Value cocreation and codestruction in artificial intelligence-enabled service interactions: literature review and research agenda Open Access

The adoption of artificial intelligence (AI) is considerably altering service encounters as it takes on a central role in the interaction between providers and customers (Larivière et al., 2017). The modifications are due to the adaptation of AI for performing various tasks and the development of different intelligences (Huang and Rust, 2021), minimizing the human presence in service interactions (Bolton, 2020; Wirtz et al., 2018). AI-enabled technologies, such as virtual agents and service robots, promote a human−AI interaction format that is progressively changing business (Huang and Rust, 2021). In the face of technological transition and transformation, researchers have highlighted research gaps related to the need for a broader understanding of service interactions, mainly considering changes in the customer−provider relationship and customer experience (Bolton, 2020; Ostrom et al., 2021; Rosenbaum and Russell-Bennett, 2021).

With the adoption of AI in service encounters, it is pertinent to revisit the value cocreation approach to understand how it will be affected, considering that service interactions are its focus of analysis. In cocreation, value is seen as increasing the well-being of actors and is (co-)created in the exchange of services through interaction between actors and the integration of differentiated resources (Vargo and Lusch, 2016). AI can change users’ perceptions of value, given less human participation in service encounters (Ameen et al., 2021). It should be noted that there are advantages and disadvantages to adopting digital and technological services, and undesirable consequences may occur (Rosenbaum and Russell-Bennett, 2021). In this sense, on the one hand, the personalization of the offer and the greater possibility of access to services are factors that can be considered favorable and, on the other hand, privacy and trust are concerns of customers in these service encounters with AI (Ostrom et al., 2019).

Given this, service interactions enabled by AI can be positive but also problematic. In this way, just as value is cooperatively created – co-created – (Vargo and Lusch, 2016), it can also be collaboratively destroyed – co-destroyed (Echeverri and Skålén, 2021; Plé and Cáceres, 2010) in these interactions. Value codestruction is “a process of interaction between service systems that results in a decline in the well-being of at least one of the systems” and occurs due to the misuse of resources (Plé and Cáceres, 2010, p. 431). Studies on value cocreation in marketing have developed considerably in recent years; however, research on codestruction is incipient (Echeverri and Skålén, 2021; Järvi et al., 2018; Laud et al., 2019).

Some literature reviews have explored value cocreation in the hotel and tourism segment (Carvalho and Alves, 2023), interactive value formation from the perspective of value codestruction (Codá and Farias, 2022), cocreation and codestruction supported by information technologies (Li and Tuunanen, 2022) and AI and robots in value cocreation (Kaartemo and Helkkula, 2018). However, no known systematic reviews have jointly investigated the value cocreation and value codestruction in AI-enabled service interactions. Considering the above, this systematic literature review sought to answer the following question: How has the literature discussed the concepts of value cocreation and value codestruction in AI-enabled service interactions considering theory, context, characteristics and methodologies? To understand the question, this work aims to review the literature on value cocreation and value codestruction in AI-enabled service interactions, considering theory, context, characteristics and methodologies.

Organizations that invest in technologies and knowledge usually increase their competitive advantage (Bilgihan and Wang, 2016). Thus, research on AI technologies is essential to assist organizations in decision-making in a context of accelerated change. In addition, the study contributes to expanding the literature that investigates value formation in management and marketing and to the growing general interest in the themes of AI, robotics and automation. The review fills a gap in the literature because, as far as we know, this is the first systematic literature review that jointly explores the concepts of cocreation and codestruction from an interaction perspective with AI. We provide an overview of the reviewed literature, theoretical and practical implications and an agenda for future research.

Next, in part 2, the theoretical framework on value cocreation and codestruction is discussed, and a brief explanation of AI as a service facilitator is provided. In section 3, the method and protocol used in the development of the review are described. Subsequently, the systematic review results are presented in part 4, followed by discussions and implications in section 5. A research agenda is proposed in part 6, and finally, the paper concludes with the conclusion and limitations in section 7.

