Potential and recognized boundary spanners in multinational corporations

Boundary spanners play an important role in multinational corporations (MNC), yet it is unclear who these valuable individuals are and why certain individuals, and not others, perform this role. We advance a ‘ recognition ’ perspective based on whether and how relevant others on both sides on the boundary experience positive impact. A dynamic integrated mixed method analysis of 118 individuals involved in headquarters-subsidiary interactions in four MNCs, shows that only a minority are ‘ recognized boundary spanners ’ , experi- enced by others to positively impact intergroup relations. We identify di ﬀ erent categories and mechanisms of recognition, and make a methodological contribution by integrating qualitative and quantitative analysis.


Introduction
Individuals who work across unit boundaries, or boundary spanners, play a pivotal role in internal collaboration and coordination within multinational corporations (MNC). They have traditionally been defined as individuals who are responsible for inter-group contacts (e.g., Aldrich & Herker, 1977;Friedman & Podolny, 1992;Johnson & Duxbury, 2010) or otherwise engage in frequent interactions across group boundaries (Richter, West, van Dick, & Dawson, 2006;Tushman & Scanlan, 1981; see also Adams, 1976 andCallister &Wall, 2001). Knowing who creates the most value in this regard is crucial input for staffing and resource allocation decisions, with consequences for central organizational processes such as inter-unit coordination, knowledge sharing, and innovation Richter et al., 2006;Tortoriello, Reagans, & McEvily, 2012). But who are these valuable individuals, and why do certain individuals, and not others, perform this role? Below we engage with this important question from a recognition perspectivefrom the perspective of how relevant others experience the actions of the boundary spanning individuals.
Previous research on boundary spanning in MNC contexts has taken significant strides towards our understanding of boundary spanners, typically focusing on their characteristics, motivations and job roles (Barner-Rasmussen, Ehrnrooth, Koveshnikov, & Mäkelä, 2014;Levina & Vaast, 2005Richter et al., 2006;Tortoriello et al., 2012). Yet, an individual's involvement in inter-unit interactions does not necessarily mean that others in that interaction context experience that this involvement improves inter-unit relations. People may experience the actions of others as not particularly or at all helpful, even if the intentions may be such. Given these potential discrepancies, we suggest that in addition to examining the boundary spanning individuals themselves, we also need to understand whether and how relevant others, who are influenced by the activity, experience a positive impact. This approach to boundary spanning in MNCs is analogous to customerand user-(e.g., Beyer & Holtzblatt, 1998;Norman & Draper, 1986;Shah, Rust, Parasuraman & Staelin, 2006) or employee-centred (e.g., Bowen & Ostroff, 2004) approaches in other fields, where the experiences of users and receivers take precedence over intended product, service, or practice characteristics.
Building on the above, we make a difference between individuals who are involved in inter-unit interactions (whom we term potential boundary spanners), and the subset of these individuals who are experienced by relevant others to facilitate, or positively impact, intergroup transactions and relations (whom we term recognized boundary spanners). The aim of our study is to explain recognized boundary spanning, and the empirical research question we ask is: "Which individuals in a given interaction context are experienced by relevant others to facilitate inter-group relations, and why?" By doing so, our study builds on and forwards the recent vein of IB research on boundary spanning that highlights the importance of deep contextual insight (Søderberg & Romani, 2017) and sensitivity to the experiences of others (Yagi & Kleinberg, 2011).
inter-unit tasks in a specific interaction context, instead of examining the characteristics of a predetermined sample of individuals assumed a priori to be boundary spanners. First, we investigate which members of the population of these potential boundary spanners in the focal context are recognized by relevant others on both sides of the boundary to facilitate intergroup transactions and relations; then, we explore why these individuals are recognized. The core difference between this approach and previous research is the requirement that relevant others need to experience and recognize positive impact in the actions of the boundary spanner. As will be discussed later, this change in perspective provides important new insights on boundary spanners and boundary spanning.
The study builds on 145 interviews with 118 individuals involved in cross-boundary interactions between the headquarters (HQ) and a strategically important subsidiary in four MNCs. The interviews were conducted over 18 months of in-depth empirical work, providing us with deep qualitative insight into all four case contexts. This combination of data volume and qualitative insight enabled us to adopt a mixed-methods research approach (Molina-Azorin, Bergh, Corley & Ketchen Jr, 2017), where we first analysed interview accounts qualitatively to identify who our interviewees recognized as positively facilitating inter-group transactions and relations, and then examined significant differences between these individuals through statistical analysis of quantified interview data. Thereafter, we sought to understand why certain individuals and not others were recognized, by combining further qualitative interpretation and statistical analysis in a dynamic integrated mixed method approach (Fetters & Molina-Azorin, 2017;Turnarosa & Glynn, 2017), following up on 'surprises' (van Maanen, Sörensen, & Mitchell, 2007) emerging at each stage of analysis. This method integrates strengths of qualitative research, such as multiple voices, including those of the researchers (Ketokivi & Mantere, 2010;Welch & Piekkari, 2017) and high sensitivity to the complexity of the local context (Johnson & Onwuegbuzie, 2004;Van Maanen, Sørensen, & Mitchell, 2007), with strengths of quantitative research, such as tests and validation that uncover broader patterns in the data (Johnson & Onwuegbuzie, 2004). Below, we first review previous research on firm-internal boundary spanning in MNCs, then explicate our data and research approach in detail, and present our analysis and results. The paper ends with a discussion of the findings and their theoretical and practical implications.

