in-use.

Our research project has several limitations. It was conducted in a specific context (Australian Co-operatives). The ADR team also conducted this research in their own practice, as researchers/practitioners who were committed to collaboration and knowledge sharing. We acknowledge that this may not always be the case, as different yet-to-be explored barriers may exist in different organizational settings. For example, VA experts external to the organization, may face trust or some other contextual issues when working with subject-matter experts for the first time. We also recognize that different multidisciplinary teams experience different team dynamics, which in turn may impact on the process of knowledge transfer. Therefore, further research is required to understand how the proposed method would work in other organizational contexts and different types of multidisciplinary VA teams (e.g. self-selected or put together). We also did not investigate the issue of knowledge retention. Given the onoing nature of our project, which is still in progress, we see knowledge retention as an important future direction of our research after the completion of the current project.

As the project continues, our project team, now led by the domain experts, continues to discover new possibilities for future VHAAC-like, VA projects in other domains. In order to set the foundations for these projects, our current and future research include design and implementation of innovative learning methods for both VA experts and non-technical VA users (i.e. ‘principal designers’) in industry and university settings.