Abstract
Keywords
Introduction
Influencer marketing has emerged as a central element of contemporary strategic communication, offering brands a powerful means to engage with audiences through social media influencers (SMIs). SMIs function as third-party actors who establish influential relationships with organizational stakeholders through content production, distribution, and interaction (Enke & Borchers, 2019). The global influencer marketing industry has experienced exponential growth, with its market size projected to surpass $32.6 billion in 2025, up from $16.4 billion in 2022 (Influencer Marketing Hub, 2025). This surge is driven by brands shifting their advertising budgets toward influencer partnerships, leveraging their ability to reach highly targeted audiences with content deemed authentic. In 2024, 69% of marketers planned to increase their influencer marketing budgets (Aspire, 2024). Short-form video platforms such as TikTok, Instagram Reels, and YouTube Shorts have further accelerated influencer-driven brand collaborations, making influencer marketing an integral part of digital advertising strategies (also in the non-profit sector, Duckwitz & Zabel, 2024).
Influencer marketing agencies (IMAs) play a critical role in this ecosystem by facilitating partnerships between brands and influencers, managing campaign execution, and providing analytics to measure impact. These agencies act as intermediaries that bridge the gap between commercial entities and SMIs. This sector has witnessed strong growth over the last years, with the number of companies projected to reach over 6,900 in 2025, up from 3,854 in 2023 (Influencer Marketing Hub, 2025). Despite this growth, IMAs face increasing market pressures. As intermediaries, they struggle with shallow value addition as their functions—such as influencer matchmaking, content amplification, and performance tracking—are increasingly commoditized. Moreover, challenges such as inflated engagement metrics, influencer fraud, and difficulty measuring long-term ROI have pressured agencies to prove their strategic relevance beyond administrative coordination. More prominent brands are exploring direct influencer partnerships, reducing their reliance on agencies.
In this situation, artificial intelligence (AI) is revolutionizing influencer marketing by enhancing efficiency and enabling data-driven decision-making (e.g., through AI-driven platforms automating influencer discovery, fraud detection, and campaign optimization, or virtual influencers, Allal-Chérif et al., 2024). 63% of industry leader expect to use AI in their influencer campaigns, with the main impacts being time savings in campaign management, its role in content creation, improvements in operational precision, and better influencer matching and selection (Influencer Marketing Hub, 2025). By automating repetitive tasks and providing actionable insights, AI empowers agencies to streamline workflows, optimize campaigns, and achieve better client outcomes. As AI reshapes the industry, IMAs must evolve by offering more profound value propositions, such as advanced data analytics, AI-powered content strategy, and real-time campaign adaptability. This is complicated by the industry structure, where most operational units are relatively small (IBISWorld, 2024). At the same time, AI can still be considered an emerging technology that evolves rapidly, and an understanding of (future) applications is not fully developed (Zahra et al., 2006). Thus, there is currently limited understanding of how agencies develop the capabilities to effectively adopt AI, a gap this study addresses.
The integration of AI within IMAs can be examined through the lens of dynamic capabilities (DCs), a framework introduced by Teece (2007) that explains how firms adapt to changing environments by sensing, seizing, and transforming opportunities. Specifically, seizing capabilities pertain to an organization's ability to capitalize on opportunities by allocating resources, making strategic investments, and restructuring internal processes to maintain a competitive edge. They are crucial for smaller companies that have to flexibly adapt to new technologies (Jafari-Sadeghi et al., 2022). The microfoundations of these capabilities—comprising decision-making processes, organizational structures, and managerial expertise—play a crucial role in determining how effectively agencies can harness AI-driven innovations. We therefore address the following research question: “How do influencer marketing agencies seize AI-based opportunities to enhance their services?”
This study explores the adoption of AI by German IMAs through an exploratory qualitative approach. Based on 18 in-depth interviews conducted in 2024 with agency leaders and influencer marketing experts across Germany, our research aims to uncover how AI-driven technologies transform industry practices. By analyzing the microfoundations of dynamic seizing capabilities, we provide insights into the strategic capabilities that enable agencies to remain competitive in an increasingly data-driven influencer marketing landscape. These insights shed light on how market intermediaries and organizations with limited organizational resources and technological domain expertise can seize the opportunities generated by emerging technologies. It also highlights how the existing business model and the position as a market intermediary influence dynamic seizing capabilities.
Literature review
Dynamic seizing capabilities in emerging technology ecosystems
The literature on DCs is extensive and has been applied across various business contexts. Helfat and Peteraf (2009) define DCs as “the capacity of an organization to purposefully create, extend, or modify its resource base” (Helfat & Peteraf, 2009, p. 1). This transformation is executed “intentionally and in alignment with strategic assumptions in such a way that a new bundle or configuration of organizational resources is created’ (Kump et al., 2019, p. 1150). DCs are particularly important in emerging technology environments, for which business models are still being developed, and often no best practices exist (Linde et al., 2021; McLaughlin, 2017; Schoemaker et al., 2018). At the same time, such insecure environments make it more challenging to routinize knowledge acquisition and resource expansion, that is, the development of dynamic skills (Helfat & Raubitschek, 2018; Linde et al., 2021; Mikalef & Pateli, 2016; Zabel et al., 2023).