Theoretical perspectives for a service-centered logic, as opposed to a goods-centered dominant logic, have resulted in consumers’ active participation in interactive value creation with companies (Prahalad and Ramaswamy, 2004; Vargo and Lusch, 2004). In this direction, Vargo and Lusch (2004) proposed the service-dominant (S-D) logic, in which service is the basis of exchange, and value cocreation occupies a central space. Value cocreation occurs when actors interact and integrate their resources, aiming for the mutual well-being of those involved (Vargo and Lusch, 2016). Value can be defined by a value-in-use approach (Grönroos and Voima, 2013) and by value in context (Chandler and Vargo, 2011). It is emphasized that value has an experiential nature, that is, “value is always uniquely and phenomenologically determined by the beneficiary” (Vargo and Lusch, 2016, p. 6). Resources are essential to the value cocreation process and are classified into operant, such as knowledge and skills, and operand, such as goods and physical resources. In short, operant resources act on operand resources to create benefits (Vargo and Lusch, 2016). Technology is an operand resource and also an operant resource, as it influences institutions, human action and the way value is determined value (Akaka and Vargo, 2014).

Despite the theoretical and practical contributions, the S-D logic is controversial (Wang et al., 2019) due to the focus only on the positive side of value formation, not considering aspects of decrease or loss of value in the interactional process or the outcome of the interaction (Echeverri and Skålén, 2011; Plé and Cáceres, 2010). For this reason, Plé and Cáceres (2010) proposed the concept of value codestruction, as they argue that value can also have a negative side, reducing the well-being of at least one of the actors. Additionally, value codestruction can be explained by the difficulty of an actor following a script expected by other actors (Järvi et al., 2020), which can occur due to the intentional or unintentional misuse of resources (Plé and Cáceres, 2010). Given the gap in understanding, Echeverri and Skålén (2011) presented the interactive value formation (IVF) framework as a theoretical support that jointly encompasses value cocreation and codestruction, assuming that there are congruent or incongruent elements in the relationship between actors.

Research on value codestruction has focused, on resources and service ecosystems on the one hand and practices on the other (Echeverri and Skålén, 2021). Furthermore, in general, there is no unifying understanding of the concept, so some researchers consider value codestruction as a negative process specific to cocreation and others as a decrease in value in a space of value variation that has a direct relationship with the cocreation (Cabiddu et al., 2019; Echeverri and Skålén, 2021). Nonetheless, despite the conceptual distinction and interrelationship, there is a consensus on the occurrence of value codestruction (Lumivalo et al., 2023). Varied reasons are given for value codestruction, such as the absence of information, insufficient trust, mistakes, lack of clear expectations (Järvi et al., 2018) and poor understanding of how to integrate resources (Laud et al., 2019).

AI is a technology that imitates human intelligence to perform tasks in functions traditionally performed by humans, such as those related to understanding, language and learning (Huang and Rust, 2018), enabling interaction through natural language (Robinson et al., 2020). Associated with other technologies such as deep learning and big data, AI goes beyond the limits of some human capabilities, analyzing, for example, large volumes of data (Bock et al., 2020). However, it generally lacks human characteristics of an emotional and social nature (Wirtz et al., 2018). It is worth highlighting that AI intelligence levels develop in the direction of mechanics, thinking and feeling (Huang and Rust, 2021).

In service encounters, AI can act through sensing, learning, decision-making and actions (Bock et al., 2020) in a physical or digital environment permeated by interrelated technologies (Larivière et al., 2017). As an executor, AI enables customer interactions to deliver services (Ostrom et al., 2019). Examples of technologies that assume this type of function are physical assistant robots (Bock et al., 2020), chatbots (conversational agents) and voice-controlled digital assistants (Larivière et al., 2017). Some AI-enabled technologies are called service robots due to their learning, autonomy, adaptation, interaction and service-offering characteristics, which may or may not have a human appearance (Wirtz et al., 2018). Notably, AI can replace employees, customers or both (Robinson et al., 2020).

