Abstract
This article examines the process of platformization from the perspective of platform complementors. It focusses on Dutch complementors—retailers, news organizations, transportation companies, and software start-ups—that developed Conversational Actions for Google Assistant. The analysis builds on a combination of digital fieldwork and interviews with Dutch developers, conversational specialists, and marketing and communication managers between 2018 and 2021. Pushing back against the dominant focus on successful platforms, we demonstrate that platformization needs to be understood as a balancing act, marked by frictions that arise in attempts to align expectations, interests, concerns, and objectives of different actors. Our findings show that complementors’ imaginaries of a Dutch assistant conflicted with Google’s global priorities, revealing the limits of localization in platform infrastructures. Moreover, we highlight the asymmetry between complementors’ expectation of a linear platform development and Google’s circular strategy of continuous reconfiguration of infrastructures and platforms to capture specific markets.
Keywords
For a few years, virtual assistants—software agents that perform tasks and services based on user commands or questions—were hyped as the next revolution in communication technologies. Operated by world’s largest tech companies, Apple, Amazon, and Google, these assistants come preinstalled on all recent smartphones and run on a wide variety of hardware, from smart watches to smart home hubs. Their rise has been associated with the development of artificial intelligence (AI). Virtual assistants employ speech processing, natural language processing (NLP), and information Retrieval (IR) to match voice or text input to executable commands and to adapt system outputs (Hirschberg and Manning, 2015; Natale and Natale, 2021; Riedl, 2019). Through these software systems, virtual assistants can mimic the human ability to communicate, allowing users to engage in personalized exchanges with the system.
While Apple, Amazon, and Google develop many key applications for their assistants themselves, they have also tried to entice third parties to offer services through these assistants. Particularly Google, from 2016 onwards, made a large effort to grow its Google Assistant into a major platform that would allow a broad range of device manufacturers, and content and service providers—from news organizations and banks to music sharing and food delivery platforms—to reach end users (Heater, 2016). Initially Google’s efforts appeared to pay off, as third parties around the globe started to develop applications or so-called Conversational Actions for the Assistant. By late 2019, there were thousands of third-party actions available in 20 different languages, including English, Hindi, French, German, Japanese, Korean, and Dutch (Thormundsson, 2022).
However, after a couple of years of experimentation, Google decided, in June 2022, to deprecate third-party actions, shifting its voice assistant efforts to app actions with Android, its leading mobile operating system. It maintained on the Google Developer blog: While Conversational Actions were an excellent way to experiment with voice, the ecosystem has evolved significantly over the last 5 years and we’ve heard some important feedback: users want to engage with their favourite apps using voice, and developers want to build upon their existing investments in Android. In response to that feedback, we’ve decided to focus our efforts on making App Actions with Android the best way for developers to create deeper, more meaningful voice-forward experiences. (Nathenson, 2022)
In other words, Google effectively abandoned its efforts to develop the Assistant as a platform, instead relying on Android, an already well-established platform. For Google Assistant developers, this had major consequences, as they had been working with Conversational Actions for the previous 5 years.
Furthermore, Google’s earlier NLP ventures have been eclipsed by significant advancements in large language models (LLMs), opening new realms of possibilities for artificial intelligence in human-computer interaction. Google’s LLM Gemini boasts a wide array of capabilities thanks to its multimodal AI model, combining image processing with voice and language processing (Rane et al., 2024). Taken together, Google has pivoted to a new voice interface strategy.
The stalled development of Google Assistant as a platform is particularly interesting, as most scholarly attention has focused on successful platform development, such as the Apple App Store or Facebook (Ghazawneh and Henfridsson, 2013; Goggin, 2014; Plantin et al., 2018; Van Dijck, 2013). Yet developing a platform involves a complex balancing act aligning interests and investments of various actors (Evans and Schmalensee, 2010; Poell et al., 2021; Van Alstyne et al., 2016). Platformization can stall when end users do not find compelling content, or when platform complementors—third-party content providers, advertisers, data intermediaries—struggle to reach end users. The platform company must justify substantial infrastructure investments, while economic returns are limited initially. It must provide resources to open the platform sufficiently to end users and complementors, while simultaneously securing it for safe communication and exchange. Given the small number of large platforms in any sector, balancing these interests is clearly challenging.
In the fields of platform studies and business studies, platformization has primarily been studied from the perspective of platform companies and their success in bringing different sides of a market together (Gawer, 2021; Helmond et al., 2019; McIntyre and Srinivasan, 2017). The considerations of other actors are rarely systematically studied, leaving us with limited insight into what balancing different interests entails in practice. Particularly, complementors’ perspectives and struggles are rarely examined systematically.
In light of these considerations, this paper aims to enhance our understanding of the role of complementors in processes of platformization, especially pertaining to the challenges faced by these actors in the development of AI. We will do so through a case study on third-party developers for Google Assistant in the Netherlands. First, how did these complementors perceive and imagine the efforts to develop Google Assistant into a leading platform in the voice space? Addressing this question, we aim to gain insight in the specific interests, expectations, and understandings of the platform from the complementor perspective and understand why these actors were initially motivated to invest in the platform. Second, we want to know how complementors worked in practice with the assistant and with Google as a company. What issues and struggles did they face? Addressing this question should enhance our understanding of how complementors try to bring their interests and investments in alignment with the platform. The complementor perspective is particularly relevant in this case, as the assistant is developed by Google/Alphabet, one of the world’s leading tech companies with an extensive platform ecosystem and major financial and technical resources at its disposal.
