The ecosystem blueprint: How firms shape the design of an ecosystem according to the surrounding conditions

Ecosystems are formed by organisations that jointly create a value proposition that a single firm could not create in isolation. To deliver this value proposition, the partners need a focal firm, the orchestrator, to be align them towards the joint value proposition. Thus, how orchestrators design the alignment structure of an ecosystem is at the very heart of the ecosystem concept – yet it has not been sufficiently addressed by extant research. This is all the more true for the question of how the design of an ecosystem is shaped depending on surrounding conditions. This paper applies a qualitative study with ten cases and, based on the attention-based view of the firm, contributes to research on ecosystems in several ways. First, it explains which ecosystem designs are beneficial under which conditions. Second, it elucidates the structure and activities within ecosystems and shows that start-ups can be just as good ecosystem orchestrators as incumbents. Third, it explains the circumstances under which single vs. multi orchestrator ecosystems occur. Fourth, it presents the conditions when incumbents or start-ups make better orchestrators. Finally, it is among the first studies to apply the attention-based view to business ecosystems, and shows that doing so yields intriguing insights into this emerging field of research.


Introduction
Recent years have seen a boom in the ecosystem concept (Adner, 2017;Jacobides et al., 2018;Moore, 1993;Rong and Shi, 2015), in fields such as strategy (e.g. Iansiti and Levien, 2004a;Moore, 1996;Parente et al., 2019), innovation (e.g. Adner and Kapoor, 2010;Dattée et al., 2018;Davis and Eisenhardt, 2011), and organisation (e.g. Davis, 2016;Kapoor and Agarwal, 2017). Expressed in numbers, the frequency of articles published on this topic in the leading strategy journals has grown sevenfold over the last five years (Autio and Thomas, 2019;Jacobides et al., 2018;Kapoor, 2018) and '[…] the term "ecosystem" has become pervasive in discussions of strategy, both scholarly and applied' (Adner, 2017, p. 39). At the same time, in industry practice, firms are increasingly building ecosystems (Alturi et al., 2017;Fuller et al., 2019;Palmié et al., 2020) and the ecosystem concept is even expected to take over traditional thinking in products and markets (Catlin et al., 2018;Rong, Hu, Lin, Shi and Guo, 2015a;Rong, Wu, Shi and Guo, 2015b), since 'individual corporations are no longer adequate to serve as the primary unit of analysis' (Baldwin, 2012a, p. 20). Instead, the key challenge for organisations will be the management of distributed activities within ecosystems (Keupp et al., 2012).
The core of the ecosystem concept is the creation of a joint value proposition for the customer that a single firm cannot achieve in isolation (Adner, 2017;Autio and Thomas, 2019;Kapoor, 2018;Moore, 1993;Parente et al., 2019;Shipilov and Gawer, 2019), and which is based on complementary modules, i.e. the organisations involved in the ecosystem need to either specifically develop or, at least, mutually adjust their respective modules for the joint value proposition to be delivered (Jacobides et al., 2018). In order to achieve this, the ecosystem partners need to be aligned towards the value proposition (Adner, 2017) by a central player, the orchestrator 1 (Adner, 2017;Altman and Tushman, 2017;Dattée et al., 2018;Jacobides et al., 2018;Kapoor, 2018;Moore, 1993Moore, , 1996Nambisan and Baron, 2013). This is essential since all ecosystem partners have individual goals and agendas but, at the same time, must develop and mutually adapt their modules in reciprocal relationships with the other partners (Adner, 2017;Jacobides et al., 2018;Kapoor, 2018;Masucci et al., 2020).
Thus, what makes ecosystems unique (Ganco et al., 2020), and what defines them, is 'the alignment structure of the multilateral set of partners that need to interact in order for a focal value proposition to materialize' (Adner, 2017, p. 40; see also Autio and Thomas, 2019;Kapoor, 2018;Shipilov and Gawer, 2019). The design of the alignment structure is particularly crucial since all partners involved in an ecosystem pursue their individual agendas, which requires joint decision-making by all partners involved (Jacobides et al., 2018). From the perspective of the attention-based view of the firm (Ocasio, 1997), such decision-making requires the decision maker to focus their attention on available information about the environment and to understand alternative possible decisions (Ocasio, 1997;Weick, 1979). Additionally, for the same reason, ecosystem actors can only create a joint value proposition if they focus their attention on relevant innovation opportunities, since individuals are unlikely to act on opportunities that do not catch their attention (Barnett, 2008;Ocasio, 1997). Attention, in turn, is influenced by the structures that decision-makers find themselves in (Ocasio, 1997; see also Gavetti et al., 2007;Joseph and Ocasio, 2012;Ocasio and Joseph, 2005). Thus, it is essential to understand how the orchestrator designs the alignment structure of an ecosystem to facilitate appropriate distribution and allocation of attention and, thus, joint decision-making and the creation of a joint value proposition. Answering this question addresses three crucial gaps in current literature on ecosystems and the attention-based view of the firm. First, to our knowledge, the attention-based view has not yet been used to study ecosystems or other forms of inter-organisational networks or meta-organisations (with the noteworthy exception of the study by Maula et al. (2013) on the saliency of industry peers and VC-funds and their influence on the attention of managers to technological chance). Second, within the ecosystem domain, insights on how firms shape the design of ecosystems is still insufficient (Adner, 2017;Dattée et al., 2018;Jacobides et al., 2018;Phillips and Ritala, 2019). This is because the majority of publications on ecosystem design are merely conceptual, thus they lack an empirical foundation (e.g. Baldwin, 2012b;Brusoni and Prencipe, 2013;Ganco et al., 2020;Teece, 2007;Williamson and de Meyer, 2012;Zahra and Nambisan, 2012). This leads to a high level of abstraction in previous findings (Kapoor and Lee, 2013) and a lack of in-depth insights into how ecosystems are being designed (Adner, 2017). Thus, researchers have repeatedly called for research along these lines (c.f. Jacobides et al., 2018;Laamanen, 2017;Phillips and Ritala, 2019). Also, these conceptual works are usually based on existing theoretical reasoning that might not be suitable to 'help explain the distinct value creation and capture dynamics within and between ecosystems' (Jacobides et al., 2018(Jacobides et al., , p. 2256. And, third, ecosystem design differs depending on the surrounding conditions (Iansiti and Levien, 2004b), i.e. different ecosystem structures are beneficial under different conditions (Jacobides et al., 2018). This raises 'the need to revisit assumptions of ecosystem uniformity and to establish a typology of ecosystem designs best suited to varying contexts' (Dattée et al., 2018, p. 493).
In order to address these gaps in the existing literature, we use a qualitative multi-case study (Eisenhardt, 1989;Yin, 2014) with ten cases. In particular, we focus on the design of ecosystems in the early stages of their lifecycles (Moore, 1993(Moore, , 1996Rong and Shi, 2015) since the emergence of ecosystems is still an unexplored field of research. However, it is particularly essential for firms to understand the design of ecosystems in these early stages since ecosystems do not emerge on their own, but must be purposefully built (Dattée et al., 2018;Fuller et al., 2019;Jacobides et al., 2018). In order to achieve comparability across our cases, all of our cases are characterised by value propositions that can be defined as digital services (c.f. Chatman and Jehn, 1994;de Wulf et al., 2001;Schoenecker and Cooper, 1998).
Our research reveals two surrounding conditions that must be considered when shaping an ecosystem's designand both playing a crucial role within the attention-based view as well: 1) the substantive uncertainty of the environment (Dosi and Egidi, 1991) and 2) the effective distance of knowledge between the orchestrator's existing knowledge and the knowledge required to align the ecosystem partners (Afuah and Tucci, 2012). We therefore develop four propositions that contribute to research on ecosystems in several ways. First, and foremost, we shed light on the under-researched topic of how the orchestrator designs the alignment structure of an ecosystem, thus providing a better understanding of how ecosystems are structured and governed. Second, we contribute to the understanding of the surrounding conditions in which ecosystems might be particularly beneficial, and address the related calls for research by Adner (2017), Dattée et al. (2018), Jacobides et al. (2018), and Phillips and Ritala (2019). Third, we help understand what makes a 'good' orchestrator, depending on the surrounding conditions of the ecosystem. Fourth, in so doing, we contradict existing findings and show that in some situations, start-ups are excellent ecosystem orchestrators. Fifth, we reveal differences in ecosystem design between single-or multi-orchestrator ecosystems. Sixth, by being among the first to apply the attention-based view of the firm to the emergent phenomenon of ecosystems, we show that ecosystems can both focus attention on novel fields of knowledge, i.e. distant search, and also overcome limitations of innovativeness when focusing attention on local domains of knowledge, i.e. local search.   Virtual Mobility 'So, we create digital copies of the world basically, that's in one sense what we do. And these digital copies entail 3D models which are very, very accurate and they also entail the ability towards a second stage interact with them, assimilate, make a basically dynamic environment which they can do predictive analysis and also enables you to incorporate third party data too, to actually overlay over the 3D models, […].' (B1) 'Yes, but we're the ones creating the ecosystem because everyone is running through us, so, without us, the ecosystem doesn't exist.' (B1) ' […] there are basically three components [to our product]; there's the whole capture side and recording the whole data, then there is the side to make it interactive or to put it into a gaming engine to make it into a whole simulation, and then there is the part of processing all the data and also visualising all this data. The first part is what we excel in, so this is our core competence in that sense.

The second part is where there is plenty of people who already have in this in terms of gaming engines, it's not something new, so it's something we don't want to build ourselves it doesn't make sense. And the third part is the computational power which is something we also need as a commodity. But it's a commodity which is costly, and it's a commodity where you want to have strong partners to get that commodity a bit cheaper or to be able to integrate better with that commodity. And one of these examples could be [Graphic Processing Units Manufacturer] a provider of graphic cards, so they are providers of computational resources, [Game Engine
Developer] is a provider of gaming engines so they're the provider of that particular part of the product, and we're the provider of the 3D model. Then we, as a company, are bringing all of these three parts together and, in the end, sell one single product to [the] car manufacturer. And why we're really pushing for this is because in the end we have really strong synergies.' (B1) 'In the end, they have to interact with each other as well because the gaming engine has to run on the computational resources and they already do; they already optimize, so they have to interact as well. If they don't interact, in the end it makes it much more difficult, or then another partner for instance, [Graphic Processing Units Manufacturer], it would make more sense who is more willing to interact in that triangle. So, the more they interact, the better it is for the product.  to specifically create new modules, or at least mutually adapt existing modules, that complement the modules provided by the other partners (Baldwin and Clark, 2000;Jacobides et al., 2018;Liu and Rong, 2015). This requires that the various actors are mutually aligned with one other towards the joint value proposition (Adner, 2017;Nambisan and Baron, 2013). The alignment is inherently multilateral, i.e. the connections among the players cannot be composed as bilateral arrangements, in order to ensure that all modules function with each other and make use of the full potential of complementarity (Adner, 2017;Autio and Thomas, 2019;Jacobides et al., 2018;Shipilov and Gawer, 2019). This implies a strong dependency among the players, especially since the mutual adaption of modules causes significant ramp-up costs (Adner, 2017;Jacobides et al., 2018;Kapoor and Lee, 2013;Ozalp et al., 2018). Consequently, if one player fails or leaves the ecosystem, it fails, or at least struggles, as a whole. This makes ecosystems all the more critical since actors within it are still independent economic actors with individual agendas and goals (Dattée et al., 2018;Moore, 1996). Hence, the design of the alignment structure between the partners is at the very heart of the ecosystem concept (Adner, 2017;Jacobides et al., 2018).

