Augmenting human service agents with intelligent technologies

Every day, service agents shape billions of customer experiences. In 2023, over 2.8 million service agents in the US managed customer inquiries in contact centers through email, live chat, and phone alone (Statista, 2024). Service interactions form the foundation of value co-creation (Grönroos, 2011), but their nature is changing rapidly. Advances in intelligent technologies such as artificial intelligence (AI), machine learning (ML), deep learning, and natural language processing (NLP) are fundamentally changing service work and customer expectations (Huang and Rust, 2018; Larivière et al., 2017).

Hitherto, a dominant focus has been on automation, where systems replace human labor, streamline operations, or deliver self-service experiences (Bowen, 2016; Chi et al., 2020). Yet, service firms continue to rely heavily on human employees (Bowen, 2024), and consumers often prefer or expect human contact, exhibiting an aversion to algorithm-provided services (Ameen et al., 2021; Dietvorst et al., 2015). As a result, attention has shifted to how technology augments service work through complementing human strengths such as empathy and judgment rather than replacing them (Bowen, 2024; De Keyser et al., 2019; Xiao and Kumar, 2019).

Recent research has conceptualized human-machine collaboration as a form of teamwork rather than replacement (Kunz et al., 2025). Despite this shift, the service literature still lacks an integrated understanding of how augmentation reshapes the work of service agents. Existing reviews either exclude technology (Walker et al., 2023), focus on automation (Chi et al., 2020), examine isolated applications (Shah et al., 2023), or consider the effects of AI at an organizational level (Bankins et al., 2023). Furthermore, previous work often addresses specific applications, rather than examining the underlying technological capabilities that transcend tools and sectors for augmenting service (Huang and Rust, 2018; Kraus et al., 2024; Marinova et al., 2016). The literature calls for a cross-disciplinary understanding that centers on service agents in technology-enabled contexts and clarifies how augmentation occurs across various interaction settings and technologies (Larivière et al., 2017).

To address this void, we adopt socio-technical systems (STS) theory (Trist and Bamforth, 1951) to interpret the interplay between technological augmentation and service customization. This perspective moves beyond functional classifications. It emphasizes that service outcomes depend on how technical and social subsystems work together in context. Service-dominant logic (SDL) complements this view and regards both employees and technologies as operant resources that co-create value during interaction (Vargo and Lusch, 2008). Accordingly, augmentation is effective only when technological inputs are interpreted and applied through human judgment in context. Together, these perspectives imply that technologies are not merely tools for automation but resources that shape how value is created in service interactions.

This focus on augmentation reflects evidence that technologies add value when they complement human skills (Huang and Rust, 2018; Larivière et al., 2017; Marinova et al., 2016). Our review builds on this perspective and focuses on the core technological capabilities that underlie service augmentation: AI as the overarching paradigm; ML and deep learning, as vehicles for perception and decision support; and NLP, as the foundation for interpreting and generating human language. Together, these capabilities constitute the foundation of intelligent technologies for service augmentation (Hirschberg and Manning, 2015; Jordan and Mitchel, 2015; LeCun et al., 2015; Russell and Norvig, 2020).

We examine how these technologies become embedded in frontline service work and how they influence processes and outcomes of service interactions (Larivière et al., 2017). This contrasts prior work that examines digital transformation broadly or focuses on self-service and automation (Chi et al., 2020; Shah et al., 2023; Van Doorn et al., 2016). To our knowledge, this review is the first to systematically synthesize empirical findings on augmenting human service agents in service interactions, contributing to the literature in at least four ways.

First, we provide a human-centric lens, positioning the service agent as a key actor in technology-enabled service delivery. This reflects recent calls to examine employee perspectives in service systems and explore how technology can complement human service qualities (Larivière et al., 2017; Marinova et al., 2016).

Second, we contribute to the understanding of technology-enabled augmentation of service agents and organize the literature into six core themes, mapped onto the stages of the customer interaction process (Huang and Rust, 2018; Zapf et al., 2003). Together, the themes show how technologies augment service work and how optimization efforts increasingly depend on an understanding of the employee experience as “the totality of cognitive, emotional, behavioral, sensorial and social responses that result from interactions with other parties (e.g. customers and technology)” (Larivière et al., 2017, p. 242).

Third, we anchor these themes in a framework derived from STS theory (Trist and Bamforth, 1951) and SDL (Vargo and Lusch, 2008), mapping them along two dimensions. The first dimension, service customization, reflects the degree to which interactions adapt to individual needs and situations (Huang and Rust, 2018). The second, technological augmentation, captures the extent to which intelligent technologies support human service work (Larivière et al., 2017; Marinova et al., 2016). Together, STS theory and SDL emphasize that service outcomes depend on how human and technological resources are aligned and integrated during the interaction. This framework reveals how frontline work evolves from manual processes to adaptive systems and highlights tensions and boundary conditions that determine whether augmentation supports or constrains employees.

Finally, the review combines insights from service research, organizational behavior, information systems, and computer science (Choi, 2018; Kraus et al., 2024). This multidisciplinary integration clarifies how underlying technologies reshape frontline roles and provides a theoretically grounded research agenda on human–technology collaboration, well-being, personalization, and service system design.

Figure 1 illustrates the article identification process following the PRISMA flowchart. In the first step, a literature search was conducted using Scopus and Web of Science, which were selected for their extensive coverage of peer-reviewed research across technical and business disciplines.