Addressing these research questions, we center on the Netherlands, a country with one of the highest smartphone penetration and Internet usage worldwide. In the period covered by the research, 2018-2021, Google Assistant was also the only major voice assistant available in Dutch. To understand how complementors positioned themselves in relation to Google and its assistant, we have done four months of digital fieldwork, attending weekly meetings of the Open Voice Event series to gain insight in conversations and concerns of the overall Dutch voice community. Furthermore, we have interviewed 11 key Dutch developers, conversational specialists, and marketing and communication managers with experience in developing actions for the assistant. These interviewees have worked with 23 different Dutch companies and public organizations.
To conceptually prepare the investigation, the next section discusses current research on platform development and on the role and position of complementors in this process. We will especially draw on platform studies and business studies. While the latter field is often ignored in media and communication research, it offers vital insights on platform development.
Current research
In business studies, we can find a lot of research on why platforms fail to take off (Akter and Iqbal, 2020; Evans and Schmalensee, 2010; Van Alstyne et al, 2016). Examining platforms’ “failure to launch,” Evans and Schmalensee (2010), for example, suggest that success or failure of a platform “may depend most importantly on both the value that the platform brings to participants as well as the steps that platform entrepreneurs take early on to push adoption past the critical mass frontier” (pp. 22–23). In other words, the key to successful platform development relies on the company’s ability to satisfy the interests of end users and complementors, and to quickly get enough participants on board. Reaching critical mass is essential for end users to have sufficient choice of content and services and for complementors to have a sufficiently large consumer base. To understand why participants do or do not join a platform, further business studies research has analyzed potential obstacles, such as interaction failure, poor matching, and congestion (Van Alstyne et al., 2016). As Akter and Iqbal (2020) observe in an extensive review of the research on platform failure, many factors appear to play a role, with no single factor being decisive.
In this study, we want to shift the focus from the strategies of platform companies to the perspective of complementors. Rather than understand the launch of platforms and the process of platformization more generally in terms of the success or failure of platform corporations, we understand this process as a balancing act that involves a variety of actors with different expectations, interests, objectives, and concerns. The role of complementors in this process is vital, as they provide the “complements”—content and services—that should draw and keep end users on a platform (Rietveld et al., 2019). Yet, just like platform companies, complementors have to take a range of strategic decisions on whether, when, and how to invest in a platform.
Taking such decisions is particularly challenging in the platform start-up phase, when it is not yet clear how a platform will develop. As Van Dijck (2013) has argued, platforms tend to go through a phase of “platform ambiguity” and “interpretative flexibility.” During this start-up phase, it is often unclear which actors will offer services, how the platform will open its infrastructural boundaries, and what its business model will be. Especially in this phase, key stakeholders’ visions of the platform are crucial. While such visions motivate investment, stakeholders must also work with the platform in practice. For complementors this is challenging. Following the ethos of “move fast and break things,” tech companies frequently change platform parameters or abandon platforms entirely. Consequently, third parties occupy a precarious position, constantly adapting to change and preparing for contingency (Cunningham and Craig, 2019; Rietveld et al., 2019). New platforms offer opportunities to produce content and services for growing user populations, but with little certainty of continued growth.
To analyze these relations, it is important to clearly define the key actors and processes involved. First, drawing on the work of Poell et al. (2021), who combine insights from platform studies, business studies, critical political economy, and cultural studies, we understand platforms as “data infrastructures that facilitate, aggregate, monetize, and govern interactions between end users and content and service providers” (p. 5). From this institutional perspective, platforms constitute multi-sided markets, data or computational infrastructures, and governance frameworks.
In turn, this leads us to the process of platformization, which can be defined as the “penetration of infrastructures, economic processes and governmental frameworks of digital platforms in different economic sectors and spheres of life, as well as the reorganization of cultural practices and imaginations around these platforms” (Poell et al., 2019). Central to this process are network effects: platforms gain traction when the growth of the presence of end users attracts complementors, and when complementor contributions in turn attract new users. Hence, complementors are vital to platformization, but their role in this process is rarely studied systematically. Pursuing such an inquiry in this paper, we draw on business studies literature to define complementors as formally independent providers of products and services through digital platforms (Gawer and Cusumano, 2002; McIntyre and Srinivasan, 2017). Existing research has already demonstrated how platformization reshapes the conditions for complementors, whether by reconfiguring creative labor within media industries (Nieborg and Poell, 2018), extending platforms through infrastructural integration (Helmond et al., 2019), orienting complementor practices around speculative futures (Egliston and Carter, 2024), or conditioning participation through governance and value distribution (Wu et al., 2022). Building on these insights, our analysis foregrounds the complementor perspective in the specific context of Google Assistant’s Dutch launch, where cultural and linguistic specificities shaped trajectories of platformization.
Developing this study, we can examine platformization as both a techno-commercial and as a socio-cultural process. In the following analysis of the relations between Google Assistant and complementors, we first focus on the socio-cultural dimension, particularly on the imaginaries that developed around the assistant. A striking feature of the development of the assistant is that both the platform company and complementors were actively constructing imaginaries of the future capabilities of virtual assistants, without most of these capabilities having been realized. In this respect, there are strong correspondences with the “myths” around the “magic” of Big Data and AI that circulated in the 2010s (Elish and boyd, 2018). To analyze the visions that emerged around Google Assistant, we draw on the work of scholars who have theorized imaginaries in the sociotechnical realm (Bucher, 2017; Jasanoff and Kim, 2015).
