A system and learning perspective on human–robot collaboration

Quickly developing artificial intelligence (AI)–based devices and digital agents (e.g. chatbots, virtual agents) offer the potential to automate a wide range of processes and operations and thereby alter how people perform their jobs (Brookings, 2023). The robot sector is predicted to expand rapidly in the next few years, including projections that it might reach a value of $260 billion by 2030. Its main applications thus far involve customer care (Statzon, 2023), because robots can offer round-the-clock help, individualized aid, effective problem-solving and tailored recommendations (Hsu and Lin, 2023; Ranieri et al., 2024), all while gathering detailed client input in a variety of industries, such as hotel, retail and healthcare (Lu et al., 2020).

Service research already offers extensive findings regarding the interaction between humans and robots. Most of these studies adopt a dyadic perspective, addressing the superiority of one or the other (e.g. Leiño Calleja et al., 2023) or stressing the implications of automation and substitution by robots (Belanche et al., 2020). Arguably though, we are at the point that we should move beyond a focus on interaction to address collaboration (Li and Zhang, 2023; Paluch et al., 2022). In cooperative, collaborative efforts, human actors and robots enhance each other’s complementary strengths, namely, the leadership, teamwork, creativity and social skills of the former and the speed, scalability and quantitative dynamic capabilities of the latter (Mele and Russo Spena, 2024). Thus, they might accomplish advanced, shared tasks that entail greater complexity and personalization, as are commonly required for effective service delivery.

Some studies of frontline employees accordingly have adopted a collaborative view, though without in-depth analyses of the impacts of such collaborations on companies and work practices. In response, one school of thought proposes the notion of a constitutive entanglement of human actors and robot actants, according to a system perspective (Mele et al., 2021; Mele and Russo Spena, 2024). An actant refers to a nonhuman entity that participates in shaping and influencing the system. From this perspective, a critical question pertains to how collaboration affects system development and dynamics (Williams, 2020). Taking a system lens also requires acknowledging multiple elements and the performative features of the (service) system. We postulate that such a view can enrich understanding of robot adoption.

The process of introducing a new robotic technology into a service system is complex, so providing continuous maintenance and support for this process, and its impacts on work practices, can be challenging (Liu et al., 2022). When robots enable service provision, human workers may become concerned about job displacement or role shifts. Furthermore, it can be difficult to scale up robot deployment across multiple service practices or extend its capabilities to include various jobs (Belanche et al., 2020). Therefore, studying human–robot collaboration (HRC) from the perspective of service employees is vital for understanding and adapting work practices. Employees’ insights can provide a fuller understanding of how roles, responsibilities and service delivery processes evolve as robotic technologies get integrated into service systems. With such understanding, researchers can develop more nuanced and practical insights for managing transitions toward HRC in service systems. A systemic perspective depicts the process of robot adoption as a series of attempts to alter the status quo to fit new schemas, processes and institutions, which result in new work practices. Therefore, in our attempt to extend scant knowledge of the robot adoption process and collaborations of robot actants with human actors in work practices, we adopt such a perspective and pose the following research question:

To address this question, we carried out a two-year longitudinal analysis of changing work practices (i.e. customer care) after the introduction of digital agents (i.e. chatbots) in an international company specializing in developing and producing natural health products. To understand and elaborate on the findings, we leverage expansive learning theory (Engeström, 2001), which emphasizes activity systems, defined as intricate networks of interconnected factors (e.g. people, tools, regulations, the division of labor). By examining how HRC evolves over time using Engeström’s model, we can trace the progressive integration of robots into employee teams and the consequent transformation of work practices.

This study in turn makes two main contributions to extant understanding of HRC within systemic activity frameworks and expansive learning. First, we conceptualize the human–robot activity system as a foundational element for enhancing customer care practices. In such a system, collaboration between humans and robots is integral and organized around shared goals; it is mediated by tools, rules and the community. Second, we explore the evolution of this activity system according to an expansive learning process, which involves a critical examination of existing work practices and an analysis of how HRC transforms such practices. This process begins by identifying and addressing the contradictions and tensions between current human work practices and robotic capabilities, which often reveal challenges and opportunities to improve HRC.

These results emphasize the systemic and dynamic nature of HRC by placing it within broader contexts of activity systems and expansive learning. As this study shows, collaborations between humans and robots entail an expansive performativity that extends beyond the traditional roles or tasks of either actor or actant. It thus spans a diverse range of objects, tools, procedures and institutional setups, culminating in transformations of working activities and innovative customer care practices.

In service research, discussions of HRC have largely replaced traditional considerations of human–robot interactions (HRI), emphasizing their cooperative and synergistic relationships. These discussions involve two main lines of research (see Table 1): (1) the dynamic roles and tasks performed by human actors and robot actants, which reflect efforts to leverage their respective strengths for enhanced performance and efficiency and (2) questions about how collaborative efforts affect work practices.