One of the most widely adopted approaches (Schilke et al., 2018) is Teece's (2007) framework, which conceptualizes DCs as a set of sensing, seizing, and transforming capabilities that enable firms to adapt to changing market conditions. Previous research has shown that all three phases are relevant for innovation success and company performance (Ali et al., 2021; Bogers et al., 2019; Jantunen et al., 2018, Zabel & O’Brien, 2024). In addition to sensing, “seizing the opportunities with for example decisions on investments, acquisitions and business models” (Maijanen, 2022, p. 56) is crucial. Given the emergent nature, a “firm must often ‘invent’ solutions to survive. Both new and established firms engage in experimentation […], learning-by-doing […], trial-and-error learning […], and improvisation […] to deal with changing demands” (Zahra et al., 2006, p. 932).
A growing body of research highlights the need to extend Teece's (2007) framework beyond the firm level to capture the dynamics of firms embedded in digital ecosystems (Helfat & Raubitschek, 2018; Suominen et al., 2019). To date, research into the microfoundations of DCs has primarily focused on platform operators (Cenamor, 2021; McIntyre & Srinivasan, 2017). The less extensive research on the (much more numerous) complementors has tended to examine established, “mature” ecosystems (Baumann, 2022; Deilen & Wiesche, 2021).
Linde et al. (2021) provide a framework for how firms can orchestrate ecosystems to sense and seize opportunities, identifying microfoundations for each of the three DC phases. Although their approach focuses on focal firms, that is, the organizations at the center of ecosystems, it has already been applied to non-focal firms, for example, the more numerous complements in an ecosystem (e.g., Zabel et al., 2023). At the heart of this framework lies the seizing phase, which determines how organizations capitalize on identified opportunities. In digital markets, where interdependent actors drive innovation—from technology providers and regulators to end-users—seizing opportunities cannot be confined to a single firm's internal processes. Instead, it must be understood within the broader ecosystem context (Adner, 2017; Jacobides et al., 2018). The ability to seize effectively requires firms to mobilize resources, develop new competencies, and align internal structures with the demands of an evolving ecosystem (Helfat & Raubitschek, 2018).
Microfoundations of dynamic seizing capabilities in emerging technology ecosystems
Linde et al. (2021) propose four microfoundations of seizing capabilities in emerging technology ecosystems. As part of value proposition development, these microfoundations comprise understanding customer pains and gains within the ecosystem—ensuring that solutions address real needs rather than technological possibilities in isolation—and experimenting with offer configuration. Successful seizing also requires defining the firm's role and responsibilities in its ecosystem. In addition, resource allocation processes have to be defined and executed. In line with Maijanen (2022) the concept of microfoundations can also be applied to complementor firms, reflecting the more complex nature of value creation which depends on multiple stakeholders co-developing solutions (Adner, 2017; Jacobides et al., 2018). Previous research has identified routines/activities specifying these microfoundations for complementor firms, which can be structured according to Linde et al.'s (2021) conceptualization (see Table 1).
Microfoundations of Dynamic Seizing Capabilities.
Source: Own analysis.
Exploiting external knowledge involves engaging in market intelligence activities such as monitoring industry reports, attending trade fairs, and analyzing publications and social media trends, aiming to complement the existing knowledge base (Jantunen et al., 2018; Roseno et al., 2013). The focus lies here on the application of these insights to seize new market opportunities (Felin & Powell, 2016), for example, by adopting best practices (Mousavi et al., 2018) or by modifying product development (Kump et al., 2019; Plattfaut et al., 2015). This can also be achieved by collaboratively generating and integrating knowledge, where organizations engage with academic and other public research institutions as well as suppliers and consultants (Khan et al., 2020). Internally, firms must ensure that existing expertise and feedback are effectively mobilized and shared across the organization (Fainshmidt & Frazier, 2017). This activation of internal knowledge requires mechanisms such as formalized meetings, task forces, or job rotations, all of which create opportunities for employees to contribute insights (Mura et al., 2024). Moreover, firms benefit from knowledge management systems that combine existing knowledge (Bornay-Barrachina et al., 2023) and facilitate both top-down directives as well as peer-to-peer collaboration and exchanges (Felin & Powell, 2016).
Seizing capabilities are based on the firm's capacity to continuously refine its value proposition in response to customer needs (Plattfaut et al., 2015), for example, by incorporating customer feedback to enhance products and services (Fainshmidt & Frazier, 2017). In addition, organizations can leverage complementary capabilities with customers (Linde et al., 2021; Wilden et al., 2013), also deciding on the boundaries of their business model (Jantunen et al., 2012). Finally, A customer-centric approach allows firms to evaluate strategic fit and market potential rapidly and better (Jantunen et al., 2012; Roseno et al., 2013). Given the high importance of technological advancements, experimenting with offer configurations englobes the exploitation of technological knowledge for process and product innovations (Kump et al., 2019). This involves investment in new technologies, outsourcing specialized expertise, and engaging in structured experimentation (Lee et al., 2012), such as trial-and-error testing (Zahra et al., 2006) and more traditional approaches like road mapping (Khan et al., 2020).