The systematic literature review method was adopted to carry out this work, which consists of the methodical selection of studies published on a given subject for understanding and synthetic explanation of the literature. Thus, it is possible to identify broader evidence and conclusions that guide and justify future research (Siddaway et al., 2019). The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol was used, which assists in detailing the criteria used in systematic reviews, following four steps to collect the review files: identification, selection, eligibility and inclusion (Moher et al., 2009). Furthermore, according to the instructions of Koseoglu and Arici (2023), content limits are explained below, considering the keywords used in the search for articles (identification of limits), selection of databases (search for limits) and reading titles, abstracts and full text (confirmation of limits).

In the identification phase of works for the composition of the collection, based on the problem and the research objective, the following search descriptors were used: ((“value co-creat*” OR “co-creat* value” OR “value co-destruct*” OR “co-destruct value” OR “Resource* integration” OR “resource* disintegration” OR “interactive value formation” OR “value formation” OR “co-creat*”) AND (“artificial intelligence” OR AI OR chatbot* OR robot* OR “virtual agent *” OR “virtual assistant*” OR “voice assistant*”)). The documents were retrieved from the Web of Science and Scopus databases, as they have many relevant publications and are, therefore, frequently used by researchers in the area of management and business (Koseoglu and Arici, 2023; Mariani et al., 2018). A search was carried out by topics, in which the defined terms could be identified in the titles, abstracts or keywords. Search engines, quotation marks and Boolean operators (OR and AND) were used to define the search filter criteria. Furthermore, “article” published in journals was defined as the type of work and “English” as the language. The searches were conducted in June 2023, and an initial date for including articles was not defined.

Aiming for reliability (Koseoglu et al., 2022), duplicate documents were extracted and verified automatically using the R software Bibliometrix package and manually in Excel spreadsheets. Thus, by eliminating duplicates, 335 articles were retrieved. Only articles belonging to journals classified in the Q1 and Q2 quartiles of the Journal Impact Factor Rank (JCR) or the SJR Scimago Rank were selected to ensure quality; therefore, 290 articles met this criterion. In the other phases, the focus of discussion for value cocreation and codestruction enabled by AI was considered an inclusion criterion. In sorting the articles by reading the titles and abstracts, 186 papers were eliminated. The full texts were read in the eligibility phase, and considering the inclusion and exclusion criteria, 48 empirical and theoretical articles were included in the sample for the review. The following (Figure 1) is an information flow diagram of the PRISMA protocol with the respective phases.

Data analysis followed a categorization of content based on the theory, context, characteristics and methodology (TCCM) framework by Paul and Rosado-Serrano (2019). At this stage, Excel software spreadsheets were used to categorize the data. Furthermore, the TCCM framework guided the presentation of results and the proposed research agenda.

This section presents the results found in the reviewed literature following the TCCM framework (Paul and Rosado-Serrano, 2019).

The reviewed literature did not present a uniform theoretical basis. However, S-D logic stood out as the most used approach in the theoretical foundation of the articles analyzed since value cocreation is an axiom of the respective theoretical lens that has occupied a relevant space in services marketing. There was also the use of literature on the theoretical construction of value cocreation from the service logic perspective (such as Grönroos and Voima, 2013). In a more limited way, the basis of the concept of codestruction was observed with the citation of the seminal articles by Plé and Cáceres (2010) and Echeverri and Skålén (2011). Considering the focus of the research, conceptual bases on AI and emerging intelligent technologies also supported the construction of the articles, as well as bases on robotics and automation. In a more fragmented way, other lenses, perspectives and concepts were also identified, such as technology adoption, customer experience, service interaction, human−computer interaction, diffusion of innovation, trust, human-machine cocreation and service innovation, among others.

With AI, the discussion of technology beyond operand resources, that is, as an operant resource, becomes evident around S-D logic. It was found that AI is simultaneously an operand resource as it is interfered with by knowledge, experience and skills to improve its performance, and it is also an operant resource, as AI learns and acts through algorithms to improve the performance of humans (Paschen et al., 2021). Thus, it is an operant resource as technology assists and expands decision-making capacity in complex contexts (Barile et al., 2021; Kondapaka et al., 2023). For Mele et al. (2021), AI can affect the agency power of actors and interaction practices in the service ecosystem, potentially expanding access to resources and improving engagement. Notably, AI acts as an operant resource when following interpretative schemes designed by humans, since only humans have the interpretative capacity related to decision-making that transcends data manipulation and adheres to human values and beliefs according to the context (Barile et al., 2021).