An early debate centered on analyzing robots’ roles and tasks. Technology can assist both customers and employees in performing activities more effectively, which leads to enhanced capabilities and richer interactions (De Keyser et al., 2019; Leiño Calleja et al., 2023). Thus, (service) robots and chatbots are increasingly deployed to enhance customer service, often substituting for human agents (e.g. Le et al., 2024; Lu et al., 2020). These studies highlight robots’ abilities to provide personalized assistance, handle routine inquiries and manage complaints, in ways that enhance the customer experience (Chung et al., 2020; Ranieri et al., 2024).

When they explore the interactions of robots and employees in specialized service tasks, scholars also emphasize demands for more collaborative efforts (Paluch et al., 2022). In this context, collaboration involves more than task execution; it includes crucial aspects of training and supervising the technologies, as performed by humans (Mozafari et al., 2022; Xiao and Kumar, 2021). The dynamic interplay of service robots and frontline employees requires humans to engage in demanding roles to augment robotic services, which in turn likely require reskilling initiatives and efforts to encourage proactive attitudes among the employees (Odekerken-Schröder et al., 2021). In retail settings, technology embedding likely involves a collaborative process of co-evolution and effective knowledge transfer to support employees’ adaptation to the new technologies. Engaging employees early in the AI integration process is crucial for limiting resistance and ensuring smoother transitions (Bonetti et al., 2023). For example, the frontline employee–robot interdependence (FLERI) framework (Le et al., 2023) identifies three structural components that encourage employee adoption: shared goals, common workflows for coordination and collaborative decision-making authority. This framework supports effective integration for service-oriented tasks and identifies collaboration between frontline employees and robots as vital for enhancing the customer experience and improving service delivery. It also recognizes the challenges involved in balancing efficiency with customer-oriented service and managing employee attitudes toward robot collaborations.

Recent research highlights the complexity of integrating AI into workplace practices and emphasizes the importance of addressing human aspects to maintain a healthy and productive work environment, underscoring that when algorithms are used alongside human managers, employees' prosocial motivation, such as the desire to help others, is reduced (Granulo et al., 2024). Employees’ willingness and customers’ readiness to collaborate with robots both are critical for successful adoption (Xiao and Kumar, 2021). Some research demonstrates that customers prefer services in which digital employees support rather than replace human employees, suggesting the need for human involvement, especially in decision-making processes (Le et al., 2024). This preference stems from increased transparency and cohesiveness provided by collaborative efforts by human and digital employees. Coordination cues that emphasize the orchestration of their tasks, team goals and joint commitments to customer service also strengthen perceptions of process fluency and seamless service experiences (Chong et al., 2021).

In healthcare contexts, substantial research has considered how HRC affects work practices by analyzing the integration of cognitive assistants (Mele et al., 2022). Employees must adapt their workflows to use the technologies effectively and allow the cognitive assistants to complement their tasks. Thus, the employees need training in using cognitive assistants, understanding their functionalities and leveraging them for improved patient care. Cognitive assistants, as boundary objects, can facilitate collaboration among various actors (patients, caregivers, healthcare professionals), resources and activities, fostering learning and knowledge sharing to increase value co-creation and patient care (Mele et al., 2022; Mele and Russo Spena, 2024).

Notwithstanding these important contributions, some shortcomings persist. First, we note insufficient consideration of the system within which activities occur. Although research investigates how robots collaborate with employees and customers to execute tasks efficiently, it frequently neglects the surrounding organizational context. Second, we lack sufficient understanding of the impact of robot adoption on work practices. The process of robot adoption is rarely investigated as a progression, aimed at adapting to new frameworks, procedures and institutional setups and then ultimately leading to novel work practices. Such an understanding requires a significant shift, to include transformative and dynamic service systems. Therefore, we adopt Engeström’s (2001) activity-system framework to ground our systemic perspective on how to leverage the capabilities of robotic technologies and optimize their integration into the workplace.

Scholars in the field of work research theorize that learning, as a transformative (cf. adaptive) process, can change and create new activities (Blackler and Regan, 2009; Engeström, 2004). Engeström’s general model of activity systems reflects collaborative efforts by a group of people to achieve shared goals. Activity is systemic by nature and comprises individual and short-term action, as well as collective interaction (Engeström, 2004). In service research, some researchers adapt this model to understand the service–activity system (SAS) (Mele and Russo Spena, 2018). As depicted in Figure 1, the SAS involves different actors (subjects) who apply specialized competencies (division of labor) to perform a set of actions involving resource integration (instrument/tool) in an actor-to-actor context (community). Collective participation is governed by institutions (rules) and enables actors to offer new value propositions (objects), representing trajectories of value in contexts of use by another activity system (beneficiary). In this sense, the outcome (i.e. element produced by the activity system, as a transformation of the object to be used by other activity systems; Engeström, 2004) reflects the essence of service provision, in that it offers potential value co-creation. Considering SAS requires understanding how elements in a system work together toward a common goal and transform themselves according to expansive learning processes. In an HRC context, an SAS implies that humans work with robots to reach goals. The division of labor is particularly crucial, because both humans and robots apply their specialized competencies (Odekerken-Schröder et al., 2021; Xiao and Kumar, 2021). Embedding AI successfully involves a collaborative process of co-evolution and effective knowledge transfer, to address challenges and support employees’ adaptation to the new technologies (Bonetti et al., 2023). However, we know of no studies that address how system-level collaborations might be mediated by elements such as tools (technological interfaces), rules (protocols and guidelines) or the community (organizational or social context).

