Tracing the evolutionary leaps and boundaries of digital platforms: a case study of Facebook

ABSTRACT

Digital platforms exert a significant influence on the creation and utilisation of products and services across the globe. Past research has shed light on digital platform establishment, strategies for growing platforms’ scale and scope, as well as the role of boundary resources in stimulating third-party development. However, we lack insight into the interaction of growth in scale and scope as traditional digital solutions are transformed into emergent platforms, and how such platforms over time evolve beyond the platform metaphor. .In this paper, we seek to understand the emergence, evolution, and boundaries of digital platforms over time through a longitudinal case study of Facebook. For this, we utilise Facebook’s publicly available documentation of its evolution as digital trace data. We discover that from 2004 to 2019, Facebook underwent four distinct evolutionary stages: interaction, integration, interconnection, and implantation. These stages were facilitated by two novel roles for digital platform boundary resources that we term as distributing and centring. Based on these findings, we theorize the role of digital artefacts’ material qualities in their evolution from applications, via platforms, to information infrastructures. The paper contributes a model explicating this traversal as well as the subsequent unbundling to a new entity repeating the same traversal.

KEYWORDS:

• Facebook
• digital platforms
• platform evolution
• boundary resources

Introduction

Digital platforms serve as a powerful model of complementary product and service delivery. Digital platforms are argued to stimulate generativity (Yoo et al., 2010), whereby ‘new capabilities unforeseeable by the platform’s original designers’ are created (Tiwana et al., 2010, p. 675). A digital platform provides a foundation for users to develop content, innovate with complementary technologies, and create new services. We define a digital platform as ‘the extensible codebase of a software-based system that provides core functionality shared by the modules that interoperate with it and the interfaces through which they interoperate’ (Tiwana et al., 2010, p. 676). Platforms’ loosely coupled, layered, modular architecture enable firms to innovate simultaneously at different layers of the architecture and compete on related services (Yoo et al., 2010). Digital platforms typically facilitate two-sided markets, allowing content providers and end-users to interact through the platform and enjoy direct and indirect network effects (Ojala & Lyytinen, 2018). Previous scholarship has explored the deliberate process of establishing a digital platform (Eaton et al., 2015; Sandberg et al., 2020; Skog et al., 2021). We also know a great deal about strategies for platform growth (typically referred to as scaling) (see, e.g., Huang et al., 2017; Varga et al., 2023). However, we know less about how growth in the scale and scope of platforms interact, and what unfolds in the subsequent stages after the successful establishment of a digital platform. To investigate these issues, we study Facebook’s evolution, from its inception as a social networking site to expanding into multiple platforms and infrastructures.

Facebook was launched on 4 February 2004, as a social network for Harvard students. Since then, it has grown into a crucial part of the global digital landscape, with close to 3 billion monthly active users. Only six other social platforms have a user base exceeding one-third of this size: YouTube, TikTok, WeChat, and Facebook’s subsidiary services Messenger, WhatsApp, and Instagram. Facebook’s remarkable growth is also due to its significant development new features (e.g., for media, commerce, gaming, micropayment, and advertising) that connect heterogeneous actors through diverse devices with Internet connectivity such as computers, smartphones, and tablets. By offering features, Facebook has steadily expanded its scope through introducing new features to scale its user base to encompass vast range of user groups. Google, Tencent, Apple and many other firms seem to develop their features in a similar manner, with an ever-expanding digital infrastructure driving new platform development (Galloway, 2017). Failed platforms such as MySpace and Friendster provide examples of how digital platforms may require strategies that extend not only initial features’ scope, but also architectural boundaries to grow and survive in the market (Sandberg et al., 2020).

Strategies for improving innovation for digital platform features have received substantial research attention (Eaton et al., 2015; Leiponen et al., 2022; Szalkowski & Mikalef, 2023), but the strategies for extending digital platform boundaries beyond the basic platform logic have eluded theorisation (Constantinides et al., 2018). Digital platforms’ have editable, interactive, reprogrammable, and distributable qualities (Chalmers et al., 2021; Kallinikos et al., 2013; Lyytinen, 2022) which enables seemingly infinite expansion of architecture, features, and their relationships (Sandberg et al., 2020). Therefore, the associated opportunities for value creation are unprecedented, but also present significant strategic challenges. We examine this through a qualitative case study of Facebook based on archival digital trace data asking: How do digital platforms emerge and evolve beyond their architectural constraints?

To explore this research question, we extend Hanseth and Lyytinen’s (2010) conceptualisation of three types of digital artefacts: application, platform, and information infrastructure. Applying this to our archival data (Romano et al., 2003), we identified four stages in Facebook’s evolution between 2004 and 2019: (i) interaction, (ii) integration, (iii) interconnection, and (iv) implantation, and how it unbundled the Messenger app, evolving through the same stages on its own. We pay particular attention to characteristics of each stage during the period, and transitions between them. Our analysis of the fundamental technological advances prompts a discussion of the need for research to revisit previously neglected roles of boundary resources and to examine opportunities for theorising digital materiality (Kallinikos et al., 2013; Lyytinen, 2022).