Based on our conceptual anchors, we restricted the review to studies that examine technologies that support, extend, or enhance human service agents, excluding research on full automation or self-service technologies. The search string incorporated core technological domains relevant to service augmentation: AI, ML, deep learning, and NLP (Hirschberg and Manning, 2015; Jordan and Mitchel, 2015; LeCun et al., 2015; Russell and Norvig, 2020). We also included broader terms such as “technology,” “automat*,” “intelligent,” and “smart” to capture the diverse language used across business and technical literature, while maintaining a focus on augmentation rather than substitution. Tool-specific terms, such as “chatbot” or “virtual agent,” were not included in the search string, as they are predominantly studied in substitution and self-service contexts (Marinova et al., 2016; Xiao and Kumar, 2019). However, studies that examined such tools in ways that aligned with augmentation and met our eligibility criteria were retained, including those on hybrid human-chatbot configurations.

Terminology differences across disciplines required special attention. Technical publications often use narrow terms such as “machine learning” or “natural language processing” and situate studies in contact center contexts (Ahmed et al., 2023; Seng and Ang, 2018). In contrast, service research tends to employ broader constructs such as “service interaction,” “customer service,” or “frontline employee” (Fan and Mattila, 2020). To bridge these disciplinary conventions, the search string was composed of two groups of terms that had to co-occur. The first group captured the technological component: (“artificial intelligence” OR “AI” OR “machine learning” OR “ML” OR “deep learning” OR “natural language processing” OR “NLP” OR “tech*” OR “automat*” OR “smart” OR “intelligen*”). The second group captured the service context: (“customer service” OR “service interaction*” OR “customer interaction*” OR “customer support” OR “service encounter*” OR “service employee*” OR “service agent*” OR “customer-employee” OR “employee-customer” OR “service rep*” OR “frontline employee*” OR “FLE” OR “call cent*” OR “contact cent*”). This ensured that the review captured applications of augmentative technology in human service interactions while minimizing the risk of missing relevant studies due to terminological or disciplinary barriers.

Articles required at least one search term from both groups in the title, abstract, or keywords, ensuring both a service context and a technological aspect. We included work published between January 2015 and April 2025, as 2015 marks the year when breakthrough advancements in deep learning and neural networks laid the foundation for today’s transformative AI technologies (Silver et al., 2016). Articles were considered only if they examined technologies explicitly designed to augment, support, or enable human service agents, rather than substitute their role in the service process. Further inclusion criteria included peer review status (fully peer-reviewed), language (English), and quality (published in Q1 or Q2 journals, as per the SCImago Scientific Journal Ranking) to ensure high-quality and impactful research (González-Pereira et al., 2010). This process resulted in 5,200 articles, of which 3,649 papers remained after duplicates were removed.

In the second step, all titles and abstracts were screened independently by two expert coders. In cases of disagreement, a third coder made the final decision following a conservative approach (i.e. including a paper when in doubt). Papers that did not meet the criteria were excluded. This screening yielded 189 papers, which underwent full-text screening to assess eligibility against predefined inclusion and exclusion criteria. Key criteria included a focus on customer service, human-to-human interactions, technology usage, and empirical methodology. This process resulted in 96 included papers, with three additional papers identified through reference list screening, totaling 99.

We conducted a systematic thematic analysis of the resulting literature, following Clarke and Braun’s (2016) six-phase approach. First, we familiarized ourselves with the studies through repeated readings to develop a comprehensive understanding. In the second phase, we developed initial codes to capture key technologies, implementation approaches, and outcomes. Third, these codes were clustered into functional subthemes, including queuing, routing, forecasting, speech recognition, emotion detection, and service agent evaluation, which represent recurring technological roles across studies. In the fourth phase, the subthemes were iteratively refined through team discussions, memo writing, and negative case analysis to ensure that they reflected functional roles rather than technical categories or specific service settings. In the fifth phase, the subthemes were consolidated into six overarching themes that capture the principal ways technology is integrated into service interactions. In the sixth phase, each paper was assigned to one primary theme to highlight its central contribution. When studies addressed multiple aspects, they were assigned the dominant theme.

In developing the themes, we examined not only the type of technology but also the way it augments service agents. We examined how technologies interact with both social and technical elements, their location within the service process, who they affected (e.g. the customer or the service agent), and the kinds of work they changed or supported. This helped us distinguish, for example, between technologies that assist service agents in understanding conversations and those that monitor their performance, even if both rely on similar underlying methods. As coding progressed, recurring patterns emerged in how technologies shape interactions, tasks, or working conditions, informing themes that reflect different ways technology is embedded in human-to-human service. Table 1 summarizes the six themes: pre-service optimization, interaction intelligence, service agent well-being, service agent monitoring, emotion work, and collaborative service, and their core descriptions. A complete list of reviewed articles per theme is provided in Web Appendix 1. A bibliometric analysis is available in Web Appendix 2.

This section explores the six themes, highlighting their contribution to understanding the broader landscape of service augmentation, their interdependencies, and boundary conditions. The themes provide a structured framework for analyzing findings and implications.

Pre-service optimization technologies align service agents with customer demand, aiming to minimize pre-conversation wait times and churn (Haenlein and Kaplan, 2020). Effective scheduling relies on call forecasting, skill-based assignments, and proactive issue resolution (Albrecht et al., 2021; Ebadi Jalal et al., 2016). Hybrid models incorporating virtual agents offer additional flexibility, maintaining customer preferences for human involvement during more complex interactions (Legros, 2021).