Based on the reviewed literature, co-creating value in AI-enabled interactions is a complex process (Paschen et al., 2021) between human and autonomous nonhuman actors (Grundner and Neuhofer, 2021; Neuhofer et al., 2021) where roles, resources, nature and quantity of contributions differ (Paschen et al., 2021). Data is essential in this service interaction (Breidbach and Maglio, 2020). Furthermore, understanding the value cocreation in AI-enabled services depends on ecosystem analysis, considering the various actors interacting (Leung and Loo, 2022; Payne et al., 2021a). In short, it is a recent and progressive phenomenon that depends on human and nonhuman behaviors (Zhu et al., 2022), requiring theoretical deepening (Wen et al., 2022).

Value codestruction processes can occur in interactions with AI, causing a reduction in the well-being of at least one actor (Castillo et al., 2021). As described by Čaić et al. (2018), AI has, at the same time, the potential for value cocreation and codestruction, causing trade-offs in the value formation since, in the case of socially assistive robots in elderly care, the beneficiary may recognize it as an invader of their privacy and, on the other hand, a family member may consider it a facilitator. This duality in value formation is portrayed by Neuhofer et al. (2021) as “ambivalent value cocreation,” as AI technologies can fail, causing annoyance, fear and increasing waiting times and, on the other hand, AI has, among other potentialities, the ability to make highly personalized offers.

Regarding the publication period (Table 1), it was found that 2018 was the first year in which publications with the search focus used were presented. The years 2021, 2022 and 2023 had the highest number of published works, showing that the theme is recent, under development and there is a growing research interest.

Based on indications of the location where the data were extracted, it was found that most of the empirical articles, approximately 40%, were developed with data collected in China. Less significantly, studies were identified in Australia, Austria, Belgium, Canada, the USA, France, India, Italy and Malta. In some studies, data were collected in several countries; in others, the data collection location was not reported. In addition, we found that the articles were mainly made in the banking, hospitality and tourism and health sectors. Only one investigation focused on the public sector.

The characteristics are divided into two categories based on the discussion focus of the analyzed articles: an approach to value cocreation and an integrated approach − value cocreation and codestruction. Of the 48 papers analyzed, 41 discussed value cocreation, and seven dealt jointly with value cocreation and codestruction.

With the evolution and massive adoption of AI, integrating the digital, physical and social domains will shape the customer experience and value cocreation in environments that tend to gradually become of high digital density (Bolton et al., 2018). Value cocreation with AI is a prominent theme in AI research in marketing (Anayat and Rasool, 2022). Chatbots, service robots, machine learning and natural language processing are some AI-based technologies influencing value cocreation (Solakis et al., 2022). In tourism research, the experience in augmented reality and virtual reality for the value cocreation process was identified as a growing thematic field (Nannelli et al., 2023). In this context, psychological distance and cognitive processing influenced the adoption of digital services (Jia et al., 2023).

AI influences the value cocreation process by expanding the capabilities of self-understanding, control and action (Mele et al., 2021) and helping decision-making in a complex environment (Barile et al., 2021). In this direction, humans and AI can act harmoniously to co-create value in decision-making (Kondapaka et al., 2023). Human actors are cited as essential for the value cocreation with AI, as the interaction depends on their domain in the development and use of technologies; in addition, they act in the continuous improvement of algorithms (Paschen et al., 2021).