Current findings related to ecosystem design
So far, research on ecosystem design has been the subject of many studies, which have looked at single elements or aspects of alignment in isolation. Links among the actors are certainly a cornerstone in this regard (Adner, 2017) and deal with the exchange of money or goods, as well as influence and, thus, the question of how ecosystems are governed. Ecosystems are less hierarchical than supply chains, yet, on the other hand, they require some hierarchy and control to ensure the alignment of actors towards the value proposition (Autio and Thomas, 2019;Jacobides et al., 2018). Additionally, other authors claim that ecosystems are loosely connected networks of actors (e.g. Clarysse et al., 2014;Iansiti and Levien, 2004b;Moore, 1993Moore, , 1996. The importance of links between actors for technological progress has been subjected to simulations and empirical research (e.g. Ganco et al., 2020;Masucci et al., 2020;Ozalp et al., 2018;Rong et al., 2015a).
In terms of the actors involved in an ecosystem, the orchestrator clearly plays a key role, as the orchestrator is the actor in charge of designing the alignment structure, as well as the main decision-maker within an ecosystem (Dattée et al., 2018;Jacobides et al., 2018). In this vein, authors have claimed that the orchestrator brings in key resources and infrastructure (e.g. Clarysse et al., 2014;Kapoor and Lee, 2013;Zahra and Nambisan, 2012) or dynamic capabilities (Teece, 2007). Thus, these works consider the orchestrator to be a large, powerful, and established company. Some articles, however, claim that the orchestrator might not always be the most powerful firm within the ecosystem, but rather one using other means to exert control based on knowledge, status, or key resources and technologies (e.g. Brusoni and Prencipe, 2013;Gulati et al., 2012;Williamson and de Meyer, 2012). In general, most works on ecosystem partners consider ecosystems to be based on specialised firms, which have very specific roles and activities within the ecosystem (e.g. Brusoni and Prencipe, 2013;Moore, 1996;Williamson and de Meyer, 2012).

Methods
A case study is an ideal approach if a (complex) phenomenon is little known and existing aspects are incomplete, fragmented, or contradictory (Eisenhardt, 1989(Eisenhardt, , 1991Yin, 2018). As shown in our introduction and literature section, this is the case for ecosystems and, even more specifically, for its design. We use a multiple-case approach for our research since it allows for the collection of comparative data and is likely to provide more accurate and generalizable insights than a single case would (Eisenhardt, 1991;Ozcan and Eisenhardt, 2009), and this approach strengthens the external validity of a case study (Eisenhardt and Graebner, 2007;Gibbert et al., 2008;Goffin et al., 2019;Yin, 2018). On top of this, since many studies in the ecosystem field remain conceptual, the rich empirical insights generated by a multi-case study may be of particular value to understand this rising phenomenon.

Case selection
We used a theoretical case sampling (Eisenhardt, 1989;Yin, 2018) based on several criteria. First, and most obviously, the cases selected needed to represent the literature's view of an ecosystem. Given the novelty of the concept, there are still different views on the ecosystem construct, which creates some confusion about what is or is not labelled an ecosystem (Adner, 2017;Jacobides et al., 2018;Oh et al., 2016;Ritala and Almpanopoulou, 2017). Another issue with many existing concepts about ecosystems is that it is often difficult to decide whether a certain company is a part of the ecosystem or not. For this reason, we apply Adner's (2017) structural stream as it is 'more clearly distinguishable from other available strategy constructs' (Adner, 2017, p. 40; see also Autio and Thomas, 2019;Ganco et al., 2020;Jacobides et al., 2018;Kapoor, 2018;Shipilov and Gawer, 2019). On the other hand, it does not rule out related research streams on ecosystems, which ensures a greater impact of the resulting findings (Adner, 2017). On top of this, recent and essential works on ecosystems (e.g. Ganco et al., 2020;Jacobides et al., 2018;Masucci et al., 2020) build upon this, which demonstrates the relevance and usability of this foundation. Thus, all of our ecosystem cases, and all of the companies within these ecosystems, needed to fulfil the following criteria derived from Adner (2017) and supplemented and reinforced by Jacobides et al. (2018, see also Autio andThomas, 2019;Kapoor, 2018;Shipilov and Gawer, 2019). In the case descriptions, as well as in the findings section, we describe the ecosystems studied in detail and show how they match these criteria: 1. All of the ecosystems studied focus on a fully developed and clearly describable value proposition for the ecosystems' customers (e. g. a product or service). 2. The modules provided to the joint value proposition by the ecosystem partners can be characterised by non-generic complementarities (i.e. unique or supermodularity complementarity). 3. We only used cases that exhibit multilateral links among the partners. This means that there is a critical interaction across the relationships between partners, i.e. they cannot be fully broken down into independent bilateral relations.
Based on these criteria, we were able to decide whether companies can be viewed as being part of an ecosystem: Companies must be engaged in multilateral relationships with the other partners, are being aligned towards the value proposition by the orchestrator, have a clearly defined position and role in this regard, and provide non-generic/complementary modules to the value proposition.
As an additional criterion for the case selection, we selected ecosystems 'with an understanding of where there was likely to be variation and where there was not' (Eisenhardt and Ott, 2017, p. 86). For instance, the cases varied by the orchestrator's background and company size. Additionally, the orchestrators of the ecosystems studied are established firms, spin-offs/joint ventures, and start-ups. All value propositions can be assigned to the same sector, digital services. This approach to comparing cases from different industries by the formation of a common cluster has been used before (e.g. Chatman and Jehn, 1994;de Wulf et al., 2001;Frankenberger and Sauer, 2019;Schoenecker and Cooper, 1998). In addition, prior research on ecosystems has also used such sampling strategies in order to obtain a more comprehensive view, whilst still making sure the findings are not distorted by cross-industry differences (e.g. Davis, 2016;Mäkinen and Dedehayir, 2013;Rong et al., 2015b).
The case selection was not entirely a straightforward process. Rather, it can be described as an iterative approach with constant adjustments between case selection, data collection, and data analysis. We started with an initial set of 20 short cases and, based on our sampling criteria, selected around five of them as an initial sample. After the first rounds of interviews and data analysis, we began searching for additional cases to extend and enrich our original sample, as well as to yield a better understanding of the emergent theory (Eisenhardt, 1989). Consequently, we added further cases but also excluded those cases from our sample that turned out to not ideally match our criteria until we reached a state where additional cases did not significantly enrich the understanding of the context anymore (Eisenhardt, 1989). Our final sample of cases is shown in Table 1.

Case description
To increase the reliability of the case study, the organisations' actual names should be used (Gibbert et al., 2008;Yin, 2018). However, in our study, this was not possible because not all informants agreed to publish their company's name or the name of the ecosystem. We are aware that this methodological constraint is not unusual in qualitative research (c.f. Dattée et al., 2018;Frankenberger and Sauer, 2019;Hannah and Eisenhardt, 2018;Velu and Stiles, 2013). By highlighting the methodological audit trail and providing in depth information on our collection and analysis processes, we aim to maintain trustworthy qualitative research (Pratt et al., 2020).

Keyless access
The value proposition in this first case is a digital solution, which allows people to open doors by using a temporary smartphone code, technology that was developed by the ecosystem orchestrator. One of the use cases of this technology is keyless access to rental cars. An ecosystem-approach was necessary for the implementation of this use case given the interplay of several players involved: The car rental company had to modify its cars for the new system and provide end customers with a mobile app, and the car fleet had to be equipped with additional hardware and software provided by other ecosystem partners.

Virtual Mobility
The orchestrator in this case develops 3D-model maps of cities with two possible value propositions for different customers: First, the 3D model enables cities to assess whether their infrastructure is ready for autonomous, self-driving cars (e.g. if kerbstones are high      enough to be recognizable to such vehicles). Second, car manufacturers have the opportunity to use the model for digital test cycles of autonomous cars. An ecosystem-approach is necessary to create such maps since an intense co-specialisation of software and hardware providers is necessary, and authorities need to facilitate the creation of 3D models of the cities involved.

Contactless Payment
The value proposition of the digital payment provider is a payment transaction scheme for mobile phones with lower transaction fees than those of credit cards. To deliver the value proposition, an ecosystem was necessary: Retailers must adapt their cash register software for the new payment method and install Bluetooth beacons on their cash desks. The payment scheme needs to be integrated into existing systems of financial services providers and banks. Finally, end customers need to install an app and link it to their bank accounts, which requires the collaboration of the banks.  Values are the mean of the rating of three independent industry experts. Scale: 1-5; Range 1-2.9 regarded as low/weak; 3-5 regarded as high/strong.

Insured Factoring
The value proposition allows SMEs to insure account receivables against non-payment and, thereby, use their subsidiary ledgers as  collateral for a flexible credit line. The orchestrator developed this value proposition through an ecosystem involving investors who provide the monetary funds for the factoring, an insurance company that covers the risks of payment default, and a software company that provides an algorithm that scans through all the claims by the respective SMEs to assess and categorise the risks of their claims.

Intelligent Insurance
The orchestrator developed a health index using data available about the personal habits of the insured person, e.g. doing sports, smoking, or eating and drinking behaviours. An ecosystem approach was necessary given the following aspect: The alignment of health insurance companies, reinsurance firms, public data banks, and the health index provided by the orchestrator is essential to create such a customer-centric insurance offering.

Simple Relocation
The orchestrator (a relocation service) developed a personally tailed relocation process. Normally, the person moving must reach out to each of the companies involved in the relocation process individually (e.g. cleaning service, transportation of furniture, etc.) and deal with each of them separately. The orchestrator completes this task and aligns all ecosystem partners involved to enable a tailormade relocation out of one hand.

Autonomous Delivery
In the Autonomous Delivery case, the value proposition is based on the transport of sensitive and high value medical materials (e.g. organs, blood, etc.) from one hospital to another. For this value proposition to be delivered, a logistics company collaborated with a flight cargo company (which later left the ecosystem) and a drone technology provider. A hospital brought in this use case and provided, in addition to the test ground, knowledge and experience of hospital logistics. Governmental authorities also became an ecosystem partner to develop a legal basis for the autonomously flying drones.

Digital Prescription
The orchestrator in this case provides a distributed ledger to store medical prescriptions, which is accessible to physicians, health insurance companies, pharmacies, and patients. The value proposition offered by the ecosystem is fraud protection, convenience, and automation (e.g. for the billing process). An ecosystem was set up in which one orchestrator provided the technological infrastructure. To benefit from automation and fraud detection, a health insurance company served as a co-orchestrator, aligning its systems to the technology. Additional partners, such as physicians and pharmacies, were involved to deliver the value proposition. All parties involved needed to adapt their systems towards the new technology.

Autonomous Busses
The value proposition is an autonomous self-driving bus for use in pedestrian zones. To realize this offer, an ecosystem with multiple co-orchestrators who each provide different capabilities (e.g. knowledge on public transportation, data analysis, test ground, etc.) had to be implemented. In a later iteration, additional ecosystem partners (e.g. fleet manager, bus manufacturer) also had to be involved and aligned.

Smart Commuting
The value proposition in this case addresses the whole mobility customer journey for commuters in cities and suburbs. The initiator of the ecosystem is a European car manufacturer. Further co-orchestrators include a globally operating consulting company that provides the management skills as well as a network of relations and a globally leading software company that brings in communication capability and data processing technologies. Additional partners are being involved to refine and extend the services offered.
Please see Table 2 for an explicit description of the value propositions and their required ecosystems.