Inspired by this scholarship, Van Es and Poell’s (2020) have developed the notion of “platform imaginaries,” which is specifically useful for our inquiry into how Google Assistant was imagined as a platform. They define platform imaginaries as “the ways in which social actors understand and organize their activities in relation to platform algorithms, interfaces, data infrastructures, moderation procedures, business models, user practices, and audiences” (Van Es and Poell, 2020: 1). Studying such imaginaries, we underline Tsing’s (2005) concept of friction, which highlights “the awkward, unequal, unstable, and creative qualities of interconnection across difference” (p. 243). This concept prepares the ground for the analysis of the particular expectations of Dutch complementors regarding a “Dutch” Google Assistant vis-a-vis the global platform strategies of Alphabet. At the same time, friction “shows us where the rubber meets the road” (Tsing, 2005: 243). That is to say, the friction between complementors and Google captures the change enacted through the “sticky engagements” (Tsing, 2005: 243) between global tech companies and stakeholders firmly situated within local cultures and business practices. Finally, important to observe is that work in the tradition of sociotechnical imaginaries, understands imaginaries as performative (Bareis and Katzenbach, 2022; Kotliar, 2025): they help to mobilize resources, investments, and commitments, while proving fragile when Google Assistant was hampered by linguistic errors and its development did not meet expectations.
In the second part of the analysis, we focus on the techno-commercial and material dimension of platformization. Here, the reality of developing for the voice platform comes into view through the infrastructural, economic, and organizational arrangements that shaped complementors’ work with Google Assistant. To study these arrangements, we draw on the concept of boundary resources, defined as the tools, data, documentation, and human support provided by platforms to third parties (Ghazawneh and Henfridsson, 2013; Helmond et al., 2019). These resources shape the conditions of production for complementors, structuring how they can design, test, and distribute services for the platform. In the case of Google Assistant, the central boundary resource was the SDK, which enabled developers to add “voice control, natural language understanding and Google’s smarts to [their] ideas” (Google Assistant SDK).
Methodology
Examining the development of the Google Assistant in the Netherlands from the perspective of complementors, we conducted exploratory digital fieldwork to understand the larger context of the Dutch voice community, beginning with the Open Voice Event series. These events helped us identify key players in the Dutch voice industry. We used LinkedIn as an additional search tool to identify relevant professionals and understand the field. The Open Voice Events also led us to the weekly “VoiceLunch NL” Zoom meetings, which we attended as observers after disclosing our research status and obtaining consent from organizers. We participated in 10 Voicelunch NL sessions between November 25, 2020, and March 25, 2021, on a biweekly basis. These events, announced through the Open Voice newsletter, were open to professionals in the Dutch voice industry and regularly drew participants from across this relatively small and tightly connected network. 1 Our participation had two aims: first, to gain a broader understanding of the popularity and thematic focus of these gatherings; second, to identify potential interview candidates. During the sessions, we took field notes on the number of attendees and on the general topics of discussion, omitting names to protect participant anonymity. These sessions covered a wide range of topics, from technical innovations to the professional trajectories of voice specialists. They also illustrated the unique character of the Dutch voice industry as a small, interconnected network in which professionals frequently crossed paths across multiple forums. This included sub-communities such as Women in Voice, which pledged to amplify women’s presence and create openings in a male-dominated field.
Second, drawing on our fieldwork, we interviewed 11 leading members of the Dutch voice community, who developed Conversational Actions for the assistant. 2 Participants were selected based on their active involvement in Actions development between 2018 and 2021, and their ability to speak from direct professional experience. While no minimum years of experience were required, all interviewees held roles that positioned them as key decision-makers or practitioners within their organizations’ Google Assistant projects. These interviewees included representatives from large companies, such as retailers, news organizations and shipping/transportation companies, as well as smaller-scale actors, such as software start-ups and freelance innovation executives and linguists (see Table 1).
Background of interviewees in the Dutch voice community.
Selecting interviewees, we built on Suri’s (2011) “purposeful sampling.” We combined the Intensity Sampling and Maximum Variation Sampling techniques to arrive at a set of interview candidates that both contributed to the development of Conversational Actions and covered key areas of expertise in the voice community. Following these criteria, we primarily approached software developers, user experience designers, and product managers, to achieve a well-rounded scope of the field. We discovered that, oftentimes, those involved with the development of Conversational actions must wear many hats and development teams frequently change composition.