The concept of expansive learning offers some ideas for how activity systems work (Engeström, 2001) and deeper insights into collaboration processes (Kallio and Lappalainen, 2015). It encapsulates the notion that profound qualitative renewals in activity—development, expansive shifts and transformation—are achievable through processes of learning focused on the expansion of the object. Rather than merely describing interactions between different entities, the theory highlights the transformations that take place within an activity system. Expansive learning also addresses the challenges and disruptions to existing schemas, processes and institutions, enabling the creation of new ways of working. It suggests that the renewal of everyday activities is possible through collective learning, where cycles of expansion and transformation occur both within individual systems and across broader horizons of opportunity (Engeström, 2001). This framework illustrates a qualitative transformation in activity systems, highlighting dynamic relations among object-oriented activities, subjects and their community (Engeström, 2001, 2016; Engeström, 1999). The key stages in the cycle are depicted in Figure 2. It starts by questioning existing work practices and analyzing the situation. According to Engeström, the continuous development of an activity system gets propelled by its intrinsic tensions and contradictions. The division of labor, including specialized competencies, creates a community with diverse viewpoints, traditions and interests that constantly generates challenges and inconsistencies. These tensions, extending beyond mere problems or conflicts, manifest as historically accumulated structural contradictions within and between activity systems. They lead to disturbances, prompting translation and negotiation actions. As a result, the further phases of the cycle, including modeling and then examining and testing new models, are evident in innovative attempts to reshape systems. In these phases, to address contradictions, participants start to collectively reevaluate and reconceptualize the object, which necessitates the construction and implementation of a new way of doing. As the process evolves, the transformation initiated by individual subjects evolves into a collective movement involving the implementation, reflection and generalization of new work practices. In Engeström’s view, expansive learning unfolds as a mix of emergent and conscious processes that cannot be known beforehand and that focus on not only specific problems at hand but also wider contexts that generate them (Engeström, 2001). The result is a qualitative transformation of all components of the system; a broadened object of activity arises, and new tools are produced to mediate intellectual and practical operations, as are novel rules and labor distributions (Engeström et al., 1999; Engeström, 2004).

Recent HRC studies also acknowledge that the role of robots extends beyond technological adoption and can accommodate new forms of work (e.g. Le et al., 2023, 2024; Mele et al., 2022). However, this research domain lacks of systemic vision. No studies address the complex tensions and challenges associated with integrating robots into work activities.

Our research design is based on a longitudinal study, to support a thorough investigation of the case while also accounting for the broader context (Stake, 2005). Moreover, it allows us to collect rich, detailed data and provides the flexibility to explore various perspectives (Mele et al., 2023). In detail, we investigated an international healthcare company, founded in 1978, specializing in the development of natural, effective and safe products. In its more than 40 years of experience, the company has established itself as a global leader in natural health and wellness products. The company has a fully vertical production chain, from the cultivation of high-quality organic raw materials to the verification of the pharmacological and clinical efficacy of its products. Its rigorous development process is deeply rooted in scientific evidence and designed in accordance with the principles of systems medicine—a comprehensive approach that considers the interconnectedness of biological systems.

Over time, this company has cultivated a work culture that places a profound emphasis on health and well-being as fundamental pillars. Central to the company’s practices is a dedication to producing products of exceptional quality while fostering pleasant services for its customers. In turn, the organization places paramount importance on its people, recognizing them as invaluable assets pivotal to its success. This relentless focus serves as a bedrock for its innovation. The company’s goals also extend beyond immediate objectives, aiming to create groundbreaking solutions that address present needs and also pave the way for a meaningful future in the healthcare industry. This dedication to excellence and foresight underscores the company’s vision to make enduring contributions that enhance healthcare and well-being on a global scale.

In pursuit of these goals, the company has embraced recent AI-based technologies (chatbots), integrating them into operational workflows to enhance both internal employee interactions and customer relationships. Specifically, it has introduced seven chatbots (Table 2), according to system-based approaches (Kunz et al., 2022), to deliver tailored solutions efficiently and maintain a consistent level of quality and performance across a wide scale.

Multiple one-to-one in-depth interviews were conducted online between August 2021 and October 2023 with 14 company employees involved in the implementation of the chatbots (Table 3). Each interview lasted around 60 min, for a total of 56 h of interview data.