Boundaries and boundary spanning in MNCs
MNCs have many advantages over other organizational forms, including enhanced opportunities to combine various bodies of knowledge both internally and externally (Kogut & Zander, 1993); yet, due to their geographical dispersion, they also face considerable internal coordination challenges (Bartlett & Ghoshal, 1987). Formal structural mechanisms created to overcome these, such as steering groups and communities of practice, cannot substitute for well-functioning regular interunit interactions in the context of the firm's core operations (Argote & Ingram, 2000). Characteristic of MNCs is that these interactions take place across many different boundaries, and often simultaneously (Carlile, 2002(Carlile, ,2004.
Boundaries have long attracted research attention as a defining organizational characteristic with important implications for the patterns of interaction among organizational actors, their inclusion and exclusion, and the allocation and distribution of authority (see Aldrich & Herker, 1977;Leifer & Huber, 1977). The existence of boundaries forces organizational actors to assume boundary roles such as information processing, which increases organizational ability to learn and adapt to environmental changes, and external representation, which enables effective organizational responses to environmental constraints and contingencies (Birkinshaw, Ambos, & Bouquet, 2017;Johnson & Duxbury, 2010;. All organizations have boundary roles for their members to take on, although these vary in terms of extent and formalization. Building on Carlile (2002),2004), we define a boundary as an organizational barrier characterized by difference (without difference there is no boundary), dependence (without dependence the boundary is not meaningful) and novelty (without unfamiliarity there is no friction). Most MNC-internal unit boundaries score high on these characteristics: they are typically separated by clear and meaningful differences in terms of organizational structure and geographical distance , and with a high degree of novelty in terms of unfamiliarity and foreignness (Zaheer, 1995). Reciprocal dependence is common in that subsidiaries depend on HQ and other units for various resources and inputs, while HQ rely on subsidiary units for key parts of the global value chain, and/or as contributors to sales revenue (Birkinshaw, Holm, Thilenius, & Arvidsson, 2000). What is more, interunit boundaries in MNCs are typically associated with strong faultlines between the two social groups they separate (Lau & Murnighan, 1998;Mael & Ashforth, 1992), each unit possessing its own localized knowledge, practices and discourses (Carlile, 2002;Yagi & Kleinberg, 2011). This is relevant for boundary spanning because most boundary spanners in the MNC context are located on either side of the boundary. In other words, they are members of one group (their in-group) and not the other (their out-group). Morgan and Kristensen (2006) conceive of the MNC as a contested social space, rife with organizational micropolitics and clashes between units, particularly between HQ and foreign subsidiaries (Kristensen & Zeitlin, 2001). Across this boundary, the actions of the other side often appear to be characterized by a weak understanding of operational realities or even "stupid and wrong" (Kristensen & Zeitlin, 2001, p. 188). Combining the above with the general tendency to view one's own in-group and its members more positively than out-groups (Erez & Earley, 1993;Kramer, 1999), interunit boundaries in MNCs are likely to influence both the behaviour of the interacting individuals and their experiences of each other.
In this context, boundary spanning is obviously challenging, but also highly important. What boundary spanners do in MNCsfostering inter-group relations by sharing knowledge, linking people and ideas, brokering between different fields and discourses, and resolving conflicts (Barner-Rasmussen et al., 2014)has been shown to contribute to a range of positive organizational outcomes Levina & Vaast, 2008;Richter et al., 2006;Schotter & Beamish, 2011;Tortoriello et al., 2012). Thus, understanding and encouraging boundary spanning behaviours and the individuals who carry them out is an important competitiveness issue for MNCs.
Previous research has linked boundary spanning to individuals' abilities, characteristics and network positions (e.g., Levina & Vaast, 2008;Marrone, Tesluk, & Carson, 2007;Tushman & Scanlan, 1981), and to organizational roles and practices (e.g., Aldrich & Herker, 1977;Birkinshaw et al., 2017;Zhao & Anand, 2013). These perspectives are partly interrelated, as individuals' network positions may reflect organizational roles. In cross-cultural settings, individuals with particular skills and capabilities, such as multicultural experience and language skills, tend to act as boundary spanners even if this is not part of their formal position or role (Barner-Rasmussen et al., 2014;Yagi & Kleinberg, 2011). Such de facto boundary spanners may be driven by intrinsic motivation and a personal inclination to connect people or knowledge, and can be found in a variety of positions and hierarchical levels (e.g., Beechler, Søndergaard, Miller, & Bird, 2004;Levina & Vaast, 2005). Inversely, individuals formally appointed to boundaryspanning positions are not always the most active or effective in such roles, wasting substantial organizational resources (Levina & Vaast, 2005). These findings challenge the view of boundary spanning as the function of a role or position, such as group or unit leader (Ancona & Caldwell, 1992;Ancona, 1990;Richter et al., 2006), high-status manager (Birkinshaw et al., 2017;Tushman & Scanlan, 1981;Wiesenfeld & Hewlin, 2003), or expatriate/inpatriate (Au & Fukuda, 2002;Kane & Levina, 2017;Reiche, 2011;Thomas, 1994). Indeed, it seems that boundary spanning can only be formalized to a certain extent (Schotter & Beamish, 2011).
These insights have resulted in recent calls to study boundary spanning at the activity level; that is, in terms of the specific actions taken by boundary spanning individuals (Birkinshaw et al., 2017;Søderberg & Romani, 2017; see also Schotter, Mudambi, Doz, & Gaur, 2017). For instance, Birkinshaw et al. (2017) identified four boundary spanning activities practiced by the HQ executives of an MNC: spearheading and facilitating focused on making connections across boundaries whereas reconciling and lubricating focused on overcoming differences in worldview across boundaries. Søderberg and Romani (2017) distinguished between three generic boundary spanning activities of Indian vendor managers in a global IT service company in the context of global IT development projects. These activities were to manage boundaries (buffering and reflecting), forge common ground (connecting and mobilizing), and develop new frontiers (weaving and transforming).
Alongside objective variation in the individual-level motivation and skills needed to act as a boundary spanner (e.g. Barner-Rasmussen et al., 2014;Levina & Vaast, 2005;Marrone et al., 2007), research has highlighted the role that 'others', i.e. those affected by the boundary spanning activity, play in actual organizational situations. For instance, Yagi and Kleinberg (2011) suggest that effective boundary spanners develop effective working relationships by fusing their identity with those of 'others', suggesting an explanation for why bicultural experience has repeatedly been found to be such an important resource for boundary spanners. Kane and Levina (2017), in turn, argue that challenges related to how 'others' perceive boundary spanners' identity may damage their boundary spanning efforts.
These findings highlight two important points for further research. For one, to detect and capture the full range of boundary spanning activity in MNCs, research must not rely solely on employees' positions or role descriptions. For the other, to establish whether a particular individual's activity can be seen as boundary spanning in the sense of positively impacting intergroup transactions and relations, the experiences of 'others' who are influenced by the activity become highly relevant. Without overlooking the possibility that some boundary spanning may also take place inadvertently or covertly, or that some activities experienced by others as boundary spanning may in fact not actually facilitate intergroup transactions and relations, this underlines the need for contextualized studies that capture how boundary-spanning efforts are received by others (Søderberg & Romani, 2017). Much in the same way as the users' or receivers' experiences determine the real value of a product or practice and have a strong impact on how they will relate to it in the future (e.g., Bowen & Ostroff, 2004), relevant others' experiences of positive impact are a key component in identifying boundary spanners who add organizational value.
In sum, previous research has generated important insights on boundary spanners, but with an emphasis on the skills and capabilities of individuals, their activities, and/or the organizational roles in which they operate. To deepen these insights, we need a complementary approach that places the actions of boundary spanners in their social context; in particular, in terms of how relevant others experience these actions. Our empirical approach addresses these lacunae by taking the difference between potential and recognized boundary spanners as a starting point. We define the former as any individual involved in interunit interactions across a focal boundary, and the latter as a subcategory experienced by others as facilitating, or positively impacting, inter-group transactions and relations. The difference between these two groups is illustrated in the conceptual framework in Fig. 1.