The semi-structured interviews with these actors were conducted between March and October of 2020. In the first few minutes of the interview, the questions were more structured and direct but at a later stage the questions became more open ended because the purpose was to access tacit knowledge with reference to the collaboration between Google as platform operator and third-party developers. We employed “elicitation techniques” (Barton, 2015), such as using field-specific diction, so that the interview candidates would feel comfortable expressing their own ideas on smart voice technologies. Interview participants were also encouraged to share presentation materials, images or text to explicate an observation or shed light on a project with which they were involved. 3
Once the interview transcripts became available, we moved to the coding stage, which followed an iterative, mixed inductive–deductive approach. Guided by Braun and Clarke’s (2006) principles of thematic analysis, we treated coding as a process of identifying patterns of meaning across the data while remaining attentive to our research questions. In the first round, both researchers independently read through the interviews, noting words and phrases that appeared significant. These preliminary observations were compared and developed into a shared codebook through discussion (Guest et al., 2012). In the second round, each researcher applied the codebook independently in ATLAS.ti, after which we discussed and resolved differences to ensure coherence and consistency. This process resonates with Saldaña’s (2016) description of first-cycle and second-cycle coding, where initial codes are refined into a more structured framework through collaborative iteration. The final codebook contained 13 codes, which were subsequently organized into three broader categories. This allowed us to capture inductively emerging insights while maintaining systematic attention to the research questions, with the overall aim to understand how complementors tried to align their investments and interests with Google Assistant as a platform. Informed by the literature and research questions, we arranged the 13 inductive codes into three categories: (1) Google Assistant Imaginaries, (2) Challenges in developing Google Assistant, (3) Future of Voice. These categories provided the scaffolding for our analysis.
Google Assistant imaginaries
When the Dutch Google Assistant was announced in July 2018, Google sought to entice complementors to join the platform by emphasizing the ease of development (Haenen, 2018). It promoted its extensive software development kit (SDK), offering “a low-level API that lets you directly manipulate the audio bytes of an Assistant request and response” (Pichai, 2018). Developers were promised access to user request transcripts and the Assistant’s textual and visual responses. In 2018, the developer preview (1.0.0) of the Google Assistant library for Python became available, supporting hotword models, queries, and responses, thereby broadening the platform’s capabilities. This development infrastructure was consolidated in the Actions Console—a web-based tool to build, deploy, and scale actions. Marketed as a practical resource, Google stated: “Whether you’re developing a way to voice control smart home devices or the next voice-driven game, the Actions console has the tools to build your Action” (Actions on Google). Through such boundary resources and promotion, Google aimed to persuade Dutch companies to become early adopters. These promotional materials projected not only technical possibilities but also suggested a linear trajectory of platform development, implying that early adoption would mature into stable capabilities and user uptake.
The Netherlands was a prime candidate for the launch. The Dutch market has been particularly accessible to international tech companies due to its near-universal Internet penetration and digitally literate population. This accessibility is reinforced by a supportive regulatory environment that encourages innovation and cross-European expansion (Invest in Holland, 2023). More specifically, in 2018, around 12% of all mergers and acquisitions in the NL/Belgium region were tech-related, up from 9% in 2015 (PwC, 2020). Coupled with its role as a leading hub for digital infrastructure and data centers, the Netherlands provides both the user base and infrastructural backbone that make it a strategic entry point into the European market (Dutch Data Center Association, 2018).
Launching the assistant, Google publicly spotlighted its launching partners, key complementors recruited during the beta-testing phase of the Google Assistant and given free access and resources to develop prototypes. Google positions established industry partners front-and-center to solidify its place within the Dutch voice market and to appease risk-averse smaller companies for which voice action development is a large investment. If large companies, such as Albert Heijn and PostNL, have “seamless” actions already available as early adopters, smaller companies might be willing to make an investment as well. A product manager we interviewed commented: The goal for Google, when they’re launching in a new country, is to have many local parties present on their application so that when they launch, they can say, “You can also talk to . . .” and then name some local retailers.
Crucially, Google also presented the assistant as a Dutch assistant: “The Google Assistant not only speaks Dutch but is Dutch” (Google Nederland). This framing was central to the assistants platform imaginary (Van Es and Poell, 2020), a vision of seamless integration designed to enroll Dutch partners. It proved effective, as 28 Dutch companies were named in the official press release. In the words of an interviewee: “We really are a Dutch company, so we don’t want to develop English conversations.”
Beyond the promise that the assistant can be integrated into Dutch company culture, the assistant was also marketed based on its technological, communicational and entrepreneurial possibilities. When coding the interviews with the Dutch developers, we noted a number of visions of what the Google Assistant could be from the perspective of complementors: (1) new layer of communication, (2) instrument to generate publicity, (3) interactive platform, (4) customer value creator. Interview participants showed great enthusiasm to be among the first to test out voice as a new channel of communication: “we see that voice is an additional layer of communication which provides new opportunities.” Getting on board early, the Dutch developers saw this as an opportunity to collectively work through the kinks of the new technology: At the start, everybody was a bit clueless, all companies were struggling. That’s kind of a good opportunity to start as well then, instead of starting after a couple of years, if there are many companies already far ahead and then you’re trying to investigate how this works. Whereas now everybody was just investigating and figuring out how to do it. We did really want to be a launching partner and be there from the start and also just learn and experience this new channel.
What is more, our interviewees imagined that the collaboration with Google could generate positive publicity. In the words of one complementor: “From our perspective, we thought, ‘It’s very nice to be a launching partner, because then you also get the free media publicity’. When they say, ‘We’re launching with . . .,’ and then you might be named with your company.” Having a company’s name appear next to the Google brand also portrays them as cutting-edge.
Beyond publicity, complementors elaborated on their ambitions for using voice actions to create multimodal, interactive interfaces. The combination of chat and voice was seen as a potential solution to the limited interactivity that screens offer when one is doing other things: [Google] enables you to use the assistants, the agents in different situations. In situations that you couldn’t before, while cooking, while driving, while doing whatever. It doesn’t really matter. And that’s why we thought that the user should always have a choice.
