2.1 Enterprise Social Platforms, Online Communities, and Knowledge Problem Solving

Enterprise social platforms are increasingly adopted by organizations to support knowledge-intensive work across distributed teams and technical domains (Leonardi et al., 2013). Within these platforms, firms often host a network of in-house online communities—virtual spaces organized around specialized technical areas or business functions (McDermott and Archibald, 2010; Kim et al., 2018; Liu et al., 2025). These communities facilitate domain-specific exchanges but also serve as structural units that can be spanned when complex problems require cross-boundary knowledge sharing.

While enterprise social platforms support localized knowledge search within communities, they also allow knowledge seekers to expand the scope of inquiry. Specifically, enterprise social media and internal online platforms allow knowledge seekers to broadcast their problems across community boundaries, enabling actors from other technical domains to selfselect into discussions (Leonardi et al., 2013; Wu, 2013; Kim, 2023). This functionality enables what prior work refers to as a broadcast search—a mechanism through which problem owners invite responses from anyone who deems themselves qualified to contribute, thereby accessing a broader knowledge base (Jeppesen and Lakhani, 2010; Afuah and Tucci, 2012).

2.2 Technical Marginality of Knowledge Contributors

Broadcast search within an enterprise presents an opportunity for leveraging both proximal and distal knowledge resources. Drawing on the concept of technical marginality (Jeppesen and Lakhani, 2010), we distinguish contributors based on their expertise relative to the focal problem domain. Internal contributors possess knowledge in the same domain as the problem; external contributors come from different technical fields. Prior research suggests that external contributors—by virtue of being removed from the core domain—may introduce novel heuristics and problem framings (Jeppesen and Lakhani, 2010; Afuah and Tucci, 2012). However, while knowledge diversity offers potential value, its benefits are not automatic.

Diversity in perspectives and expertise can hinder knowledge integration, particularly when contributors lack shared interpretive frameworks (Dougherty, 1992; Bechky, 2003; Carlile, 2004; Cronin and Weingart, 2007). These barriers are especially pronounced when contributors draw on different professional vocabularies, norms, or mental models. Therefore, effective knowledge seeker learning requires not just exposure to diverse perspectives but also integration into the seeker’s frame of understanding (Mors et al., 2018; Ritala et al., 2023).

Prior work suggests that absorptive capacity—such as the knowledge seeker’s own experience in the external domain—could mitigate this tension (Cohen and Levinthal, 2000). However, the open-ended nature of broadcast search often exposes the seeker to knowledge that spans multiple unfamiliar domains. In such cases, integration may depend on the presence of actors who can elaborate or translate external contributions into the seeker’s frame of reference. This leads us to consider how the structure of informal relationships among contributors—particularly the density of connections—affects the interplay between knowledge diversity and integration.

2.3 Network Density and Informal Structures

Knowledge contributors participating in broadcast search discussions are embedded in informal networks of relationships established through prior interactions within the platform (Kim et al., 2018). We capture these informal ties using a network approach: if two contributors have previously co-participated in online discussions (even outside the focal thread), we treat them as having a prior tie. The overall density of such ties among contributors is used to characterize the structure of the discussion network.

Within enterprise social platforms, these informal relationships materialize over time as contributors repeatedly engage in problem-solving discussions, forming networks whose structures—sparse or dense—shape the ways in which knowledge is accessed, shared, and integrated. One of the most widely studied properties of such networks is density, referring to the extent to which actors are connected to one another (Borgatti et al., 2013). In our study, sparse networks indicate that most contributors have had limited prior interactions with one another, allowing access to nonredundant knowledge by spanning structural holes (Burt, 1992, 2000). In contrast, dense networks suggest more extensive prior interaction history among participants, facilitating integration through cohesion and shared understanding (Reagans and Mc Evily, 2003).

While prior research often treats sparse and dense structures as mutually exclusive configurations, a more nuanced view suggests that these structures can coexist within subgroups of a larger network. To address this tension, we build on literatures that emphasize both within-group cohesion and cross-group bridging (Reagans and Mc Evily, 2003; Mors et al., 2018), and extend them by decomposing the network of contributors into subgroups based on technical marginality (internal vs. external), and analyzing within- and between-subgroup density (See Figure 1).