Routing strategies direct customers to appropriate agents or channels, considering urgency, expertise, and customer characteristics (Marín Díaz et al., 2025). Traditional routing often favors familiar agents, creating biases and inertia effects that impact overall performance (Schecter et al., 2021). Automated routing, using text analytics, agent skills, or customer personality traits, improves resolution rates, customer satisfaction, and efficient use of service agent capacity (Borg et al., 2021; Ilk et al., 2020).

Customer service conversations contain valuable information. In phone-based interactions, speech transcription technologies facilitate information acquisition (Plaza et al., 2021). After transcription, NLP techniques enable more detailed analysis. Topic modeling categorizes conversations into predefined categories, identifying common customer questions and issues (Bost et al., 2015). Based on these topics, transcripts can be automatically classified, suggesting similar behavior or follow-up steps per group (Papadia et al., 2023). Similarly, intent classification can detect the underlying customer goals from interactive voice responses or the call itself (Cai et al., 2025). Furthermore, conversations can be summarized, enabling faster, more compact information use (Lin et al., 2023). Additionally, customers can be classified into distinct personas, enabling a more personalized approach (Hathaway et al., 2022; Marín Díaz et al., 2025). More generally, frameworks can combine methods to enhance analysis information (Fan and Ilk, 2020).

In summary, interaction intelligence technologies analyze and interpret service interactions (Shahin et al., 2024). Extracting detailed patterns enhances the understanding of communication dynamics. This analytical capacity also connects to other themes, such as service agent well-being (Theme 3) and service agent monitoring (Theme 4), and it supports Theme 1 by feeding early-stage service processes with information from previous interactions.

Service agent turnover rates are high (Zito et al., 2018), and training new service agents is both time-consuming and costly (Hillmer et al., 2004). Here, technology has been introduced to support service agents in terms of workload and well-being (Choi and Kim, 2025; Pacella et al., 2024).

Stress levels play a crucial role in shaping both performance and overall well-being (De Ruyter et al., 2001), and they can be predicted in real-time based on customer emotion patterns (Bromuri et al., 2021). ML can evaluate the emotional workload of service agents, also providing managers with insight into stress levels and enabling timely interventions (Park et al., 2024). Additionally, technologies, such as service robots, can improve well-being by adjusting the physical and purpose-related work of service agents through task allocation strategies (Phillips et al., 2025).

However, technology can create new stressors for service agents (Henkel et al., 2020). Greater system availability shifts responsibilities to digital platforms, pressuring employees to stay constantly connected (Breit et al., 2020). Increased transparency requires cautious communication, as it may backfire for service agents. AI awareness can trigger negative emotional responses, including counterproductive work behaviors (Zhou et al., 2024).

Some features can mitigate these risks. Anthropomorphized robots increase service agent awareness of technologies, aiding collaboration, but reducing emotional warmth and motivation (Yang et al., 2024). AI integration also shows mixed effects. When perceived as a positive challenge, it boosts proactive behaviors, yet when raising job insecurity, it harms them (Huang and Gursoy, 2024).

Generative AI tools can increase emotional labor among low-skilled workers and limit their ability to contextualize these challenges within broader labor market dynamics (Oder and Béland, 2025). Support during implementation, including education and upskilling, is essential as inadequate support can elevate strain, weaken coping responses, and undermine consistent service quality (De Ruyter et al., 2001).

Service agents have highly structured jobs, balancing multiple responsibilities (Tovar, 2020; Zapf et al., 2003), bridging operations, sales, marketing, and technology (Choi, 2018). To support this multifaceted role, organizations increasingly use technological solutions that monitor and enhance service agent performance.

Service agent behavior can be visualized to highlight the strengths, weaknesses, trends, and training opportunities (Rees et al., 2021). Understanding these behaviors reduces negative customer interactions. For example, service agent malpractice and service sabotage can be detected automatically (Ma and Ye, 2022; Obinna Iheme and Ozan, 2022). This underscores the need for a supportive work environment, where negative behavior is addressed early and constructively. Given the limitations of manual feedback, such as delays and inconsistency, technologies have been introduced to classify service agent productivity and identify subjective calls, enabling evaluation that reduces subjective variation through predefined criteria (Ahmed et al., 2021, 2024).

This fourth theme links to Theme 2, as both leverage data-driven analysis to improve interaction quality. While interaction intelligence analyzes the conversational content of customers, this theme evaluates the behavior and performance of service agents. Integrating them provides a holistic view of how service agent actions align with customer expectations. This supports targeted coaching and continuous improvement, enhancing service effectiveness.

Another relationship exists between Themes 3 and 4. Monitoring service agents enhance well-being through timely, personalized feedback and targeted interventions that reduce stress and improve job satisfaction (Park et al., 2024). However, pervasive surveillance may undermine a sense of agency, leading to anxiety and decreased motivation due to constant observation and evaluation (Ma and Ye, 2022). Therefore, studying these themes together is crucial for understanding how monitoring practices affect both psychological health and performance.

Customer emotions significantly influence service outcomes (Mattila and Enz, 2002). Therefore, interpersonal emotion regulation skills of service agents are crucial in shaping customer interactions (Zaki and Williams, 2013). However, effectively managing these emotions first requires identifying them. Here, two approaches are reflected in the literature, sentiment detection and emotion recognition.