Another interviewee delineated their vision to use the Google Assistant as “a full-service shopping and cooking assistant, by adding the intents, which are available on Google Home to the customer service assistant and get the ball in all our channels.” Moreover, interviewees hoped that the Google Assistant could alleviate some of the load born by customer service employees and offer end users hassle-free answers to their queries: “I think voice can be a nice channel to get your answers to your customer service question, instead of having to call or something like that.”
Some of the most aspirational visions dealt with lowering the threshold for access for users of different age groups, using the Google Assistant to create social impact. For instance, interviewees envisioned using voice as a communication channel for children to limit their screen time: “parents were very enthusiastic about smart speakers because they saw it as an alternative to screen time, and that it really helps stimulate the imagination of children if they listen to a story.” On the opposite end of the spectrum, our informants remarked that elderly users can also profit significantly from voice actions. Whereas screen interfaces are not intuitive for older users “one of the advantages of voice. . .is that it’s very natural. So it overcomes these technical challenges. Because you can talk to a smart speaker and it talks back.” Some interviewees also maintained that voice interactions could prove critical in emergencies: “my mom lives alone. What if she has a speaker and when she’s lying on her hard floor with a broken hip and she can say, hey, Google, call my son. I love that idea.”
These visions reveal how complementors co-constructed imaginaries alongside Google. They often imagined their investments as cumulative, expecting that early investments lead from experimentation to consolidation. As we shall see, these expectations were later undermined by Google’s strategy of redeploying infrastructures and narratives. What Jasanoff and Kim (2015) call “performative scripts,” played a vital role in coaxing leading Dutch brands to get on board early, despite the limited capabilities of Google Assistant during the starting phase. One interviewee compared their company’s application integration through the assistant to the music recognition application ‘Shazam’: Shazam used to be a discrete app that you went to and then you had a lovely logo, and it had a name that you had to remember and it was on your screen. Now it’s built into Siri. And so you can use it to recognize music, but you don’t even need to know that Shazam is there. All right. So I remember thinking, oh, I don’t know if lots of things that we consider to be necessary today will be necessary in the future, they just become kind of part of the furniture and voice.
Even though Google’s partners did not have a clear image of the Assistant’s capabilities during the onboarding process, the idea of the assistant as a platform that enables voice communication with clients was enough to get complementors on board. Thus, platform imaginaries do not just project possible futures; they nourish expectations and motivate the mobilization of resources, despite persistent uncertainty.
Challenges
While both Google and Dutch complementors imagined Google Assistant as a culturally embedded and technically seamless platform, in practice, developing the assistant was rife with challenges. In this section, we examine these challenges through the eyes of our interviewees and their distinct professional roles in Actions projects. Coding the interviews, we identified the following themes: a) linguistic errors, b) a shrinking market, c) lack of feedback, d) problems with “Actions,” and e) gatekeeping data. These challenges demonstrate how localization strategies and governance arrangements shaped complementors’ ability to sustain investment in the platform. Following Popiel and Vasudevan (2024), they can also be read as forms of friction: uneven encounters that emerge when global infrastructures are scaled across national markets.
As a multinational behemoth, Google prioritizes international over local markets. Faced with limited resources, they first invested in the more lucrative and data-rich English market. This elicited disappointed reactions from Dutch complementors: “I can imagine that their priorities are somewhere else. Right. . .So why would they enable exploration of third-party apps?” According to another interviewee, it takes “a year or longer before new developments get to the Netherlands. That’s difficult.”
This delay had a negative impact on Dutch voice recognition models, causing linguistic errors. When English models were adapted to Dutch, there was a more restricted training set, and the accuracy of speech-to-text was lower. A UX designer explained that their company’s brand was compromised by flawed translations: sentences were translated “very strangely” and were “grammatically incorrect,” which led to widespread dissatisfaction even at the early stages of development. For professionals working with copy, “satisfactory accuracy” was not enough; errors became frictions that compromised their labor and the reputational stakes of their companies. Yet, software developers in the same firm disagreed: “we’re happy and satisfied with accuracy.” This divergence illustrates the fragility of localization imaginaries and the frictions of cultural translation: what seemed a minor technical compromise to engineers was perceived as a brand-damaging obstacle by copywriters and managers. As Kotliar (2020: 930) reminds us, global systems often assume that cultural differences can be collapsed into shared “lifestyles.” In the case of Google Assistant, this assumption, embodied in English-first infrastructures, produced frictions that eroded trust in the cultural specificity that the “Dutch” Assistant had promised. In the case of Google Assistant, cultural translation errors were not anomalies but predictable frictions of scaling English-first infrastructures across smaller language markets.
These frictions highlight a paradox. All actors had an interest in the platformization process to succeed: Google wanted its Assistant embedded on as many devices as possible to channel users to its wider ecosystem, while complementors hoped voice could open a new layer of communication for content and services. Yet, in this process these actors had different priorities. Google, as in the case of all its platform services, aimed to standardize and automate support to be able to scale and get as many end users and complementors on board as possible without having to heavily invest in human support. Many Dutch complementors, however, struggled to get the new technology to work, needing direct communication with Google staff and labor-intensive localization to connect with end users.