3.1 Research Setting and Data

We tested our hypotheses using archival and survey data collected from a global energy company known for its enterprise-wide use of internal social platforms to support technical problem-solving and knowledge sharing. The company maintained an extensive network of online communities, each aligned with a specific technical or operational domain (e.g., chemicals, subsea integrity, defect elimination, geomodeling). These communities were intended to foster domain-specific exchanges while contributing to broader organizational learning. At the time of data collection, more than 10,000 employees were registered across 100+ communities. Employees, who were globally distributed across business streams, belonged to an average of 2.36 communities (std = 2), with strong alignment between community membership, work domain, and knowledge expertise.

Each community had its own discussion forum, where employees could post questions and technical problems. Discussions were non-anonymous and work-focused. While it was technically possible for any employee to post in any community, doing so without membership was rare. The platform also supported cross-linked discussion threads, which allowed a thread initiated in one community to appear on the forums of additional communities. This enabled broader participation by members of different domains. A thread could be cross-linked either by the knowledge seeker or the community leader (typically a senior, experienced manager) of the host community. Knowledge seekers were advised to use cross-linking selectively—only when their problem was expected to be relevant and valuable to experts in other communities. Appendix A shows examples of cross-linked discussion topics.

We collected two types of data. First, we compiled archival data on user profiles (including name, position, business unit, and location) and activity logs (such as community membership and posting behavior). Second, we administered a survey to knowledge seekers who had initiated cross-linked discussions during the observation period. The goal of the survey was to capture the extent to which the seekers experienced learning through the discussion.

The unit of analysis was the cross-community discussion thread. For each thread, we identified the knowledge seeker and the contributors who posted responses. We applied three screening criteria to define our analytical sample. First, we excluded threads with two or fewer replies. While a single reply could provide useful information, our theoretical focus is on how seekers learn through multi-actor interaction and diverse contributions, which short threads do not capture. Second, we excluded threads that involved announcements or simple requests for static content (e.g., reports, templates). Third, we limited the sample to threads initiated within seven months prior to the survey to ensure that seekers could recall the discussion accurately.

The survey was administered in collaboration with the company’s platform manager and community coordinators. Each knowledge seeker received an email invitation with a link to the survey, which was customized to reference the specific thread they had initiated. Seekers were asked to review the full thread and then respond to a short questionnaire assessing their perceived learning and the value of the discussion. Two rounds of reminders were sent to increase the response rate.

In total, we identified 310 eligible threads posted by 255 unique knowledge seekers. From this pool, we received 195 complete survey responses, each linked to one cross-linked discussion thread. These threads involved participants from 34 different host communities and included a total of 1,005 unique contributors.

3.2 Measures

4.1 Descriptive Statistics

The descriptive statistics of all variables are presented in Table 3

Because our model included two latent constructs—knowledge seeker’s learning and expertise level—we assessed construct validity and reliability prior to estimation. All survey items loaded strongly onto their intended constructs (loadings $> .80, p < .001$), and item-to-total correlations ranged from .73 to .81, exceeding the .40 threshold for convergent validity. The square root of the average variance extracted (AVE) for each construct exceeded inter-construct correlations, supporting discriminant validity. Additionally, composite reliability, AVE, and Cronbach’s alpha were all above .70 (Learning: CR = 0.93, AVE = 0.72, $\alpha = 0.90$; Expertise: CR = 0.94, AVE = 0.83, $\alpha = 0.90$).

4.2 Hypothesis Testing

To test the hypotheses, we ran ordinary least squares (OLS) regression with robust standard errors clustered by host community. The clustering accounts for potential correlation among threads initiated in the same community. As shown in Table 4, we built models stepwise: starting with control variables (Model 1), then introducing individual main effects (Models 2–4), all main effects (Model 5), two-way interactions (Models 6–9), and the three-way interaction (Model 10). Network density variables were mean-centered before generating interaction terms (Aiken and West, 1991).

Hypothesis 1 proposed that lower density among internal contributors is positively associated with the knowledge seeker’s learning. As shown in Model 5 of Table 4, the coefficient of reversed internal-internal density is positive and significant ($\beta = .25, p < .05$), supporting H1. This suggests that when internal contributors are less embedded in tightly connected networks—yet still share technical common ground—they are more likely to bring diverse and complementary inputs (Brown and Duguid, 2001; Cronin and Weingart, 2007).