Theoretical framing

Digital platform evolution

Digital platforms enable coordination by providing a ‘blueprint that describes how the ecosystem is partitioned into a relatively stable platform and a complementary set of modules that are encouraged to vary’ (Tiwana et al., 2010, p. 677). Hence, digital platforms trigger shifts from discrete products and services with fixed boundaries to collections of interoperable but self-contained modules that comprise their digital ecosystems (Jacobides et al., 2018). Digital platform firms have attractive positions within digital ecosystems as coordination enables them to benefit from diverse external resources and skills, and strongly influences their development (Gawer, 2022). However, they must also contend with unforeseen platform uses and extensions, and accommodate new, rapidly emerging digital capabilities and devices (Eaton et al., 2015; Leiponen et al., 2022).

Popular accounts of digital platform firms’ strategizing are strongly influenced by strategic management research, primarily rooted in the economic perspective (Gawer, 2022). They show that success depends on the ability to scale the user base by growing large, heterogeneous networks and harness positive feedback (Huang et al., 2017; Varga et al., 2023). This has also led scholars to develop concepts such as platform leadership (Cusumano & Gawer, 2002; Leong et al., 2019). Here, platforms are seen reducing transaction costs between user groups, which platform firms capture as rent. As the platform effectively serves more market sides, its value increases, and strategies for leveraging both direct and indirect network effects have been explored (Parker et al., 2016).

In the engineering perspective, success hinges on digital platform firms fostering innovation by offering third parties to develop complementary products and services (Tiwana et al., 2010). This has prompted development of concepts such as platform boundary resources (Ghazawneh & Henfridsson, 2013) and investigation of platform firms’ modification and continuous renewal of boundaries between platforms’ core and peripheral components (Eaton et al., 2015). Scholars have highlighted how value is generated by continuously generating novel modes of user interactions (Spagnoletti et al., 2015). Here, boundary resources are used to adjust platforms’ degree of openness and stimulating generativity by offering attractive APIs and SDKs to third-party developers (Karhu et al., 2018; Thomas et al., 2024). Studied aspects encompass how digital platform firms expand their user base until reaching critical mass, utilise it to fend off and disrupt competitors, and adapt to the rapid emergence of new digital capabilities and devices to sustain the vibrancy of their digital ecosystems (e.g., Parker et al., 2016). One drawback of these foci is that digital platform firms’ emphasis on capitalisation and governance in value creation within digital ecosystems has garnered more scrutiny than their impact on the resultant dynamics (e.g., Huber et al., 2017).

Boundary resources are pivotal in facilitating the broadening of platforms’ feature range and diversity (resourcing) (Ghazawneh & Henfridsson, 2013). Boundary resources are also critial for directing the trajectories of digital ecosystems (securing) which are recognised as mechanisms for minimising variability in core components (Ghazawneh & Henfridsson, 2013). This can lead to notable tensions between the core and peripheral elements of a digital platform, which might manifest as boundary resources are adaptively reorganised through ‘distributed tuning’ (Eaton et al., 2015). Theory predominantly explains how boundary resources facilitate the longevity of digital platforms through evolutionary processes, ensuring that core components maintain a state of ‘relative stability’ (Tiwana et al., 2010). As a result, there remains a lack of understanding of how digital platform companies such as Facebook leverage their digital ecosystems’ dependence over time. This is particularly true with regards to the use of technological resources that they generate to connect different stakeholders and transform the core platform (Constantinides et al., 2018). Unfortunately, the digital platform literature falls short in delivering frameworks to study the long term emergence and evolution of digital platforms beyond their architectural constraints.

The digital artifact spectrum

Hanseth and Lyytinen’s (2010) seminal paper on dynamic complexity in information infrastructures preceded the explosion of digital platform scholarship and is generally not understood as part of that literature. However, the paper does conceptualise the notion of platforms as a class within a spectrum of digital artefacts. Hanseth and Lyytinen (2010) recognised three classes of digital artefacts with different degrees of complexity: application, platform, and information infrastructure (see Table 1). Each class of artefacts has distinct emergent and structural properties that shape its development.

Table 1. Hanseth and Lyytinen’s (2010) three classes of digital artefacts.

Property/Class	Application	Platform	Information infrastructure
Emergent properties
Shared	Yes, locally and through specified functions	Yes, across involved user communities and a set of IT capabilities	Yes, universally and across multiple IT capabilities
Open	No, closed by user group and functionality	Partially, depending on design choices and managerial policies	Yes, universally allowing unlimited connections to user communities and new IT capabilities
Heterogeneous	Yes, partially and mainly by involved social groups	Partially, mainly by social groups but also technical connections	Yes, increasingly heterogeneous both technically and socially
Evolving	Yes, but limited by time horizon and user community	Yes, and limited by architectural choice and functional closure	Yes, unlimited by time or user community
	Linear growth	Mostly linear growth	Both linear and nonlinear growth
	Evolution bounded and context-free	Evolution path-dependent	Evolution path-dependent
Structural properties
Organizing principle	Direct composition of IT capabilities within a homogeneous platform	Direct composition of a set of horizontal IT capabilities within a set of homogeneous platforms	Recursive composition of IT capabilities, platforms, and infrastructures over time
Control	Centralized	Centralized	Distributed and dynamically negotiated Can involve only basic organising principles (standards) and rely on installed base inertia