To sustain long-term success, organizations must embrace business model innovation and adaptation (Jantunen et al., 2012). External resources may improve value proposition design (Heider et al., 2021). This involves exploring alternative revenue models by fostering closer collaboration with suppliers (Khan et al., 2020) and experimenting with different business approaches to identify the most effective strategies. Developing a strong knowledge base requires a commitment to capacity building, which includes investing in innovation-related employee training (Mousavi et al., 2018) and recruitment (Weaven et al., 2021), managing intellectual property (Katkalo et al., 2010), and continuously expanding organizational expertise. These efforts ensure that firms remain agile, adaptable, and well equipped to navigate technological and market shifts. Closely related is the challenge of managing co-specialization (Mousavi et al., 2018), which requires organizations to identify strategic partners and establish clear ecosystem boundaries (Khan et al., 2020; Linde et al., 2021). Effective co-specialization ensures that all participants in a business ecosystem contribute distinct yet complementary capabilities (Quayson et al., 2023). Integration is facilitated by developing a shared understanding and, in the extreme form, even a “sense of kinship” (Ince & Hahn, 2020).
Underpinning these strategic efforts is the need for optimized decision-making and governance structures. Organizations must prioritize opportunities, balance centralized and decentralized decision-making, allocate resources, and restructure governance frameworks (Khan et al., 2020) to effectively define roles, responsibilities, and operational rules (Eisenhardt & Martin, 2000; Jantunen et al., 2018). “Codified processes and tools” (Bornay-Barrachina et al., 2023, p. 5) allow for more rapid reactions to market changes. They may vary for internationally active organizations (Matysiak et al., 2018) but should be executed rapidly and timely (Wilden et al., 2019). Firms must allocate human resources effectively through dedicated teams or part-time employees or leveraging external expertise to support technological development and strategic initiatives (Linde et al., 2021; Zabel et al., 2024). Finally, fostering an adaptive organizational culture is essential for sustaining innovation. Creating transparent communication channels and clear collaboration rules builds trust among stakeholders (Fainshmidt & Frazier, 2017). A co-innovation approach may strengthen operational autonomy and shared values among contributors (Lee et al., 2012), encouraging knowledge exchange. These may also be affected by individual managers’ social and cognitive capabilities (Helfat & Peteraf, 2015) and changes in leadership practices (Jantunen et al., 2018). A culture that embraces experimentation and iterative learning allows firms to remain flexible and responsive in the face of uncertainty, ensuring their long-term success in an increasingly dynamic business landscape.
IMAs as market intermediaries
IMAs face several inherent challenges related to their organizational capabilities, learning processes, and business models, which are often exacerbated by their relatively small size. First, they serve as intermediaries connecting brands with SMIs to promote products and services in a triadic business relationship (Das & Teng, 2002). As intermediaries, they alleviate market inefficiencies. Their approach of structured flexibility—“formalized agreements in which a brand provides an organizational scheme designed to suit its campaign goals yet open enough for influencers to satisfy their unique needs” (Stoldt et al., 2019, p. 5)—allows for adaptation, but generates substantial transaction costs for negotiating, monitoring and adapting agreements (Luo & Donthu, 2007). The large number of partners on each side further increases these outlays: According to industry sources, campaigns, on average, included 15 SMIs (Aspire, 2024). Since each campaign requires tailored strategies, negotiations, and coordination, even standardized processes result in high operational workload and complexity.
IMAs create value by reducing transaction friction and providing expertise (Zabel, 2024), thus justifying their cut. They develop formal best practices and infrastructures that would be costly for brands to maintain independently. Nonetheless, they must continuously innovate this value-add. It is complicated that most IMAs also often operate with a limited scope of services, making it challenging for IMAs to differentiate themselves in a competitive market, potentially leading to the commoditization of their services (Lynch, 2019). Also, agencies can only generate limited customer lock-in, as demonstrated by the long-established customer policy of regularly tender contracts for renewal (Hill & Johnson, 2004). In addition, many IMAs are small enterprises, which can limit their capacity to invest in comprehensive competency development. This can be (partially) alleviated by becoming part of a larger international network of specialized agencies (Ceccotti et al., 2024; Li, 2020). Whereas developing technology tools or intellectual properties (such as databases) can be considered options to evolve the business model, the critical resource of advertising agencies lies with employees and their knowledge, especially creativity (Lynch, 2019). Here, IMAs experience high turnover and look for experiences related to customer industries, which hampers the ability to attract and retain talent with specialized skills (Ceccotti et al., 2024). This limitation can impede organizational learning and adaptation to market changes, affecting long-term sustainability.