To ensure reliability, the authors who conducted the interviews used a consistent protocol (Denzin and Lincoln, 1998) that specified five key interview objectives: (1) to analyze existing routines, (2) to examine the objectives and processes for introducing chatbots, (3) to understand the problems and challenges experienced by employees, (4) to identify positive versus negatives outcomes and (5) to identify new activity systems and work practices. Furthermore, we transcribed each interview verbatim and meticulously crafted individual documents and notes about each discussion. Beyond conducting interviews, the researchers systematically amassed key performance indicators (KPIs) and pertinent quantitative metrics, provided by the technology provider (see Table 2). We then used these metrics to compare with statements by employees. By contextualizing the interview data to specific scenarios or settings, we sought to extract comprehensive, detailed descriptions, drawing inspiration from Geertz’s (2008) thick description concept. This comprehensive approach facilitated a deeper understanding of the nuances, subtleties and broader implications embedded in these data.

The juxtaposition of qualitative interview insights with quantitative performance metrics provides a holistic perspective that can enrich our interpretation of the study findings. We integrated “systematic data collection, coding, and analysis with theoretical sampling to formulate an integrated theory that is closely aligned with the data and expressed in a clear format for subsequent verification” (Conrad et al., 1993, p. 280). This method entails scrutinizing each interpretation emerging from the data analysis against existing findings (Glaser and Strauss, 2017) to discern various facets of the studied routines and their transformations since the introduction of chatbots.

During the research period, our analysis encompassed four distinct phases associated with the introduction of the different chatbots into daily routines. We delved into the diverse perspectives of employees interacting with the chatbot to understand its impact on the SAS. We prioritized resultant effects—namely, the emergence of new work practices—by evaluating alterations in behavioral patterns and available resources among the involved participants, their strategies for integrating resources and the outcomes regarding collaborative value creation.

The coding process, aimed at achieving a clearer comprehension of the core phenomena, encompassed a structured three-tier analysis, outlined in Table 4: open, axial and selective coding (Corbin and Strauss, 2008). In the initial phase of open coding, we sought to discern patterns in the responses and continually refine these patterns through ongoing analyses of newly acquired data (Boeije, 2002). This iterative process persisted until the incoming data corroborated and solidified existing codes, leading to theoretical saturation. Subsequently, the axial coding phase involved an ongoing process of comparisons and synthesis (Kolb, 2012). Finally, in the selective coding stage, we pinpointed core categories, guided by our theoretical frameworks of activity systems and expansive learning. This methodological approach supports systematic, comprehensive analyses and enables a deeper understanding of the underlying dynamics and interrelations across the data. After completion of the coding process, we convened meetings with key stakeholders, including the technology partner, customer care and marketing managers, to elicit and compare their perspectives.

Our results offer a detailed account of how the adoption of multiple chatbots and generative AI can foster continuous adjustments to the work practices within an SAS. The expansive learning phases—the five distinct yet interconnected stages of (1) questioning and analysis the current practice, (2) modeling, (3) examining and testing, (4) implementing and reflecting a new solution and (5) consolidating the new practice—serve as crucial milestones in the evolution of approaches to work practices. The transitions between the phases mark pivotal moments that also mirror corresponding tensions encountered while striving to meet and exceed the evolving expectations of the company’s customers. These tensions are more than mere obstacles; they constitute intrinsic elements of emerging human–robot activity systems, in which interactions take meaning from the activities they perform. Opportunities for learning, growth and transformative development shape new work practices for customer care.

An initial analysis of the company’s SAS reveals substantial hurdles related to the structure of customer care. The company’s expansion to various online platforms, such as its corporate website, online communities and social networks, had escalated the volume of customer inquiries transmitted through digital channels. The company was inundated with a wave of unfavorable reviews, pointing to the inefficacies plaguing its customer care. Contradictions emerged between the service promised and its effective care practices (tension 1), as customers expressed frustration over prolonged response times, detailed experiences of waiting extensively on the phone, and then encountered the additional inconvenience of being transferred to human operators. A collective understanding of the need for change arose, motivating and directing a process of questioning traditional customer care activities:

The pronounced volume of customer inquiries grew increasingly demanding, due to the customer care department’s dependence on conventional communication methods such as emails, phone calls and online forms. This reliance exacerbated their workload, because employees struggled to address inquiries promptly across the multiple channels for which they were responsible. An emerging tension 2 related to inefficient coordination of human and technical elements across the organization’s existing resources. These first two tensions created a cause-and-effect loop: Inadequacies in the customer service system (tension 1) exacerbated operational challenges (tension 2), and then frustrations expressed by customers and employees related to response times and handling capacity highlighted the inefficiency of the current system and its outdated methods. This scenario further stressed the system, amplifying the need to alter the management of customer inquiries. This discernible need for a thorough reassessment and substantial adaptation of the company’s operational structure, to integrate and harness modern technologies effectively, enhanced the conversion management.

The contradictions in managing employee activities for customer care and the imperative to introduce changes became evident in this step. Despite the clear patterns of repetitive, similar customer inquiries, efforts to explore solutions to these tensions that maintained the firm’s current status introduced a broader, complex layer related to compliance and regulatory standards, reflecting the stringent rules surrounding national healthcare institutions. Any failure to adhere to their rigorous guidelines could lead to additional and substantial tensions involving the company’s broader network (tension 3), as well as the risk of substantial damage to multiple actors (companies, customers, regulatory institutions). The implications of possible noncompliance (tension 3) put additional pressure on customer service (tension 1) and human–technical resource coordination (tension 2) efforts that informed both the prospect and necessity of making changes:

Figure 3 depicts the focal SAS and some tensions arising from the analysis conducted during this phase. In response, the company was driven to find new solutions to improve the HRC.