Data and methods
Our empirical study builds on 145 qualitative interviews with 118 individuals, yielding circa 1800 single-spaced pages of transcripts and 936,000 words. The data were collected in the course of a study on inter-unit knowledge and competence sharing in four MNCs, below referred to as MNC1, MNC2, MNC3 and MNC4. Here we focus on interunit interactions between their HQs (all in Finland) and major subsidiary units in China (MNC1, MNC2) or Russia (MNC3, MNC4). The criteria for selecting the subsidiaries was that the case MNCs considered them to be of high strategic importance, and effective interaction between them and the HQ a key enabler of value creation. All four contexts fulfilled Carlile (2004) criteria for boundaries in terms of difference (HQ versus subsidiary unit, located in Finland versus Russia/ China), dependence (subsidiaries depended on HQ for resources, HQ depended on subsidiaries for product design and/or production and major sales revenue in strategically important markets), and novelty (HQ and subsidiaries were located in different cultural and linguistic local environments, and had different traditions of doing business). These relatively clear-cut boundaries coincided with a range of interunit interaction contexts and situations for the participating individuals, yielding a rich, varied data set. Basic data on the four MNCs are given in Table 1.
As part of our preparation, we asked our contact persons (located in HR or corporate development) for the names of all individuals who were involved in the focal interunit interactions. New names were added whenever our interviewees mentioned new people. This was a key methodological learning point for us, as we quickly realized that our HQ company contacts did not have a full picture of everyone who was involved, prompting us to broaden our search until no new names were mentioned by interviewees. We believe such lack of a complete picture is typical for large organizations, and was particularly evident in the subsidiary interviews, which indicated that the lists we got from HQ often lacked a number of people who were not only involved in interunit interactions but actually played key roles in it; our contacts at HQ were just not aware of this. Second, we found that accounts of who interacted across the focal boundaries and who did not varied across interviewees, meaning that no source alone could be classified as 'objective'. Indeed, the incompleteness of lists provided by one source only supports the relevance of perceptions in boundary spanning.
This process provided us with rich data, enabled triangulation to capture multiple voices (Welch & Piekkari, 2017), and meant that we were able to interview all individuals that were identifiably and regularly involved in the focal interactionsthe potential boundary spanners. The interviews lasted 45-120 minutes. To get as rich data as possible, most of them were conducted in the preferred language of the interviewee (English, Finnish, Swedish, Russian, Mandarin or Cantonese), facilitated by a native-speaking or highly fluent member of our eight-person research team. 1 Interviews conducted in Russian, Mandarin and Cantonese were translated into English so that the whole research team could access them.
The interviews centred on issues the interviewees deemed relevant to inter-unit interactions. In the first round, we focused on the overall challenges, successes and failures of interactions. We drew on these initial interviews to identify topics and practical examples to be probed more deeply in subsequent interviews. When the interviewees gave examples, we dug deeper to better understand what happened, but did 1 In addition to the four co-authors, the larger research program team members consisted of two senior Principal Investigators who oversaw the overall research program but did not participate in the writing of this paper, and two native Chinese speaking research assistants, who helped us with our interviews in China. not guide the discussion by using the term 'boundary spanning' or asking upfront about the actions of specific individuals. References to individual boundary spanners emerged inductively from interviewees' examples.

Analysisan overview
We adopted a dynamic integrated mixed-method approach to our semi-structured, yet highly narrative and extensive qualitative data. Mixed methods research designs are still relatively uncommon in management studies (Molina-Azorin, Bergh, Corley, & Ketchen, 2017), but can combine strengths of both qualitative and quantitative research (Johnson & Onwuegbuzie, 2004) by improving the question-asking process (Molina-Azorin et al., 2017, p. 184) and helping build theory through interactive elaboration and interpretation of findings (Gibson, 2017). Mixed methods are often used for triangulation, corroborating qualitative analysis with quantitative (Gibson, 2017;Johnson, Onwuegbuzie, & Turner, 2007;Molina-Azorin et al., 2017), but the scope of our data set and our deep familiarity with the case companies enabled a more innovative, explorative and dynamic approach. Rather than collecting qualitative and quantitative data separately (sequentially, in parallel, or concurrently), our approach corresponded to what Teddlie and Tashakkori (2006), p. 15) refer to as "conversion mixed design": We applied qualitative and quantitative analysis to the same original interview data, alternating between the two methods as the analysis process unfolded, each step focusing on answering new questions rather than specifically on triangulation. This allowed for a dynamic and interactive use of mixed methods (Turnarosa & Glynn, 2017, p. 235) that were integrated "in the data analysis dimension" (Fetters & Molina-Azorin, 2017).
By using qualitative and quantitative analysis methods in an explorative, dynamic fashion, moving back and forth between the methods and building on their unique strengths to answer emergent questions at each step of the research process, our mixed methods approach allowed us to balance cognitive and computational aspects of scientific reasoning (Mantere & Ketokivi, 2013). Our research process had deductive elements in that both the qualitative and the quantitative analyses were based partly on literature. Yet largely its nature was abductive, in that it involved a flexible and continuous movement between "concepts, conjectures and data" (Van Maanen et al., 2007: 1146, combined with a similar and parallel movement between methods of analysis. Throughout, we drew heavily on our in-depth understanding of the four MNCs, the interaction contexts, and the focal individuals, digging deeper into emerging insights and posing further questions when initial findings appeared unsatisfactory, unclear or counterintuitive. This digging was often inductive, leaning strongly on the "co-production" of interpretive insights based on triangulating many voices (Welch & Piekkari, 2017, p. 721). To us, this dynamic and integrative analysis process follows the spirit of pragmatism which, in its many forms, offers a paradigmatic positioning of mixed methods (Feilzer, 2010;Johnson, Onwuegbusie, & Turner, 2007;Morgan, 2007) that distances itself both from "positivist" ideals and language such as "objective detachment" (Welch & Piekkari, 2017, p. 721), and also from "purist" interpretive positions (Johnson & Onwuegbusie, 2004, p. 14). Such pragmatist methodological pluralism goes beyond quasi-essentialist views of scientific research by accepting not only the strengths of the two broad scientific reasoning strategies of idealization and contextualization, but also the "uneasy balance between them" (Ketokivi & Mantere, 2010, p. 320), without assuming that any form of scientific inquiry "is an outcome in and of itself" (Mantere & Ketokivi, 2013: 75;cf. Morgan, 2007).
We addressed our overall research question "Which individuals in a given interaction context are experienced by relevant others to facilitate inter-group relations, and why?" in two empirical steps. In Step 1, we identified several subgroups of individuals based on how they were recognized by others, on both sides of the boundary. In Step 2, we Table 1 The case firms and the distribution of respondents.

MNC Key statistics
Description Focus of present study

Interviews
In Finland/ at HQ In Russia/ China MNC1 40,000 employees in 40 countries World-leading process manufacturer Chinese operation including regional HQ functions for Asia Pacific, major production facilities, a local sales and marketing operation and a competence centre focusing on regional competence development addressed the question of why some individuals, and not others, were recognized, examining both individual-level factors (hierarchical position, expatriate job role, structural and relational embeddedness), and the organizational context.