Methodology
To uncover how AI-driven technologies are transforming the practices in IMAs, we conduct an exploratory study based on expert interviews with agency leaders and influencer marketing experts. The expert interview is particularly suitable for gaining insights into the topic beyond the results of the literature analysis (Bogner et al. 2009). It helps to identify the logic of actions in organizations and to explore emerging structures (Blöbaum et al., 2016).
The agency market structure in Germany is very scattered, with around 23,600 advertising agencies (Statistisches Bundesamt, 2022). IMAs constitute only a fraction of this market. Almost 90% of agencies employ fewer than ten people. No agency generates more than 5% of industry revenue (IbisWorld, 2024). In addition to full-service or digital agencies that have integrated influencer marketing as a unit or offer it as a service, some agencies specialize solely in influencer marketing. Since the basic population of IMAs is unknown, the selection was based on the list of the top-selling IMAs from reputed industry sources (iBusiness, 2024) on the one hand and experts from full-service and digital agencies on the other. Table 2 provides an overview of the interview partners.
List of Respondents.
The data for this study was collected between May and June 2024 through online interviews conducted in German by the authors and trained student researchers who had previously familiarized themselves with the interview process based on initial discussions (see the Acknowledgments section). With participants’ consent, all interviews were recorded and subsequently transcribed. The interview process followed a structured yet flexible approach. Each session began with gathering general background information about the interviewee and their organization, including their role, the company’s size, and its business focus. The discussion then shifted toward the company’s approach to AI adoption, covering four key areas: how IMAs gather AI-related information, how they integrate AI into internal processes, the partnerships formed to facilitate AI implementation, and the organizational changes undertaken in response to AI adoption. While a predefined set of questions served as a framework, interviewees were encouraged to expand upon their responses, allowing for spontaneous explorations of additional relevant topics. Any clarifications were provided only when necessary to ensure the interviewee fully understood the questions. The interview questions were informed by previous research.
Once the interviews were completed, transcripts underwent a quality review before being coded using MAXQDA. Following the Gioia methodology, a thematic analysis approach was applied to structure the data (Gioia et al., 2013). In the initial coding phase, first-order codes were derived inductively from the data based on insights from the literature. These first-order concepts were then consolidated into broader second-order themes, which emerged through an iterative process of analyzing patterns and grouping conceptually related codes thematically, according to the routines/activity categories identified in the literature review. Given the varying structure of interview responses, statements made in one section of the interview but relevant to another topic were reassigned to the appropriate category.
The coding categories were aligned with the microfoundations of dynamic seizing capabilities, as conceptualized by Teece (2007). An initial set of subcodes was generated through an independent review of the interview transcripts and notes. Coding was performed by at least two researchers, ensuring rigor through an iterative process of refinement and discussion. Instead of employing quantitative measures of intercoder reliability, a group consensus approach was adopted, in line with Saldaña (2021), to ensure a nuanced and comprehensive interpretation of the data. Following the coding phase, the data was further examined for recurring themes, with weekly discussions to refine the categorization and interpretation of findings.
Results
The results correspond to the four microfoundations identified in the literature review. To avoid repetitions, superseding aspects are reported where most appropriate.
Addressing pains and gains for customers
The analysis of the interviews shows that the employees of IMAs use a variety of external sources to expand their AI knowledge and bring it into the organization. Some are based on the employees’ initiative; others are organized by the agency. The employees use general journalistic sources such as magazines and newspapers, as well as specialized information resources such as weblogs, newsletters, whitepapers, podcasts, and tutorials from AI experts. Some respondents also use networks like LinkedIn, Slack, and Reddit to share ideas with other experts. Industry events, trade fairs, and local networking meetings are also mentioned. A third of the agencies surveyed offer further training courses, e-learning, or training from external coaches.
An important component of knowledge building is the use of AI tools themselves, with agencies using LLMs such as ChatGPT, Perplexity, Midjourney, and DALL-E for research, conception, and content production of texts, images, and videos. Some agencies work with tool providers to develop customized and individual solutions (see below). Overall, the respondents report intense pressure to innovate, to adapt quickly to AI market trends, and to secure a competitive advantage, underscoring the importance of this seizing microfoundation: “What are you trying to do as a company? You usually try to be a first mover or early adopter or something like that and build up as much expertise as possible as quickly as possible so that you have a competitive advantage. Moreover, that's why, and because generative AI, for example, is on everyone's lips, agencies in our industry are simply trying to build up this knowledge very quickly. […] That drives us, of course” (I 10).
However, cooperation with external knowledge or market partners is limited. The results of the interview analysis show that most agencies work with technology and tool providers (see below) but not with universities and research institutions—mirroring findings of other small complementors in emergent media technology markets, such as the metaverse. Overall, agencies instead rely on their knowledge acquisition.