The challenges that emerged from the initial phase, focused on questioning and analyzing the existing SAS and customer care, prompted the company to seek more efficient solutions. The introduction of the first chatbot marked a significant turning point, shifting the company’s work practices from traditional call centers and email-based approaches to a more technologically advanced design that relied on HRC. The implementation of the chatbot introduced some immediate advantages, including resolving service inefficiencies and reducing both delays for customers and cumbersome customer interactions for employees:

Once overwhelmed by their excessive workload, employees experienced at least modest alleviation and could rely on collaboration with chatbots. The KPIs, including the total number of customer conversations with the chatbot and the number in which it autonomously resolved problems, substantiate this initial positive trend. However, these improvements also introduced new tensions that affected both customers and employees.

Customers interacting with the chatbot received prompt responses to basic queries but encountered limitations when seeking more detailed exchanges; the chatbot seemed unable to respond effectively—a limitation linked to the constraints of its training (tension 1). This issue led to inefficiencies in customer care and negative experiences in interactions with the chatbot:

Thus, new challenges arose for organizing work activities, putting extra burdens on employees (tension 2). Although the chatbot reduced the volume of emails and requests directed to employees through the call center switchboard, its limited performance subsequently increased the workload and stress levels of those who had to manage interactions with frustrated customers. That is, the integration of robots into employee workflows inadvertently increased pressures rather than alleviating them:

As a result, a new need emerged for additional training procedures for the chatbot. New tensions also emerged, related to establishing new rules and work routines (tension 3). Initially, human operators had to invest significant energy and time to ensure the chatbot was prepared to address customer queries and handle diverse situations autonomously. They also needed to acquire new knowledge and competencies, because the training process was complex and multifaceted. The challenge of preventing the chatbot from confusing different training programs increased the workload of the human staff:

In addition, tensions resulted as direct consequences of inadequate collaboration across broader organizational settings (tension 4). Their root cause might be traced to different skills and competences across departments, which became particularly evident during the training process. Strict coordination among employees from various departments was necessary to provide timely and accurate information; furthermore, to keep chatbots up to date with novel product and service introductions, they required frequent, time-consuming updates. Any lack of collaboration between human tasks and robot tasks led to information imbalances, which grew over time and intensified existing tensions. To employees, the chatbot seemed to operate disconnected from the organization, functioning merely as an additional tool available to customers. Consequently, employees experienced diminished trust in the technology and expressed less commitment to exploiting its potential:

Figure 4 illustrates the modifications thus implemented in the SAS, delineating the emerging tensions associated with modeling a new solution based on a chatbot. Recognizing these challenges, the company pursued another transformative change in work practices.

The introduction of the first chatbot solution illuminated some unforeseen organizational challenges. As the product range expanded and more tasks were assigned to the chatbot, communication effectiveness declined; it confused product details, often giving inaccurate or incorrect answers. To address this issue, the company overhauled its customer care process, shifting from reliance on a single chatbot to deploying a network of seven specialized chatbots, focused on different activities: the main website, events and marketing, or specific product lines. The chatbots were deployed only for particular instances and locations upon request, whereas employees handled more complex inquiries. Although the chatbots could efficiently answer straightforward queries, specific to their tasks, their limitations were evident, leading to significant new issues in customer care practices (tension 1). Handling vast amounts of information across multiple chatbots again sometimes resulted in outdated data:

Overlapping information across the different chatbots also led to employee errors, due to the inconsistencies and challenges in keeping the chatbots updated. The training effort required to manage them demanded increasing attention, and updating the same information across multiple chatbots was exceedingly time-consuming. These operational limitations of specialized chatbots (tension 1) directly contributed to complexities in managing them, leading to a new set of tensions (tension 2).

Despite improvements in the efficiency and relevance of customer interactions through tailored responses, the new multi-chatbot system did not evoke positive feedback from employees. Its promising effectiveness, such that each chatbot would perfectly answer questions about its dedicated activity, never came to fruition:

In addition, the persistent tensions affected work activities. The new system required active participation from employees from various departments during training processes, in the hope of providing chatbots with a comprehensive understanding of the organization’s activities. But the lack of specialized expertise dedicated to developing and integrating the seven chatbots complicated their implementation (tension 3). The chatbots struggled to adjust their independent responses to internal changes, which reduced their ability to provide current information and also highlighted the need for improved supervision over their design and implementation.

Figure 5 presents the modifications implemented during this implementing phase. A renewed impetus arose to reevaluate and enhance the SAS, particularly to achieve a more integrated and streamlined system for customer care that leveraged the novel capabilities of chatbots, especially following the introduction of generative AI.