Hypothesis 2 posited that the effect of external contributor density is contingent on the density of ties between internal and external contributors. The main effect of reversed external-external density is not significant (Model 5: $\beta = .20$, n.s.), but its interaction with internal-external density becomes positive and highly significant in Model 9 ($\beta = .28, p < .001$), supporting H2. As shown in Figure 2

Hypothesis 3 argued that learning is maximized when within-group density is low for both internal and external contributors, and between-group density is high. As shown in Model 10 of Table 4, the three-way interaction term is positive and significant ($\beta = 1.17, p < .05$), providing support for H3. Figure 3

Together, the results show that the network structure surrounding a cross-community discussion thread significantly shapes the seeker’s learning. Sparse intra-community ties, when combined with strong inter-community bridges, promote the integration of diverse perspectives. Conversely, diversity without bridging produces little benefit—or even harm—underscoring the importance of connection in realizing the value of distributed knowledge.

4.3 Robustness Checks

Robustness checks were conducted to validate the findings (See Table 5

5.1 Theoretical Implications

Our findings reveal that the knowledge seeker’s learning is greatest when contributors from the same knowledge subgroup—either internal or external—are sparsely connected among themselves, while connections between internal and external contributors are dense. This structural configuration maximizes access to diverse, nonredundant knowledge while supporting its integration into a coherent understanding. The results offer several contributions to research on knowledge sharing, intra-organizational broadcast search, and organizational networks.

First, we extend the technical marginality view (Jeppesen and Lakhani, 2010; Afuah and Tucci, 2012) by showing that access to contributors from distant domains does not automatically lead to greater learning. Prior research has shown that exposure to diverse perspectives can help reframe problems, but only if those perspectives can be interpreted and synthesized (Carlile and Rebentisch, 2003; Reagans and Mc Evily, 2003). Our results provide evidence that pre-existing connections between internal and external contributors—reflecting familiarity or shared context—play a critical role in enabling that synthesis. Thus, the benefit of marginality is not inherent but contingent on the presence of integration mechanisms.

Second, our study complicates the common assumption that internal contributors—those close to the problem domain—offer only redundant knowledge. The literature often characterizes internal contributors as informationally close to the seeker, and therefore less valuable in surfacing novel insights (Jeppesen and Lakhani, 2010). However, our results suggest that internal contributors also vary in how connected they are to one another. When they are sparsely connected, they are more likely to bring diverse, nonredundant perspectives—even if they are technically close to the seeker. This aligns with prior findings that weak ties and structural diversity support novel idea generation (Burt, 1992; Reagans and Zuckerman, 2001).

Third, this study offers a network-based explanation for inconsistent findings in prior research regarding the relationship between contributor diversity and outcomes such as idea usefulness or perceived value. For example, some studies show that diverse contributors generate novel responses but not necessarily more useful ones (Kudaravalli and Faraj, 2008), while others find an inverted-U relationship between contributor diversity and knowledge-related outcomes (Singh et al., 2011). Our results suggest that such inconsistencies may stem from the challenge of integrating diverse knowledge. While sparse networks promote access to novel, nonredundant inputs, dense networks support the coordination and integration of those inputs into a coherent understanding (Reagans and Mc Evily, 2003; Tortoriello and Krackhardt, 2010). In our context, dense internal-external ties appear to function as boundary-spanning mechanisms that enable cognitive integration, thereby enhancing the seeker’s learning.

Fourth, the study contributes to the growing literature on the ecology of online communities. Prior work has shown that shared members—individuals who participate in multiple communities—can facilitate knowledge flows across community boundaries (Kim et al., 2018; Butler and Wang, 2012; Gu et al., 2007). Our findings add nuance by highlighting that such shared members play an especially important role when they serve as bridges between internal and external contributors. These boundary-spanning actors make it possible to blend unconnected knowledge into coherent understandings, enabling intercommunity learning. This mechanism helps explain how enterprise social media platforms can generate cross-domain value even within a single firm.

Finally, this study speaks to broader debates in network research regarding the tension between sparse and dense structures. Whereas sparse networks are associated with idea generation and dense networks with trust and coordination, these benefits are often assumed to be mutually exclusive (Obstfeld, 2005). Our findings show that when network structure is decomposed into subgroups defined by technical marginality, it is possible to realize the benefits of both sparsity and density. That is, sparse ties within subgroups support diversity, and dense ties across subgroups enable integration. This perspective invites future research to consider how multidimensional network segmentation—by technical field, function, or geography—may unlock complementary network benefits.

5.2 Practical Implications

This study underscores that effective learning through broadcast search in enterprise social platforms requires more than broad participation from contributors across the organization. It depends critically on whether the diverse knowledge brought into cross-community discussions can be meaningfully interpreted and integrated. In other words, knowledge seekers benefit not just from exposure to novel perspectives but from the presence of contributors who can bridge those perspectives with the focal problem.