Step 1 analysis and findings: identification of recognized boundary spanners
Our first step was to search the transcripts for any references to specific individuals, who others on either side of the boundary experienced to facilitate inter-group transactions and relations between the focal subsidiaries and HQ units (Richter et al., 2006). Following Barner-Rasmussen et al. (2014), experiences of facilitation, or positive impact, were qualitatively coded based on whether an individual was described by an interviewee to (i) exchange knowledge or information; (ii) link previously unconnected people; (iii) aid the interactions of others (e.g., by functioning as a communication channel and/or an interpreter); or (iv) actively intervene to resolve conflicts and solve problems across the focal boundary. Other types of mentions were not coded as facilitation, such as neutral mentions of interactions ("Mrs. A visited Finland"), or negative mentions referring to harmful interactions, gossip, or explicit lack of positive impact ("Mr. X did not help us", "He started to blame [individuals on the other side]").
The coding was done collaboratively by the four co-authors of this paper. Each of us first coded the same interviews separately. To achieve inter-subjective comparability (cf. Welch & Piekkari, 2017), we then compared and contrasted our efforts and progressively adjusted the coding rules until we achieved 90-percent comparability, determined by cross-checking the codes each of us had assigned for the same interview. Thereafter, we focused on one company each, discussing and agreeing collectively on all unclear instances to ensure reliability and equivalence (Hult et al., 2008). All experienced instances of facilitation were recorded in an Excel file together with the respective citation and information on the interviewee and the boundary spanner, including their respective locations. Mentions by in-group (i.e., people located on the same side of the boundary) and out-group (i.e., people located on the other side of the boundary) members were coded separately, and self-references were excluded. Tables 2 (data coding examples) and 3 (definitions and operationalizations) provide more detail on the coding scheme for experienced facilitation.
Next, we sought to identify potential differences in terms of the extent to which individuals were recognized by others. We found cluster analysis to be an ideal method to discern such differences in an inductive yet quantitative manner (Hair, Black, Babin, & Anderson, 2008). For this, we used the number of instances in which our interviewees mentioned someone (else than themselves) engaging in the above behaviours as a count measure of experienced facilitation. We examined both the breadth and the depth of this evidence. Breadth was operationalized as the number of in-and out-group members who mentioned a specific individual as engaging in any of the above boundaryspanning behaviours, while depth was operationalized as the total number of times the individual was mentioned by all in-and out-group members, respectively. Accounting for these two distinct dimensions of recognition together grew from an insight of our preliminary qualitative analysis: If an individual gets a large number of mentions from only one or a few persons, these may be biased by personal, political or other factors beyond actual facilitation. We will discuss such a case below. Similarly, relying solely on the total number of people mentioning may induce another bias; for example, a low number of passing mentions by a large number of people may be driven by social desirability bias, halo effects, or name-dropping rather than genuine recognition of boundary spanning (Guenther, 2009;Palmer & Loveland, 2008). Accounting for both dimensions mitigates the risk of both biases, substantially increasing the validity of our findings, and constitutes an important methodological learning element for future work in this area.
To ensure comparability across the four companies, we then converted the arithmetic counts into relative measures (see Table 3). As a result, each individual who was mentioned by others was given a percentage score for in-group breadth (percentage of in-group members who experienced this individual at least once to have facilitated inter-unit relations), out-group breadth (percentage of out-group members), ingroup depth (percentage of the total number of in-group mentions), and Interviewee recognizes Anton to actively engage with people at the Russian unit to ensure that the relations between the HQ and the Russian unit run smoothly Coded as One piece of evidence of 'actively intervening to resolve conflicts and solve problems' recorded for Anton out-group depth (percentage of the total number of out-group mentions). Using these relative percentage-based measures, we then used hierarchical cluster analysis to identify potential sub-groups among the individuals who had been recognized by others (Hair et al., 2008). Deploying a four-cluster model, which fitted with the four measurement dimensions used in the analysis, we found that potential boundary spanners clustered in the following subgroups that differed significantly in terms of whether their boundary spanning was recognized and by whom.
Firstly, in a noteworthy departure from research equating people in boundary-crossing positions with boundary spanners, 38 of the 118 individuals involved in the focal inter-unit interactions (32%) were not recognized at all by others to facilitate inter-group relations. The remaining 80 individuals (68%), who fit the definition of recognized boundary spanners in that they were experienced by others to facilitate inter-group relations, clustered in three separate subgroups. Cluster 1, here termed weak boundary spanners, included 66 individuals who were recognized to engage in facilitation, but not to a statistically significant extent. Cluster 2, strong but unilateral boundary spanners, consisted of 8 individuals who were recognized to facilitate inter-group relations to a statistically significant extent, but only by members of their own in-group. Finally, Cluster 3 or strong and bilateral boundary spanners consisted of 6 individuals who were recognized by both inand out-group members to facilitate inter-group relations to a statistically significant extent. These subgroups are illustrated in Fig. 2 and incorporated into the revised conceptual framework presented in Fig. 3. A final cluster consisting of one person mentioned only by the outgroup was omitted from further analysis as it was driven by a high number of mentions by just one interviewee, an ex-subordinate of the boundary spanner. While recognition only by out-group members is theoretically possible, the fact that we only observed one such outlier suggests that boundary spanners recognized by out-groups also tend to be recognized by their own in-group. In practice, lack of in-group clout is likely to limit boundary spanning ability.

6.
Step 2 analysis and findings: explaining recognized boundary spanning The central question we sought to address in Step 2 was why some individuals were more recognized than others, specifically why 14 (8 + 6) potential boundary spanners were recognized by others by a statistically significant extent of mentions, whereas 66 were recognized but not to a statistically significant extent, and 38 were not recognized at all. Table 4 presents detailed characteristics of the 8 individuals in Cluster 2 (strong but unilaterally recognized boundary spanners) and the 6 individuals in Cluster 3 (strong and bilaterally recognized boundary spanners). Fig. 4 gives an overview of the findings detailed in the next section.

Hierarchical position and structural and relational embeddedness
We began Step 2 with an initial quantitative and deductive examination of individual-level characteristics identified in prior research to be associated with boundary spanning, namely expatriate job role and hierarchical position (for operationalizations, see Table 3). We also examined whether there were differences between the four firms, or geographical/cultural contexts (Finland-China vs. Finland-Russia). We tested the explanatory power of these variables by using the Fisher-Freeman-Halton non-parametric test of significance for differences between groups, which compares the expected and observed values of two cross-tabulated categorical variables. This test was deemed particularly appropriate for our long-tail patterned data, as it does not carry any assumptions with regard to normal distribution or group sizes.
The analyses uncovered no statistically significant differences between firms (Chi-Sq. 6.83; Fisher's Exact = .41) or locations / cultural contexts (Russia vs. China Chi-Sq. 1.93; Fisher's Exact = .41). We return to the issue of culture in the Discussion. There was also no evidence that individuals were more likely to be expatriates in some cluster(s) than in others (Chi-Sq. 1.04; Fisher's Exact = .41). The effect of hierarchical position differed across the clusters to a statistically significant extent (Chi-Sq. 16.31; Fisher's Exact = .002**). Post hoc analyses showed that Cluster 3 differed significantly from the others (Fisher's Exact = .001***), while Clusters 1 and 2 did not (Fisher's Exact = .37). Qualitative examination suggested that CEOs/unit heads were more prevalent in Cluster 3, whereas middle managers (function or department heads) were more evenly spread across the three groups. However, we also observed that not everyone in Clusters 2 and 3 was in a high hierarchical position and there were high-level managers in Cluster 1.
As these quantitative analyses did not explain our clusters fully, we followed them up by qualitatively and inductively re-examining our interviews with the significantly recognized boundary spanners in Clusters 2 and 3. This analysis included all references to these individuals by others on both sides of the boundary, including those that had not been coded as facilitation in Step 1 but shed light on the personal characteristics or circumstances of the focal individuals. We read, discussed, and re-read these references in several rounds to gain more insight into potential explanations. From this qualitative analysis, Table 3 Definitions and operationalizations.