First, sentiment detection provides insights into customer affective states (Ahmed et al., 2023; Labat et al., 2024). A large-scale study of call center interactions found that positive customer sentiment significantly affects satisfaction and intention to recommend. In contrast, negative customer emotions have a greater (negative) effect on recommendations than on satisfaction, while service agent sentiment has less influence, and emotional matching is generally beneficial (De Cleen et al., 2025).

Second, emotion recognition studies adopt more diverse data modalities and methods (Labat et al., 2024). Recent works introduced an approach that combines video and audio (Guo et al., 2024; Seng and Ang, 2018). Taking it one step further, ML can analyze customer emotion patterns (Bromuri et al., 2021), aiding service agents in enhancing their effectiveness in regulating emotions during interactions (Henkel et al., 2020).

This fifth theme complements Themes 1 and 2 by adding an emotional dimension. Whereas Theme 1 addresses pre-call strategies, Theme 2 analyzes content and patterns during the call, and Theme 5 captures how customers express themselves emotionally. This emotional information can then be linked back to Theme 1, enabling the anticipation of potential emotional states and optimizing wait times and routing accordingly. Together, these themes offer a comprehensive view of the customer experience, from anticipating needs to managing emotional dynamics during interactions.

Theme 5 also connects with Theme 3, addressing the emotional dimensions of interactions. Emotion recognition technologies enable the real-time detection of customer affect, equipping service agents to tailor responses and manage challenging interactions. This emotional insight can reduce uncertainty and support emotion regulation, thereby enhancing emotional resilience and job satisfaction. However, they may also heighten emotional labor, as service agents are expected to continually respond to emotional cues, potentially leading to increased fatigue and burnout.

Understanding human-AI collaboration is crucial for enhancing all aspects of customer service across contexts (Le et al., 2024). In retail, employee-robot teams show that increased attention from robots does not always lead to higher sales. Effectiveness depends on employee characteristics and motivation, with individuals holding positive attitudes and lower anxiety collaborating more successfully (De Gauquier et al., 2023; Lin, 2025). Frontline robots with high automated social presence enhance impressions of employee competence, warmth, and teamwork quality (Leiño Calleja et al., 2025).

Similar dynamics emerge in human-chatbot collaboration, where hybrid configurations support humans and demonstrate that disclosing human involvement improves customer responses. However, it can increase handoffs and workload. Customers prefer arrangements where humans remain central, and artificial agents are evaluated more positively when paired with creative employees (Gnewuch et al., 2023; Le et al., 2024; Huang et al., 2024).

Additionally, collaborative systems enhance employee awareness, service outcomes, and work meaning (Blaurock et al., 2024). Digital encounters, mixed-reality systems, and virtual reality support decision-making, workflow, and transparency (Dolata et al., 2020; Pöyry et al., 2024). Conversational agents can boost productivity, performance, and work experience (Sheng et al., 2024). Service robots similarly augment human work and contribute to positive customer outcomes (Moliner-Tena et al., 2024).

Across these technologies, collaborative service acts as a central connector. Tools from each theme can be applied alone or combined to empower service agents and co-create value. Although themes highlight distinct functions, their boundaries often converge in practice. For example, emotion recognition (Theme 5) closely links to service agent well-being (Theme 3) and monitoring (Theme 4). Likewise, collaborative service shapes and is shaped by other themes. This interconnectedness underscores the importance of integrated socio-technical approaches for achieving both operational excellence and positive human outcomes. Building on these insights, the following section develops a conceptual framework that synthesizes technological augmentation and service customization, providing a foundation for theory development and future research.

To structure our findings, we developed a conceptual framework that links the six themes to two underlying dimensions. The dimensions emerged from inductive patterns in the reviewed studies and are grounded in STS theory (Trist and Bamforth, 1951) and SDL (Vargo and Lusch, 2008). Prior work has often grouped technologies by tool or function, fragmenting the literature and overlooking the central role of the service agent in value creation (De Keyser et al., 2019; Larivière et al., 2017). Focusing instead on augmentation and customization, we integrate the evidence into a framework that reflects how technologies and employees interact during service interactions.

Technological augmentation captures the extent to which the technical subsystem contributes during the interaction. At low levels, tools remain in the background, stabilizing processes without influencing the exchange itself. At high levels, systems guide communication, surface insights, or carry out subtasks alongside the employee in real time. In STS theory, higher augmentation refers to a tighter coupling between social and technical subsystems, altering how coordination, error detection, and feedback loops unfold during interactions (Trist and Bamforth, 1951). In SDL, augmentation refers to the idea that technology provides operant resources not only before or after the service but also within the interaction, directly influencing how value is co-created in context (Vargo and Lusch, 2008). Prior service research shows that augmentation can enhance (weaken) outcomes when technological inputs complement (displace) human strengths such as empathy and judgment, which makes alignment central to effective augmentation (Larivière et al., 2017; Marinova et al., 2016).