Perceived personalization, autonomy, community identity, trust and self-efficacy influence consumer value cocreation behaviors toward AI (Wen et al., 2022), as well as efficiency, degree of control, content quality and information security (Zhu et al., 2022). Also, Gao et al. (2023) found that the interactivity perceived by customers based on AI stimuli positively affects value cocreation. Solakis et al. (2022) pointed out that customer perceptions, attitudes, confidence, social influence, hedonic motivations, anthropomorphism and experience can influence the value cocreation process. Moreover, AI service quality can promote a more pleasant cocreation experience (Yang, 2023). Furthermore, data-centric business models impose ethical challenges and change actors’ roles in value networks (Breidbach and Maglio, 2020).

A considerable part of the reviewed studies discussed using service robots in the hospitality and tourism areas. In the hotel segment, the intention of co-creating value for customers was observed by the very novelty of the robotic service (Xie et al., 2022b), as well as by the high proactivity of tourist service robots, favored by empathy with robotics (Xie et al., 2022a). However, the dehumanization of the process is also evidenced (Buhalis et al., 2019). The perceived privacy, functional benefits and appearance of service robots are drivers of value cocreation (Lin and Mattila, 2021). The robotic services available in restaurants were seen as contributors to the value cocreation process with the customer (Jain et al., 2023), positively affecting customer attitudes (Chuah et al., 2022). One research with a restaurant service robot showed that the attributes role meaning, competence, social presence, cordiality, autonomy and adaptability promoted value facilitation effects in the process of cocreation (Zhang et al., 2022). In addition, customers can have improved experiences by integrating intelligent technologies into the gastronomic service network (Leung and Loo, 2022).

In the health area, trust is a central element for value cocreation, in which AI innovations can favor well-being from a transparent approach (Megaro et al., 2023). Furthermore, the human−technology interactions provided by cognitive assistants intensify the value cocreation through the possibility of sharing and integrating resources from various network actors (Mele et al., 2022). According to Mabillard et al. (2022), AI has a strong potential to support diagnosis and treatment in the health area, and it should be used to strengthen the cocreation relationship between the doctor and the patient without replacing human interactions. In health services, the well-being of actors can be enhanced through shared control, increased dialogue and access to information (Dodds et al., 2022), as well as through collaborative decisions with AI (Liu et al., 2023).

Technological discomfort influenced the adoption of robo-advisors in the banking/financial sector due to their simplicity and lower requirement for customer participation (Flavian et al., 2022). In AI-enabled mobile banking systems, the service delivery configuration, safety and security influenced value cocreation (Payne et al., 2021b). In B2B segments, AI enables value cocreation when network actors, especially end users, are involved in developing technological solutions (Leone et al., 2021; Li et al., 2021). Research on AI-enabled smart TV has revealed that cocreation is an antecedent of perceived usefulness (Gao and Huang, 2019), and two-way communication and personalization affect cocreation (Gao and Huang, 2021).

Regarding virtual assistants, it was found that dialogue and access to information are benefits of these technologies in the cocreation process and influence consumers’ online trust (Castellano et al., 2018). Voice assistants can support cocreation through a more personalized service (Buhalis and Moldavska, 2022); in that direction, the interaction between people with disabilities and voice assistants resulted in well-being (Vieira et al., 2022). Super-functionality, personalization and convenience influenced consumers’ perception of value cocreation in the use of service chatbots (Lalicic and Weismayer, 2021). On the other hand, high emotional complexity was highlighted as a criterion for human agents to take over a conversation between customers and AI (Saviano et al., 2023). In public service, four dimensions of value in use were identified from citizens’ interactions with virtual agents: path to human support, engagement/interaction, connection and personalization (Scutella et al., 2022).

On the dual potential of value cocreation and codestruction, Čaić et al. (2018) identified that socially assistive robots in elderly care can assume different roles in the value network, for example, as a facilitator, intruder, ally and substitute. In another research, Čaić et al. (2019) mention that the potential for value cocreation/codestruction depends on dimensions of social cognition. As mentioned by Grundner and Neuhofer (2021), AI can minimize the customer experience as social contact is reduced, leading to the value codestruction due to the massive use of technology; on the other hand, when AI provides personalized offers and information in real time, the experience can be positive by promoting the value cocreation.