Hanseth and Lyytinen (2010) define platforms as utilised by multiple user groups, possessing diverse capabilities, and exhibiting degrees of openness and heterogeneity. Their evolution is constrained by architectural decisions and functional closures, resulting in linear and path-dependent growth. These characteristics are contingent on structural properties of platform architectures, notably their organising principles and control mechanisms. In contrast, applications are architecturally simple. While they have the potential to grow in both scale and scope, their generativity is often limited. This limitation arises because the underlying capabilities, which refers to users’ abilities or rights ‘to perform a set of actions on a computational object or process’ (Hanseth & Lyytinen, 2010, p. 2), are typically functionally decomposed and organised hierarchically. This hierarchical organisation restricts the possibility of extensions and results in homogeneous uses (Henfridsson et al., 2014). Information infrastructures are, however, distinguished by their complexity. They comprise diverse capabilities organised recursively and distributed among heterogeneous actors and exhibit nearly unlimited potential for growth and generativity (Henfridsson & Bygstad, 2013). This expansive capability stems from their universal sharing, both socially and technologically, making them accessible to almost everyone for use. This accessibility facilitates the extension and recombination of technological resources in unforeseen ways. Therefore, their growth follows a non-linear trajectory as they, in contrast to digital platforms, lack a centralised control point (Hanseth & Lyytinen, 2010). While initially proposed as a taxonomy for classifying digital artefacts, we propose that Hanseth and Lyytinen (2010) offer a valuable framework for comprehending how digital platforms emerge, transcend their architectural limitations and evolve.

Methodology

Information Systems scholarship has increasingly turned to digital trace data to understand the evolving dynamics of today’s digital landscape (Grisold et al., 2024; Miranda et al., 2022) as they offer developing precise and nuanced analysis of sociotechnical phenomena. Digital trace data capture digitally mediated actions and events over time with precision and detail (Pentland et al., 2020) which surpasses that of traditional data collection methods like observations and interviews (Grisold et al., 2024). Scholars can therefore extract highly specific insights into the characteristics and dynamics of sociotechnical phenomena. This is particularly relevant when studying digital platform evolution.

We utilise digital traces to explore the evolution of Facebook for two principal reasons: (1) digital trace data are abundant and cover extensive periods providing detailed yet broad insights into the evolution of digital platforms and infrastructures (Østerlund et al., 2020) and (2) digital trace data contain temporal information indicating the timing of specific activities (Pentland et al., 2020), making them particularly valuable for visualising and analyzing the dynamics surrounding the evolution of digital platforms. We analyse such data to construct contextually grounded explanations of the unique formulation of a digital platform’s emergence and evolution. However, the utilisation of trace data also presents methodological challenges. Scholars need to be aware of the social and technical processes involved in their generation and that they are closely tied to the technical infrastructure that captures them (Hanseth & Lyytinen, 2010). Trace data should not be accepted as ready-made evidence, rather scholars must invest significant effort in preparing the data before analysis (Østerlund et al., 2020). Therefore, to comprehend the processes by which digital platform firms achieve and maintain their dominant position and investigate their evolution, we conducted a qualitative case study of Facebook (Yin, 2014).

Our decision to select Facebook as a case was influenced by three key factors: (i) our existing familiarity with the platform, (ii) the availability of public data related to Facebook, and (iii) its well-established position as a global leader. Selecting a single, comprehensive case such as Facebook is an effective approach for developing theory through a qualitative case study, particularly when the findings can be compared to broader industry trends or established theoretical frameworks (Yin, 2014). Given the potential limitations in accessing internal data, we relied on publicly available online data (Romano et al., 2003) to trace Facebook’s evolution aligning with established practices when access to internal data is restricted (Alaimo et al., 2020; Eaton et al., 2015; Ghazawneh & Henfridsson, 2013). Gaps in existing theory motivated our case study design. While Facebook’s rapid ascent to platform leadership aligned with established ideas of feature integration (Cusumano & Gawer, 2002), it also surpassed them. Facebook evolved from being merely a platform to becoming distributed across the overall Internet infrastructure. This proliferation of connections resembled the concept of ‘coring’ (Cusumano & Gawer, 2002), enabling the integration of user-generated content and facilitating the introduction and expansion of independent applications. These observations prompted an expansion of Hanseth and Lyytinen’s (2010) framework, which categorises applications, platforms, and information infrastructure. This longitudinal case study, which spans the period from 2004 to 2019, employed four stages (interaction, integration, interconnection, and implantation) to collect trace data. This extended framework was utilised to illuminate the technological advancements that underpin Facebook’s dominance.

Data collection

The case study was delineated using two key datasets. The first dataset covered Facebook’s trajectory from its launch in 2004 to its integration into the broader Internet landscape by 2011, spanning a period of seven years. The second dataset focused on Facebook Messenger, a highly successful application launched in 2011, and traced its evolution up to 2019, encompassing another seven-year period. Our examination included all four stages: interaction, integration, interconnection, and implantation. We used similar methods to collect both sets, but for conciseness and clarity, we provide detail on the process for the initial, larger dataset. Figure 1 offers a detailed summary of the data collection and analysis process for this dataset.

Figure 1. Summary of data collection and analysis.

First, we reviewed publicly available information published in a section of the Facebook-website called the Newsroom (Box A, Figure 1). The Newsroom contained several subsections – including several blogs written by Facebook staff, official press releases, and information on advertising, terms and conditions, etc. Following initial review of the structure and content of the Newsroom, we elicited data (Arrow 1, Figure 1) (Miles et al., 2014; Romano et al., 2003) from three sub-sections particularly relevant to our research: Announcements, Timeline, and Facebook Blog (Box B, Figure 1).