Service customization captures how much service adapts to individual needs and contexts. In STS theory, customization reflects how social and technical subsystems are aligned to enable situational adjustment. Loose coupling preserves autonomy but limits flexibility, making interactions efficient yet difficult to adapt (Trist and Bamforth, 1951). In SDL, value is co-created in use and is determined phenomenologically by both customers and employees, which makes their context central to effective customization (Vargo and Lusch, 2008). The literature shows that customization depends on three conditions. First, systems must represent context at a helpful level of detail, whether in terms of customer histories, emotional states, or employee workload signals (Payne and Frow, 2005; Peppers and Rogers, 1993). Second, insights must be surfaced at the right time so they can be acted upon (Rust and Huang, 2014). Third, discretion must be maintained where human judgment adds value, as rigid prescriptions risk undermining adaptation (Marinova et al., 2016). When these conditions hold, customization can support differentiated experiences that enhance outcomes for customers while also sustaining employee meaning and engagement. When they do not, efficiency may rise, but without genuine adaptation, leaving value creation constrained.

Together, the two dimensions clarify how technology and human judgment interact. Augmentation changes how the technical subsystem participates in the encounter, while customization reflects how the interaction adjusts to the situation at hand. Linking STS and SDL shows that these dimensions operate independently. More augmentation does not guarantee more customization, while more customization does not require extensive technology. What matters is how information, discretion, and support are aligned in practice.

Existing frameworks distinguish automation from augmentation (Huang and Rust, 2018), highlight the expanding role of technology in frontline work (Marinova et al., 2016), or emphasize that service systems must integrate multiple actors (Larivière et al., 2017). Our review extends these perspectives and shows that augmentation and customization are independent rather than sequential. This orthogonality explains why some service firms utilize advanced systems to enforce uniformity, while others achieve personalization without relying on technology. It also clarifies why effects on service quality and employee experience vary across settings. The axes reveal when technology strengthens human work versus when it undermines it, providing a framework of tensions to guide future research.

Placing the axes together yields four configurations that capture current practice and likely evolution. The framework presented in Figure 2 serves as an analytic lens rather than a rigid classification. The placement of themes should not be interpreted as fixed boundaries but rather as indicative patterns that highlight how the functional roles of technology have been treated in the literature to date and how they are expected to shift as systems evolve.

Each theme in Figure 2 occupies the quadrant that best reflects its dominant role in the reviewed studies. Pre-service optimization and interaction intelligence sit in more standardized settings as most applications focus on forecasting, routing, transcription, and classification, which raises efficiency without tailoring interactions. Emotion work and service agent well-being occur in human-led contexts, as empirical evidence shows that emotion work, stress regulation, and recovery practices are still primarily shaped by human judgment rather than algorithmic guidance. Service agent monitoring resides in standardized, technology-augmented settings, where its primary functions are reliability and compliance rather than adaptation. Collaborative service, in contrast, falls within the high-augmentation, high-customization quadrant, reflecting its joint emphasis on human and technological inputs. The lighter circles in Figure 2 indicate expected movements as technologies mature. Mainly, interaction intelligence is expected to shift upward as real-time analytics support personalization, and emotion work and service agent well-being toward greater augmentation as affective computing becomes more sophisticated, and adaptive systems proactively balance workload.

Quadrant 1 describes service interactions through routine processes with little technological involvement. Service agents follow predefined scripts, while technology remains in the background. The focus is on efficiency and reliability, reducing training requirements and ensuring consistency. The trade-off is limited flexibility as individual needs are often left unmet because processes are designed for uniform throughput rather than adaptation (Trist and Bamforth, 1951; Payne and Frow, 2005).

In socio-technical terms, this quadrant reflects a loose coupling between technical and social subsystems. Technology stabilizes the workflow but does not intervene during the interaction. Employees carry the primary responsibility, operating within the boundaries of standard scripts. In SDL, value is created through these uniform exchanges, but it remains constrained because the interaction adapts little to the context of either the customer or the employee (Vargo and Lusch, 2008).

Examples from the review demonstrate how pre-service optimization in this quadrant frequently takes the form of static forecasting and simple routing rules that reduce queues but do not cater to individual requests (Gans et al., 2003). Elements of interaction intelligence appear only after the conversation (e.g. transcription or keyword analysis that feeds reporting; Fan and Ilk, 2020). Service interactions in this configuration exemplify the classical paradigm of mass service as predictable, scalable, and cost-efficient, yet limited in its capacity to accommodate diverse needs or to support employees beyond basic role execution. This loose coupling enhances efficiency but restricts human judgment, reducing value creation to uniform exchanges with little scope for adaptation (Vargo and Lusch, 2008). Loose coupling protects service agent autonomy but hampers flexibility. The condition that sustains this model is volume-driven demand, where efficiency outweighs the need for adaptation.

Quadrant 2 reflects service where technology plays a visible but standardized role. Systems automate tasks to ensure consistency and scale, with emphasis on reliability and throughput. In STS theory, this quadrant represents stronger coupling between the technical and social subsystems, with technology exerting a more direct influence. Predictive tools and automated routines support employees, but their emphasis on standardization restricts the system’s ability to adapt to context (Marinova et al., 2016; Trist and Bamforth, 1951). According to SDL, value is co-created with technology, providing operant resources in real time; however, these contributions remain generic rather than context-sensitive (Vargo and Lusch, 2008).

The review indicates that pre-service optimization includes predictive routing and skill-based allocation, which improve efficiency but treat cases uniformly (Ilk et al., 2020). Interaction intelligence contains summarization, intent recognition, or topic classification with generalized models that inform quality assurance but do not personalize interactions (Papadia et al., 2023). Service agent monitoring is prominent here, with electronic performance systems providing standardized evaluation and compliance feedback. Meta-analytic evidence suggests that such monitoring can enhance measurable reliability, but it also raises strain and undermines perceptions of autonomy when framed as control (Ravid et al., 2023). This illustrates how augmentation with a focus on standardization secures efficiency but turns into control, reducing autonomy and engagement.