Construct Dimension Definition Operationalization
Boundary spanning Depth of recognition The number of times an individual's boundary-spanning behaviour was mentioned in the interviews by in-group and out-group members. boundary spanners' connections and the quality of their relationships emerged as two broad themes, leading us to abductively infer that the distinction between structural and relational embeddedness (Moran, 2005;Nahapiet & Ghoshal, 1998) might explain important variance in facilitation by boundary spanners. Moran (2005), following Nahapiet and Ghoshal's (1998) view of structural embeddedness as impersonal linkages between people or units, and relational embeddedness as relationships between specific persons based on prior interactions, empirically found both types of embeddedness to exert a positive influence on managerial performance in a MNC context. Using similar procedures as in Step 1, but now focusing on boundary spanners' structural and relational embeddedness, we deductively coded the data following Moran's (2005) framework. Structural embeddedness (operationalized as connectedness) was coded when an individual was mentioned as having connections with colleagues on the other side of the boundary. Relational embeddedness (operationalized as relationship quality) was coded when an individual was mentioned as having high-quality relationships with colleagues on the other side of the boundary (Moran, 2005; see also Nahapiet & Ghoshal, 1998). We then used the Kruskal-Wallis non-parametric tests of mean rank group differences to assess the importance of embeddedness. We found statistically significant differences between the three clusters for both types of embeddedness (structural: Chi-square 14.593; p = 001***; relational: Chi-square 12.509; p = .002**). Post hoc tests showed that it was again Cluster 3 that deviated significantly from the two others. Then, to evaluate the relative importance of different levels of embeddedness, we categorized the embeddedness scores as high (H; > one standard deviation above the mean in the respective MNC), medium (M; within one standard deviation of the mean), and low (L; < one standard deviation below the mean), and added them into Table 4.

Fig. 2. Three clusters of recognized boundary spanners.
Notes. Breadth is denoted on the x (out-group breadth) and y (in-group breadth) axes; the size of the bubble denotes combined in/out-group depth (value included in the label, together with the company of the individual); the three large dashed circles illustrate our clusters, and full details of individuals in Clusters 2 and 3 are provided in Table IV. These analyses highlighted that hierarchical position and structural and relational embeddedness contributed to recognition. Nevertheless, we still found it surprising and interesting that not all high-level managers were recognized, that some of those who were significantly recognized were not high-level managers, that some strongly embedded individuals were not recognized, and that some recognized individuals were not highly embedded (see Table 4). To better understand the mechanisms and possible contextual factors underlying our statistical findings, we returned to our in-depth qualitative data. This part of the analysis is particularly illustrative of the advantages of our dynamic and largely abductive, integrative use of mixed methods. It was the "interplay of [the] theoretical and empirical elements" (Turnarosa & Glynn, 2017: 224)not just the interplay of "concepts, conjectures and data" (Van Maanen et al., 2007, p. 1146) but also the combination both qualitative and quantitative reasoning strategies (Ketokivi & Mantere, 2010;Mantere & Ketokivi, 2013) that led us from the insights above to the ones below. By combining our quantitative analyses with and qualitative insights arising from the narrative interviews, we gradually reached the interpretation that the importance of hierarchical position and embeddedness suggested by the statistical analyses was not important per se. Rather, the underlying mechanism was the organizational knowledge that was often associated with hierarchical position and embeddedness. This knowledge made it easier for high-level managers and highly embedded individuals to engage in behaviours that actually were helpful and beneficial to others (as opposed to behaviours that were intended to be helpful but actually weren't, behaviours that were just interaction, or behaviours the intentions of which were selfish or one-sided), resulting in positive experiences of facilitation. We will now elaborate on the qualitative evidence for this, focusing first on hierarchical position and then on embeddedness.

From hierarchical position to organizational knowledge and scope of influence
The following quote illustrates how some individuals were more regularly exposed to and thus had better access to organizational knowledge thanks to their position. Mr. Y, one of our strong and bilateral boundary spanners, played a key role in his company's expansion and investment into Russia: "Once in a month there is what is called Russia Management meeting when Mr. Y and a few other key people come over here [to Russia] and we go through the whole business, function by function, present the results and discuss the next steps." (A Russian interviewee about Mr Y, a high-level manager in Finland) Because of such visits and the exposure to organizational knowledge gained from them, Mr. Y developed a deep understanding of the overall situation in the Russian subsidiary, its challenges and prospects. This was recognized by several respondents. This knowledge and understanding of the local business realities in Russia enabled him to intervene and facilitate trust-related issues between the Finnish and Russian sides: "Mr. Y [insisted] that all Finnish persons based in Russia have advisory roles. Their task is to support the Russian management in reaching both operational and strategic goals. This kind of clarification really helped [in resolving the trust issues between the Russian and Finnish managers] … Since then I think everything has been settled." (A Russian interviewee about Mr. Y's role in solving inter-unit conflicts)  Due to such exposure, higher-level managers such as Mr. Y were more likely to know better than lower-level employees how their organizations worked and who did what on either side of the boundary. Higher-level managers were also more knowledgeable about and motivated to achieve the goals and address the challenges of both the organization as a whole and its different constituent parts. This enabled them to engage in actions and behaviours that others experienced as facilitation.
Managers in senior hierarchical positions also held a greater scope of influence, meaning that they were able and motivated to create a broader impact. The following quote describes how the hierarchical position of a Finnish top manager in a Chinese subsidiary, allowed him to be more effective than a Chinese colleague in promoting this colleague's initiative across the boundary.
"Mr. K has taken quite an active role in [promoting our initiative] to the corporate office and to other European colleagues… he has promoted the initiative very much certainly, because he [is a high-level manager] … [promoting it] is very difficult for me, even though it is my initiative… to be honest, I won't be recognized." (A Chinese interviewee about Mr. K, his/her Finnish superior at the Chinese unit) This facilitation was also recognized on the Finnish side of the boundary: "Since Mr. K promotes the [initiative]… our trust in the ability of the Chinese to run projects has increased significantly…. it is clear that the Chinese can manage projects much better than us." (A Finnish interviewee about Mr. K, his/her Finnish colleague at the Chinese unit) At the same time, hierarchical position in itself did not guarantee positive impact. Some high-level managers were not just not recognized to facilitate inter-unit relations, but their involvement was, in fact, experienced by others as negative. For instance, Mr. N, a top manager from the Finnish HQ, who from the very start of the company's operations in Russia was actively and formally involved in coordinating the relationship between the units, was indeed very influential by being one of the key connecting nodes between these. His influence was acknowledged by one Russian colleague, who stated that "in fact, the whole perception or opinion [in the Finnish HQ] about what is the management style here [in Russia] is created by Mr. N". Yet, the problem was that despite his formal position and good intentions, Mr. N was not recognized at the Russian side to facilitate inter-unit relations, because he did not have the organizational knowledge to do things that were actually helpful. Hence, hierarchical position enabled such managers as Mr. Y and Mr. K, but not Mr. N, to acquire pivotal organizational knowledge and have a scope of influence that allowed them to do things that were beneficial to many others. This is important, because even if the intentions of an individual are good, if s/he takes actions that are not actually helpful to others, they will not be experienced positively. Similarly, even if an individual engages in facilitation, s/he will contribute less to inter-group relations, the narrower the group of people who benefit.
From embeddedness to organizational knowledge and scope of influence Structural and relational embeddedness had a similar dynamic as above. Regardless of hierarchical position, structurally and relationally embedded individuals had a better understanding of the goals and needs of the other unit, and the roles and situations of people on the other side, making it easier for them to engage in actions experienced as facilitation. For example, the following quote by a Chinese interviewee refers to the deep structural embeddedness of a Finnish colleague, Mr. C, who had been working with the local Chinese organization for many years: "Mr. C has been working in China for many years and he knows a lot [about our organization here] and plays an important role here [in the Chinese unit]." (A Chinese interviewee about Mr. C, a Finnish colleague) Another Chinese colleague remarked that this embeddedness in the local organization in China allowed Mr. C to understand the local situation and the local organization well: "Mr. C is one of the few persons from Europe who understand… China. Compared with others, he is much closer to the Chinese organization with his experience." (A Chinese interviewee about Mr. C, a Finnish colleague) Referring to another strong and bilateral boundary spanner in our sample, a Russian interviewee describes how he started working on an important investment project in Russia together with a Finnish colleague from HQ. Their relationship become increasingly strong and trustful over time, and through this relational embeddedness, the Finnish colleague became very knowledgeable about the Russian unit: "We started doing [this investment project] with [the Finnish colleague] … I think it has worked in a perfect way. I think he understands [the Russian unit], because he has worked here." (A Russian interviewee about a colleague from the Finnish HQ) These trustful relationships enabled him to expand his scope of influence in helping the local organization "to avoid many problems" in the HQ-subsidiary interactions. This facilitation was recognized by others: "So this [person] has our respect and we [always] considered him as a member of our management team. And I always got support… from [his] side. And I am very thankful to [him] for this, because when we had some difficulties… I asked [him] to help me… his interference helped us to avoid many problems." (A Russian interviewee about her colleague from the Finnish HQ) Another strong and bilateral boundary spanner, Ms. S was not in a high hierarchical position, but deep structural and relational embeddednessdriven by her knowledge of the subsidiary host country language and culturehad enabled her to develop a thorough organizational knowledge of the subsidiary's business. She also had a broad scope of influence as she was acknowledged to act as a glue between the key Russian and Finnish decision makers. This enabled her to engage in behaviours that both HQ and subsidiary colleagues experienced as increasingly important for the interunit relationship: Like hierarchical position, structural and relational embeddedness provided individuals with deeper organizational knowledge and a broader scope of influence. More and better linkages across a boundary enabled them to engage in facilitation that was experienced positively by a larger number of others. A less embedded individual may also engage in facilitation, but it will contribute less to overall intergroup relations if only a limited number of others experience it. And given the fundamentally social nature of inter-unit relations, if others don't experience positive impactor have negative experiencesboundary spanners' actions will matter little. In fact, it is reasonable to ask whether any boundary spanning has taken place if relevant others either do not experience it or experience it negatively. On a related note, it is possible and even likely that interpersonal affect (Lefkowitz, 2000) may bias recognition in such a way that the actions of close connections are experienced more positively than those of others. However, given that we have interviewed all individuals involved in the focal interactions, the completeness of the picture should mitigate the effect of any such biases.