The announcements section contained all of Facebook’s press releases. They covered a wide array of topics, ranging from new feature announcements, via integrity issues, to business partnerships. We captured all 55 press releases published from April 2006 to December 2011. The timeline section contained short facts on number of users, major investments, and global expansion at different points in time. It contained 41 data points covering the period 2004–2011. The blog section contained several blogs focused on various topics. Facebook staff mainly writes the posts published in these blogs. The Facebook blog is one of them. In December 2011, it contained 455 posts by 254 different Facebook employees. We chose this blog as it contained the same wide variety of content as the announcements-section, while often providing discussion and reflections about those announcements. The posts are written in an informal and relaxed tone and language.

We sought to elicit data to support examining changes in platform architecture, governance, and environmental dynamics (Tiwana et al., 2010). We reduced (Arrow 2, Figure 1) the number of data points by excluding data that fell outside of this scope (Miles et al., 2014; Romano et al., 2003). While all data points in timeline (41) and announcements (55) were deemed relevant, several blog posts were excluded. For example, we excluded guest blog posts (e.g., by politicians and community leaders), posts containing personal reflections (e.g., about an upcoming Holiday), and staff leisure events (e.g., parties). This reduced the Facebook blog data points from 455 to 301.

Data analysis

Our data analysis progressed through three principal stages (see Stage A–C in Figure 1). In Stage A, we applied data-driven, grounded codes (Box C) (Corbin & Strauss, 2008). Initial open coding produced a list of 224 codes (Charmaz, 2006). Reviewing these codes, we engaged in further data reduction by merging similar and overlapping codes (Miles et al., 2014; Romano et al., 2003). This reduced the codes from 224 to 173. Next, we created code definitions, resulting in further merging of redundant codes, ultimately arriving at 156 codes. We then reviewed all codes and code definitions again (Corbin & Strauss, 2008) and built five code categories categories: app/feature (63 codes), user information (13 codes), organizations (34 codes), language (4 codes), and growth (4 codes). The first three categories were deemed particularly interesting given our research question. They contained descriptions of specific features and information about when they were introduced (app/feature), which third-party actors that Facebook engaged with over time and how (organisations), as well as Facebook’s response to privacy and integrity arising due to platform changes (user information).

Having gained this initial overview, we revisited the coded segments in these three categories, further exploring, delineating, and defining particular key events and processes. Visual mapping of these events and processes facilitated our sensemaking of their role in shaping Facebook’s evolution (Langley, 1999). This first visualization (Arrow 3, Figure 1). The first version of the timeline was created in February 2012. It became a boundary object for the co-authors and was continuously revised. It served as a foundation for the results and analysis presented in the paper. The final version of the Timeline is presented in the paper (see Figure 2) enabled us to develop and verify our theoretical ideas, shifting the analysis from a descriptive to a conceptual level (Romano et al., 2003). As we identified similarities, continuations, and discrepancies, three distinct patterns emerged. To further investigate the meaning of these patterns and gain additional insight, the data coded thus far was triangulated (Arrow 4) with additional data sources (Box D) (Yin, 2014).

Figure 2. Timeline of Facebook’s early development.

The first stage of our analysis had enabled us to gain an initial macrolevel understanding of Facebook’s evolution. However, we observed that the three patterns identified were not the equivalent of discrete chronological phases. Rather they were reoccurring and interrelated. To explore those interrelations, we needed a theoretical vocabulary that could support us in shifting attention towards microlevel analysis. Therefore, Stage B was initiated with the decision to apply an evolutionary lens (Arrow 5; Box E) (Nelson & Winter, 1982; Simon, 1996). In this stage, theory-driven microlevel analysis was combined with a temporal bracketing strategy (Langley, 1999). This approach allowed us to not only conceptualise and clearly delineate the three patterns as relatively stable, cumulative stages; it also revealed how the transitions between these stages occurred. It led us to create a second, more refined visualization (Arrow 6) (Romano et al., 2003). Our analysis revealed that, viewed from evolutionary theories, Facebook, its users, and partners co-evolved under three different paradigms of variation, selection, and retention that produced different classes of features. Informed by the coevolutionary dynamics suggested in the three patterns, we labelled them interaction, integration, and interconnection.

Having isolated the micro-level mechanisms for each pattern, Stage C was geared towards examining the role of digital materiality in each stage (Kallinikos et al., 2013). Applying a theory-driven multilevel analysis-approach, we were able to theorise how the class of features retained in each stage affected the locus of innovation and control in distributed tuning of the platform (Eaton et al., 2015). This analysis, along with a third visualization (Arrow 7) sharpened our understanding of both the different mechanisms, and the process by which new patterns emerged.

We applied the same approach to the second dataset, focusing on Facebook Messenger (Figure 3). This allowed us to compare the technological advancements driving architectural shifts in both Facebook and its standalone app.

Figure 3. Timeline of Messenger’s subsequent development.