While Google provided Actions on Google, Dialogflow, and SDKs with training materials, the implementation process was shaped by opaque and uneven feedback channels. For some interviewees this did not matter. A marketing executive recalled: The documentation of Google was really useful for me when I started, I wasn’t a programmer, I couldn’t read this documentation, it was so technical. So I really started at the basics, seeing the video, seeing what they did exactly on the screen.
Yet others struggled with the absence of feedback. Especially smaller firms only had access to generic support: “it’s kind of hard to get in touch with someone meaningful from Google. . .Maybe a customer service employee. But that’s as far as you can get.” One participant described the communication with Google bluntly: “The randomness of it, and the complete disillusionment it is to work with them.” Another recalled needing “five submissions before we got it approved.” These kinds of frictions expose the unevenness of encounters between dominant platforms and dependent actors (Poell et al., 2021). Here, opaque quality controls and disjointed communication channels undermined the process of platformization.
Data access presented another layer of friction. Complementors received only partial transcripts and basic usage counts. One interviewee lamented: “They sell the smart speakers. They get the data, they make money, they make money on smart speakers and they make money with the data they receive.” Another added that the dashboard, their main access point, was “often inaccurate. . .You could go back only one day or one week, and then it was broken again.” Some firms resorted to designing their own systems to track activity. These restrictions exemplify frictions in platform market relations, where information asymmetry entrenches dependence (Popiel and Vasudevan, 2024).
Furthermore, neither Google nor complementors foresaw that end users were not ready to get on board. Instead of growing, as one interviewee observed, “the market within the Google Assistant seems to shrink” past the launch period. With fewer potential users, complementors had little incentive to develop complex Actions. As another interviewee remarked: “I think that’s the reason why we’re currently not doing more development on this channel. Because we see that the traffic remains kind of low. Google also acknowledges that.” By 2020, launch partners began slowing down third-party development. One interviewee noted: “we all work part time on the project. . .except for the business developer who was leading the project. But now she’s also part time on the project.” Another explained: “in smaller markets like the Netherlands, as soon as the primary supermarket and the primary e-commerce players stop developing actions, then the utility goes down.” Google consequently reduced investments, further decreasing engagement: “People don’t use it because there’s not enough utility on the platform and that goes in a vicious circle.”
This cycle illustrates the challenges in sustaining platform network effects, which depend on complementors profiting from end users getting on board and vice versa (Nieborg and Poell, 2018). On the assistant, complementors’ services, in contrast to social media platforms, remained invisible unless explicitly invoked, limiting discovery by end users and undermining both complementors’ visibility and Google’s ability to harness network effects. Interviewees associated this with the stagnation of user growth. One product developer pointed out that, in 2020, “the average reach” per week verged on “1,300 people.” “On a population of 17 million, that’s nothing. So it’s not a mass medium yet.” Here, the absence of ambient visibility became a structural friction that disadvantaged both sides of the market.
Underpinning the different priorities and strategies of Google and complementors is a deeper asymmetry in trajectories of investment and development. The Dutch complementors invested along linear paths, expecting a steady uptake of the platform to sustain their projects. One start-up founder noted that: “we started like a company that develops Conversational Actions because we were really thinking okay, Google Assistant will grow so much.” Google, by contrast, as it has frequently done with its other platform services, developed a circular strategy, redeploying infrastructures and resources when momentum slowed. This asymmetry produced a sense of precarity among the Dutch complementors. One of our interviewees captured this precarity vividly in a statement where they questioned the future of the Google Assistant as a platform: you really are almost completely at the mercy of the platform. And there’s no effort to make that an international standard or allow a user to defer their conversation to another agent or use a different piece of technology. And I don’t know how that’s going to play out.
These reflections make clear that precarity was experienced as dependence on a single platform, with no standards or alternatives to safeguard complementors’ position.
In sum, our analysis of the complementor perspective suggests new directions for a more nuanced understanding of what business scholars call “platform failure.” Based on our findings, complementors relied on Google in three fundamental ways: 1) access to boundary resources that allowed them to run their applications, 2) stable communication channels with the platform host, and 3) access to user data. When Google fell short on these promises, complementors could not produce viable services, which led to a shrinking market over time. From the perspective of Google, declining end-user uptake and complementor investment meant that a change in strategy was needed: cut losses and reinvest elsewhere. Hence, what transpires is not simply a story of unmet expectations but a structural asymmetry between complementors’ linear investments and Google’s circular strategies.
Repositioning
In June 2022, Google announced that it would terminate Conversational Actions on the Google Assistant. Subsequently, in December 2023, Google launched its proprietary large language model (LLM), initially named Bard and later rebranded as Gemini, to compete with OpenAI, Anthropic, and others. As a company undergoing ongoing digital transformation, Google’s success relies on sensing paradigm shifts. Declaring itself an “AI first company,” it timed the release of its LLM strategically, redistributing users to build critical mass for its new product (Akter and Iqbal, 2020). To date, Google has managed to “steer redefinition of its internal and external boundaries” (Plekhanov et al., 2023), shaping a renewed landscape where complementor partnerships may again become profitable. Ultimately, the Google Assistant platform failed to sustain user engagement and complementor commitment. In response, Google shifted its voice strategy toward Android, its highly successful mobile operating system. The impetus for this strategic pivot became even more clear in March 2025 when Google stated: “Android infuses Gemini directly into its core, to make the entire OS layer a powerful, personal, and secure AI experience for each user” (“The OS designated for AI”). Thus, we can see a thread running from Google’s earlier promotion of voice as a communication modality – via the relatively modest Google Actions Directory – to the current AI moment. The question is how complementors responded to these changes. As Akter and Iqbal (2020) argue, platform companies often recalibrate their strategies in response to shifting adoption curves. Google’s move illustrates this capacity for recalibration; earlier voice infrastructures and partnerships were not abandoned but reorganized within a wider strategic framework.