Data collection
We collected our data along several steps. First, we conducted initial interviews (between 15 and 60 min) with the main contact person at the ecosystem orchestrator in order to find out whether the ecosystem fulfilled the sampling criteria and to get an initial understanding of these criteria. In a second step, we conducted an interview with one executive from the orchestrator who had an excellent overview of the ecosystem. This semi-structured interview was based on a questionnaire with a threefold structure: 1) Details about the orchestrating firm, such as number of employees, revenues, background, and company history; 2) description of the ecosystem; and 3) surrounding conditions of the ecosystem, such as environmental uncertainty, industry characteristics, or the competitive situation. This initial questionnaire helped to increase internal validity due to its formal method (Gibbert et al., 2008; see also Spieth et al., 2019) and is shown in Appendix 1. The interviews lasted between 60 and 90 min. We took detailed notes during each interview. Jointly with the respondents, we sketched graphical representations of the ecosystems (c.f. Dattée et al., 2018). All interviews were conducted by four researchers, two of which are the first and second authors of this paper, while the others two are members of the same research group and, while familiar with the paper, were only involved in its data collection (c.f. Dattée et al., 2018). In all cases, the interviewers independently took notes and analysed the information. When possible, the interviews were recorded and transcribed to maintain quality of data (Gibbert et al., 2008;Mayring, 2007; see also Fernandez et al., 2018;Frankenberger and Sauer, 2019;Velu, 2017;Velu and Stiles, 2013). The authors guaranteed all informants who agreed to publish their titles/role within their firms in Table 1 that they would remain anonymous (Gioia et al., 2013).
As a next step, we collected extensive supplementary data: External and (orchestrator-) internal documents, as well as those provided from third parties about the ecosystem and the aspects addressed in the initial questionnaire. These documents included press releases, media reports, company homepages, annual reports, and internal presentations and reports. This allowed us to enrich the insights we gained in through the first interviews, as well as for validating and triangulating those insights (Jick, 1979;Yin, 2018). Collecting this supplementary data also helped us to detect potential inconsistencies between statements from the initial interviews and the internal and external documents.
Next, we conducted additional interviews with the orchestrator to deepen our existing findings and disentangle inconsistencies or misunderstandings. We also approached our interviewees by email and with short follow-up calls whenever needed (e.g. after the first rounds of analysis to validate constructs from literature in each case). We then initiated further interviews with additional informants from both the orchestrator firms as well as from other companies within the ecosystems in order to validate and triangulate our findings (c.f. Hoppmann et al., 2019). Table 1 gives an overview of our data sources per case.
We took great care to mitigate potential biases in our data collection process, which are the usual biases related to case study research as well as those that arise from the ecosystem context. Regarding the usual biases, we followed several recommendations from previous works on qualitative research (e.g. Eisenhardt, 1989;Eisenhardt, 1991;Eisenhardt and Graebner, 2007;Gioia et al., 2013;Yin, 2018). In order to avoid respondent biases, we did not reveal any theoretical insights or reasoning to our respondents (Huber, 1985;Huber and Power, 1985), and we avoided questions about specific theoretical constructs (Ozcan and Eisenhardt, 2009). We also used interview techniques, such as event tracking, to make sure respondents mentally returned to the situations the questions referred to, which is a means of increasing the accuracy of information (Eisenhardt, 1989;Ozcan and Eisenhardt, 2009). Even though we used semi-structured interviews for the first two interviews per case, we avoided asking very broad questions. If respondents provided us with vague answers/information, we initiated a series of follow-up questions to encourage them to go into greater detail and to be as specific as possible. In general, rather than focusing exclusively on the questions in our questionnaire, we asked interviewees for additional information surrounding these topics. This allowed us to put the answers into a wider context and to check for consistency between the answers and additional background information provided.
One key to achieving a higher construct validity of our findings was extensive data triangulation with multiple sources of information, such as interviews, documentation, archival data, and, in few cases (e.g. the simple relocation, smart commuting, and digital prescription cases), direct observation and participant observation in workshops (Eisenhardt, 1989;Jick, 1979;Patton, 2015;Yin, 2018). As for the interviews, in every case, we had multiple informants from at least two companies involved in the ecosystem. In most of our cases, information on the core aspects of the ecosystem was publicly available. For instance, companies described their joint value propositions on their homepages or in sales brochures. Likewise, partners and collaborations were often publicly announced or, in the case of M&A or venturing activities within the ecosystems, described in annual reports. On top of this, several of the ecosystem activities studied received attention from the public, for instance, given their innovativeness or potential for disruption. In these cases, third party reports existed from newspapers or media in general. Thus, the availability of such additional material allowed for extensive triangulation and, in particular, to mitigate retrospective recall biases (Davis and Eisenhardt, 2011;Eisenhardt and Graebner, 2007;Frankenberger and Sauer, 2019;Gioia et al., 2013;Ozcan and Eisenhardt, 2009;Velu, 2017;Yin, 2018). Also, we were able to check whether key words and constructs from the first rounds of interviews were frequently mentioned in the additional documents. If we found additional key words in the reports that had not been mentioned in our interviews, we addressed them in short follow-up calls or emails with the respondents.
Another potential bias resulted from the ecosystem construct per se. Our paper is written from the perspective of the ecosystem orchestrator, or the company initially driving the ecosystem, since this actor is considered to be the designer of the alignment structure (Moore, 1993;Teece, 2007Teece, , 2016. Thus, the core interviews were only conducted with managers from an ecosystem's orchestrator firm. This could potentially lead to two different biases. First, a manager of an orchestrator firm might be tempted to present their firm and its role within the ecosystem in the best light. Second, these managers might not be aware of all activities within the ecosystem or they might misinterpret some of them. The risks arising from the first potential bias can be considered less significant since none of the aspects under consideration are particularly sensitive nor would they affect the perception of the orchestrator by other firms or people. Also, all of the firm names for all of the cases are used anonymously, which, arguably, reduces the incentive for respondents to present their firms in a specific way. The second potential bias, misinterpretations (c.f. Fernandez et al., 2018) of ecosystem partners and activities by the orchestrator, was mitigated by several means. First, the core elements of our research, such as the value proposition, activities of partners, or links, were documented in the underlying contracts among the partners, thus clearly documented. Second, shaping and managing all these core aspects of the ecosystem is the main responsibility of the orchestrator as the key actor within the ecosystem (Moore, 1993;Teece, 2007Teece, , 2016. Thus, there is no other company within the ecosystem that can be considered to have the same insights into these aspects as the orchestrator. Third, we primarily interviewed managers who are in charge of managing the ecosystem, which means that they deal with the core aspects under consideration on a daily basis. In order to additionally address both potential biases, we used extensive data triangulation as described above. In all cases, the core statements of the interview respondents could be validated by this data, which significantly strengthened their credibility. Further, we purposefully established time lags of at least a few weeks up to several months between the interviews and follow-ups, and we did not provide respondents with transcripts or information from previous interviews in between. This allowed us to ask several questions from previous interviews in the course of the follow-ups and to check for consistency of the respective statements (c.f. Fernandez et al., 2018). In all of our cases, we did not detect significant discrepancies between the statements, which additionally confirmed the credibility of our respondents. Finally, in some of our cases, to further check for consistency, we conducted additional interviews with other firms involved in the ecosystems or with third parties.
All information (e.g. recorded interviews, transcripts, archival data, questionnaires, etc.) was stored in a database (Yin, 2018). As some cases were collected by different researchers, the academic research group uses detailed protocols for each case (write-up of a narrative, overview of which interviews were conducted with which researchers). This ensures reliability (Gibbert et al., 2008). Table 1 provides an overview of our data sources.

Data analysis
Inspired by recent publications (Frankenberger and Sauer, 2019;Lewis et al., 2011), our data analysis was based on two stages of 1) inductive analysis (Eisenhardt, 1989) and 2) deductive-inductive analysis (c.f. Kuckartz, 2018). This allowed us to approach our research questions with an open mind and to be open to emerging theoretical aspects, as well as to further deepen exiting theory on ecosystems (Siggelkow, 2007;Yin, 2014).

Inductive analysis
As a starting point, we used our interview transcripts to create individual cases, which we enriched with the additional data sources (Yin, 2014). When information was missing, we asked our respondents for further information or conducted additional desk research. The individual cases were the foundation for the cross-case analysis, which we started with an open mind and without pre-defined constructs, theories, or hypotheses (Gehman et al., 2018). We searched for high-level themes in our data that referred to the ecosystems' design or to prevailing surrounding conditions, e.g. 'value proposition', 'partner', 'handover', 'input', 'module', or 'coordination'. Additionally, we used narratives drawings, tables, and other forms of visualization (e.g. post-it for recombination) to gain an understanding of our cases and an overview of our content, and to detect patterns in our data more easily (Miles and Huberman, 1994;Yin, 2014). For the same reason, we did case pairings to understand similarities and differences between the cases (Ozcan and Eisenhardt, 2009). Thus grounded, we followed the coding procedure of Gioia et al. (2013) and developed first themes in an iterative and recursive manner (e.g. Velu, 2017), representing the elements of design, as well as the surrounding conditions of the ecosystem. We used these themes to form first relationships between them and the surrounding conditions (e.g. strong substantive uncertainty leads to fewer actors being involved in the ecosystem) and discussed the causes of these effects (c.f. Frankenberger and Sauer, 2019). The resulting relations were cross-checked with the other cases to verify their occurrence (Ozcan and Eisenhardt, 2009) by using the themes for open coding from our transcripts (we highlighted and used the comment function within MS Word). This initial analysis was conducted by two of the three authors independently from each other, followed by a discussion between the authors and a synthesis. The primary idea of this first stage of inductive analysis was to detect interesting and relevant constructs (1st order constructs) that we intended to deepen further and extend in the second stage of our research. By returning to literature (c.f. Velu and Stiles, 2013), we identified two key conditions: 1) the substantive uncertainty of the environment (Dosi and Egidi, 1991) and 2) the effective distance of knowledge (Afuah and Tucci, 2012). We used these factors from literature and searched through all of our cases for respective constructs. In this iterative process, we recognized that these conditions influence the ecosystem structure.
At the beginning of our analysis, we discussed whether these two factors are exogenous, since uncertainty can be reduced and knowledge can be gained. In the same vein, the degree of knowledge might influence the uncertainty as well, making these factors dependent upon each other. However, in the course of our analysis, we were able to sharpen the theoretical understanding of these two factors, based on our emerging findings and the related studies in literature. In section 3.5, we explain why these two factors are independent and exogenous in the context of our study. These initial findings, and especially the two surrounding conditions, pointed towards the attention-based view of the firm as an appropriate theoretical lens. After having gained such an initial understanding of the theoretical frame, constructs, and themes, we started the second stage of deductive analysis.

Deductive/inductive analysis
As a starting point for the second stage of analysis, we conducted a thorough literature analysis on ecosystems and the attentionbased view of the firm, with a special focus on the theoretical aspects we derived from our first, inductive, round of analysis. Following Yin's (2014) recommendation, we developed a theoretical framework and outlined patterns of expected results prior to the next stage of our research. This 'will enable the complete research design to provide surprisingly strong guidance in determining the data to collect and the strategies for analysing the data' (Yin, 2014, p. 38). Another 'benefit is a stronger design and heightened ability to interpret your eventual data' (Yin, 2014, p. 38). Thus grounded, we used axial coding (Strauss and Corbett, 1998) to search for key words in our interviews that represent the patterns of expected results and the theoretical constructs we had previously defined. We highlighted the respective sentences in our transcript (a MS Word file), adding the relating codes (1st order concepts) as a comment. In MS Excel, we copied the representative quote and its code. By sorting the cases according to the respective codes, we gained a better overview of our 1st order concepts. This allowed us to build 2nd order themes, before finally defining the aggregate dimensions. We illustrate the final coding scheme in Fig. 1. Table 3 displays the proof quote scheme for all cases and elements of the coding scheme.