The Facebook case

Interaction: Facebook as a social-network application (service innovation)

Facebook was launched in 2004 as a website that allowed registered Harvard students to create Profiles, search for other users and add them as Friends who they could interact with through profile pages. Within the first year, Facebook launched the additional features Wall and Groups. The Wall, a guestbook on the user profile, initially supported text-based content, but was gradually refined to increase interactions’ richness and diversity. For example, Facebook subsequently added support for attachments and HTML-like standards through which users could embed external content such as Flash applications. Groups allowed interactions centred on topics rather than users, providing open fora where Facebook users could engage with and meet new friends. Facebook rapidly grew through direct network effects. The social network service was gradually extended to students at other US universities until its launch to the general public in September 2006. Facebook representative Carolyn Abram (Facebook blog, 26 September 2006) argued that they learned “[that Facebook] was useful to everyone – not just to Harvard students, not just to college students’ and ‘kept growing to accommodate this fact”.¹

Facebook launched two new features in conjunction with its public launch: News-feed and Mini-feed. Facebook co-worker Ruchi Sanghvi (The Facebook Blog, 05 September 2006) explained to the user community how the News-feed would highlight stories about² ‘what’s happening in your social circles on Facebook’, e.g., if ‘Mark adds Britney Spears to his Favorites’ or a ‘crush is single again’. Before this, Facebook users navigated Facebook through profile pages, the wall, and groups. Subsequently, they saw a list of automatic updates ‘generated by the activity of your friends and social groups’ upon login. Mini-feed was similar but kept user profiles were central as content focused ‘around one person’. When visiting a profile, it gave an overview of what that user activities or events, including changes ‘in their profile’ such as attending events ‘and what content (notes, photos, etc.) they’ve added’.³ Facebook’s architecture remained stable during this period. However, as the social-network service scaled globally and users amassed more friends, new features were needed, and the application became more complex. News-feed and Mini-feed caused the first observed major dissatisfaction regarding users’ privacy and integrity. Interactions that previously happened in a certain context, on a friend’s Wall or in a Group discussion, were now exposed to wider audiences. At that stage, it appears that Facebook deemed such exposure necessary to make the social-network service more engaging. However, complaints from users who found the feeds intrusive elicited two public responses from Mark Zuckerberg. In the first, entitled ‘Calm down. Breathe. We hear you’, Zuckerberg tries to minimise the problem and argued: ‘None of your information is visible to anyone who couldn’t see it before.^4,5 In the second public response, entitled ‘An open letter from Mark Zuckerberg’, he appeared to concede to the intrusive nature of feeds:

We really messed this one up. When we launched News Feed and Mini-Feed we were trying to provide you with a stream of information about your social world. Instead, we did a bad job of explaining what the new features were and an even worse job of giving you control of them.⁶

To retain users, Facebook implemented privacy and security settings that could mitigate the new user liabilities. These events reflect Facebook’s development as an application during its first two years: First, it allowed users to interact in increasingly novel ways, and next, users could interact around the interactions themselves.

In the following years, Facebook launched many additional features to improve interactions that would stimulate the growth of the application in a similar way. Through Inbox and Chat, Facebook evolved new features such as email and instant messaging. The introduction of Comments and Likes made the feeds more engaging by prompting interactions around events in the feeds. The scope of the feed structure was also expanded as such, e.g., Live-feed (launched in 2009) enabled users to see what was currently happening, in contrast to News-feed, which presented algorithmically ranked activities since the users last logins. In 2011, this was complemented by Ticker, which displayed short versions of updates in the website’s top-right column, just before the Wall was replaced with a new Timeline. By the time Twitter had successfully scaled its user base, Facebook introduced a Subscribe feature through which users could follow updates from users who were not their friends. To summarise, technology developments for service innovation in the interaction stage focused on features that changed how, and with whom, the users interacted.

Integration: Facebook as a digital platform (platform innovation)

In May 2007, Facebook announced its first boundary resource, the Facebook Platform API. It had specifically invited 65 partners (including Amazon and Microsoft) to develop applications on the platform and stated that anyone could develop and submit applications. Four months later, Facebook launched fbFund,⁷ a venture capital fund ‘focused on continuing to create incentives for the development of applications on [the] Platform’.⁸ Allocating $10 million in capital to the fund, Facebook provided conditions for stimulating the growth of a developer community, and possibly an ecosystem around the platform. Meanwhile, the Facebook Platform API reduced technological barriers to entry for third parties. In the first six months, 7000 third-party applications were launched and by May 2010 there were more than 550,000 applications, and 1 million developers had registered.

Continued rapid scaling of the user base followed Facebook Platform API’s launch. In contrast to the first stage, the growth was now driven by indirect network effects. Facebook also started monetising its social network by targeting new user groups – in late 2007 Pages and Ads enabled integration of commercial user content for the first time. The change was communicated in a blog post where Facebook stated that it wanted to make clear what’s changing – and what’s not – for you’.⁹ In contrast to the first stage, Facebook now communicated the change proactively to the user base, ensuring that the user experience would not be compromised because of these two new features. It ensured that users would retain control of their data but were offered the option to interact ‘with products, businesses, bands, celebrities [etc.]’, and ‘share actions you take on other sites with your friends on Facebook’. Facebook’s increasing focus on integration to attract new user groups was also illustrated in its proprietary applications such as Marketplace, which essentially emulated features popularised in the USA by eBay and Craigslist. However, the Social games application category had the strongest impact on the network. Although third-party developers initially attempted to ‘game’ the News-feed by spamming updates to Facebook users, social games made Facebook’s platform hugely successful. For example, Zynga created two of the most widely played games : Mafia Wars (2008) and FarmVille (2009). In 2010, Facebook announced a five-year strategic partnership with Zynga.