Quadrant 3 represents service interactions where customization is achieved primarily through human skill and judgment, with technology playing a secondary role. Employees tailor their interactions by reading and adapting to situational, cultural, and emotional cues. Technology provides support through databases or basic tools, but it does not drive real-time adaptation.

From a socio-technical perspective, adaptive work in this quadrant is carried primarily by the social subsystem, while the technical subsystem plays an enabling, secondary role (Trist and Bamforth, 1951). In SDL, value emerges from the capacity of service agents to integrate knowledge and empathy into the interaction, with technology providing limited operant resources (Vargo and Lusch, 2008).

The review shows that emotional work aligns closely with this quadrant. Service agents recognize and regulate affect, consistent with research on emotional labor and its links to both performance and strain (Grandey, 2000; Hülsheger and Schewe, 2011). Service agent well-being also fits here when supervisors tailor support, recovery, and coaching to the individual rather than through automated systems. This reflects job demands-resources theory, which posits that human resources buffer demands and sustain engagement (Bakker and Demerouti, 2007).

The strength of this quadrant lies in its capacity for deep relational value and its ability to adapt to complex or atypical situations. However, such interactions are costly, complex to scale, and variable in quality across employees and cases. When resources do not match the heightened emotional and cognitive demands, customization can lead to overload, weakening both service quality and employee well-being (Hülsheger and Schewe, 2011). Human-led customization thrives in contexts with high relational intensity but fails under volume pressure.

Quadrant 4 represents service interactions where technologies and employees collaborate to deliver real-time, tailored services. Intelligent tools, such as live analytics, emotion detection, and conversational assistance, integrate into the interaction, supporting service agents as they adapt to individual needs.

STS theory views this quadrant as the most aligned social and technical subsystems. Technology and human skill are closely intertwined, with each reinforcing the other to produce adaptive outcomes (Trist and Bamforth, 1951). In SDL, value is created during interaction through the integration of human empathy and technological capabilities, enabling customization at scale while keeping judgment with the employee (Vargo and Lusch, 2008).

The review reveals that interaction intelligence evolves from static transcription toward dynamic guidance that informs decisions during the interaction. Emotion work detects affect and surface cues, helping employees to choose appropriate strategies without removing control. Service agent well-being aligns here when adaptive systems monitor workload and suggest redistribution or micro-breaks that sustain performance, consistent with job demands-resources logic applied to digital environments (Bakker et al., 2023). Collaborative service is most visible here, where technologies act as partners, extending employee capacity. Mixed-reality systems, intelligent assistants, or frontline robots provide real-time input while employees retain framing and relational tasks (Wirtz et al., 2018; Wu et al., 2015).

The strength of this quadrant lies in adaptive service at scale. Yet tools that reduce uncertainty when presented as options can create technostress and role confusion when perceived as control, making the design of control and support mechanisms central to determining whether augmentation supports employees or adds strain (Dietvorst et al., 2015; Tarafdar et al., 2015). Customization delivers adaptive service at scale but requires trust, role clarity, and resource buffers to avoid technostress.

Taken together, the quadrants show that the value of augmentation depends not only on technological capability but also on how support and feedback shape the role of human judgment in service interactions. Boundary conditions are inherent to every configuration, indicating that efficiency gains can readily create new strains when resources are misaligned with demand (Bakker and Demerouti, 2007; Tarafdar et al., 2015). This affirms the socio-technical view that augmentation succeeds only when technology and human judgment evolve together (Trist and Bamforth, 1951). SDL adds that value emerges in context through the joint integration of human and technological resources (Vargo and Lusch, 2008). These insights form the foundation for a research agenda that extends the framework and guides future work on technology-augmented service interactions.

The future research agenda builds on the six themes, organizing open questions along the dual dimensions of augmentation and customization. This structure responds to calls for a more straightforward human-centered pathway in technology-enabled service (Larivière et al., 2017; Marinova et al., 2016). Rather than separating technology, employee outcomes, and augmentation practices, the agenda links themes to questions on how technologies can better support service employees. Monitoring illustrates these tensions most clearly, as augmentation can increase output reliability while reducing autonomy, thereby creating paradoxes that call for theoretical inquiry (Bakker and Demerouti, 2007; Ravid et al., 2023). Table 2 summarizes the six themes and links each to its position in Figure 2, which determines the tensions and boundary conditions shaping future research questions. Quadrants with high augmentation raise questions about human judgment and control, while quadrants with high customization raise questions about context sensitivity and emotional demands.

The transition from human-led to technology-augmented service signals a move toward hybrid, real-time, collaborative environments. This section outlines key avenues for studying how technologies can amplify human strengths and how employees can build the competencies, trust, and fluency to work effectively with them. To guide this transition, Figure 2 highlights where each theme shifts on the augmentation–customization plane, and Table 2 links these shifts to boundary conditions that shape future research opportunities.

At the far end of the human-led to technology-augmented continuum, cyborgs integrate human and machine functions through wearable or bio-integrated technologies (Garry and Harwood, 2019; Grewal et al., 2020; Nyberg, 2009). This fusion transforms service roles into hybrid entities, fostering deeper collaboration between human and machine intelligence (Theme 6: Collaborative Service). Key research challenges involve identifying feasible cyborg technologies and assessing their ethical and psychological implications (Kies et al., 2024). Cyborg employees may face issues of autonomy, stress, identity, and biological risk, which highlights the need for ethical frameworks that protect service agents (Theme 3: Service Agent Well-Being).