Regarding the leveraging of internal knowledge, there are various measures to disseminate AI knowledge within the organization. Ten of the experts explicitly mention peer-to-peer exchange between employees. Slack, digital manuals, and e-learning are used as internal knowledge repositories. Regular internal workshops and meetings where employees present best practices, use cases, or new AI tools are established instruments in all the surveyed agencies. “We currently have a session like this every Monday. So once a week, where we exchange ideas: Who has created a cool use case with AI? How was it achieved? Which tool was used?” (I_11) “The task force workshops show that you can really get people excited about it too, once you show the potential, then you also get employees to give it a bit more of a chance” (I_9). “So there are always a few colleagues in the individual teams who are very committed because they find it so interesting, who then simply come up with new tools or use cases on their own and then simply test them.” (I_14)
Experimenting with offer configuration
Regarding the impulse to include AI in the IMA services, active customer demand plays a surprisingly minor role—a result grounded in the IMA business model (see below). Only six agencies report this as a primary driver for AI adoption. In two cases, customer needs strongly influence the use, as customers demand the use of AI tools to increase the agency's productivity (I_8) or the use of AI is specifically desired (I_12). While some clients demand the use of AI to appear innovative (I_3), others merely show a general interest (I_1). Consequently, there are no cases of customer co-creation regarding AI, which starkly deviates from other service-oriented B2B markets facing emerging media technologies, such as XR/VR.
However, when considering customer requirements regarding AI implementation, data protection and data security are the top priorities. Twelve of the agencies surveyed report that their clients have data protection rules that they must observe when using and implementing AI tools that restrict the possibilities for use. One expert emphasizes the challenge of having to comply with legal requirements on the one hand and drive innovation on the other (I_16). Some experts also point out that the dynamic development of data protection guidelines requires regular review. Ten agencies have developed internal regulations to ensure the data protection-compliant use of AI. One expert also mentions the obligation to label AI-generated content (I_17).
“Our motto is try, try, try. No one is going to come in from outside and take me by the hand. … it's a learning-by-doing approach.” (I_9)
These experiments’ findings flow directly into optimizing their processes, which suggests iterative learning processes.
Regarding technology, the agencies primarily rely on integrating external products and developments adapted to the respective needs instead of having to build up R&D resources themselves. While all agencies use AI tools from software providers, eight of them state that they work with them on individual adaptations that mainly concern data security or data sets, for example, in the area of target group and sentiment analyses (I_14), prioritization and processing of emails (I_6), search and management of influencers (I_4) and as part of the development and integration of AI algorithms in influencer management software (I_16). However, developing own technical solutions is not the goal (I_6, I_10).
When it comes to using AI for adapting the existing business model, the agencies link AI to their image of innovative ability, which is perceived as an absolute competitive advantage: “In the end, social media agencies thrive on being innovative and being first movers before corporations or other organizations can catch up. And there will certainly be a lot rolling towards us, and I think that we should be at the forefront.” (I_7)
As clients increasingly expect the cost of services such as research, content creation, and analytics to fall with AI, the adoption of AI puts more pressure on IMAs: agencies need to convince their clients that human work—especially in consulting and creative services—is of higher quality than AI. Several interviewees question whether the prevailing revenue model in agencies, where payment is based on time spent, is still viable. “Time has never been a good business model. So that's where we're stuck, so to speak.” (I_9)
Some interviewees also see potential to change the architecture of value creation by expanding the range of services through AI, for example, services such as translations and product placements. Another business model extension is developing and distributing virtual, AI-generated influencers, which have advantages in brand safety, even if these would not work in all application contexts. The aim is to establish outcome-oriented revenue models that charge for results rather than output calculated in hours. “And yes, if that helps us to get away from that a bit and really talk more about the result and say, what is the concrete result worth to you? That would be great, of course.” (I_7) “It's clear that it will change the business model. We are all finding out together just how far it will go. And the developments come so quickly that sometimes you don't even realize that something new has just happened. (I_7)”
Defining (ecosystem) roles and responsibilities
Given the tacit nature of AI expertise, one could assume that IMAs engage in the internal development of specialized AI experts to gain a competitive advantage. However, the special recruitment of employees with AI expertise does not play a role at the agencies surveyed. Only one of the experts states that employees with AI skills had been hired (I_18). Instead, the focus is on building up the knowledge and training of existing employees. Consequently, the agencies are also not considering replacing employees with AI or creating other jobs. Instead, the agencies hope that employees will be able to devote more time to their actual area of responsibility, that is, consulting and creation, if routine tasks are taken over by AI (I_11, I_14).
The partnering of IMAs in AI focuses on cooperation with software providers. In addition to pure AI software solutions such as ChatGPT, Midjourney, StableDiffusion, Perplexity, Google Gemini, Network AI, and Jasper.AI, providers that offer individual GPT solutions are also mentioned. In addition, established software providers are integrating AI functions into their existing products, including Salesforce, HubSpot, Miro, Canva, CreatorlQ, Infludata, Adobe, and Microsoft. These are typical complementor cooperations where off-the-shelf services are used. As already noted, eight companies also report that they are in direct discussions with the larger tool/platform providers, for example, exchanging ideas with employees about possible AI application areas or having individual solutions developed for the agency's needs. Software manufacturers, for their part, gather feedback from agencies as users to develop their products further.