Internal interaction structure
The final insight in our quest to understand why some individuals were more recognized than others arose from the following two interrelated observations, again enabled by the interplay of qualitative and quantitative analysis. First, in qualitatively reflecting on the results of our quantitative analyses thus far, we noted that the only individual not from MNC3 or MNC4 in Cluster 3 (Manager 6 from MNC2, see Table 4), did not fit with the explanations above. He had a high-level hierarchical position and was structurally somewhat embedded on the other side, but there was no evidence of him being relationally embedded in the sense of having high-quality relationships with outgroup members. In fact, many interviewees talked about his interactions in a less positive manner (these instances were recorded, but not coded as facilitation). To understand why he still was the only person from MNC2 to be recognized on both sides, we went back to our qualitative interviews and holistic understanding of the focal interaction contexts, and realized that this pattern was driven by the overall structure of the interaction between the two units in MNC2. This structure was such that all important interactions were channelled through Manager 6, which meant that he was the only one who could be recognized, even if there was evidence that as a person he was not creating much positive value. In other words, the internal interaction structure both allowed Manager 6 to be recognized, and prohibited it for all other individuals involved in inter-unit work in MNC2. This pattern is depicted in Fig. 5 (Image 1).
This observation then alerted us to the point that neither firm nor location had been statistically significant in our Fisher-Freeman-Halton test for differences between groups, yet all but one individual in Cluster 3 were from MNC3 and MNC4. Combining these insights, we realized that the internal interaction structure influenced recognition patterns in all four companies, but in different ways. Unlike MNC2 (and MNC1, see below), the internal interaction structures of MNC3 and MNC4 exhibited a multitude of overlapping ties: many people on both sides were interacting across the boundary, at and across different hierarchical levels and functions (see Fig. 5, Image 2). In MNC3, both HQ and subsidiary top management believed that all managers "with their own areas of responsibility should have their own [cross-border] networks" to support effective cross-border interaction. In MNC4, multiple crossborder ties between multiple actors had developed over time following the initiative of top management, as the following quote illustrates: "In the beginning, I was of course involved in many operational issues [with the Finnish HQ] and there was a lot of kind of daily communication, but then when we built this organization and I was able to delegate down to my managers, there was no need for me to participate in this anymore." (General Manager of the Russian unit) Consequently, in MNC3 and MNC4, many people were exposed to and therefore able to experience the facilitation that took place. We believe that this was the primary reason why most people in Clusters 2 and 3 (9 of 14), and in particular most of those in Cluster 3 (5 of 6), were from MNC3 and MNC4.
In MNC1, by contrast, interactions took place in functional silos. There were no boundary spanners from this firm in Cluster 3. Two out of the three MNC1 boundary spanners in Cluster 2 were employed in the same function (a competence centre), and while in-group members had many experiences of their boundary spanning, they were not significantly recognized in the out-group. This seemed to be driven by a silo effect in that most inter-unit interactions took place between people within the same function and/or on the same hierarchical level, which restricted the extent to which any facilitation could be recognized on the other side. This is how one of the top managers explains the structure of cross-border interactions in the company (see Fig. 5, image 3): "We have internally divided our roles specifically so that, for instance, Mr. P, Ms. Q, and Mr. R have their own sort of functional divisions where they act as the key contact persons [for Chinese managers]." (A top manager at the Finnish HQ) We validated these qualitative findings quantitatively by comparing the number of interviews in the four companies (MNC1 = 44; MNC2 = 37; MNC3 = 31; MNC4 = 33) to the total number of boundary spanners identified in these interviews (MNC1 = 10; MNC2 = 15; MNC3 = 26; MNC4 = 29). Logically, a larger number of interviews should yield a larger number of identified boundary spanners as different interviewees will likely have different experiences, but we found the reverse. MNC1 and MNC2, in which we had conducted more interviews, had clearly fewer boundary spanners than MNC3 and MNC4 (see Fig. 6 below)a pattern fully consistent with our qualitative understanding of the focal interaction contexts.
In sum, our findings suggest that the inter-unit interaction structure exerts a substantial prior-order influence on whether, and which, boundary spanners can be recognized, forming pipelines in which individual-level boundary spanning can take place. The individual-level mechanisms we identifiedorganizational knowledge and scope of influence, enabled by hierarchical position and embeddednessonly come into play later, explaining relative differences within the same interaction context.