These events reflect Facebook platform’s shift from technological developments towards indirect network effects through integration of complementary innovations that expanded the digital ecosystem and enabled further scaling. To integrate third-party features Facebook had to ensure that they could be securely resourced. For example, when users complained that developers were spamming their News-feeds, Facebook changed the terms and conditions and reconfigured its app-ranking system to penalise such behaviour. Facebook continued to integrate features to accommodate the needs of its increasingly heterogeneous users and introduced refined APIs and SDKs through which third parties could access Facebook’s core components and exploit its scale. Continued improvement of user interactions, and high interactivity of the most popular applications, enhanced the social network’s core market proposition, and integration enabled Facebook to stimulate further scaling by inviting users to interact with or through various tools (e.g., videos and games).

Interconnection: Facebook as an information infrastructure (ecosystem innovation)

In April 2010, Facebook launched Open Graph and Social plug-ins. Open Graph enabled any website or mobile app to interconnect with the social network and structure interactions according to Facebook’s logic. When users shared what they were doing they created stories that included actions, indicating people they were with, and places where they occurred. In Open Graph, these were structured accordingly, enabling any service or platform to inherit Facebook applications’ functionalities, furthering third parties’ ability to promote engagement, distribution, and growth through deeper integration with Facebook as part of their ecosystem innovation. Through Open Graph, Facebook could target users beyond the website’s immediate scope. In 2006, Facebook launched Share, which enabled websites to embed code allowing users to share content on such websites with Facebook friends by clicking a so-called share button. Similarly, Connect (launched in 2008) enabled third parties to use Facebook credentials to identify users. Open Graph extended these features by enabling websites and mobile apps to generate and publish content in users’ News-feed, Timeline, and/or Ticker. This enabled standardised capture of any type of interaction (e.g., listen to a song, recommend a book, buy a ticket, or visit an event). Social plug-ins by comparison enhanced third-party developers’ ability to make their own services platforms more engaging through core Facebook features such as Like buttons and Comment fields. Over 100,000 websites adopted the Social plug-in feature in a few weeks.¹⁰

Before the launch of Social plug-ins, Facebook again formed exclusive partnerships, e.g., with media firms. Facebook wrote, ‘News sites have implemented social plugins to help surface individualized content for readers, and in the process seen significant increases in daily referral traffic from Facebook’.¹¹ According to Facebook, Facebook referral traffic had increased by 250% for ABC News, and 80% for Canada’s largest-circulation national newspaper. Users who liked their Pages commented, shared, and read more external content. IMDB.com and NHL reported similar increases in Facebook referral-based traffic and user engagement with their content.¹² These events effectively illustrate Facebook’s development into an information infrastructure with technological resources connecting heterogeneous actors on and off Facebook.

Compared to the identified frictions with the established user base identified in Stage 1 and Stage 2, the launch of both Open graph and Social plug-ins appear to have generated little criticism. However, the new feature Beacon caused controversy. Similar to Open Graph, it allowed Facebook to immediately capture and structure external user interactions at high resolution. For example, if Jane Doe bought Age of Facebook from Amazon, the interaction could be published on her profile, including a link to Amazon’s website. Massive criticism ensued, causing (for example) Coca-Cola to pull out shortly after becoming a strategic partner.¹³ To appease critics Facebook switched from an opt-out to an opt-in model. In a blogpost entitled ‘Thoughts on Beacon’¹⁴ Mark Zuckerberg tried to explain, to little avail, that the ‘goal was to build a simple product to let people share information across sites with their friends’, which had to be light enough to avoid hindering users, but ‘clear enough so people would be able to easily control what they shared’. Beacon supposedly solved this as ‘a lot of information people want to share is not on Facebook’ and its algorithm would facilitate sharing: ‘people wouldn’t have to touch it for it to work’. Two class action lawsuits later, Beacon was terminated in late 2009. In a blogpost from 2011, entitled ‘Our commitment to the Facebook Community’ Zuckerberg called Beacon a ‘high profile mistake’.¹⁵

Implantation: Facebook as a launchpad and incubator (platform dominance)

The previous sections report on how Facebook developed from a social-network application into a global information infrastructure, enabling the launch and incubation of standalone applications. This strategy became clear when Facebook in 2011 ‘unbundled’ the Facebook application for iPhone and Android. Facebook Messenger, the first feature to become a standalone application, was soon followed by others, such as Poke, Camera, and Pages. Zuckerberg saw high potential for growing this way, ‘whether that’s Groups or Events or things like that can reach a lot of people’, because most users used Facebook every day.¹⁶ Previously Facebook had faced little competition, but the ease of building and scaling new social applications on mobile platforms had increased. These features’ development as standalone applications could enable continuing domination of the mobile space. While theoretically simple, this strategy initially proved quite challenging. Despite a large user base and low switching costs (downloading the app and login with Facebook credentials was enough to get started) Facebook users proved hesitant. Significant user adoption of Messenger only occurred after Facebook announced in 2014 that messaging would no longer be available in the main Facebook application. Three years later, Messenger had over 1 billion monthly active users and was becoming a digital platform in its own right.