The majority of the agencies see the need to work with external development partners to create and customize AI software: “Partnering, partnering, partnering, because in the end you won't be able to do it in-house. In other words, it comes down to partner deals. So, just like you have a performance marketing partner or a CRM partner, ideally, you already have AI service providers whom you know that you used to commission for data topics or analysis topics; they can usually also do AI or specialized players or consulting.” (I_9)
Interestingly, there are no collaborations with competitors besides professional exchange at industry events or via online and social media platforms. Collaborations with universities or research institutions enhance reputation or employer branding more than a joint development partnership. One respondent (I_10) reported hackathons, where developing new ideas and solutions reveals new partnering opportunities.
However, the selection of partners depends heavily on compliance with data security criteria, all the more so as the agencies’ client structure requires this. This again underscores the directional role of the business model for dynamic seizing capability formation.
Establishing resource allocation processes
Two different types of decisions regarding the implementation of AI tools can be identified from the interviews: a quick, flexible decision made via short decision-making channels based on ad hoc analyses (7 experts) and a structured decision made across several hierarchical levels based on a systematic evaluation (11 experts). The type does not seem to depend on the organization's size; sometimes, units are in tension between ad hoc decisions and systematic processes (I_16).
Accordingly, different decision criteria can also be observed, ranging from a subjective gut feeling to carefully examining cost-benefit ratios and legal framework conditions. For agencies working for data-sensitive clients, these often lead to the exclusion of AI tools, while others try to find a balance between benefits and legal security. In addition, one expert points out that legal frameworks, particularly around copyright and licenses, change frequently, which is why regular reviews are necessary, especially when working with large corporations. Overall, data protection and security, as well as AI regulation, play a significant role, causing concerns on the part of customers and sometimes preventing AI integration. “The implementation of AI tools also depends on the specific use cases in which they can be used, how the output quality is assessed, and how error-prone they are. Several respondents also emphasized user-friendliness. “If it cannot be adapted quickly, it is discarded, and a new one is tried out.” (I_6) “Which work annoys you the most, and which takes you the most time, and then we made a list of unloved tasks, let's say. And then we sorted this list again, according to which of them could possibly be taken over by an AI, or which of them could simply be automated.” (I_11) “AI saves time, but AI is still work. You have to deal with it, you have to train people, you have to integrate the tools into the company, and the tools themselves also cost money, and these costs have to be recouped somehow, and we must have the feeling that what we invest in AI, we will get back.” (I_7) “The responsibility for deciding whether to purchase an AI tool for a fee lies with the management level, but they usually follow the recommendations of the team or the AI task force. Whether and to what extent the individual employees use the AI tools in their day-to-day work is usually up to them and depends heavily on the specific area of responsibility.” (I_12) “So there are no criteria in that sense. Everyone simply decides for themselves what is relevant for me, what I can work well with or not.” (I_8)
Overall, the most frequently mentioned decision criteria are increasing creativity (10 experts), availability of resources (5 experts), functionality, and user-friendliness (4 experts).
Regarding governance optimization, integrating AI tools in agencies requires careful consideration of both external and internal regulations. Twelve of the agencies surveyed refer to external regulations relating to data security. These are also reflected in the internal AI guidelines of 10 agencies, for example, on documenting the use of AI, not to use private accounts (I_6), and the mandatory use of spellcheckers (I_1). The targeted introduction of AI guidelines shows how agencies are creating structures to control the use of AI tools and optimize integration in line with internal and external requirements.
Most agencies allocate resources for AI implementation through task forces, or they identify individual AI experts who can focus on potential AI projects. Even though all employees should be provided with the necessary resources to continue their training, the intensity of involvement with AI depends on the motivation and individual decisions of individual employees as to whether the use of AI makes sense for the use case at hand.
However, strategic road mapping can hardly be identified apart from the overarching goal of using AI tools to remain competitive. Human resources are primarily invested to test whether specific AI tools make existing processes more efficient so that the investment can be amortized quickly (I_6). The size of the agency or the network in which the agency is integrated does not appear to play a role here either: “Perhaps to answer this comprehensively, there is currently no cross-company strategy to say that AI must be implemented here in this way or that way, because I think it depends very much on the department or profession as to which tools and processes are used (I_17).”
In general, IMAs consider themselves well-positioned in adapting to organizational culture. According to the interviewees, their industry constantly changes, meaning that agencies must always react quickly to changes. In the case of AI, it is consistently pointed out that the early use of AI contributes to their image of being innovative. In addition, the competitive pressure to reduce costs through AI is palpable.
The experts see the prevailing work culture in agencies as a particular advantage in adopting AI. The perceived “lively communication culture” (I_14) and “open feedback culture” (I_4) ensure that valuable insights and best practices from individual employees and teams are regularly passed on and discussed, which supports the continuous improvement of work processes. “Agencies have a bit of a reputation for being a bit more relaxed, but basically, you can tell people what to do, but people do what they do. In the end, you try to manage it. (I_9)
Discussion
Our study illustrates how IMAs, as market intermediaries, develop dynamic seizing capabilities to reconfigure (or, more precisely, defend) their role in the value chain. It provides three main contributions to theory and practice.