Even with significant advances in technology and organizational practices, employees struggled with effective time management, primarily due to the difficulty of integrating renewed customer care activity into the multi-chatbot system. To address these issues, the company implemented generative AI to summarize chatbot–customer interactions and thereby give employees insights into these conversations, from which they might identify areas in which chatbots underperform. In seeking to realize its full potential, new challenges again arose, including the need to analyze even more extensive customer data (tension 1) and integrate the chatbot system into all working activities. Yet the potential improvements could enhance the efficiency of customer analysis and streamline workflow activities, leading to more responsive and personalized care services.

By distilling conversations into concise summaries, the generative AI solutions empowered human operators to focus on nuanced problem-solving, rather than grappling with the sheer volume of customer data. This transformation required employees to engage in ongoing education and training (tension 2) to utilize new generative AI functionalities and improve their problem-solving skills, for which human insight remained crucial (Figure 6):

As the generative AI gained a refined understanding of customer queries and chatbot responses, it fostered a dynamic and evolving collaboration that improved the quality of customer care and also highlighted employees’ commitment to using technology to identify potential areas for improvement. By automatically gathering and analyzing customer data insights, the system uncovered specific areas that required enhancement or topics suitable for personalized marketing initiatives:

Beyond efficacy, the benefits extended to creating a work environment where humans and AI complemented each other, fostering more flexible and harmonious HRC (Figure 7). Robotic technology was recognized as a powerful tool that, when thoughtfully integrated, enhanced human abilities and enriches job roles, rather than just replacing humans for repetitive tasks.

Because the company embraced the tensions during the initial customer care activities, the customer care practice transformed. Figure 8 provides a summary of all modifications made to the human–robot activity system throughout the expansive learning process, as also detailed in Table 6. The initial transformative phase centered on establishing a shared comprehension of the imperative for change, serving as both a motivator and guiding force. Subsequently, the second phase featured meticulous analysis and exposition of the contradictions inherent in current activities, aiming to facilitate critical examinations. In the third phase, a novel, chatbot-based activity system took shape, offering recommendations for the expansion of customer care practices. The fourth phase saw the actualization of the new activity system, executed with the implementation of multiple chatbot systems enhanced by AI, as well as continuous refinements in real-world settings. The fifth phase, dissemination and consolidation of the new human–robot activity system, reflected the outcomes of expansive learning developed through tension reflections and solutions, ensuring its widespread adoption and enduring integration.

The process was dynamic, not static, such that the introduction of various chatbots into workflows revealed complex interplays of solutions and subsequent tensions, as well as their systemic impacts on activities. After the first phase, when chatbots were identified as an immediate remedy to alleviate staffing pressures, their integration affected HRC (involving both customers and employees) and work rules. New tensions emerged from the effort to balance automated responses with human oversight. To address these issues, learning expanded, on the basis of additional questioning about how to integrate advanced chatbots programmed with sophisticated algorithms to handle diverse customer interactions. However, as chatbots became more sophisticated, needs for ongoing training and refinement of both human and robot capabilities kept increasing too, which strained resources and demanded further investments in technology and personnel training.

Each step in the process of integrating chatbots thus initiated a cycle of problem-solving and new challenges, necessitating an expansive learning process of dynamic and systemic transformations. In each phase, by identifying and exploring emerging tensions and contradictions—such as discrepancies between customer needs and company provision or between chatbots’ capabilities and employees’ expectations—the organization also engaged in expansive learning. Once chatbots could handle routine inquiries, employees spent more time on tasks that required empathy, critical thinking and complex problem-solving. New employee roles emerged too, such as chatbot trainers and AI system supervisors, in which employees worked closely with AI developers to improve the chatbot algorithms and performance. This continuous process extended beyond adapting or integrating new robots; it entailed transforming employee–robot collaborations and the entire customer care practice to establish a hybrid system that could combine robot and human abilities. This system then could improve response times and operational efficiency (Table 5) while also enhancing the customer care service process by providing immediate support for simple queries and specialized attention for complex issues.

In framing the notion of collaboration between human actors and robotic actants, we build on an activity systems framework and the dynamics of expansive learning (Engeström, 2001, 2016; Mele and Russo Spena, 2018). The comprehensive view of HRC, we offer extends beyond typical dyadic HRI to encompass a systemic perspective that can facilitate the adoption of robotic technologies in customer care practices. We make two main contributions to HRC discussions: first, enriching extant studies on the evolving roles and tasks of humans and robots, and second, exploring how collaborative efforts reshape work practices.

First, we conceptualize the human–robot activity system as a specific case of the service-activity system (SAS) framework (Mele and Russo Spena, 2018), designed to enhance customer care practices. Within this shared system, HRC emerges as a dynamic collaboration where human actors and robotic actants operate, mediated by tools, rules and the community (Engeström, 2004). The focus of HRC—enhancing customer care—serves as the motivating objective of this system. Introducing new technology-based actants (e.g. chatbots) initiates a transformative process involving all the SAS components. Our findings highlight that the positive impacts of HRC extend beyond operational efficiency, accommodating new forms of working practices where humans and robots leverage their complementary strengths. This result adds to previous studies showcasing how collaboration drives greater efficiency and innovation in service provision (De Keyser and Kunz, 2022; Odekerken-Schröder et al., 2021).