While external contributors—those from different technical domains—are vital for injecting new ideas, our findings demonstrate that internal contributors, whose expertise is closer to the seeker’s domain, play an equally critical role. These internal contributors can help translate and contextualize external inputs, facilitating the integration necessary for learning. Thus, organizations should view internal and external contributors as complementary—not competing—sources of value in collaborative knowledge sharing.

Enterprise social platforms allow members to participate in multiple communities of practice, and this flexibility presents both opportunities and risks. Wang et al. (2013) argue that community managers should consider restricting members from participating too broadly, to ensure that they devote sufficient time and attention to their primary community. This perspective reflects a concern with resource dilution. However, our findings offer a counterpoint: time spent in other communities can build the integrative social ties necessary for effective knowledge recombination across boundaries. Rather than limiting participation, organizations may benefit from encouraging strategic overlap—where members engage across a range of communities, particularly those that span complementary domains.

Recent evidence from corporate online communities confirms that this remains a significant challenge. Liu et al. (2025) report that departmental boundaries continue to constrain knowledge sharing, even on enterprise social platforms designed to enable cross-boundary collaboration. Employees still show a strong inclination toward intradepartmental exchanges. In other words, while digital platforms overcome physical and temporal constraints, organizational divisions continue to fragment online interaction and limit the potential of cross-community learning.

Our study suggests that organizations must go beyond providing technological infrastructure. They need to actively cultivate the social and structural conditions that enable integration across knowledge domains. Network analysis of contributors’ participation histories can help identify employees who are not only exposed to diverse domains but also positioned to facilitate knowledge integration. Community managers should leverage such insights to involve the right mix of internal and external contributors in discussions, especially when progress stalls due to misaligned frames of reference.

Ultimately, the success of broadcast search in enterprise social platforms hinges not just on who contributes, but also on how those contributors are connected. Our findings show that learning is maximized when contributors are sparsely connected within their respective groups—internal or external—but densely connected across those groups. This network configuration provides access to diverse, non-redundant knowledge while also supporting the integration necessary for constructing a coherent understanding of the problem. Building and supporting cross-community ties that enable such synthesis is key to unlocking the full value of distributed expertise. By identifying this synergistic network structure, the study clarifies the conditions under which intra-organizational broadcast search fosters meaningful cognitive learning from distributed expertise.

5.3 Limitations and Directions for Future Research

This study has several limitations that suggest opportunities for future research. First, our findings are situated in the context of cross-community discussions on enterprise social platforms, where contributors can observe each other’s inputs and engage in follow-up dialogue. The mechanisms we identify—access to diverse knowledge and its integration through network connections—may function differently in settings where contributors respond independently or where problem solving involves more routine or localized knowledge. Future research should examine how broadcast search operates under varying organizational, technological, and task-related conditions.

Second, we infer social connections among contributors based on community co-membership and co-participation in threaded discussions. While reasonable in this context—since most contributors posted primarily in communities where they were registered—this approach may miss shared understanding developed outside the platform. Although we controlled for offline communication between seekers and contributors, we could not observe potential offline ties among contributors. Future work should explore how online and offline networks jointly shape knowledge integration, potentially using multi-modal data or ethnographic methods (Zhang and Venkatesh, 2013; Howison et al., 2011).

Third, although we use network density as a proxy for access to diverse (i.e., non-redundant) knowledge and the capacity for integration, we do not directly measure cognitive processes. While prior work supports the link between sparse structures and diversity, and dense structures and integration (Burt, 2000; Reagans and Zuckerman, 2001), stronger evidence could come from analyzing the content of interactions. Future research should combine network analysis with content analysis or natural language processing to assess the richness, novelty, and integration of exchanged ideas.

Fourth, our analysis focused on the network structure of contributors and controlled for seeker attributes such as experience and prior contributions. However, seekers can also shape outcomes—for example, by posing clear problems, replying to others, or synthesizing inputs. Interviewees in our preliminary study emphasized the value of well-articulated questions. Prior research shows that question quality influences interaction (Ahern et al., 1992), and our unreported analyses found that question length and seeker engagement were positively associated with the number of contributors, though not with learning outcomes. Notably, only 16% of seekers engaged in their threads. Future research should examine how seeker behaviors—such as clarifying questions, follow-up engagement, or synthesis efforts—not only influence participation but also interact with contributor network structure to affect knowledge integration and learning.