Advancing technology-augmented customer services involves overcoming key challenges in building intelligent systems. This includes developing multimodal, real-time technologies to create a coherent understanding of interactions. These systems support service agents through emotional attunement (Theme 5), monitoring of well-being (Theme 3) and performance (Theme 4), and responsive, context-aware collaboration. Generative AI adds the ability to produce personalized content (Theme 6), although its integration raises new research questions. Addressing issues such as data fusion, contextual interpretation, and deployment in immersive environments, including robotics and augmented reality, is essential for truly adaptive, human-centered services (Wirtz et al., 2018). Importantly, explainable AI is necessary to support decision-making through transparent, interpretable outputs. Few papers in our review addressed transparency, explainability, or bias, highlighting a critical gap in the literature. Equally important is embedding these tools into employee workflows in a way that preserves user agency and avoids cognitive overload. Future research should examine how service agents adjust routines, interaction strategies, and decision-making when they work with increasingly sophisticated intelligent technologies.

Our research agenda focuses on the shift from standardized to customized, human-centric service. This subsection examines this transformation and highlights key research directions on the role of technology in tailoring customer service interactions. Figure 2 locates where each theme shifts along this axis, and Table 2 outlines the boundary conditions that determine when customization enhances or constrains value creation.

A central trajectory in the shift from standardized to personalized service is the increasing use of data-driven personalization. These approaches tailor interactions to meet the needs of individual customers and service agents, which can enhance relevance, satisfaction, and engagement (Theme 2: Interaction Intelligence) (Ameen et al., 2021). Here, predictive personalization (i.e. anticipating customer needs before expression) is particularly promising. However, integrating behavioral, contextual, and historical data into actionable customer profiles remains a technical challenge (Themes 1 and 2). Additionally, overly aggressive predictions or unexpected insights can be perceived as intrusive or manipulative (Themes 3 and 6), which can undermine the trust that personalization aims to establish (Nishant et al., 2024). Personalized approaches also raise concerns about monitoring agent behavior, performance, and emotional states (Themes 3 and 4), which may increase stress and reduce autonomy, highlighting the need to evaluate monitoring practices with attention to human agency and transparency.

Building on the complexities of personalized service, trust emerges as a fundamental factor shaping the relationship between service agents, technology, and organization (Theme 6: Collaborative Service). Trust influences willingness to rely on AI systems and the perceptions of fairness, transparency, and support in hybrid work environments. With evolving roles and increasing human-machine interdependence, understanding trust is critical. Future research should explore the mechanisms for fostering trust, considering technological design and organizational practices that align with professional values and psychological needs of service agents.

Technology integration reshapes autonomy, identity, and work dynamics of service agents (Theme 3: Service Agent Well-Being). As tasks become increasingly augmented, they face shifting boundaries in decision-making and responsibility, raising concerns about professional identity and potential deskilling (Theme 6: Collaborative Service). Studies highlight that this shift increases cognitive demands and requires service agents to continuously adapt to novel human-technology work arrangements (Gnewuch et al., 2023). Future research should examine training, onboarding, and interface design to support adoption, foster appropriate trust in AI, and preserve autonomy and purpose while minimizing displacement concerns.

As customization deepens, it raises critical questions around data, transparency, and fairness (Wirtz et al., 2022). The growing reliance on data necessitates compliance with evolving regulations, such as the GDPR and the European AI Act, particularly as personalization overlaps with high-risk categories (e.g. affective computing) (Kusche, 2024). Research should examine how organizations navigate regulations and how these shape technology design. Efforts should focus on bias mitigation and explainable AI to ensure compliance, build trust, and deliver equitable services (Nishant et al., 2024). It is essential to distinguish between technology augmentation and complete automation in service environments (Theme 6: Collaborative Service). Confusing the two risks, overlooking human expertise and managerial choices, particularly when automation is overused in ways that weaken agent autonomy and service quality. Future work should clarify which tasks are best suited for augmentation versus automation, and how these boundaries shift as technologies evolve. Together, these questions indicate that the same personalization mechanisms can enhance customer relevance while increasing employee strain, underscoring the importance of understanding how augmentation and customization jointly shape the service experience.

Looking ahead, the development of frontline service is likely to move further into hybrid constellations in which conversational agents and human employees collaborate. Industry projections estimate that the worldwide conversational AI software services market will approach USD 32 billion by 2028 (IDC, 2024). This anticipated growth underscores the urgency for research on role division, customer acceptance, and employee well-being in hybrid service environments.

This review advances our understanding of how technology can augment rather than replace human service agents. Recent research conceptualizes human–machine collaboration as teamwork rather than substitution (Kunz et al., 2025). Our review develops this perspective through six themes of augmentation, organized along the dimensions of technological augmentation and service customization. This structure provides a lens for interpreting the evolving role of technology in service interactions and for clarifying how augmentation differs from automation.