Neuhofer et al. (2021) mentioned that the degree of AI interference in the customer experience, such as authoritarian intrusion or encouraging assistance, can lead, respectively, to value codestruction or cocreation. According to Hsu et al. (2021), when customers consider themselves responsible for poor resource integration, they tend to continually co-create value with the company; however, they are prone to co-destroy value when they realize that the company is at fault. Hottat et al. (2023) point out that the profile of the public and the context are determinants of the potential for cocreation/codestruction; for example, automation is not well accepted in high-risk services (hospital services) and high-priced services (luxury restaurants and hotels); in these cases, human interaction is preferred.

Castillo et al. (2021) more specifically addressed the codestruction process in AI-enabled service interactions, drawing on the value-cocreation literature. The authors identified authenticity problems, cognition challenges, affective problems, functionality problems and integration conflicts as antecedents of codestruction from a client perspective using chatbots. They also reported that negative interactions trigger losses of customer resources, leading to the choice of more expensive service channels, service termination and public complaints about the company (Castillo et al., 2021).

Of the works analyzed, more than 70% consist of empirical studies, followed by theoretical-conceptual studies, systematic literature reviews and bibliometrics. As for the methodological approach, it was found that 56% of the studies used a qualitative approach, 33% quantitative and 10% mixed. It should be noted that some empirical research was configured as theoretical-empirical, as it involved the development of a theoretical or conceptual model. In qualitative empirical studies, data collection was predominantly carried out through interviews and focus groups, and data were analyzed using thematic analysis and coding. On the other hand, quantitative empirical research is primarily carried out through surveys, in which the most recurrent data analysis was structural equation modeling.

AI disruptively changes the conventional structure of relationships between companies, customers and other service actors in a service ecosystem, implying different interaction formats by replacing one or more human actors in service encounters (Robinson et al., 2020). This technology has been discussed as an operant resource of one of the actors, commonly the service provider (Castillo et al., 2021). However, the discussion goes beyond aspects of framing AI as operand and operant resources and gains evidence as an actor integrating these resources and interacts socially using increasingly natural language. Given the above, AI is considered an autonomous nonhuman actor (Grundner and Neuhofer, 2021; Neuhofer et al., 2021), which imposes complexity on service interactions and value formation (Paschen et al., 2021).

In progressively digital and technological service ecosystems, interactions between human actors and AI-enabled autonomous nonhuman actors (human−AI interaction) can result in value cocreation or value codestruction, or both. These value formation phenomena are also likely to occur in service encounters where more than one actor is replaced by AI (AI−AI interaction). AI, as a nonhuman actor, will increasingly act independently, considering that it is in the early stages of development. As the intelligence levels of this nonhuman actor evolve (Huang and Rust, 2021), it will become more autonomous and popular in service encounters. Consequently, the greater the development of AI, the greater its ability to achieve the premises of value cocreation, moving closer to successful human interactions. It is noteworthy that AI autonomy (Neuhofer et al., 2021) is guided by human actors and follows human interpretative systems (Barile et al., 2021), which is why we reinforce that AI needs to be guided by principles that minimize prejudices, discrimination and undue favoritism.

AI expands the potential of value cocreation due, for example, to scaled service, convenience, availability and personalization (Buhalis and Moldavska, 2022; Lalicic and Weismayer, 2021; Neuhofer et al., 2021; Payne et al., 2021b). However, risks and challenges related to value codestruction are evident, such as a lack of human touch, empathy, emotional intelligence, transparency, security, data privacy, and the existence of algorithmic bias (Bolton et al., 2018; Breidbach and Maglio, 2020; Castillo et al., 2021; Wen et al., 2022; Zhu et al., 2022). Furthermore, the resources integrated by AI are sometimes limited, and it may not progress in conversation and resolve user demands (Saviano et al., 2023).

Data are essential resources for AI to operate and are provided by actors consciously or unconsciously (Breidbach and Maglio, 2020). In this aspect, the lack of trust for integrating resources in the service relationship with technology, for example, due to ethical issues, can threaten the value cocreation process (Paschen et al., 2021). Therefore, the value formation in AI-enabled interactions inevitably presupposes complete and reliable data integration; thus, organizational and legal instruments that guarantee transparency and secure data sharing are needed.