First, our study highlights how the business model type and position in the industry value chain impact the development of seizing capability. As small actors in the digital marketing market, IMAs possess limited strategic power. They might be considered at the mercy of larger platform operators like Google or Meta (Cenamor, 2021). At the same time, they perceive strategic pressure not only from these platforms or their competitors but also from their customers and the commoditization of their services (Lynch, 2019). Therefore, the build-up of requisite seizing capabilities is strongly governed by the existing business model and the position of IMAs as market intermediaries. Clients limit the options of IMAs embracing AI through stringent requirements (such as data protection), not sufficiently rewarding AI product innovation from IMAs, and limiting customer lock-in options. Therefore, seizing capability development is strongly oriented towards efficiency and increasing the perception of innovativeness, which might help to attract and retain employees and customers. IMAs have to tread a fine line here since the revenue model based on work volume creates the implicit threat that customers will appropriate the generated savings in future contract negotiations. This is highlighted by the fact that customer co-creation could not be observed—a feature that is commonplace in other service-oriented B2B sectors, but where firms possess a higher ability to appropriate a part of the generated returns (for XR/VR, see Zabel et al., 2024; Zabel & Telkmann, 2024). This dilemma might increase since AI adoption will expose IMA's core services to Baumol's cost disease, thus making them relatively more expensive. A similar development can be stipulated for other media sectors with high labor inputs, such as investigative journalism (Wellbrock, 2024). While in previous technology adoption circles, agencies could leverage emergent media technologies (e.g., the web, mobile, social media) to create new markets while serving their existing clients, AI, as a general-purpose technology, does not offer this option. The active development of new business approaches, collaboration with suppliers, and experimentation with revenue models to respond to environmental changes described in the literature (Jantunen et al., 2018; Khan et al., 2020; Mousavi et al., 2018; Mura et al., 2024; Roseno et al., 2013) was therefore only observed at a hypothetical stage. IMAs thus engage in what could be termed ‘intermediary seizing’: Leveraging AI opportunities (like AI-based content tools) for more efficient value delivery, whereas capabilities focused on new or alternative forms of value creation are much less developed. This extends the existing literature on DC formation in digital markets, which often examines large organizations (Kroh et al., 2024; Selander et al., 2013) or firms with a deep value chain integration (Aramand & Valliere, 2012; Arend, 2013; Berman et al., 2024) by focusing on mid-chain actors like agencies.
From a theoretical perspective, the absence of customer co-creation observed in our study also contrasts sharply with expectations derived from Service-Dominant (S-D) Logic, which posits that value in service ecosystems is co-created through the integration of resources among multiple actors (Vargo & Lusch, 2004). In marketing and advertising services, co-creation with clients is typically central—agencies rely on close client collaboration to co-develop campaign goals, creative direction, and brand messaging (Hughes et al., 2018). Anyhow, if the service is sold as an outcome (as in the case of AI adoption), the process co-creation seems to be less central to success. Thus, our study helps to direct attention towards the boundaries of co-creation. Such conditions include lack of trust or common ground, fears of opportunism or lock-in, and asymmetrical dependence (Hawlitschek & Hodapp, 2024). The study adds to the literature on value co-creation in service ecosystems by providing a real-world counterexample to the implicit assumption that service firms naturally co-create with their customers, by demonstrating that asymmetric power relationships and associated risks might limit co-creation not only from the customer side (Appiah et al., 2021).
Second, the study's context of influencer marketing—a creative, fast-moving media domain—adds to the DC literature in creative industries. Prior research in media businesses noted that innovation capability and adaptability are key to performance (Jantunen et al., 2018; Zabel & O’Brien, 2024). Regarding the nature of dynamic seizing capabilities, systematic, agile, decentralized, and iterative processes could be observed in knowledge building, decision-making, and resource allocation, which are sometimes initiated top-down and sometimes bottom-up in the same organization. This aligns with Teece's (2007) theory, which emphasizes the importance of flexibility in the early phases of technological development before a dominant design is established. Whereas elaborate decision protocols such as cost–benefit analyses could be observed, subjective criteria, such as gut feeling and personal assessments, dominated, especially for creative processes where quantitative targets are challenging to define. This practice aligns with DCs for new business ventures facing similar uncertainty and resource constraints (Roseno et al., 2013). These findings could also be replicated for the partnering dimension. Here, analytical approaches again give way to informal and unstructured collaborations. Instead of processes or routines, trust and shared values (Ince & Hahn, 2020) and personal connections (Kleinbaum & Stuart, 2014) play a decisive role. This keeps partnerships unbureaucratic, flexible, and time- and cost-saving—at least in the early-testing stages, and when overall market turbulence is very high (Zabel et al., 2024). One reason may be the fast-moving, often project-based, and interconnected agency business, which does not correspond to the long-term partnership models of research institutions or that of technologically driven complementors (Khan et al., 2020; Zabel et al., 2023). The study thus underscores that DCs can be developed even by resource-constrained, dependent actors through ecosystem connections (Bicen & Johnson, 2015; Zabel & Telkmann, 2024), an insight that both complements and complicates traditional DC theory, which often focuses on firm-internal processes. From a broader perspective, the study underscores the central role of organizational culture as a DC in media organizations facing emerging media technologies (Jantunen et al., 2018; Küng, 2024). For the IMAs, AI adoption was seen as a differentiator for competing with other agencies and attracting talent (in addition to leveraging efficiencies). The IMAs could build on an agile organizational culture facilitated by the agency business model, which rewards flexibility and interconnectedness. This allowed IMAs to embrace AI despite limited resources and a market position severely limiting the potential for appropriating returns from AI innovations. Anyhow, the seizing microfoundations do not seem adequate to develop “deeper” uses of AI; the retention of key human resources poses a challenge.