Messenger was an offshoot of a venture that Facebook had acquired. In Zuckerberg’s words, Facebook unbundled Messenger because as ‘as a second-class thing inside the Facebook app’ it increased ‘friction to replying to messages’.¹⁷ Facebook’s VP of Growth and Analytics, Javier Olivan, also reported that trying ‘to put too many things in the same app at the same time’ had created huge challenges—’in mobile you have to be fast all the time’.¹⁸ Initially, Messenger simply copied the Chat feature from the website. However, new features were soon introduced that enabled novel modes of interaction, such as Voice calling, Video calling, and Money transfers between friends, and in 2015 the Messenger platform API was launched (see Figure 3 for a timeline).¹⁹ The applications were initially simple but subsequently included a partnerships with carsharing service Uber and AR camera effects for brands, giving Messenger a distinct scope from the social-network. The improved access to user-data boosted Facebook’s ad network and data monetisation. This also boosted Facebook’s platform dominance as most Facebook’s revenue came from the mobile space.

Messenger’s development into a digital platform illustrates how Facebook recognised the value of ‘creating single-purpose first-class experiences’ to drive and further ‘unbundle the big blue app’, as Zuckerberg said in 2014.²⁰ Chatbots’ launch in 2016 showed that Facebook would strive to become further interconnected in this process. Chatbots opened the Messenger platform up to new extensions and uses. Messenger and WhatsApp jointly served 60 billion messages monthly (three times more than SMS), so the potential demand was enormous. Messenger Executive David Markus noted the feature enabled ‘a richer form of communication with different types of other parties’²¹ and promised to make them secure and personalised: ‘your identity is preserved, you never need to use password and username combinations’. Similarly, Seth Rosenberg (developer blog, 12 April 2016) commented that Chatbots ‘are for anyone who’s trying to reach people on mobile – no matter how big or small your company or idea is, or what problem you’re trying to solve’.²² Through open APIs Chatbots can update users about the weather, use GPS to send reminders, and offer sophisticated consumer experiences. As TechCrunch noted , this enabled developers to create ‘utilities that work on top of Messenger to aid with things like customer service, e-commerce transactions, and other interactive experiences’ and improve them using Facebook’s machine learning.²³ The shift towards interconnection was completed in 2017 when Facebook launched Messenger Web plug-ins enabling third parties to offer Messenger functionalities like these on their websites.²⁴

These events illustrate Facebook’s ability to increase interconnection by implanting standalone features that expanded the scope of the social network and fostered innovation. Once ‘unbundled’ as standalone applications, features could develop more freely and connect heterogeneous users in new ways. The primary example of this fractal growth pattern, Facebook Messenger, shows that digital platform firms can benefit greatly by extending services beyond platforms’ architectural limits, but this is difficult. Other applications, e.g., Camera, did not experience similar growth and Facebook subsequently acted more defensively in this space by enveloping features from Instagram and the rival Snapchat into the Messenger platform. Nevertheless, the unbundling strategy has significantly affected Facebook’s development with features increasingly introduced outside of the main Facebook application. This impact seems likely to increase as Facebook prepares for competition in the micro-payment space by testing features and developing boundary resources that promise to make Messenger a channel for users to send money to friends, split bills, and pay third parties. Hence, implantation has enabled Facebook to maintain dominance by unbundling features that facilitate branching into new application domains and leverage the digital ecosystem’s dependence on resources it has developed to ward off competition to boost their platform dominance.

Discussion

Our longitudinal Facebook case study refutes the assumption that digital platforms are relatively stable (de Reuver et al., 2018; Tiwana et al., 2010). To answer our research question by investigating how digital platforms emerge and evolve beyond their architectural constraints we found value in extending Hanseth and Lyytinen’s (2010) classification of digital artefacts and treating them as a continuum. Our findings provide a detailed and rich account of how Facebook developed from a social network service application, into a platform, then an information infrastructure. Based on these findings from the Facebook case, we extend previous literature on digital platforms by illuminating pre-, and proto-stages of digital platform evolution. We conceptualise this process in a model of the emergence, evolution, and extension of digital platforms (see Figure 4).

Figure 4. Emergence, evolution, and extension of digital platforms.

The model shows how a specific actor is targeted in each stage: (1) users, (2) third-party developers, and (3) websites and digital services. It also illustrates how boundary resources do not solely enable resourcing and securing (Ghazawneh & Henfridsson, 2013), despite their importance as (for example) both Facebook and Messenger developed from applications into platforms. Boundary resources such as Share, Beacon, Connect, Open Graph, Social Plug-ins (in Facebook’s case), Web Plug-ins, and AR camera (in Messenger’s case) proved as important in their development from platforms into information infrastructures. Past research, including Ghazawneh and Henfridsson (2013) shows how cultivating boundary resources enables digital platform firms to access external resources and skills, and securing boundary resources enables them to prevent third parties violating their platform architecture’s internal coherence. In contrast, the boundary resources we discovered in the Facebook case played distinct roles that we call distributing and centring.

The distributing role is illustrated by the rapid embedding of core Facebook components in myriads of webpages through Open Graph, Social Plug-ins, and Web Plug-ins. This routed traffic data and content into Facebook, ensuring that users always had plenty to interact about . It also increased the digital ecosystem’s dependence on the digital platform, thus highlighting the centring role; the imposition of informal control by encouraging external actors to make their products and services as compatible with Facebook as possible. In contrast, securing relies on formal means of control, e.g., terms and conditions for using SDKs and hard-coded restrictions in APIs. The object-action model Facebook developed (partly due to Beacon’s failed launch) provides a prime example as its adoption made Facebook’s data structure a community standard. In combination, these distributing and centring roles help to explain how Facebook successfully implanted standalone applications once it had developed an information infrastructure.