This review advances service theory by framing technology as a dynamic resource interacting with human judgment along two independent yet complementary dimensions, technological augmentation and service customization. STS theory emphasizes that performance depends on optimizing social and technical subsystems (Trist and Bamforth, 1951), whereas SDL treats employees and technologies as operant resources that co-create value in context (Vargo and Lusch, 2008). The framework clarifies when technology strengthens or undermines outcomes. It explains why monitoring in high-volume settings often feels like surveillance, while emotion or interaction guidance in advisory contexts is experienced as support. These contrasts remain invisible in single-axis perspectives and show how different combinations of augmentation and customization shape value creation. As Table 2 shows, these tensions determine whether augmentation supports employee well-being, authenticity, and performance, providing a lens to understanding how technology can sustain human strengths rather than diminish them.

First, the review offers a human-centered perspective. Prior work has often emphasized automation, where systems replace human labor and streamline processes (Castelo et al., 2023; Chi et al., 2020; Huang and Rust, 2018). In contrast, our synthesis demonstrates how technologies complement human strengths, such as empathy, judgment, and contextual adaptation. Treating augmentation and customization as complementary dimensions shows how technology can strengthen rather than weaken employee capabilities. This extends earlier accounts of digital transformation and responds to repeated calls to theorize human-technology collaboration in service interactions (Larivière et al., 2017; Marinova et al., 2016).

Second, the review integrates cognitive, behavioral, and emotional dimensions of frontline work. The six themes show that augmentation reaches beyond information processing and task efficiency. Technologies also shape emotion regulation, monitoring, and well-being, which are central to the employee experience (Bakker and Demerouti, 2007; Grandey, 2000; Hülsheger and Schewe, 2011). STS theory explains these outcomes as the result of new couplings between technical tools and human practices. In contrast, SDL emphasizes that value creation unfolds not only for customers but also for employees within a specific context. The framework, therefore, broadens service theory by linking technological change directly to the lived experience of frontline employees.

Third, the review offers a structured approach to interpreting variation across service settings. The framework treats technological augmentation and customization as independent axes, which clarifies why some organizations achieve efficiency gains without personalization, while others deliver deep customization without heavy reliance on technology. This resolves inconsistencies in the literature, where findings on technology adoption, employee strain, and customer outcomes often appear fragmented and conflicting. It also provides a foundation for comparative studies that examine how different combinations of technological augmentation and customization influence value creation for both firms and employees (Edvardsson et al., 2011; Payne and Frow, 2005).

Finally, the review establishes a cross-disciplinary foundation for future service research. Core capabilities from computer science are connected to service theory through the framework (Russell and Norvig, 2020). This integration links technical progress with human-centered theories of value co-creation, showing how underlying methods reshape the work of service agents and the design of service systems. In doing so, the review positions service research to engage more fully with advances in adjacent fields while retaining its focus on the human role in value creation.

The framework also clarifies why prior findings sometimes diverge. For instance, monitoring has been found to raise reliability while also increasing strain (Ravid et al., 2023), and emotion recognition technologies reduce uncertainty in some studies but heighten concerns about authenticity in others (McStay, 2018). Organizing these tensions along augmentation and customization reveals that they are not contradictions, but rather outcomes of different design choices. The framework further specifies paradoxes that emerge in service contexts. The same intervention can increase reliability while reducing human judgment, strengthen personalization while undermining authenticity, and enhance efficiency while heightening strain. In transactional service settings, augmentation is often experienced as surveillance because systems prioritize compliance and throughput. In advisory or mixed-reality environments, augmentation is typically interpreted as support because it reduces uncertainty and aids judgment. These contrasts help explain why similar tools evoke different reactions across settings.

This review yields several implications for service practice. The framework shows that the benefits of technology depend not only on technical capability but also on the balance between augmentation and customization in the design of frontline service. Managers need to recognize that technologies can support service agents while preserving human judgment. Still, they can also reduce effectiveness when tools impose rigid rules or create surveillance without adequate support.

First, investment decisions should focus on technologies that extend rather than replace employee capabilities. Tools that deliver collaborative support strengthen both efficiency and service quality, whereas misdirected automation strategies risk reducing engagement and trust (Wirtz et al., 2018; Wu et al., 2015). Managers should therefore evaluate new systems not only for throughput gains but also for how they enable employees to integrate technical input with empathy and contextual understanding.

Second, monitoring and transparency practices must be managed together. Technologies that track workload, emotional signals, or performance can either protect or undermine engagement depending on how they are designed and implemented. When feedback is framed as developmental support, systems can enhance motivation and lower strain. Framed as control, the same systems increase pressure and turnover risk (Bakker and Demerouti, 2007; Ravid et al., 2023). Managers should therefore embed monitoring within broader practices that provide resources and opportunities for recovery.

Third, interaction intelligence should be deployed beyond post-interaction reporting. Real-time interaction support and insights allow employees to tailor responses and handle (emotionally) complex interactions more effectively. This requires careful design to ensure that cues are presented as options rather than prescriptions, enabling employees to integrate technological input with their own judgment (Dietvorst et al., 2015).

Finally, sustaining the value of augmentation depends on more than technical capability. Managers need to recognize that no configuration offers a universal solution and that an organization’s position on the two axes will evolve as technologies and expectations change. Deliberately adjusting investments enables firms to balance efficiency and human strengths. At the same time, maintaining customer trust requires transparency and responsible use of data. Technologies that detect emotion or personalize recommendations rely on sensitive information and must be governed by clear ethical safeguards to show that augmentation is designed to support both customers and employees rather than to displace or manipulate them. Strategically and responsibly approaching these developments enables firms to preserve service quality while protecting employee well-being and customer confidence.

The supplementary material for this article can be found online.