Third, from a practical perspective, the study highlights the challenges and potential remedies for IMAs as specialized service providers. Due to their service character, their central function is to offer customers added value in a targeted manner through their in-depth expertise and extensive networks (Cecotti et al., 2025). Agencies could use their strengths as networking service providers and advise clients on suitable network partners for individual needs and trusting partnerships. The seizing capability can be expressed not only in the in-house development of technologies but also in the skillful integration and adaptation of external solutions, which is particularly important in the dynamic field of influencer marketing. At the same time, the evolving demands of AI-related activities—where the firms in our study are in the early stages of seizing—might require a more structured approach to allow for the build-up of technological expertise or the introduction of AI innovations in other fields. This could include strategic road mapping, developing coherent models for assessing long-term success instead of pure trial and error and strengthening internal knowledge building by establishing AI task forces of experts and leveraging external partnering.
Conclusion, limitations, and further research
As intermediaries in a dynamic market environment, IMAs face the challenge of adapting quickly to changing technological conditions. AI has the potential to speed up processes and change working methods. The results show that IMAs are in the early stages of seizing, where various (mixed) DCs can be observed. The microfoundations of dynamic seizing are characterized by a decentralized, agile and flexible approach to absorbing knowledge, decision-making and integrating new processes and technologies. The focus is on practical evaluation criteria such as increased efficiency, quality, and user experience. However, the aspects of data security and data protection play a significant role, with top-down and systematic rules also being implemented. The interviewees hope to evolve their business model by shifting value proposition towards more creativity and consulting, by changing the revenue model from time to value, and by expanding value creation. These changes are expected to shield IMAs from increasing efficiency pressures, thus protecting and even increasing long-term profitability (Ceccotti et al., 2024). Alas, these changes have not yet been implemented by the companies in the study. On the contrary, IMAs may fear increasing pressure from clients aiming to appropriate a part of the efficiency gains generated by AI.
Despite its contributions, the study has several limitations that open avenues for further inquiry. The research focused on a specific segment—IMAs—which are often small, marketing-focused firms. The findings, while rich, are based on this niche context and may not directly generalize to other types of platform complementors (e.g., app developers or larger digital agencies). The sample of IMAs was limited (both in size and geography). Local market conditions might influence the DCs observed. Future studies should be cautious in extending these results to all small firms or all emerging technology ecosystems without further validation. From a methodological perspective, there is an inherent subjectivity and potential respondent bias. Agency leaders enthusiastic about AI might overstate their adaptability, while those struggling may have been underrepresented. The cross-sectional nature of interviews provides a snapshot in time; however, DCs develop over time. Finally, the study is grounded in DCs, a broad lens. This focus might overlook other explanatory factors, such as leadership traits or even luck in hiring a key tech-savvy employee—factors not deeply theorized by the DC framework. Additionally, the platform-centric nature of IMAs means that platform governance (e.g., sudden changes in API access or algorithms) can critically impact them.
Future research could compare DC development in different types of emergent technology ecosystems. Do small complementors in other ecosystems (for example, independent game developers on app stores, or third-party sellers on e-commerce platforms) show similar sensing patterns and seizing with emerging technologies? Comparative studies could reveal which findings are unique to IMAs or other intermediaries representing generalizable patterns of small-firm innovation in platform economies. This research stream could also analyze the effect of different forms of corporate integration (e.g., international agency networks). A valuable extension would be longitudinal studies that track IMAs (and similar firms) over time as they adopt and integrate AI. This would illuminate the process of capability building—how do sensing capabilities lead to seizing actions, and how do those translate into transformative changes or performance outcomes down the line? Long-term data could also capture how these agencies respond to major ecosystem shifts (e.g., a new social platform rising or a significant AI breakthrough) and whether initial early adoption of AI leads to sustained competitive advantage or temporary efficiency gains. Finally, the interdependence between IMAs and platforms could be extended to include the triadic business relationships. This could involve interviews on the platform side or multi-actor case studies, including interviews with brands, influencers, and platform representatives. It would extend the current research by situating the agency's capabilities in a broader network of relations, aligning with ecosystem strategy theories.
Footnotes
Acknowledgments
ORCID iDs