Rather than perceiving robots as simple replacements, we emphasize their entanglement with human actors, where both parties leverage their respective strengths to enhance operational effectiveness and performance. This perspective enriches theoretical discussions on the dynamic roles of humans and robots in work practices, moving beyond simple task automation to highlight their interdependent collaboration within evolving work environments (Mozafari et al., 2022; Le et al., 2024). Humans and robots are interrelated, forming a system where collective development takes place through engagement in working activities. However, this process is neither smooth nor static. As our study shows, established schemas and specialized competences can either hinder or facilitate collaboration and outcomes. Our results confirm claims that human abilities and willingness are crucial to the successful integration and acceptance of robotic technologies in organizational settings (Xiao and Kumar, 2021). Involving employees early in the implementation phase can help ensure that the design of technology-based tools aligns with their practical needs, mitigates resistance and encourages a smoother transition (Bonetti et al., 2023).

Second, building on expansive learning theory (Engeström, 2001), we explain how the human–robot activity system evolves, addressing the concept of expansive performativity that emerges when HRC actively transforms the system in which they operate. As human actors and robotic actants collaborate, contradictions and tensions between current work practices and robotic capabilities arise. Most traditional studies on HRI do not address these contradictions, often portraying human–robot work practices as mere reproductions of existing activities, with the only difference being the introduction of a nonhuman actant. Instead, expansive learning emerges as a transformative process in which HRC not only adjusts to, but actively drives, a continuous and dynamic refinement of roles, tasks and institutional setups. This process of resolving contradictions and tensions triggers the co-evolution of human competencies and robotic functionalities, thereby reshaping work practices. The outcome is not only mastery of existing human skills but also the creation of new ways of working and organizational transformation (Le et al., 2023; Leiño Calleja et al., 2023).

Our findings enrich theoretical discussions on performativity and organizational transformation, revealing how HRC reshapes work systems, redistributes tasks and redefines institutional rules to accommodate new forms of work (Bonetti et al., 2023; Xiao and Kumar, 2021). The entanglement between humans and robots entails an expansive performativity that transcends the traditional role boundaries of either actor or actant, creating new ways of working and communicating between humans (employees and customers) and robots (e.g. multi-chatbot systems). This transformation creates new opportunities, redefines institutional structures and broadens the scope of collaboration.

Finally, this study offers a system-based and learning-oriented perspective on HRC. The increasing complexity of service tasks and the demand for personalized services are likely to foster continued collaboration between service employees and robots, while also driving employees to enhance their skills for creatively engaging with and adapting to a robot-enhanced environment. This environment is marked by advanced tools, new rules and procedures to accommodate change. Their complementary strengths suggest a future in which humans and robots collaborate to reorganize service activities, upskill human abilities and innovate service provision.

We offer several insights for managers regarding how to embrace a systems perspective to facilitate the transformative adoption of robotic technologies in work practices. First, HRC can produce a significant improvement in operational efficiency. Robots handle routine questions, reducing the workload on human employees and decreasing response times by improving customer service efficiency and satisfaction through timely and accurate responses to customers’ inquiries. The expansive learning process we investigated illustrates how customer care tasks are increasingly being delegated to chatbots. At the examined company, by refining the training and capabilities of chatbots, customers receive more accurate and timely responses about products without the need for human intervention. As such, we recommend that managers work to leverage similar efficiencies to optimize their resource allocation and streamline service processes.

Second, the adoption of robot actants necessitates a careful redefinition of roles and responsibilities, including an appropriate division of labor. Employees must adapt to new roles as technology trainers or AI system supervisors, which means they must develop new competencies and problem-solving skills. Our investigated case illustrates how employees transition toward becoming technology collaborators. Employees have continuously enhanced their skills through customized workshops, learning how to use, train and understand the full potential of the technology. They have also adapted their activities to share their workload with the technology. We suggest that managers implement comprehensive training programs to ensure employees are proficient in using and managing the technologies, which should foster a collaborative environment where human and robotic capabilities combine synergistically.

Third, the integration of robots into service systems often leads to tensions, conflicts and contradictions, including mismatches between chatbot capabilities and customer expectations or increased workloads due to initial inefficiencies in chatbot performance. At the examined company, a feedback loop emerges where changes made in the service activity system to resolve known tensions often generate new ones. For instance, the introduction of additional chatbots provides more accurate responses to customers but creates challenges for employees who must manually manage various overarching training programs. Accordingly, we encourage managers to adopt a proactive approach to identifying and resolving such tensions through iterative problem-solving and expansive learning processes, doing so would involve continuously evaluating and improving the HRC, while also ensuring alignment with organizational goals and customer needs. In addition, Table 6, presented in the findings, can serve as a guide for the firm to identify tensions and assess potential actions, helping it navigate through the various phases.