Definition and validation of the two surrounding conditions
It is important to understand that both surrounding conditions are independent from each other and exogenous as well. Regarding the uncertainty, we apply the concept of exogenous substantive uncertainty (Dosi and Egidi, 1991, also see Beckman et al., 2004). In this vein, in a situation of strong substantive uncertainty, the uncertainty is inherent in the environment/the situation at hand and, thus, cannot be reduced. For instance, in the Virtual Mobility case, the ecosystem creates a digital map that a city can use to simulate urban traffic involving autonomous cars. The market success of such a simulation is almost entirely dependent on the evolution of autonomous driving in general, which, in turn, is dependent on a variety of factors, including regulation, technological developments, co-evolution of solution providers, end-user acceptance, and so forth. Given the complexity of the situation, the orchestrator in this case is unable to assess the market and its development. Instead, the company can only 'wait and see' what happens and try to adapt to it. Thus, even by designing the alignment structure or getting access to novel sources of information, the uncertainty cannot be reduced in such a situation of strong substantive uncertainty. On the other hand, if weak substantive uncertainty prevails, it is possible to conduct research or collect information in order to reduce the uncertainty (Dosi and Egidi, 1991). Thus, we use substantive uncertainty as an exogenous factor. If substantive uncertainty is exogenous, it won't be influenced by characteristics of the ecosystem or the orchestrator, which, in turn, makes it independent from the knowledge base of the orchestrator as well. Therefore, exogenous substantive uncertainty is independent from the effective distance of knowledge.
In terms of the knowledge distance, it is important to note that an ecosystem is defined by its value proposition (Adner, 2017;Jacobides et al., 2018). In this vein, the effective distance of knowledge (Afuah and Tucci, 2012) refers to the orchestrator's knowledge related to the value proposition/the respective field of business that the ecosystem and its value proposition is focused on in relation to the knowledge needed to build an ecosystem in this field. An orchestrator can decrease the knowledge distance through, for instance, organisational learning or hiring expertshowever, building new knowledge in unrelated fields is difficult and takes time (Birkinshaw et al., 2007;Meulman, Reymen, Podoynitsyna, & L. Romme, 2018;Rosenkopf and Nerkar, 2001). Therefore, the distance of knowledge is given exogenously in nascent ecosystems. This is explained in detail in Fig. 2. In the following section, we explain the points of the framework one by one: 1. Initially, there is the decision to enter a specific field of business by using an ecosystem or not. Typically, the top management of a corporation takes this decision, which makes the decision exogenous for the managers who are assigned to implement the ecosystem, as highlighted in the Autonomous Delivery case: 3 The representation of our cases follows the approach of Gioia (Corley and Gioia, 2004;Gehman et al., 2018;Gioia et al., 2013). The second-order themes are represented as rows. Within each row, the first-order concepts are represented in italic to label first-order concepts ("Characteristics"). In the "Quote/Argument" column, are proof quotes supporting the concepts for each case. The letter-number codes identify the individuals designated (see Table 1). The strength for each attitude is indicated below: Strong evidence means that all principal informants agreed with an attribute. Moderate evidence means that more than one of the confirming informants agreed with an attribute. For the dimensions "structure", "activities" and "number of decision-makers", we used additional sketches of the ecosystem to validate our findings

'Such projects actually need approval at C-level, it doesn't work because some logistics specialist says "yes, come on, a drone is cool". Because if C-Level's not convinced, you'll never get this thing to fly. And we were really lucky in [City of the Hospital]
, we were always able to communicate with the CEO in person.' (G1). What's more, in all companies, regardless of whether they are corporations or start-ups, this decision is being made because there is a significant market opportunity that is obviously worth the effort of building an ecosystemotherwise the company would not start such an initiative. Our Intelligent Insurance case is a nice example of this (c.f. section 3.2.5 and Table 2): The start-up invested around CHF 50 Million into developing their health score to improve the underwriting of health insurance. The implementation of this underlying technology, however, would not have been possible without an ecosystem approach: 'We [Data Analyst] have developed a health index […]. If something consists of 100 data points, then you already know that nobody does it alone. […] That means that a huge ecosystem is being created within itself.' (E1). Additionally, the insurance companies (complementors) benefit from the core value proposition, as they require partners to make the opportunity materialize (as they can't do it alone): 'Let me put it this way: there are technology partners that we will need, because many of these efficiency improvements are, of course, only possible if we provide technical support.' (E3) Therefore, not building an ecosystem would not have been an option in this case. Which means that the decision to enter the field of business is exogenous for the people assigned to build the respective ecosystem. In any case, our paper does not focus on the decision of whether to enter a field of business or not, so this issue is beyond the scope of our investigations.
2. The field of business the ecosystem is focusing on requires a certain type and extent of knowledge (for instance, technologies, market players, customer demands, etc.). Obviously, this knowledge depends on the respective field of business and is, therefore, exogenous from the perspective of the company building the ecosystem. For example, in the Smart Commuting case, the respective field of business in the context of smart mobility required knowledge about mobility solutions, the underlying IT and data analytics requirements, and the specifics of cities and people commuting within these cities.
3. The knowledge base of the company at the time it starts the ecosystem initiative is exogenous as well, since building up new knowledge in unrelated fields is difficult and, even if the knowledge can be acquired, takes a long time (Birkinshaw et al., 2007;Meulman et al., 2018;Rosenkopf and Nerkar, 2001). Since our study focuses on ecosystems in nascent phases, the knowledge cannot change (quickly) at the time of the ecosystem initiative and is therefore exogenous at the moment the orchestrators are designing the ecosystem. As an example, in the Smart Commuting case, the project manager stated: 'As a mobility company, we did not have the necessary knowledge in the context of IT and smart cities. Building up this knowledge in such unrelated fields would have been almost impossible for us. Apart from us, our top management would never have accepted hiring people with these backgrounds since these profiles would not match with our company and our core competencies.' (J1). Thus, if the company intends to enter a field of business, it must deal with its existing knowledge base. The alternative would be to wait for an indeterminate amount of time in order to build up the knowledge needed for that field of business. But, arguably, this is not a realistic option, especially not in the fast-paced environments and nascent phases in which our case companies operate.
4 As a result, the effective distance of knowledge is exogenously given in the short term our study examines. 5 As our findings reveal, the design of the ecosystem is driven by the effective distance of knowledge. 6 The ecosystem design might influence the knowledge base of the orchestrator (Velu, 2015). Such an effect, however, will only happen in the long term since building up knowledge in unrelated fields is difficult and takes time (Birkinshaw et al., 2007;Meulman et al., 2018;Rosenkopf and Nerkar, 2001).
As noted above, both factors could change over time: Effective distance of knowledge can be actively reduced in the long term, and substantive uncertainty of the environment can decrease. Accordingly, we study these two conditions at the moment the orchestrator designs the ecosystem, thus viewing knowledge distance and uncertainty as antecedents of the design of the alignment structure. In the long term, novel knowledge can be acquired, for instance by designing an appropriate alignment structure accordingly (which makes it endogenours, c. f. Velu, 2015). Thus, in the long term, organisational design can be seen as an antecedent of knowledge distance. In addition, the uncertainty might decrease over time. Thus, it would be interesting to study if and how the design of the alignment structure changes over time when uncertainty and knowledge distance are decreasing. We will return to this call for further research in the discussion section of the paper.
Since the surrounding conditions play a major role in our paper, we wanted to back up these aspects with an additional analysis. For this purpose, two of the three authors and two additional researchers, who have not been involved in this research project, estimated the degree of substantive uncertainty (Dosi and Egidi, 1991) and effective distance of knowledge (Afuah and Tucci, 2012) for each case. Afterwards, we discussed our individual estimations and came to final conclusions. Please see the proof quotes for the surrounding conditions in Table 4 (Gioia et al., 2013;Pratt, 2009;see as examples Hoppmann et al., 2019;Souitaris et al., 2012;Velu, 2017).
In order to validate our classifications, we invited industry experts to rate our cases along our identified surrounding conditions. Each case was rated by three different industry experts individually and independently from each other. Afterwards, we calculated the average estimation of these three independent evaluations. This ensured an objective verification of our classification (c.f. Crossland and Hambrick, 2011;Frankenberger and Sauer, 2019;Hambrick, 1982;Hambrick and Abrahamson, 1995;Souitaris et al., 2012). The evaluations of the industry experts corresponded perfectly with our initial estimation, apart from slight but insignificant derivations. The results of the industry experts' ratings are summarized in Table 5.

Validation of results and overview of applied constructs
The results of this study (i.e. findings, discussion, and practical implications) were presented to different audiences in order to review the researchers' results and to strengthen the construct validity (c.f. Gibbert et al., 2008;Goffin et al., 2019;Yin, 2018). The different audiences consisted of 1) researchers from the same research group, who also could review transcripts; 2) researchers in seminars (some of whom provided reviews on this manuscript) and at academic conferences; 3) Informants (from industry practice), who received the transcripts and participated in a workshop. In a few cases (e.g. someone was not able to participate at a workshop), one of the authors presented and discussed the findings with an informant in a bilateral setting; and 4) practitioners building their own ecosystems. They received some anonymized cases and were asked to develop some of our findings by themselves and/or explain them to each other in a workshop setting.