Discussion
Our aim in this paper was to explain recognized boundary spanning by addressing the research question of which individuals in a given interaction context are experienced by relevant others to facilitate intergroup relations, and why. Rather than focusing on the characteristics, skills or motivations of these individuals, we have taken a view of boundary spanning that is analogous to customer-, user-or employeecentred approaches in other fields (Beyer & Holtzblatt, 1998;Bowen & Ostroff, 2004;Norman & Draper, 1986;Shah, Rust, Parasuraman, Staelin, & Day, 2006) by highlighting the actual lived experience of relevant others. In presenting evidence of extensive variation in others' recognition of boundary spanners, and of how this recognition is influenced by contextual features, our approach addresses a significant lacuna of existing research on boundary spanning in IB. Since people's experiences have a strong impact on their subsequent actions (e.g., Bowen & Ostroff, 2004), this variation is likely to influence the success of boundary spanning efforts and thereby influence the quality of MNC interunit relationships in powerful yet hitherto largely ignored ways. by our study is that strongly recognized boundary spanners are rare birds indeed. Most of the 118 potential boundary spanners in our data were either not recognized at all (n = 38) or only weakly recognized (n = 66) by relevant others on either side of the boundary. Only 14 individuals (12% of 118) were recognized as strong boundary spanners, and only 6 of these (5% of 118) were recognized bilaterally, beyond their own in-group.
These findings indicate the presence of a power distribution, and they alone have important ramifications for future work on boundary spanners, lending empirical support to the literature questioning the assumption that everyone engaged in inter-group contact fosters intergroup relations with similar effectiveness (Kane & Levina, 2017;Levina & Vaast, 2005;Schotter & Beamish, 2011). Our findings add that boundary spanners cannot be recognized solely based on their individual-level characteristics, and neither it is their intentions that make a differencebut rather whether their actions are experienced by others as facilitation: Do they engage in behaviors that are actually recognized as helpful and beneficial on both sides of the boundary? Not all behaviors are: even those intended by the actor to be helpful can be experienced by others as simply interaction, one-sided, or even downright selfish or harmful. The last is clearly not boundary spanning, but a recognition perspective enables us to make a distinction between intentions of the focal individuals and experiences of others, and between the experiences of in-and out-group members.
Indeed, in-and out-group members live in partially different realities. As the findings of Step 1 of our analysis show, over half of the 14 strong boundary spanners identified by our respondents were recognized only by their own ingroup. This is logical given that in-group members interact more frequently with their own colleagues, but only very few boundary spanners are able to reach out of their own 'bubble'. This structural effect can be further exacerbated by a tendency to view the actions of one's in-group members in a more positive light than is warranted, and more positively than the actions of out-group members (Mael & Ashforth, 1992;Tajfel & Turner, 1986).

Drivers and mechanisms of recognition
Our second contribution is to anchor boundary spanning in its context, underlining that context matters to a greater extent than previously recognized. Our findings complement recent research on the cultural context of boundary spanning (Søderberg & Romani, 2017) by highlighting the role of another dimension of context, the organizational one. We trace how the opportunities for recognized boundary spanning are shaped by the internal interaction structure of MNCs, and how the recognition of boundary spanners is enabled by their organizational knowledge and scope of influence. This detailed view of the organizational underpinnings of recognized boundary spanning shifts the focus away from individual-level characteristics and onto organizational solutions such as structures and means of organizing; this is important, as organizational structure and processes may at times be easier to influence than personal characteristics and competencies of individuals. The inter-unit interaction structure exerts an important influence on who will be recognized as a boundary spanner in the first place, and our findings suggest that physical distance between HQ and subsidiaries can successfully be transcended by organizational means. The cost of doing this in terms of time and money naturally increases with distance, but distance does not per se seem to constitute an obstacle to successful boundary spanning. Inversely, our results from MNC1 and MNC2 indicate that organizational structure can seriously constrain employees' ability to act as boundary spanners.
Given that the subsidiary units of MNC3 and MNC4 are located in St. Petersburg, which is physically near the Finnish border, we recognize that this aspect of our findings may to some extent be driven by the shorter physical distance between Finland and Russia compared to China. Geographic proximity may influence the ease and to some extent the cost of physical travel between HQ and subsidiary. Time zone differences might influence the frequency of synchronous mediated communication such as phone conversations and video meetings, and the timeliness of responses to asynchronous mediated communications such as email messages. However, Russia and Finland have been shown to be culturally and historically far removed as a result of bloody wars in 1939-1940 and 1941-1944 that have had a significant impact on the attitudes and cultural stereotypes of both nations in relation to each other (Koveshnikov, Vaara, & Ehrnrooth, 2016). We therefore do not think that physical proximity is enough to explain our results, an interpretation that is buttressed by our statistical analyses which uncovered no significant differences between locations. Accordingly, we strongly believe that a larger number of boundary spanners would have been identified in MNC1 and MNC2 had they been organized like MNC3 or MNC4.
Based on our detailed understanding of the four firms, we believe that their respective interaction contexts are largely of an emergent nature, rather than strategically planned or even realized/understood from an interaction perspective. Neither our data nor the scope of this paper allow us to discuss the reasons behind the existing setups, highlighting organization structure and especially its potentially unintended consequences on boundary spanning as a highly interesting question for further research. Future empirical work should also include objective measures of physical trips and mediated communications undertaken by potential boundary spanners to test if and to what extent these factors impact recognition in ways above and beyond the interaction structure.

Methodological contribution: dynamic integrated mixed method analysis
We also contribute methodologically by introducing a dynamic integrated mixed method approach. Mixed methods have several advantages, including stronger triangulated evidence for conclusions and the generation of insights that might have been missed using a single method (Gibson, 2017;Johnson & Onwuegbuzie, 2004;Molina-Azorin et al., 2017). We add to these by advancing a largely explorative use of a dynamic conversion mixed methods design (Teddlie & Tashakkori, 2006), strongly integrated in the analytical dimension by "using the results of one analysis to approach or inform …the other" (Fetters & Molina-Azorin, 2017, p. 301). The power of this pragmatic approach (Feilzer, 2010;Johnson, Onwuegbusie & Turner, 2007;Morgan, 2007), the lies in leveraging and balancing the strengths of each method and related reasoning strategies (Ketokivi & Mantere, 2010;cf. Welch & Piekkari, 2017) in the context of an emergent, explorative and largely abductive research process (Van Maanen et al., 2007). We find it unlikely that our results could have been reached by either qualitative or quantitative method alone.
Relatedly, our findings suggest that there is clearly a need for caution in the measurement of boundary spanning, given the potentially strong effect on the results depending on who is asked. We argue that future work on boundary spannerswhether oriented toward research or practiceshould move away from focusing on the boundary spanners themselves, to examining facilitation from the recipients' perspective, on both sides of the boundary. As an alternative to querying in-and outgroups directly, boundary spanning could also be empirically captured through the observation of outcomes.