Our case also demonstrates that when digital platforms become dominant, digital ecosystem dynamics enable the unbundling of core platform components into standalone applications that allow digital platform firms to branch into new application domains. Facebook shows that a massive user base alone does not guarantee success, but Messenger shows that such implants can potentially develop into dominant digital platforms in their own right. The recognised stages capture this idea and preserve the fractal pattern of growth we observed. Whereas information infrastructures are characterised by recursive growth (Hanseth & Lyytinen, 2010), the fractal growth pattern suggests that a strong form of centrality may be maintained as capabilities are distributed. Thus, a platform that develops into an information infrastructure as a digital platform firm distributes technological resources connecting heterogenous actors deep into the digital ecosystem will likely evolve differently from an information infrastructure such as the web.

A possible approach for theorising such distinctions is to explore differences in digital technology architectures’ digital materiality, particularly their editability, interactivity, reprogrammability, and distributability (Alaimo et al., 2020; Kallinikos et al., 2013; Lyytinen, 2022; Yoo et al., 2010). Table 2 shows how these qualities ere leveraged across the three stages.

Table 2. Material qualities of digital artefacts foregrounded in each stage of Facebook’s evolution.

Quality	Application	Platform	Information Infrastructure
Editable	Editable under a limited scope	Editable both at the solution’s content and service level	Editable at multiple levels and under different timescales using many open and rich formats and protocols.
Interactive	Interactions are limited to basic functionality in the application	Interactions between different actors of many types that may vary in their relation over time	Solutions are interacted with as digital bundles within an infrastructure with integrations of various depth
Reprogrammable	Only reprogrammable by the application owner	New scopes for editability and interactivity through modular design principles and APIs	Reprogrammable in a multitude of infrastructural components
Distributable	Restricted to a closed application scope	Boundary resources distribute control and innovation while securing the platform core	Boundary resources leverage distributable qualities while serving to centre infrastructure traffic to the platform’s core

Our findings show that during the interaction stage, the application limited editability to users’ own contents, interactions were limited by the application’s basic functionalities, reprogrammability largely restricted to incremental cultivation by the application owner, and the application was mainly distributable within limits dictated by tight local integration. Digital platforms, by contrast, are editable at both content and service levels, enabling interactions between users of multiple types with constantly changing relations. Accordingly, digital platforms are reprogrammable in ways that may extend the scope for editability and interactivity via modular design principles and APIs etc.

Such theorisation offers avenues for more systematic elucidation of boundary resources’ nature and roles in shifts in digital technology architecture. Facebook’s shift from application to platform was driven by a need for more editable and interactive digital materiality. Closer examination of how boundary resources’ development affords the selection, modularisation, and integration of core components to solve market problems could illuminate this mechanism. By contrast, the need for more reprogrammable and distributable digital materiality drove Facebook’s shift from a platform to an information infrastructure. Thus, closer examination of how boundary resources help disintegrate digital platform cores and form boundaries at varying depths and degrees of integration is needed to unravel the mechanisms and strategies involved in their use to escape digital platform architectures’ initial limits. Our study suggests that distributing and centring enabled Facebook to maintain dominance by embedding core platform components and rendering external sites Facebook-compatible, but further investigation may reveal other pertinent strategies.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. https://www.facebook.com/notes/facebook/welcome-to-facebook-everyone/2210227130/.

2. https://www.facebook.com/notes/263466884972254/?_rdr.

3. https://www.facebook.com/notes/facebook/facebook-gets-a-facelift/2207967130/.

4. https://www.facebook.com/notes/facebook/calm-down-breathe-we-hear-you/2208197130/.

5. https://www.facebook.com/notes/2832474983634072/.

6. https://www.facebook.com/notes/facebook/an-open-letter-from-mark-zuckerberg/2208562130/.

7. http://newsroom.fb.com/news/2007/09/announcement-fbfund/.

8. https://about.fb.com/news/2007/09/announcement-fbfund/.

9. https://www.facebook.com/notes/facebook/facebook-ads/6972252130/.

10. http://tcrn.ch/cgrmOb.

11. https://developers.facebook.com/blog/post/382.

12. https://developers.facebook.com/blog/post/382/.

13. http://techcrunch.com/2007/12/01/facebook-beacon-isnt-on-the-coke-side-of-life/.

14. https://www.facebook.com/notes/facebook/thoughts-on-beacon/7584397130/.

15. https://www.facebook.com/notes/facebook/our-commitment-to-the-facebook-community/10150378701937131/.

16. https://techcrunch.com/2014/01/29/one-app-at-a-time/.

17. https://techcrunch.com/2014/04/09/facebook-messenger-or-the-highway/.

18. https://techcrunch.com/2014/01/29/one-app-at-a-time/.

19. https://developers.facebook.com/blog/post/2015/03/25/F8_2015_Roundup/.

20. https://archive.nytimes.com/bits.blogs.nytimes.com/2014/04/16/can-facebook-innovate-a-conversation-with-mark-zuckerberg/.

21. http://www.usatoday.com/story/tech/2016/04/13/what-bots-can-do-you/82970056/.

22. https://developers.facebook.com/blog/post/2016/04/12/bots-for-messenger/.

23. https://techcrunch.com/2016/04/12/facebook-messenger-and-whatsapp-combined-see-3-times-more-messages-than-sms/.

24. https://techcrunch.com/2017/11/07/facebook-introduces-a-messenger-plugin-for-business-websites/.