The design of ecosystems depending on the degree of substantive uncertainty
In some of our cases (e.g. Insured Factoring, Intelligent Insurance, Simple Relocation), firms were confronted with a weak substantial uncertainty, i.e. they were able to assess the market potential and foresee the events that would have an impact on the ecosystems' future business. The Intelligent Insurance case demonstrates this nicely: The orchestrator already knew which countries mandated health insurance and whether basic health insurance (or only supplementary policies) can be priced individually. In addition, industrialised countries in particular will face the challenge of exploding healthcare costs in the coming years. Therefore, it is certain that the financing of the health care system will have to be redefinedeven if it is not yet clear what form this will take (there are already loud voices calling for a stronger polluter-pays principle).
Even when not all the information needed was available at the beginning, such as in the Smart Commuting and Autonomous Busses cases, the orchestrator was able to conduct market studies or ask experts in order to find all the information necessary to predict the future course of the ecosystem. In other cases, however, the situation was highly uncertain and necessary information on technologies or markets did not exist and could not be obtained. In the Autonomous Delivery case, for example, the decision-makers were not able to assess the future stance of the regulatory authorities regarding the drone transportation that the whole value proposition was built upon. At one point, a drone carrying blood samples crashed into a lake, which led to the shutdown of the transports for approximately two and half months. During that time it was unclear when, or even if, the business could continue, since the regulation authority involved in the ecosystem was using the business case to develop the legal framework for autonomous drones. This marked a major and unforeseeable threat to the ecosystem's business case. In the Keyless Access case, underlying technologies were still very immature at the beginning of the ecosystem initiative and it was not possible to predict which technology would become the future standard, as the CEO of the orchestrator firm stated: 'But from my knowledge, the framework conditions were that, first, we did not know which technical standard would be the right one. Is it Bluetooth? Is it NFC? Is it online, is it offline?' (A2). All of our cases that are characterised by weaker substantive uncertainty, such as the Insured Factoring case (Fig. 3) or the Autonomous Busses case (Fig. 4), serve as nice examples of ecosystems in which actors are intensively linked with each other by a multitude of connections and, therefore, occupy positions in several flows of activities at the same time. This allows the actors to mutually adapt their respective modules with several partners. For instance, in the Autonomous Busses case, the bus manufacturer, mobility provider, and regulatory authorities are closely collaborating to build autonomous busses that not only match the requirements of the public transportation provider, but also fulfil the regulatory requirements to get an exemption permit for the test phase, such as speed limitations, having an emergency brake button, as well as ensuring that during the test period a security escort is on board and able to stop the bus manually. On the other hand, given the novelty of the subject, the regulatory authorities had to develop specific requirements for autonomous self-driving passenger buses, which forced them to constantly exchange information with the bus manufacturer and the public transportation provider. In the Intelligent Insurance case, the data analyst, the primary insurance company, the secondary insurance company, and the public health institutions all have to collaborate with each other very closely: The primary insurance company provides the white label app (developed by the data analyst) to its customers to data of its insurees to the data analyst. The latter thereby acquires metadata, allowing it to increase the accuracy of the health scores it had previously received from public health institutions (which collaborated before the ecosystem was in place with primary insurance companies and reinsurance companies, e.g. by providing life tables), and in return, the data analyst provides reports. The primary insurance company had to integrate the data analyst's health scores into its actuarial system, and then this had to be reconciled with the reinsurance company's systems so that it could evaluate the valuation of the adjusted actuarial values. This ensures that the algorithms and calculations of the health index work perfectly with the data provided by all parties and, most importantly, that they can be fully integrated into the primary and secondary insurance companies.
Additionally, these cases that are characterised by weaker substantive uncertainty, nicely demonstrate two different ways of revenue sharing and contribution of activities among the actors involved. On the one hand, actors may have a direct exchange of goods against goods or goods against money. In the Intelligent Insurance case, for instance, the insurance company receives the app and software from the orchestrator who, in return, directly receives money from the insurance company. On the other hand, we can also observe complex connections involving several actors, for instance, between the insurance company, re-insurance company, and the data analyst in this case. The insurance company provides data analytics to the re-insurance company, which allows the re-insurance company to adjust, i.e. lower, the fees being paid by the insurance company to the re-insurance company. Finally, the insurance company forwards a share of these savings to the data analyst as a financial compensation for the data analytics it provided to the reinsurance company. Overall, in all of our cases characterised by weaker substantive uncertainty, the actors appear to have tight and intertwined channels between each other, in order to best adjust their respective modules to each other and to make the best use of the complementarity effects among them. In the Simple Relocation case, many more actors were integrated given the low degree of uncertainty: Since the management capacity could be fully directed towards partner acquisition rather than towards the management of uncertainty, the value proposition became more extensive as the additional players increasingly made the relocation process more and more tailor-made for the customers' needs. This allowed the offer to be better marketed, and additional partners were attracted.
The opposite appears to be true for cases with stronger substantive uncertainty. It becomes apparent that simple links involving just two actors prevail (e.g. as shown by the Autonomous Delivery case in Fig. 5). Typically, two actors within an ecosystem directly exchange goods for goods or goods for money without involving additional actors. The connections are still multilateral, since even these direct exchanges are influenced by, or not possible without, the contributions provided by the other actors in the ecosystem and their mutual connections (Adner, 2017). However, the ecosystems are based on a set of comparably simple links and connections between the actors.
The same is true for the co-creation and adaption of modules among the partners in the other cases characterised by stronger substantive uncertainty. As shown by the Autonomous Delivery case, the development of the value proposition took place primarily between the logistics company that brought in funds and knowledge on general distribution processes and the drone technology provider specialised in autonomous flying, which provided the technology. Also in the Virtual Mobility case, the main effort and cocreation takes place between the orchestrator and the graphic processing unit manufacturerthe city merely provides the playground for the tests, whilst the game developer provides the graphic engine that does not need to be aligned with the other modules on a daily basis. For the Keyless Access case, the modules for the value proposition can be easily disassembled: The access provider supplies the software to open the locks via the box, the box supplier configures the box to the vehicle type, installs it, and tests its functionality. The cloud service developed a special app for the car rental company to provide the functions provided to the car renters (cf. Fig. 6).
The Contactless Payment case seems to be an exception, as the actors are tightly interwoven despite the prevalence of stronger substantive uncertainty. This seems to contradict our findings as described above. However, the actors were already tightly connected to each other through the collaboration necessary to run conventional payments via credit cards, which is the foundation of the value proposition of the newly founded ecosystem. Thus, only a few very simple links had to be implemented between the orchestrator (the Digital Payment Provider) and the existing partners. Overall, in cases with strong substantive uncertainty, actors appear to be less tightly connected to each other and only adapt their respective modules with one or very few actors.
In addition to affecting the links between actors, the type and degree of uncertainty appears to have an influence on the number of actors and, thus, also on the number of activities performed in the ecosystem overall. In every case characterised by weaker substantive uncertainty, the ecosystem began with an initial group of partners necessary to realize the value proposition. Subsequently, these ecosystems have grown through the integration of additional partners, who either provide additional modules for the existing value proposition (to make it 'better') or extend the value proposition with additional offerings for the customer. In the Intelligent Insurance case, for instance, the initial value proposition of a health insurance policy with an individual fee was implemented by an initial set of actors. Subsequently, a reward partner was integrated to extend the original value proposition by offering people rewards for improving their health scores, which provides an incentive for healthier living. This additional module is complementary to the existing modules for the initial value proposition for several reasons, as it increases the value of the health index for the customer and creates a more healthy portfolio of insurees for the insurance company.
Similar findings can be observed in the ecosystems in the Autonomous Busses, Smart Commuting, Insured Factoring, and Simple Relocation cases. In the Simple Relocation case, which offers a tailor-made relocation service, the integration of more partners with additional modules extended the original value proposition: In addition to simply offering a relocation service, the ecosystem began offering additional insurances for the customers' furniture as well as complementary services, such as storage of belongings or integrated shopping for new furniture. All of these additional modules increase the value of the existing ones and contribute to greater product differentiation within the market, which leads to customers being willing to pay higher prices for the services.
However, this comes at a price: More actors and activities, as well as more links between the actors, arguably increases the orchestrator's efforts. For that reason, in all of our cases characterised by weaker substantive uncertainty, the orchestrators implemented technological platforms to coordinate the actors and to automatize the transfer of information and funds between them. For instance, in the Intelligent Insurance case, data from an insuree is forwarded from their wearable device to the app, and is then automatically processed by the data analyst company. The same process coordinates the exchange of data between the orchestrator and the reward partner or the reinsurance company. Thus, the effort of orchestration required to run the whole ecosystem is greatly reduced.
Whilst the orchestrators in ecosystems with weaker substantive uncertainty were constantly involving additional actors in order to extend or improve the value proposition through the additional modules and activities provided (as described in the Simple Relocation case above), this did not appear to happen in ecosystems that face stronger substantive uncertainty.
These findings can be explained using several arguments. First, more actors in an ecosystem implies that there are more individual agendas to be aligned (Jacobides et al., 2018), and more issues and answers that actors need to attend to within the ecosystem. A greater number of links implies a more complex decision-making process, as well as more attention channels decision-makers would have to engage in. On the other hand, situations with stronger substantive uncertainty are generally more difficult to manage and significantly increase the required management effort due to the unpredictability of events (Galbraith, 1974;Ofek et al., 2007). Also, higher uncertainty generally impedes decision-making and focusing of attention (Cyert & March 1963;Gavetti et al., 2007). Thus, orchestrators might be reluctant to embrace a more complex ecosystem with additional partners and complex links between them when the situation is already difficult to manage and it is difficult to focus attention on the most relevant information. This is nicely demonstrated by the Keyless Access case. Once the box (the key hardware module) is integrated into a car, it is very expensive to exchange: 'And for example, the car box: Once it is finally integrated into a car an exchange towards another car is very costly. Principally, I would start with the integration from scratch. Therefore, instead we have this value added [approach].' (A1). However, given the prevailing uncertainty, the orchestrator decided to tackle this issue by finding additional engineering services (rather than integrating additional actors): ' […] we were providing further integrations and product functionalities. […] so, the added value was created; in this sense, we were doing so by providing engineering services and not by integrating further partners into our ecosystem. […] The key limitation really is the question of whether the integrator, manufacturer etc. Is willing to bring the solution to the market.' (A1). Second, situations with stronger substantive uncertainty require organisations to remain flexible in order to swiftly react to changes in the environment (Galbraith, 1974). The quotes above demonstrate this: The orchestrator tried to retain a small ecosystem with as few partners as possible, as long as stronger substantive uncertainty prevailed. It is the same for all of our cases involving orchestrators in ecosystems surrounded that are by stronger substantive uncertainty; they seem to refrain from involving additional partners, even though this limits the chances of improving or extending the value proposition.
In the Autonomous Delivery case, the ecosystem tackled the challenge of technological uncertainty by doing extensive test flights and waiting both for the technologies to become more mature as well as for the regulatory situation to become more clear. In the end, the actual use case was brought in by a new partner, the hospital: They asked for the opportunity to transport laboratory samples using the drones after watching the drone tests on a TV news programme. In the Keyless Access case, the orchestrator began to build parallel ecosystems by relying on the same technology instead of trying to extend or improve the existing value proposition: In addition to equipping rental cars, the orchestrator started to build a value proposition targeting hotels. Given the fact that the actors were constantly changing, it was not feasible to implement advanced technological solutions to connect them. This led to a greater orchestration efforts.
Thus, regimes with stronger substantive uncertainty favour simple ecosystem designs that include few actors and activities to manage, and only simple links between the partners. On the other hand, in situations of weaker substantive uncertainty, it is possible to have a more complex design with more actors. As the ecosystems in situations of weaker substantive uncertainty evolve over time, however, the existing value propositions are extended or improvedoften by increasing the number of links and partners involved. This leads to the following propositions: Proposition 1a. The weaker the substantive uncertainty of the environment, the larger the extent of the structure of the ecosystem should be.
Proposition 1b. The weaker the substantive uncertainty of the environment, the more multilateral the structure of the ecosystem should be.

The design of ecosystems depending on the effective distance between the orchestrator's existing knowledge and the knowledge relevant for the ecosystem's field of business
When building the ecosystems in the Keyless Access, Contactless Payment, Virtual Mobility, Insured Factoring, Intelligent Insurance, and Simple Relocation cases, from the very beginning the respective orchestrators possessed knowledge and connections to networks related to the ecosystem's field of business. Due to their existing knowledge and networks, the orchestrators were able to define the value proposition by themselves and to understand which partners, activities, and skills were required for its instantiation. For instance, in the Contactless Payment ecosystem, the orchestrator's top management had a clear plan of the value proposition and the resulting ecosystem, as explained by the head of marketing: ' […] principally, the idea was to bring the [bank] card into the new world [of smartphones]. And then we said: we have to make it with external partners. Actually, not necessarily because of the idea of partnership. But it was more IT driven, which required external IT skills.' (C2). The co-founders of the orchestrator in the Insured Factoring ecosystem, who have a background in banking (treasury and cash management, insurance sales in banking), seemed to be quite familiar with the niche for their value proposition and simply had to integrate partners who were already part of their personal networks: 'The platform is built in every country the same way, and it is available in four languages at the moment. And our bank would be active in the same areas. On the side of distribution, we would have to find new partners, but also there we have a network.' (D1). Due to their knowledge and networks, all of these orchestrators started with a clearly defined value proposition. They were also able to determine through analysis which modules and activities were needed to deliver the value proposition, and to identify which partner profiles would be best suited to providing missing elements. Thus, partners were chosen based on their specific abilities to perform the activities needed to implement the value proposition. A perfect example of this is the Intelligent Insurance case: The (primary) insurance company creates the insurance solution, whilst the reinsurance company provides a tailor-made reinsurance solution for the primary insurance company. The public health organisation provides meta-data and the reward partner delivers the reward system. Thus, every actor brings in a clearly defined module based on its very specific competencies. It's a similar process in the Virtual Mobility case. In addition to capturing data and providing the final models, which was done by the orchestrator, actors focused on two skills and activities only: Computational power to render the 3D-model maps and an engine to process and visualise the data. Therefore, they engaged partners specifically for their contributions, as the founder of the 3D cartographer highlights: [Graphic Processing Units Manufacturer] a provider of graphic cards, so they are providers of computational resources, [Game Engine Developer] is a provider of gaming engines so they're the provider of that particular part of the product, and we're the provider of the 3D model.' (B1). In the Simple Relocation case, the orchestrator went a step further and made clear that the modules from the partners must also meet an agreed quality measure: ' […] we set a quality standard that must be maintained. If not, then we do not work together with the companies.' (F4) In other words, the orchestrators' understanding of the field of business allowed them to pre-define: the specific modules and partner profiles necessary, the positions the actors should have within the ecosystem, and how they should interact with each other. This allowed the orchestrators to define all of the key elements of the ecosystem design, even using clearly defined contracts that specify all structures and activities, as highlighted by an interview from the Intelligent Insurance case: 'First of all, these are long-term contracts for all of the ecosystem partners involved. There is no one who enters the ecosystem and could say he can cancel within two weeks. Nobody could do that. We have something like a turnkey-ready ecosystem. Everything is based on a contract and new partners just sign this contract with us and everything is defined in it.' (E1). Additionally, in the Contactless Payment case, actors entering the ecosystem had to deliver specific modules that were pre-defined, e.g. the banks' needed to provide clients data towards the orchestrator or the retailers had to give access to the payment system for the acquirer. The head of marketing in the orchestrator firm stated: 'For [the payment ecosystem] there exists a huge rule book defining all the rules for the partners involved. And everyone who takes part in the scheme has to sign the rule book. […] These are books full of rules for them.' (C2). Obviously, the ex-ante definition of the ecosystem design allowed the orchestrator to build the ecosystem through a comparably straightforward process and to have well-structured discussions with potential partners, based on clearly defined structure and activities within the ecosystem.
Conversely, in the Autonomous Busses, Smart Commuting, Autonomous Delivery, and Digital Prescription cases, the orchestrators lack the knowledge and network connections for the field of business the ecosystem focuses on. In these cases, the orchestrators were unable to define a specific value propositionthey had to define it together with partners. This is nicely explained by the head of open innovation and venturing in the Autonomous Busses case: 'We identified a vision: we did not want to be a public transportation provider anymore; instead we wanted to be a multi-modal integrated mobility provider. This was really the vision. And we identified some immediate challenges on the path to this vision. But we knew that alone we could not manage to further develop our vision.' (I1). A project manager within the same ecosystem stated: 'No one knew in which direction it [the ecosystem] went. Kind of a value proposition existed but we did not see it clearly. ' (I2) Due to the lack of knowledge, the actors did not know which modules were required for further development of the ecosystems. Beyond that, the co-orchestrators had to perform additional activities with their partners, which could not have been foreseen at the start of the ecosystem initiative. For instance, in the Digital Prescription case, fulfilling basic functionsfrom entering the prescription from the doctor to automated billing by the health insurance companywas outlined in advance as a goal. Nevertheless, it was obvious to the participants that they had to implement additional activities (e.g. financial incentives so physicians have a greater motivation to participate, which is very important, as physicians are the most critical players in the ecosystem) in order to increase the attractiveness of the ecosystem, which makes it more established and thus more efficient. Efficiency will increase when more physicians participate, as the following quote from a product manager shows: 'Therefore, perhaps the physician is the only one who can receive an incentive from the others in this model, who would ultimately benefit from an increase in efficiency. ' (H2) This was confirmed by the chief innovation officer: 'That's why you look at the ecosystem at the beginning, take it up as it exists today. And you see if you can build it more efficiently through different iterations.' (H1) In the Autonomous Busses ecosystem, it turned out the one of the co-orchestrators had to take-over the role of mediator between other partners:

'Partially, we [public transportation company] took over the translator role; imagine the official from [Country A] with the young, startup entrepreneur from [Country B
]. This does not work. Simply not only translating the language, but the mentality.' (I1). By mediating, the co-orchestrator ensured that the activities within the ecosystem remained flexible enough to enable the value proposition. In particular, the government authorities were critical for testing, so the mediation activity was key to influencing (link) the respective partner to agree to cooperate. It is the same in the Smart Commuting case, where the consulting company, as the coorchestrator, performed a multitude of tasks: providing manpower, project management, doing analysis, and so forth. These orchestrators had to be flexible and agile enough to respond whenever tasks that had not been defined at the beginning had to be performed. According to the respondent from the car manufacturer: 'They [Consulting firm] had no specific product or ability to bring in. They took over a multitude of tasks, whatever was needed. Doing project management, developing the governance model, and so on.'(J1). Given the unspecified value proposition at the beginning of the ecosystem initiative and the broad and unforeseeable activities of the co-orchestrators, it was not possible to define these activities beforehand. Thus, the initial partners in such cases agreed only on a framework contract or memorandum of understanding, which merely expressed the willingness of the various parties to collaborate without going into detail. The reason for these types of contracts was that, due to the lack of knowledge, no activities, and therefore no transfers among the ecosystem actors, could be predefined. As the head of the Autonomous Busses ecosystem stated: ''You are flexible [relating your activities]. There's a framework agreement, but it's very rough.' (I2). Consequently, there is no clear hierarchy among the initial co-orchestrators as nicely illustrated by a quote from the same case: 'Everyone [of the co-orchestrators] has an employee detached for the coordination of the ecosystem. But as we are the primus inter pares among the initial partners, […] it is our employee who pushes the whole thing. So, there is no hierarchy, all coordinators are equal, there is not one boss. Yet it is our employee who is the primus inter pares, who is driving the whole thing a bit more than the others.' (I1). The framework agreement specifies structure elements only, as the actors and their organisational forms (positions) were stipulated in advance. Given the lack of a detailed contract specifying the form of contributions (activities), in almost all of the cases, the underlying foundation for the collaboration among the co-orchestrators turned out to be strong mutual trust in the ecosystems' activities. Thus, it is no surprise that there had been prior collaborations and personal relationships (typically at the top-management level) amongst the co-orchestrators. For instance, in the Autonomous Busses case, the respondent stated: 'Such disruptive topics, like autonomous driving, we do not manage alone. Therefore, it is important to find the right partners to work on it. We were already a partner of [University 1] and naturally, when there is data analytics, we said [University 1] is very interesting and so we continued with them.' (I1). We made similar observations in the Logistics case. From the beginning, there was an intention to conduct deliveries with drones. The logistics company therefore integrated an air-cargo company so that the ecosystem was able to cover all the activities necessary to realize the value proposition. For the ecosystem, the logistics company chose a known partner who was already pushing forward in the same field of business: 'The [Cargo Company], we brought in. We knew that [Cargo Company] worked on the same topics and we knew that they had an exchange with [Drone Technology Provider]. So, we said let's do it together.'(G2). The Smart Commuting case sheds light on the role of trust and the lack of hierarchy among co-orchestrators. The automotive OEM that acted as the initial orchestrator and driver behind this case's ecosystem started initial partnerships with a consulting company and a software company to jointly orchestrate the smart city ecosystem. This was due to the access provider's (the orchestrator) large distance of knowledge related to smart cities. Naturally, the original orchestrator had clear expectations regarding the collaboration between the partners: 'If you steer an ecosystem jointly with two other orchestrators, you do expect that the whole profit-and-loss thing is fair and equal.' (J1) However, the initial partners in the Smart Commuting case (unlike those in the Autonomous Busses or Autonomous Delivery cases) did not come together based on previous relationships or trust. Rather, the collaborators met at an industry fair. Given the lack of established trust, they attempted to pre-define their future collaboration using clearly defined contracts. This led to the whole ecosystem project being delayed for about a year, since it turned out to be highly difficult to agree on a contract about future returns and future activities related to an ecosystem that did not yet exist and whose value proposition had not even been clearly defined.
'It turned out that [Consulting firm], beyond the idea of the ecosystem, was on an acquisition tour as well and tried to sell projects. It was funny; they used, I think, a classical sales tactic. At the beginning, they worked for free for the ecosystem and showed us how important they are. But later, they wanted to be paid for their man-hours. So, in order to define the contract, we had to calculate our contributions as well. And this was excessively difficult, since we contributed technologies and knowledge, for example. How could we quantify this? As a result, we got stuck in these negotiations.' (J1). Due to the distance of knowledge, an ex-ante definition of activities is difficult to achieve and can therefore only be defined (if at all) by broad and general framework agreements. However, if the distance of knowledge is small and the required activities are known, these activities can be defined ex-ante by using formal contracts. These findings lead to the following propositions: Proposition 2. The larger the effective distance of the initial orchestrator's existing knowledge and the field of business the ecosystem focuses on, the more the activities remain flexible. The smaller the effective distance of knowledge, the more the activities are pre-defined and inflexible.
Proposition 3. The larger the effective distance of the initial orchestrator's existing knowledge and the field of business the ecosystem focuses on, the more ecosystem actors act as co-orchestrators of the ecosystem.
In our last finding, we delve deeper into the characteristics of the orchestrator(s). All cases showed the importance of the orchestrators remaining as flexible as possible. In the Contactless Payment case, the partner that initiated the ecosystem did so by founding a spin-off of an incumbent corporation, in order to be able to act as the orchestrators, since the spin-off has more flexibility and fewer compliance rules than its parent corporation. The same is true for the Insured Factoring case, as our respondent in the latter case stated: 'Our partners are large firms; they have their structures and rules. As an orchestrator, you need to adapt to your partners. If one of these big companies orchestrated the ecosystem, they would not be able to adapt to other big corporations. […]. We do not have structures, so we can adapt to everyone.' (D1). On top of this, our cases show the significance of this firm having a very fast pace of product development, agility, and innovativeness. For instance, as one of the respondents in the Intelligent Insurance case stated: 'The insurance company, they're sluggish, slow, and so on. And we are trying to be super-fast, super agile, so we do everything as rapid prototyping and wire frames. This also means we can actually generate a product within four weeks or in a two-week sprint, and that would take a classic insurance company three to five years.' (E1). Highly motivated employees are essential for this, as the CEO of the Intelligent Insurance orchestrator highlights: 'Well, the agility, the innovations… Of course, you have people, who are ready to work extremely long and a lot when it matters. […] We work day and night.' (E1). He went on to describe the organisation of his company: 'We are simply more mobile, agile. These aren't just buzzwords; it is actually like this.' (E1) Agility, highly motivated employees, and speed of development are features typically associated with start-ups or smaller companies. Thus, it is no surprise that in the cases characterised by small distance of knowledge, the orchestrator firm was either a start-up (e.g. Insured Factoring, Virtual Mobility), a spin-off from an incumbent with start-up-like organisation (e.g. Contactless Payment), or a young high-growth company (Intelligent Insurance). This is understandable since the ecosystems in these cases had the sole purpose of implementing the value propositions that the orchestrator had previously defined. The faster the orchestrator drive this development, the shorter the time-to-market will be. Thus, our cases clearly show that start-ups and firms with start-up-like characteristics make excellent orchestrators. Apparently, resources or superior power are not always necessary to orchestrate an ecosystem. For instance, in the Insured Factoring case, the orchestrator is a start-up with ten employees, which does not bring in any resources or take over any tasks beyond the mere orchestration of the ecosystem.
However, in all cases with larger effective distances of knowledge, it became apparent that resources provided by the orchestrator play an important role, in addition to the aspects already mentioned above. In the Virtual Mobility case, the initiator of the ecosystem is the primus inter pares among the orchestrators, since it provides the budgets for the development activities. In the Smart Commuting case, the consulting company is a co-orchestrator, as one of the respondents explained: 'We would not have had the manpower to orchestrate this ecosystem. They [Consulting Firm] had around ten people working for this project, and we do not have that kind of budget.' (J1). Also, in all of these cases, the time-to-market, from defining the value proposition jointly with the other orchestrators, searching for and integrating further partners, and implementing the value proposition, took a very long time, typically more than two years. In the cases in which one orchestrator was driving the ecosystem, however, the same process lasted around one to one and a half years. Thus, it would be difficult for a start-up with a large distance of knowledge to: first, receive funding without a fully developed value proposition; second, provide the resources for the activities needed to define the value proposition jointly with the partners; and, third, have the credibility to attract other companies for such a project. Accordingly, and for these reasons, in all of these cases (Autonomous Busses, Autonomous Delivery, Smart Commuting, and Digital Prescription), the orchestrators were incumbents. Finally, this long process of identifying the issues and answers necessary to define the value proposition and searching for the partners needed for its implementation requires a larger attentional capacity, which is also true for larger firms with more manpower. On the contrary, if the effective distance of knowledge is small, the orchestrator needs to act quickly in order to address the market in a timely fashion. This requires delivering answers to issues with great speed, leaving out formalities and ordinary processes to easily adapt towards the partner. Such flexibility is required for start-ups and therefore a typical characteristic (c.f. Baert et al., 2016;Cainarca et al., 1992;Palmié et al., 2016b;Sirmon et al., 2011).
Therefore, this leads to the following proposition regarding the orchestrator as the firm that designs the ecosystem and is the core actor within it: Proposition 4. The larger the effective distance of knowledge, the more it is beneficial if the orchestrator(s) is/are characterised by solid funding, substantial resources, and a strong reputationall of which are typical for incumbents. The smaller the effective distance of knowledge, the more it is beneficial if the orchestrator(s) is/are characterised by flexibility and speed of product developmentwhich is typical for start-ups.