Following the termination of Actions, some complementors cut investments into voice technologies altogether, others started to develop voice assistant technology independently from Google’s boundary resources, while some followed Google Assistant into the Android landscape with renewed investments. These three pathways represent different visions and expectations pertaining to the future of voice technologies. They also complicate existing academic discourse around the dynamics engendered by the impact of platform governance on complementor strategies by revealing a greater level of agency on the side of complementors in processes of digital transformation (Plekhanov et al., 2023; Rietveld and Schilling, 2021).
In the first category, we encounter companies that discontinued experimenting with voice applications. A consultant that has advised multiple Dutch complementors elaborated on his predictions about their hesitance to encourage further collaborations with platform host companies, doubting that “third party services are really going to work in voice. I think it’s fundamentally broken.” They further explained that, whether through Android or other means, “there are some very big design infrastructure and security problems before we can get to an era where you can do something off the cuff with a third-party service.” Similarly, another Dutch complementor we interviewed stopped supporting voice assistant development projects indefinitely, citing low traffic on voice applications as the main reason. They note that every year more people get a smart speaker and are using voice assistance. But the number of users really talking to third parties on these devices is still quite low. That makes it difficult for us to generate money from this platform.
This indicates that, although a market opportunity exists for smart speakers, “product-related causes” stand in the way of platform model innovation, resulting in a flawed platform that does not meet user expectations. As third-party complementors flock to a platform and increase the value of the overall ecosystem “demand at the individual complement level” (Rietveld et al., 2019: 4) tends to decrease. In the previous section we documented these challenges: complementors are provided with limited customer data, they face delays in technical support and software updates and encounter opaque feedback pathways. These responses correspond with Van der Vlist et al.’s (2024) characterization of extractive platform practices, highlighting how data asymmetries and infrastructural constraints systematically disadvantage complementors.
In the second category, there are Dutch complementors who opted for developing artificial intelligence-powered voice and conversation technology independently from Google’s platform ecosystem. Instead, their focus was on local investments within the Dutch market. One complementor describes how their company diverted the budget they originally intended for Google Assistant actions into their own proprietary customer service chat assistant through a key partnership with Dutch public institutions. They note that this assistant will be “available on our website. It’s built on a different platform. We don’t use any Google tools for that. But we built our own conversational AI architecture, which is suitable for both chat and voice.” In 2020, another complementor spoke to us about the process of developing a proprietary assistant for their company, including software and hardware: There was no Google Assistant involved. You could say, “Hey, BOTNAME,” and then talk . . . through that speaker. We tested it at a festival . . . This was quite a success. We tested it among 800 visitors of the festival, who could test the smart speaker.
The same complementor ended their partnership with Google on voice projects in 2022 and has not implemented their voice technology through the Android platform.
Another interviewee working as a voice evangelist expressed their ambition to foster a self-sufficient Dutch voice market. Their core criticism of the leading AI companies was that “data models are lacking, they are not doing dialects,” which reduces the usability of voice assistants in Dutch. To that end, they started “Voice Commons” in collaboration with Dutch industry stakeholders to provide an open-source data bank integrating “retail, finance, the banks . . . academia, several universities. Everything except the agricultural market.” The purpose of the Voice Commons is for Dutch companies and institutions to collaborate on a national, multi-assistant voice infrastructure 4 . Hence, this category reveals that Dutch complementors who shift budget from Google Assistant to proprietary conversational AI are accepting higher upfront infrastructural costs to avoid dependence on Google’s tools and ecosystem. They can do this either through government funding or through their status as a leading incumbent in the Dutch market. By building their own architecture for chat and voice, these actors retain strategic autonomy over a core customer service channel, creating a credible exit strategy from the platform and strengthening their bargaining power in the long term (Van der Vlist et al., 2024).
The third category, which refers to complementors who continue to partner with Google via the Android platform, is largely speculative because public information is limited regarding ongoing Google collaborations with Dutch companies. Recently, Google has integrated a voice layer into Gemini through the Android platform as the canonical access point to third-party complements. At least one complementor we interviewed has since then restructured their business around the revamped Google Assistant SDK for the Android platform. They switched from developing actions for the Google Assistant to using the Google Assistant SDK in a web application format: “as a web application, you can embed in Qualtrics, let’s say, and you can use all the benefits of speech effects.” While well‑resourced Dutch complementors can reduce their dependence on Google by investing in their own infrastructure, this third strategy, primarily pursued by smaller firms, shows that they are more likely to remain tied into Google’s Android and Gemini voice stack because they lack the capital and risk tolerance to build independent systems. This uneven capacity leads to a stratified form of infrastructural dependence, where less powerful complementors are more exposed to extractive practices such as opaque data capture, enclosure of user relations, and shifting commercial terms imposed by Google (Uzunca et al., 2022).