Fourth, to mitigate resistance and ensure a smooth transition, managers should engage employees early in the design and deployment process, to help ensure the robotic technologies meet their practical needs and to foster a sense of ownership and acceptance. A participatory approach can address concerns about job displacement and build a collaborative work environment. In addition, a culture of continuous learning and adaptation can encourage employees to engage in ongoing training and development to stay abreast of technological advancements and optimize the collaborative potential of HRC systems. The investigated expansive learning process highlights the need for a feedback loop where insights from HRI are continuously integrated into system design and training. Employees continuously monitored and adapted AI training, suggesting new features and interfaces to enhance the technology’s effectiveness while staying updated on its developments. These recommendations further suggest the need for a feedback loop, in which insights gained from HRI get continuously fed back into the system design and training processes.

By embracing such implications, practitioners can achieve innovation and transformation by enhancing operational efficiency, redefining roles, addressing tensions, promoting continuous learning and improving customer care. A proactive, inclusive approach to HRC paves the way for a future in which human and robotic capabilities get integrated harmoniously, creating dynamic and resilient service systems.

This study has some limitations that can guide further research. First, it is based on a longitudinal analysis of a single international healthcare company specializing in natural health products. The findings provide valuable insights into the integration of HRC in this specific context, but the generalizability of the results to other industries and organizational settings may be limited. Continued studies should consider diverse organizational contexts to validate and expand on these findings, as well as to address specific research questions such as: How can the balance between automated service delivery and human touch be optimized in different customer care settings to maximize customer satisfaction and operational efficiency? What specific training programs and support mechanisms are most effective in helping employees adapt to working alongside robotic technologies? In what ways can organizations identify and resolve tensions between human work practices and robotic capabilities to achieve smoother collaboration and enhance system functionality?

In addition, a potential avenue for future research could involve studying how expansive learning cycles unfold in different industries or settings, including how contradictions between current work practices and robotic capabilities lead to improved collaboration and work efficiency. Specific questions could be: How do expansive learning processes vary across different sectors, and what factors influence the successful integration of HRC? What are the key triggers and facilitators of expansive learning in collaborative human–robot systems across various industries?

Second, we examine chatbots and virtual assistants primarily within the customer care domain, so we might have overlooked other robotic and AI technologies that could have more significant roles in other service contexts. Further research could explore a broader range of technologies, including cognitive assistants, robotic process automation and physical service robots, to provide a more comprehensive understanding of HRC. Example questions include: How do cognitive assistants and robotic process automation impact different domains of service delivery, beyond customer care? What challenges and opportunities arise when integrating physical service robots into human workflows across various service sectors ? What specific tensions emerge when integrating multiple AI technologies within a single service organization, and how do these tensions influence customer care practices?

Third, investigating the social and psychological aspects of HRI might provide valuable insights into the acceptance, trust and collaboration dynamics between human employees and robots. Methodologies from social sciences or psychology might help reveal these aspects, offering a more holistic view of HRC. For example, How do social and psychological factors, such as trust and perceived reliability, affect the willingness of employees to collaborate with robots in the workplace? What role do HRI design features play in shaping the social acceptance of HRC across different organizational cultures? What role do cultural differences play in shaping attitudes toward HRC, and how can these differences be addressed to minimize tensions in collaboration?

Fourth, though we identify improvements in customer satisfaction, this finding reflects internal performance metrics and employee reports. Direct feedback from customers regarding their experiences with the integrated HRC systems is limited; comprehensive customer surveys and feedback mechanisms could reveal the end-user experience and satisfaction levels. Specific questions could include: What are the key factors that influence customer satisfaction and acceptance when interacting with a combined human–robot service system? How does the quality of human–robot collaboration impact the perceived value of service from a customer perspective? How can customer feedback help address specific tensions, such as inadequacies in the customer service system? How can customer insights help organizations balance automation and human intervention, reducing service inefficiencies?

Fifth, as the adoption of AI and robotics in service industries grows, rules and regulatory considerations become increasingly important. Therefore, we call for further research to address key issues such as data privacy, algorithmic bias and regulatory compliance and thereby help ensure the responsible and ethical use of robotic technologies. For example, How do existing regulatory frameworks affect the implementation and operation of HRC systems? What measures can be taken to reduce algorithmic bias in service robots and ensure fair and ethical treatment of all customer segments?

Sixth, we do not consider economic impacts. Assessing the economic implications of HRC, according to cost–benefit analyses, productivity gains, or impacts on employment, can provide valuable information for organizational decision-makers. Research that integrates economic evaluations could determine the financial viability and benefits of HRC investments and produce answers to the following questions: What are the economic benefits and trade-offs associated with implementing HRC in various service sectors? How do initial investments in HRC compare to the long-term financial benefits for service organizations, particularly in terms of efficiency and cost savings? How does the adoption of HRC technologies affect workforce structures, employment patterns and overall productivity within organizations?

By addressing these questions, continued research can provide even more comprehensive insights into the multifaceted aspects of human–robot collaboration, enhancing its implementation and effectiveness in various service contexts.