Introducing a framework of ecosystem design from an attention-based view
Our findings are summarized in the framework displayed in Fig. 7. This framework illustrates how the surrounding conditions of substantive uncertainty and effective distance of knowledge affect an ecosystem's design, and how the ecosystem perspective (Adner, 2017;Jacobides et al., 2018), in combination with the attention-based view of the firm (Ocasio, 1997), jointly elucidate the underlying mechanisms that link ecosystem design with these two surrounding conditions. Thus, the framework contributes to the respective call for research (c.f. Adner, 2017;Dattée et al., 2018;Jacobides et al., 2018;Laamanen, 2017;Phillips and Ritala, 2019) by providing an explanation for ecosystem design.
Before we discuss the framework, here is a brief recapitulation of the constructs contained in it: The surrounding conditions that influence the design of ecosystems are first summarized. 'Substantive uncertainty' (Dosi and Egidi, 1991) describes a state in which there is a lack of information. In cases of weak substantive uncertainty, the exact state remains unknown but can be estimated, whereas in cases of strong substantive uncertainty, the missing information cannot be assessed. 'Effective distance of knowledge' (Afuah and Tucci, 2012) is the distance between the knowledge held by an orchestrator and the knowledge required to solve a problem, such as the creation of the value proposition. The distance consists of two components: the expertise of the orchestrator for the ecosystem value proposition, and the required in-depth knowledge for the value proposition compared to the knowledge the orchestrator has.
The design of an ecosystem canlike any organisational designbe examined via its structures and activities (c.f. Delbecq, 1967;Hall, 1972;Nadler and Tushman, 1970; for a pioneering introduction into the ecosystems concept please see Rong and Shi, 2015). Structures are driven by formal organisational arrangements, such as members and their links, while activities are the procedural actions of the actors for the provision of the value proposition, which is key for an ecosystem (c.f. Adner, 2017;Hannah and Eisenhardt, 2018;Parente et al., 2018).
Finally, we look at the ecosystem orchestrator. This is the key decision-maker as this firm determines the structure and activities and, thus, the design of an ecosystem (c.f. Autio and Thomas, 2019;Moore, 1996;Williamson and de Meyer, 2012). As displayed by the framework, our attention-based view on ecosystem design embraces the three elements of structure, activities, and the orchestrator. Whilst the first is solely influenced by the prevailing uncertainty, the other two are affected by the effective distance of knowledge.
Regarding the first, as our cases show, it would be beneficial from an ecosystem perspective to design an ecosystem characterised by a large and multilateral structure in order to achieve a stronger joint value proposition. However, this would increase the number of issues and answers decision-makers need to focus their attention on, as well as the number of attention channels they would need to be involved in. Given the difficulties of decision-making and attention distribution in situations of higher uncertainty (Ocasio, 1997), it becomes apparent that regimes of stronger uncertainty thus limit the size and the degree of multilaterality of ecosystem structure (Propositions 1a, b).
Activities are influenced by the distance of knowledge (Proposition 2), which shows a trade-off between an appropriate distribution of attention and the design requirements of an ecosystem. From the perspective of the distribution of attention, it would be beneficial to pre-define ecosystem activity in order to allow for preparator attention and, thus, a faster and more accurate distribution of attention (Ocasio, 1997). However, from an ecosystem perspective, such preparatory attention is difficult to achieve if the orchestrator lacks the knowledge to pre-define necessary activities.
Also, effective distance of knowledge influences the orchestrator as the key decision-maker within the ecosystem (Adner, 2017;Autio and Thomas, 2019;Dattée et al., 2018;Jacobides et al., 2018;Masucci et al., 2020). The greater the distance of knowledge, the more actors need to take over the role of co-orchestrator in order to bring in missing knowledge components, whilst a smaller knowledge distance allows the initial orchestrator to lead the ecosystem by itself (Proposition 3). Also, the greater the distance of knowledge, the more the orchestrators are involved in exploring suitable value propositions and opportunities for the ecosystem, which requires them to have a larger attentional capacity, and a broad and flexible focus of attention (Barnett, 2008;Ocasio, 1997). In this case, it more beneficial if there are more orchestrators involved and if these orchestrators are large companies with substantial resources (Proposition 4). This also allows orchestrators to cover a broader range of activities and react flexibly if certain activities are required. However, if these aspects are less significant, i.e. in situations with a smaller distance of knowledge, from an ecosystem perspective, it is more beneficial if fewer actors are involved as co-orchestrators, as this reduces the need for co-operation among them. Likewise, start-ups as orchestrators can bring in the required flexibility and speed of product development, which are both beneficial for the development of the ecosystem's joint value proposition.
In the following, we explain how our propositions and the resulting model contribute to research on ecosystems.