Limitations
Our research design prioritizes breadth and analytical scope above depth in either a qualitative sense (e.g. through more extensive contextualization; Ketokivi & Mantere, 2010;Welch & Piekkari, 2017) or a quantitative sense (e.g. through more extensive validation of constructs; Bono & McNamara, 2011). While recognizing this limitation, we argue that this tradeoff is nevertheless worthwhile and has generated insights not enabled by either method alone, or even by a more traditional triangulation procedure.
We also do not claim to have captured all aspects of boundary spanning. Our focus here is on one type of boundary (inter-unit boundaries), but in many cases, multiple boundaries and organizational identities will coincide and interact with each other in ways we have not been able to examine. Also, it is unclear if similar dynamics apply to two units at the same hierarchical level, in inter-organizational interactions, or in interactions in other types of organizations than MNCs (although there are indications that this may be so; see e.g., Johnson & Duxbury, 2010, on boundary spanning in the context of Canadian diplomats' work). Relatedly, we cannot be certain that our recognition approach fully captures all the boundary spanning that is going on. Some may occur inadvertently or behind the scenes, or inter-unit relations may have improved without the interaction partners being aware of it, given that negative effects tend to attract more attention than positive ones (Stahl & Tung, 2015). These limitations pose interesting challenges for future work.
The context of our study is inter-unit relationships between HQs in Finland and subsidiaries in Russia and China, and we have no data from other geographical and cultural contexts. We acknowledge that national culture may influence who gets recognized as a boundary spanner. For example, culture may bias recognition of 'true' boundary spanning, so that individuals from high power distance cultures may be more likely to mention high-level managers or tone down the role of individuals relative to groups. Cultural effects may also amplify or weaken the drivers and mechanisms of recognition; for example, the effect of structural and relational embeddedness may be stronger in collectivist cultures, while facilitation may be less visible in cultures that value silence. Also, although culture is unlikely to influence the significant variation we find within each in-or out-group, or within one and the same interaction context (we emphasize again that we did not find significant differences between China and Russia), we cannot exclude indirect effects. For example, culture-specific preferences that could otherwise have been more visible in the data may have been weakened by Western influence in our case subsidiaries, which were foreign-owned units with Westernized HR practices, or by the international experience of many of our Russian and Chinese respondents.
We have approached national culture as a methodological issue rather than a potential explanatory factor for several reasons. First, HQs were Finnish in all four cases, so possible cultural variation in our data could logically only stem from systematic differences between Russian and Chinese interviews. These constitute less than two-thirds of all interviews, which already in itself reduces the possible impact of cultural differences on the results. Second, our data do not reflect the considerable regional variation in either Russia or China. Third, not all of our interviewees could unequivocally be described as 'Russian' or 'Chinese'; several were expatriates (although we did not find this to be a significant driver of recognition), and we did not have data on biculturalism of varying degrees. We sought to lower the risk of cultural bias by conducting all interviews in the preferred language of the interviewees, and by researchers who were themselves members of the same culture, but could not include a comprehensive cross-cultural analysis in the scope of the paper. This is an important future avenue for research.

Managerial relevance
Assuming that recognized boundary spanning is, indeed, as important for MNCs as we have argued, a natural follow-up question is how it can be promoted in practice. Formal coordination mechanisms, such as steering groups and liaison roles, are unlikely to meet their goals just by virtue of their existence. Our results indicate that better inter-unit collaboration will ensure if decision makers can make groups more visible to each other through boundary spanning or other means; communicate that boundary spanning is important, expected and respected in the organization; and acknowledge and appreciate strong boundary spanners publicly. Our key message on the importance of 'others' in identifying recognized boundary spanners has direct practical implications for human resource management practices: MNCs need to know who their strong recognized boundary spanners are, and for this information, collegial input is crucial. We believe that a HR methodology already used by most major firms -360-degree feedbackmay be a suitable vehicle for this. The 360-degree evaluation methodology already captures feedback from different levels of the organization (superiors, peers, subordinates), and we would add that it is important to make sure that they also draw feedback from people in other units than just one's own, to capture the experiences of 'relevant others' across key boundaries.
In terms of how to increase the number of recognized boundary spanners, we note that extensive organizational knowledge and scope of influence are preconditions for boundary spanning. We are not able to say whether one of these characteristics predates the other, but familiarity with the organization is likely to be a good first step. One of the ways in which HRM practices such as job rotation, different forms of visits, and collaborative teams and projects exert their well-documented integrative influence on MNCs is that they help create a pool of individuals with the organizational knowledge needed to act as boundary spanners. We also note that two of our case firms -MNC3 and MNC4both had many boundary spanners, driven by an internal interaction structure characterized by rich and multiplex ties. The internal interaction structures of these firms had developed differently and over different periods of time: MNC3 had expanded in Russia through acquisitions over a period of more than a decade, whereas the Russian subsidiary of MNC4 was a greenfield operation established at record pace. Still, we believe that the actions of these firms in terms of structuring the organization, systematically transferring individuals between units, and creating networking opportunities, offer worthwhile learning points for others. Our findings do not negate the importance of personality characteristics and personal orientations, but rather broaden the managerial toolbox for finding and developing boundary spanners via increasing organizational knowledge and through different means of organising. These may often be easier to change than relatively fixed personal characteristics. In addition to organizational-level means of creating collaborative structures, firms should also look after their strong (and particularly bilaterally recognized) boundary spanners, and strive to develop more of them. All inter-group interaction is not boundary spanning, and only some boundary spanners were highly recognized. These people are disproportionally important for internal coordination, and firms and their leaders would do well to be sensitive to who they are: they should be acknowledged and rewarded in, for example, talent-management programs. Many of them are already top managers, but importantly, not all top managers were significantly recognized as boundary spanners by both in-and out-group members. The same seems to apply to expatriates: They are not recognized as boundary spanners simply by virtue of their position, but some of them may be through their organizational knowledge and scope of influence, enabled by a combination of position and personal embeddedness. Finally, it may be that there are situations in which organizations are better off with a smaller number of boundary spanners (such as organizations that deal with strictly confidential data); the reverse of our findings points to means such as the interaction structure that could be used in these situations.

Conclusion
From the above, we draw the following three conclusions that underline the significance of our inquiry and highlights its implications within IB as well as beyond.
Firstly, our study does negate the importance of prior work on boundary spanning as driven by formal roles or individual-level characteristics, but complements it by drawing attention to how boundary spanning efforts are experienced by relevant others. Bilaterally recognized boundary spanning builds on actions and behaviours that are experienced to contribute towards mutual benefit and constructive development of the relationship between two parties. This is crucial to ensure working interunit relations in potentially fragmented organizations such as MNCs (Carlile, 2002(Carlile, ,2004Kristensen & Zeitlin, 2001;Morgan & Kristensen, 2006;Tortoriello et al., 2012). Yet while an individual and their fellow in-group members may view particular actions as fostering inter-group relations, outgroup members may experience merely contact, or even self-interested pursuit of conflicting goals. Then even the best intentions on behalf of the boundary spanner will not facilitate inter-group relations and may even influence them negatively, as indeed could be seen also in our data. We believe that to be recognized on both sideswhether HQ and subsidiary, or two different geographical locationsboundary spanning should not have a 'direction', but should genuinely aim for mutual benefit. This suggests the importance of perspective-taking, or 'actively contemplating others' psychological experiences' (Todd, Galinsky, & Bodenhausen, 2012, p. 95) as a relevant aspect of future research on boundary spanning from a recognition perspective. Likewise, and relatedly, a better understanding of how to meld the partially different realities of in-and out-group members is needed. Inquiries into these topics seem relevant also beyond the corporate sphere.
Secondly, and pertaining to the longstanding question of to what extent boundary spanning can be formalized (Schotter & Beamish, 2011), our results suggest that it is still amenable to hierarchy and structure, albeit at the organizational level. That levels of boundary spanning can be influenced by interunit interaction structure is no panacea, but good news as a first step for managers looking to increase levels of boundary spanning in their organizations, andwith a nod to one of the classic debates within both management studies and IB, namely that of strategy versus structure (e.g., Chandler, 1962;Egelhoff, 1982)reminds us that even in this time of porous MNC boundaries (e.g, Coviello, Kano, & Liesch, 2017), structure still matters.
Last but not least, this study hints at the potential of eschewing traditional qualitative-quantitative dichotomies and epistemological partisanship in favour of more flexible and dynamic approaches such as the dynamic integrated mixed method approach (Fetters & Molina-Azorin, 2017;Turnarosa & Glynn, 2017) deployed in this study. Based on our own challenging yet highly stimulating experience of following up on the many 'surprises' (Van Maanen et al., 2007) that emerged in the course of the present inquiry, we strongly encourage future researchers to leverage both the detail and the volume of available data in their quest to advance beyond established wisdom.