The integration of Gemini into Google Assistant marks a significant inflection point in voice interaction technologies, reshaping the ecosystem for third-party complementors while advancing Google’s platformization ambitions. This technological shift exemplifies AI-driven infrastructural control (Van der Vlist et al., 2024), transforming complementors from standalone entities into dependent parts of Google’s ecosystem (Plekhanov et al., 2023). For these complementors, Gemini represents both opportunity and potential displacement, as advanced AI functionalities may reduce the need for specialized third-party applications while further consolidating Google’s position. When examined as part of a unified platformization strategy, these developments underline the broader dynamics examined in this paper: the fragility of localization strategies, the asymmetry between linear and circular investments, and the varied strategies through which complementors reposition themselves in response.
Conclusion
Through a case study on third-party developers for Google Assistant in the Netherlands, we have examined the role of complementors in processes of platformization. From this complementor perspective, platformization can be understood as a balancing act that relies on aligning expectations, interests, concerns, and objectives of a variety of actors. Complementors play a key role in this process, as their content and services form the connective tissue between global platforms and end users within specific cultural settings. Thus, the article highlights the importance of studying platformization not just as an economic, infrastructural, and governance process but also as a socio-cultural one. The analysis of this socio-cultural dimension adds a vital layer to our understanding of platformization and platform power, as continuously negotiated.
Documenting the experiences of Dutch complementors during the start-up phase of the Google Assistant, we have shown that Google did not succeed in navigating the different socio-cultural and techno-commercial challenges, culminating in the termination of third-party Actions in 2022. In turn, we observed three distinct responses on the part of the Dutch complementors: discontinuation of voice investments, development of independent voice technologies, and adaptation to Google’s new ecosystem. These divergent trajectories demonstrate greater complementor agency than is often acknowledged in platform studies, complicating existing accounts of how platform power and platform governance shape complementor strategies (Plekhanov et al., 2023). We also noted that market power structures how complementors can respond. Better‑resourced complementors had the option to develop and maintain their own infrastructures, thereby mitigating some dependencies, whereas smaller firms remained tied to Google’s evolving stack, exposing them more acutely to extractive data practices and unilateral changes. These different modes of platform-dependence need to be further investigated in future research; comparative, longitudinal studies on third parties with different resources should show how market power and infrastructural resources shape complementors’ agency, dependence, and exit options.
In addition, our case study demonstrates the fragility of platform imaginaries. Google’s narrative of a “Dutch” assistant, culturally specific yet built on English-first global infrastructures, succeeded in getting local partners on board but quickly unraveled in practice. Our findings show that these imaginaries were persuasive enough to attract investment, but unstable once confronted with linguistic errors, limited user uptake, and lack of support. This points to the structural risks faced by complementors when entering partnerships with global platform companies. The difficulties faced by Dutch third-party Actions developers should not be understood as isolated “bugs” in an otherwise functional system, but as symptomatic of the conditions under which platform complementors operate. As Popiel and Vasudevan (2024) remind us, “neither platform capitalism nor the subordinate processes of platformization are frictionless . . . Domination is never seamless. Instead, platform power must be continually exercised in the service of scaling” (pp. 10–11). From this perspective, the frictions that marked Google Assistant’s development in the Netherlands reveal the uneven terrain on which platform power is enacted and contested. Rather than frictionless adoption, the encounter between global platform strategies and local socio-cultural needs generated unstable alignments that shaped the trajectory of platformization in the Dutch market (Tsing, 2005). In turn, these frictions have inspired new AI initiatives in the Netherlands. A prominent example is GPT‑NL, a collaborative effort of the Netherlands Organization for Applied Scientific Research (TNO), SURF (the non-profit cooperative organization for education and research institutions in the Netherlands) and Netherlands Forensic Institute (NFI) to develop a Dutch language model. The aim is to strengthen the “strategic autonomy” of the Netherlands in the AI space (Lensink, 2025).
Taken together, developing the complementor perspective, the article contributes to the study of platformization in two keyways. First, it demonstrates the limits of accommodating cultural and linguistic specificity in global platform infrastructures: complementors’ local needs and imaginaries of a “Dutch” assistant were at odds with Google’s English-first priorities and limited support for third parties. Second, while Dutch complementors invested in the Assistant platform with the expectation that the platform would follow a linear development, platform companies often fundamentally reconfigure platform infrastructures and narratives to become or remain competitive in specific markets. This also happened in the case of Google Assistant. When the assistant failed to catch on with end users and complementors, Google did not double down and invest more resources but instead shifted its voice strategy toward the Android platform. Hence, we can observe a fundamental asymmetry between complementors’ linear investments and the circular strategies of tech companies, which frequently reshuffle resources and redirect platform development. To understand platformization as a balancing act, it is vital to analyze this asymmetry.
Footnotes
Acknowledgements
The authors would like to thank Prof. dr. T.B. (Theo) Araujo for his feedback and support in the early stages of this research project.
Data availability statement
The interview transcripts and qualitative coding datasets generated during and/or analyzed during the current study are not publicly available to protect the privacy of interview participants but are available from the corresponding author on reasonable request.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
Ethical approval and informed consent
This study was approved by the Ethics Review Board of the University of Amsterdam. The data analyzed were fully anonymized, and ethical considerations were in accordance with institutional guidelines.