Implications for existing and future research on ecosystems and the attention-based view of the firm
With our Propositions 1a and 1b, we show that the substantive uncertainty inherent in the environment is an important influencing factor for the structure of an ecosystem (i.e. the number of actors and modules provided), as well as for the density of links among the actors. Previous works on ecosystems, such as Iansiti and Levien (2004b), as well as Clarysse et al. (2014), view ecosystems as networks of loosely connected actors. Our Proposition 1 specifies these findings: Actors in ecosystems tend to be loosely connected only in situations of stronger substantive uncertainty. This, in combination with the number of actors and activities, which depend on the uncertainty, has important implications for the value the ecosystem concept is able to create: An ecosystem's strength is providing a design through which complementarities among non-generic modules can be obtained (Jacobides et al., 2018). Such complementarities lead to higher returns than the sum of investments from all partners involved, or to the same investments with lower costs for all partners (Arora and Gambardella, 1990;Cassiman and Veugelers, 2006). Arguably, the greater the number of actors involved in an ecosystem, the greater the number of modules provided. When more, complementary modules are provided, the overall effect of complementarity is stronger. On the other hand, since modules in an ecosystem are mostly non-generic, they need to be mutually adapted (Jacobides et al., 2018). The stronger the links between ecosystem actors, the easier it is for them to mutually adapt their modules.
According to our Propositions 1a and 1b, situations of weaker substantive uncertainty make it possible to involve more actors in an ecosystem and build more and more complex links between them, i.e. to extend the structure and activities of the ecosystem. Thus, only under these conditions, can the potential of an ecosystem to achieve complementarity effects among non-generic modules be obtained to the fullest extent. This is a significant implication for literature, since 'a litmus test for knowing where an ecosystem approachor any approachadds value is having clarity on where it does not add value' (Adner, 2017, p. 56). According to Jacobides et al. (2018), ecosystems are most likely to occur in industries characterised by strong modularity, e.g. the gaming industry. Based on our Propositions 1a and 1b, we extend this finding by assuming that ecosystems might be particularly beneficial, and have a higher likelihood to emerge, in those industries or situations characterised by weaker substantive uncertainty.
This also demonstrates a significant difference between the logic of ecosystems and the stance of the literature on networks and alliances. The latter literature typically views such constructs from the perspective of information and knowledge exchange (Ahuja, 2000;Burt, 1992;McEvily and Zaheer, 1999;Muthusamy and White, 2005;Powell, 1998;Powell et al., 1996;Tsai and Ghoshal, 1998;Uzzi, 1996Uzzi, , 1997: The more actors involved and the stronger the links between them, the better the exchange of information and knowledge will be (Mariotti and Delbridge, 2012). Thus, large and tightly linked networks are particularly beneficial in situations of strong uncertainty, whilst they are less necessary in situations of weak uncertainty (Burt, 1992). The ecosystem logic, however, focuses on the aspect of how partners can jointly create a value proposition, with very different results: The weaker the uncertainty, the greater the number of partnersand the greater the linkages between themwho can be involved in the ecosystem, providing more modules to the joint value proposition in general, which increases its overall benefit for the customer.
On the other hand, mutual dependency among the actors has been mentioned as a core aspect of ecosystems (Adner, 2017;Christensen and Rosenbloom, 1995;Ganco et al., 2020;Moore, 1996). According to our Propositions 1a and 1b, this dependency is particularly strong in situations of weaker substantive uncertainty, given the dense connections between partners involved, whilst it is less pronounced in situations of high uncertainty. Thus, mutual dependency as a key challenge of an ecosystem is less pronounced in situations of stronger substantive uncertainty.
Proposition 2 present a surprising finding in terms of whether the design of an ecosystem is pre-defined by the orchestrator or shaped by the situation. Based on conventional wisdom, one might expect uncertainty to be the key influencing factor in this regardthe higher the uncertainty, the more essential flexibility and adaption to changing environments become, especially in inter-firm settings, such as alliances or supply chains (Albers et al., 2016;Teece et al., 2016). In contrary to these findings, our proposition shows that in ecosystems uncertainty is not an influencing factor in this regard. If an orchestrator has a small distance of knowledge, it will pre-design the alignment structure and maintain it over time, regardless of the uncertainty at hand. As Propositions 1a and 1b show, uncertainty only has an influence on the size (number of actors) of the ecosystem and the density of links deriving from the positions. Therefore, the uncertainty at hand dictates the structure of an ecosystem, while the design of activities depends on the effective distance of knowledge.
The reason for this finding is the modular structure of an ecosystem (Jacobides et al., 2018): The partners are brought in to contribute specific modules to the value proposition, modules that only function in conjunction with complementary modules provided by the other partners. Arguably, the more the design can be defined ex-ante, the more it is possible to make best use of the complementarity effects among the modules and to involve only those actors who can provide modules in the most efficient and effective ways possible. This is another difference between ecosystems and alliances and networks, since those latter concepts lack the ecosystem-specific aspects of a joint value proposition based on complementary modules (Adner, 2017;Jacobides et al., 2018).
Propositions 2 and 3 discuss another important but under-researched topic: The question of how strictly the ecosystem partners are aligned towards the value proposition, which is directly linked to the aspect of hierarchy in ecosystems. This question has been discussed (with great controversy) in previous literature (see Jacobides et al. (2018) for an overview). Some authors (e.g. Gulati et al., 2012;Jacobides et al., 2018) consider ecosystems not to be hierarchically managedothers (e.g. Alexy et al., 2013;Baldwin, 2012a;Brusoni and Prencipe, 2013;Leten et al., 2013;Nambisan and Baron, 2013;Teece, 2016) claim formal mechanisms and rules are used by the orchestrator to steer ecosystem members, even in situations of high uncertainty (Furr and Shipilov, 2018). Our empirical findings help to disentangle these contradictory findings: In all of our cases, the orchestrator(s) are clearly designing the alignment structure and are, thus, hierarchically leading the complementors within the ecosystem, regardless of the prevailing uncertainty. However, in cases where the initial orchestrator has a larger distance of knowledge and is, therefore, not able to steer the ecosystem alone, there is no hierarchy among the co-orchestratorsyet there is still a clear hierarchy between orchestrators and complementors.
These findings show another aspect in which ecosystems are different from alliances. Previous findings on alliances (e.g. Gulati and Singh, 1998;Mowery et al., 1996;Williamson, 1991) have stated that they should be managed more hierarchically in situations of higher uncertainty. In ecosystems, however, the orchestrator tries to formalise the design and to govern the partners hierarchically if the orchestrator possesses the network and knowledge necessary to do so, regardless of the uncertainty at hand. This finding can be explained based on the modular nature of an ecosystem and the complementarity effects among the modules (Jacobides et al., 2018): In order to make best use of complementarity effects, and to cope with the problem of dependency among the actors (Adner, 2017), a clearly designed alignment structure and, thus, strict governance, seems to be key.
In general, our Propositions 2 and 3 provide insights into one of the key aspects that distinguishes ecosystems from networks or alliances: Ecosystems are characterised by a joint value proposition, which can only be delivered if all actors are aligned, i.e. stick to the designed structure and activities (Adner, 2017;Jacobides et al., 2018). This, in turn, implies the significance of the orchestrator as the central player ensuring this alignment. Propositions 2 and 3 shed light on the question of how and why the allocation of roles regarding orchestrators and complementors takes place, which is a unique feature of an ecosystem, as opposed to alliances and networks (Adner, 2017;Jacobides et al., 2018).
Propositions 2 and 3 also delve deeper into the characteristics of the actors involved in an ecosystem. An ecosystem consists of actors interacting in order for a value proposition to come true (Adner, 2017). As our propositions highlight, ecosystems can profit from specialisation (which can be regarded as initial capabilities) but they don't necessarily have to. Brusoni and Prencipe (2013) and Williamson and de Meyer (2012) found that ecosystems consist of specialised actors taking over specific roles and activities, especially when uncertainty is high. On top of this Brusoni and Prencipe (2013), as well as Williamson and de Meyer (2012), highlight the importance of specialised actors, particularly for tackling complex problems. This runs counter to the findings of Hannah and Eisenhardt (2018, p. 3190) who found that in more complex strategic settings, like ecosystems, specialisation (initial capabilities) have fewer consequences than strategy does, as some required activities could not have been foreseen ex-ante. Our Proposition 2 resolves this contradiction. The Autonomous Busses, Smart Commuting, Digital Prescription, and Autonomous Delivery cases have ecosystems specifically focused on the development of joint solutions for highly complex problems, e.g. autonomous flying and driving or smart cities. These ecosystems rely on co-orchestrators being actors with the ability to assume multiple roles and activities within the ecosystem, as well as specialised complementors being involved to implement the joint value proposition. The requirement of specialised activities in an ecosystem is influenced by the initial orchestrator's distance of knowledge, regardless of the prevailing uncertainty; this contradicts the findings of Brusoni and Prencipe (2013) and Williamson and de Meyer (2012), yet explains the findings of Hannah and Eisenhardt (2018).
Our Proposition 3 sheds light on a largely under-researched topic: The question of whether ecosystems have one or several orchestrators. Whilst the existence of an orchestrator is one of the fundamental characteristics of an ecosystem (Adner, 2017;Altman and Tushman, 2017;Iansiti and Levien, 2004a;Jacobides et al., 2018;Moore, 1996), until now it has been unclear whether ecosystems can be successfully run by several co-orchestrators, and there has been uncertainty about the conditions under which such multi-orchestrator ecosystems emerge (c.f. Autio and Thomas, 2019;Hannah and Eisenhardt, 2018). Proposition 3 identifies the initial orchestrator's effective distance of knowledge as a key influencing factor that defines whether or not it is beneficial for this firm to involve additional firms as co-orchestrators.
This raises several important questions for future research. Ecosystems are not considered to be stable constructs, but are expected to change over time (Moore, 1993(Moore, , 1996Rong and Shi, 2015). This might be particularly true for ecosystems with several orchestrators. The initial orchestrator might learn over time, thereby reducing its distance of knowledge (over time), which, in turn, reduces the need to involve co-orchestrators as a way of overcoming that distance. Yet, how do such multi-orchestrator ecosystems develop over time? Is there a fight for power until one orchestrator takes over? Which orchestrator leads the ecosystem into the next stage of development? And how stable are such constructsdo they split into several sub-ecosystems, each of them led by one orchestrator? Our research clearly shows under which conditions such multi-orchestrator ecosystems exist. But more research is needed to gain a deeper understanding of such constructs when ecosystems mature and reach the next stages of their lifecycles. The same is true given the general focus of our paper on the alignment structure of ecosystems in early stages of the lifecycle. Thus, we call for research on alignment structures in later stages of the lifecycle. This is particularly interesting since, as described in our methods section, the alignment structure might influence the orchestrator's distance of knowledge over timeleading to a cycle of knowledge distance influencing alignment structure and alignment structure, in turn, influencing knowledge distance in the long term.
Proposition 4 shed new light on the role and profile of the orchestrator. Some previous works have highlighted orchestrators as being established companies that provide key resources and commercial infrastructures (e.g. Clarysse et al., 2014;Zahra and Nambisan, 2012). Others claim that the orchestrator does not have to be the largest player in the ecosystem, but could be one that uses smart power, problem framing, or informal authority (e.g. Brusoni and Prencipe, 2013;Gulati et al., 2012;Williamson and de Meyer, 2012). Our findings differentiate these findings more clearly by elucidating that orchestrators with a smaller distance of knowledge are typically start-ups or firms exhibiting start-up like organisation. Thus, resources, power, or firm size are less relevant in these cases compared to typical features of start-ups, such as the pace of product development, neutrality, and flexibility. Only in ecosystems led by several co-orchestrators do the firms tend to be large and established in order to have the resources in place (monetary as well as work force) to develop the joint value proposition and survive the comparably long time until the value proposition is developed and brought to the market. These new findings open up a new pathway of research, which may explain how start-ups orchestrate ecosystems, even though they face limitations, such as resource constraints or lack of reputation and brand name, factors that, arguably, are not ideal when orchestrating several interlaced companies that are dependent upon the orchestrator.
Proposition 4 also demonstrates another difference between ecosystems and networks/alliances. Start-ups typically enter alliances and networks for access to resources held by corporations, whilst start-ups bring innovativeness to the alliance that corporations often lack (Fang, 2008;Hu et al., 2017;Rothaermel and Deeds, 2004;Wuyts et al., 2004). As our findings reveal, this is also true for ecosystems: Most of the start-ups possess innovative technologies that are important enablers of the innovation the ecosystem is striving to create. However, a key factor distinguishing ecosystems from alliances and networks is the strong focus on the creation of a joint value proposition (Adner, 2017;Jacobides et al., 2018). This leads to a start-up taking on an additional role when it is the orchestrator: Organising the development of the joint value proposition and ensuring alignment. As our findings and Proposition 4 demonstrate, the start-up's main advantage in this role, compared to corporations, is its higher flexibility, agility, and neutrality, which are important qualities of an orchestrator. Thus, and especially in cases such as Insured Factoring, where the start-up does not possess any technologies but is merely the orchestrator of the ecosystem, the start-up is a key player in the ecosystem simply because of its ability to orchestrate. This finding adds an important contribution to the literature on start-ups in inter-firm collaborations and opens up interesting pathways for future research, such as: How can start-ups be evaluated if their mere purpose is the orchestration of an ecosystem? What are the resulting investment strategies? And which factors influence whether a start-up becomes the orchestrator of an ecosystem or not? For instance, do such start-ups display effectuation or a causation approach to decision-making (e.g. Palmié et al., 2019;Sarasvathy, 2001)? What kind of investors would be interested in such start-ups? And does the risk of failure for start-ups decrease or increase if they are orchestrating ecosystems? There are also questions from the perspective of the complementors in a start-up-led ecosystem: How can partners secure the success of the start-up and avoid failure of the ecosystem as a whole? What are (financial) strategies to possibly integrate such start-ups in later stages? How can corporations use start-ups to drive innovation in ecosystems? These are just a few questions that emerge from our fourth proposition, which, we hope, opens up a new stream of research in the entrepreneurship literature and beyond.
Additionally, applying the attention-based view to ecosystems helps to gain a better understanding of this emerging phenomenon and might open up intriguing pathways for future research along these lines. In this vein, our propositions reveal two different types of ecosystems. First, in the case of a smaller effective distance of knowledge between the orchestrator and the field of business the ecosystem is focused on, the orchestrator focuses its attention on partners and potential value propositions closely related to the existing domain of experience and knowledge, which can be seen as local search (Afuah and Tucci, 2012;Cyert & March 1963;March and Simon, 1958). In such situations, the ecosystem is clearly focused on the realization of the joint value proposition and further actors are involved solely for this objective. Second, in the case of a larger effective distance of knowledge, the ecosystem involves some actors for the materialization of the value proposition, and involves others to serve as co-orchestrators, who mostly provide the knowledge and network to understand the ecosystem's field of business. Given the larger effective distance of knowledge, the orchestrator has to conduct distant search, since the field of business the ecosystem is focused on is not related to its prior experience and knowledge base (Afuah and Tucci, 2012;Cyert & March 1963;March and Simon, 1958). Extant research in the context of attention has argued that local search offers several advantages over distant search, since it allows firms to pay attention to and utilise novel knowledge and information faster and more accurately (Ocasio, 1997; please, also see Anand et al., 2016;Denrell & March 2001;Helfat, 1994;Knudsen and Levinthal, 2007). Even though our qualitative research design does not allow for a performance evaluation of the different types of ecosystems identified in this study, we argue that the orchestrator does achieve a higher financial return when the effective distance of knowledge is smaller. In such situations, ecosystem actors are brought in because they are specialised in the activities and contributions that help achieve the value proposition, which, arguably, increases the ecosystem performance compared to ecosystems where actors are not specialised in their specific activities. Also, if additional orchestrators are involved in the ecosystem, they will demand their share of the overall returns, which reduces the initial orchestrator's share. Thus, we argue that local search is advantageous in an ecosystem context (and not just in classical one-firm settings).
Local search is also associated with a lower degree of innovation (e.g. Fleming, 2001;Fleming and Sorenson, 2001;Katila and Ahuja, 2002) and a lower potential for substantial performance improvements in complex environments, since building novel solutions upon existing knowledge reduces the extent of knowledge recombination (Fleming, 2001). In addition, innovative solutions are less likely to be found in the vicinity of the existing experiences (Barnett, 2008;Levinthal, 1997). Yet our study shows that ecosystems are an exception to these findings. Even though the orchestrator conducts local search in situations with a smaller effective distance of knowledge, our cases clearly show that the ecosystem is able to create a highly innovative solution. In fact, creating a value proposition that is based on non-generic, thus very innovative, modules is a core aspect of the ecosystem concept (Jacobides et al., 2018). Therefore, ecosystems might be an alternative way for firms to step into novel fields of business and create substantial innovations without leaving the existing domain of knowledge and without being forced to conduct distant search.
Previous research has introduced various means of conducting distant search, since it is considered to be more difficult to perform than local search (Cohen and Levinthal, 1990;Cyert & March 1963). Research has shown that drawing on actors from outside of an organisation may facilitate broader search (Rosenkopf and Nerkar, 2001;Schildt et al., 2005) (please see Posen et al., 2018 for an overview). Our research introduces ecosystems as a further means of conducting distant search. This offers an intriguing pathway for future research to explore how the distinctive features of ecosystems, such as collaborations between actors from different backgrounds on a joint value proposition, mutual adjustments of unique or supermodular modules, or multilateral collaborations among actors (Adner, 2017;Jacobides et al., 2018), affect distant search. In general, if the single firm is no longer adequate as the primary level of analysis, and ecosystems are replacing traditional thinking in firms and single organisations, so too should the attention-based view of the firm develop into an attention-based view of ecosystems. We hope this study, as one of the first to apply this theory to ecosystems, will spark future research along these lines.

Practical implications
Given the growing significance of ecosystems both in research and also in industry practice, we would like to channel our findings into several implications for practitioners, especially for companies serving as ecosystem orchestrators.
Our first proposition emphasises the significance of the substantive uncertainty of the environment as a core influencing factor on the design of an ecosystem. The weaker the uncertainty, the larger the ecosystem (i.e. more actors and positions) and the greater the links among the actors. This leads to an ecosystem with greater, stronger complementarity effects among the modules provided by the actors involved and, thus, provides greater benefits to the ecosystem's customers. Therefore, ecosystems might be particularly beneficial in situations of weak substantive uncertainty (c.f. Proposition 1).
According to Propositions 2, 3, and 4, the effective distance of knowledge of the initial orchestrator driving the ecosystem is a key influencing factor on how the ecosystem will be designed and shaped. The smaller the effective distance of knowledge, the stronger of an alignment structure the orchestrator will design, and, thus, will govern the ecosystem by itself, resulting in a single orchestrator ecosystem. The orchestrator will also greatly pre-define the ecosystem structure and activities. Ideally, the orchestrator shows characteristics of a start-up, namely speed of product development, agility, and flexible forms of organisation. Thus, if the orchestrator of such an ecosystem appears to be a corporation, it might consider founding a spin-off or acquiring a start-up or a small firm as an orchestrator for the ecosystem. The larger the initial orchestrator's distance of knowledge, the more likely it will be to involve additional partners as orchestrators, and a multi-orchestrator ecosystem will emerge. The activities will be flexible and adjusted to the situation at hand, with the orchestrator ideally showing characteristics of a corporationmostly solid funding and a reputation as a trustworthy partner, despite there being no clearly defined value proposition at the beginning.