In comparison, often in a relationship between humans, even if some resources integrated by the service provider are limited, human communication and emotional intelligence can alleviate the user’s discontent and calm tensions. On the other hand, AI is promising for its broad cognitive capabilities, but a challenge for value formation lies in developing its affective aspects, especially when customers are disappointed with the lack of understanding and empathy for human emotions (Castillo et al., 2021). Therefore, AI technologies must evolve in affective and subjective aspects to provide a more satisfactory user experience and minimize the chances of value codestruction.

Considering that AI can make humans’ operant resources more potent by improving their decision-making capabilities (Barile et al., 2021), we reinforce that the combination between them can be promising for value cocreation. Services must be supervised and, when necessary, carried out or completed by humans who utilize AI as partners, enabling the enhancement of human intelligence. In short, a synergy between humans and AI tends to promote superior value outcomes in service ecosystems, as the capabilities of both can be leveraged.

The articles identified in this review focused mainly on value cocreation, and few articles addressed value codestruction. Despite the findings in the results, there is no broad and deep understanding of the antecedents, manifestations, causes and processes of value cocreation enabled by AI and even less about value codestruction due to the low volume of publications.

Regarding the theoretical foundations, Vargo and Lusch (2017) point the S-D logic as a cohesive general approach; however, they do not address the possibility of value codestruction (Echeverri and Skålén, 2021; Plé and Cáceres, 2010). Therefore, we endorse the need to expand the scope of the S-D logic axioms (Plé and Cáceres, 2010; Wang et al., 2019) based on the recognition that the interaction and integration of resources are subject to uncertainties and complexities that can cause value codestruction, especially in a period of rapid and intense adoption of AI. Thus, if the S-D logic encompassed the codestruction construct, the perspective could offer a more balanced and integrative analysis lens that aligns with reality and organizational practice. Echeverri and Skålén (2021) recognize that the value cocreation and codestruction are part of the same analysis structure, the IVF. Nonetheless, theoretical advances are needed regarding AI, as a nonhuman actor, being involved in IVF and affecting the conceptual structure of IVF in terms of transition between constructs.

This research responds to calls for investigation made by Bolton (2020), Ostrom et al. (2021) and Rosenbaum and Russell-Bennett (2021), considering the rapid diffusion and adoption of emerging technologies. This way, we contribute with researchers and academics by presenting an overview of the literature on value cocreation and codestruction in AI-enabled interactions. To our knowledge, this is the first systematic literature review that jointly explores the respective concepts related to AI usage. Furthermore, we collaborate with the understanding of the themes, especially when considering the negative side of value formation. The results and research agenda were presented following the TCCM structure (Paul and Rosado-Serrano, 2019), presenting the theory, context, characteristics and methodology of the articles analyzed. The suggestions and research questions aim to collaborate with developing new studies and the field’s maturation.

This research provides organizations with information that guides them in adopting and managing AI technologies for customer interaction and customer service. In a period of progressive use of AI and changes in business processes, organizations must know how to use it ethically and strategically. Therefore, the presented results help managers assess the potential for value cocreation/codestruction in the adoption and application of AI, providing knowledge that supports organizational policies and assists in decision-making. For example, when implementing AI technologies, managers need to evaluate the potential for value cocreation according to the public profile, as in some segments, the technologies may not be well accepted (Hottat et al., 2023), while in others, there are advantages that strengthen the customer experience.

There are many possibilities for applying AI to customer service, interaction and service provision, such as the use of virtual assistants and service robots, as well as through the improvement of self-service and automation platforms. However, there are challenges and risks related to the loss of well-being, which require planning, responsible monitoring and focus on the beneficiary. There is, for example, the need to develop cognitive and affective aspects, including empathy and emotional intelligence. In addition, it is noteworthy that some sectors, such as hospitality and tourism, can benefit more from the findings already highlighted in the literature while facing the ongoing challenge of closely monitoring sector trends.

Table 2 summarizes the research conclusions and implications.

It was found that there is a need for greater understanding and theoretical and empirical deepening related to the value cocreation and codestruction in AI-enabled service interactions. For this reason, a research agenda is proposed below with lines and questions for investigation, following the TCCM framework (Paul and Rosado-Serrano, 2019).

Theoretical foundations – Toward a comprehensive understanding of value formation in interaction with AI, we suggest the integrated use of the S-D logic theoretical perspective with the concept of value codestruction and also with the IVF structure to, from this, jointly investigate the constructs value cocreation and value codestruction, verifying how the transition between them occurs. We encourage the expansion of research on AI as an operant resource and, more importantly, as an autonomous actor integrating resources. We also suggest investigating how AI affects value formation at the ecosystem level. As well as examining the intentional and unintentional value codestruction in interactions with AI. Given the above, we present below some research questions (RQ):

Other consolidated theoretical lenses can help to understand the phenomenon, such as the actor-network theory (ANT), in which controversies, negotiations and conflicts exist in a network whose interaction can occur between human actors or also between human actors and nonhuman actors (Latour, 2007). It is up to the study of AI to understand the extent to which technology and algorithms (or even algorithmic biases) exercise agency (“distributed agency”) in decisions over which one would expect that only the human actor could exercise. However, as in the machine learning process, the machine is taught (and learns), and then we transfer to it, in a certain way, the power of agency in certain processes of interaction and dispute resolution. Gutiérrez (2023, p. 4) opines that “ANT can help understand how actants and human actors are mutually dependent and influential in a network of interactions that shape the outcomes of their joint endeavors”. ANT shows that relationships between humans and AI (nonhuman actant) are not passive. Actors may use shortcuts or hacks to take advantage of programs, consciously and critically/actively explore algorithm actions, or surrender to them. It is necessary to recognize that the system adjusts its approach according to feedback from human actors (Gutiérrez, 2023). ANT is a theory that can serve as a theoretical-philosophical-conceptual basis and even a methodological approach for studying algorithmic bias, equity, relationship of forces, controversies, and ethical, technological and systemic challenges in implementing AI. Some research questions can guide this discussion:

Furthermore, we suggest the lens of institutional theory (North, 1990; Scott, 2013) to help understand how institutions can promote well-being in AI-intensive service ecosystems. Institutions are rules, norms and cognitive-cultural beliefs that shape organizational and social behavior (Scott, 2013). Therefore, this theoretical current can serve in the investigation of value formation and can guide or shape the development and adoption of AI, such as, for example, institutional influence (Scott, 2013), stability, institutional change (North, 1990), legitimacy and isomorphism (DiMaggio and Powell, 1983; Scott, 2013). Thus, we raise some more questions for future research:

Context – We encourage the replication of research in different territories, as most research has been carried out in Asia, emphasizing China. It should be noted that different characteristics between emerging and developed countries, with different maturity levels in applying AI, can bring results that contribute to a detailed understanding of the subject. It is recommended that investigations be expanded beyond the hospitality and tourism, banking and health sectors so that other sectors also benefit from knowledge about value formation made possible by AI. As previously mentioned, the public sector has a significant research gap. Furthermore, adopting AI affects the labor market in different sectors; thus, the notion of social value cocreation/codestruction in the context of AI can assist in more sustainable value processes, analyzing the impacts for various actors in the ecosystem. We also recommend studies with micro and small companies so that they can be favored with AI and not be disadvantaged compared to the technological investment capacity of medium and large companies. We suggest investigating different aspects of user−AI interaction in virtual and physical environments associated with the Internet of Things (IoT) and related to domestic, commercial and public use. We also suggest that researchers evaluate the value cocreation/codestruction in different technologies that apply AI. As research questions, we formulate:

Characteristics – In service interactions between the human actor and AI, there are two almost unexplored gaps in the literature:

1. the specific discussion of value codestruction; and

2. the integrated discussion of value cocreation with value codestruction.

Because of the above, it is recommended that research be carried out that explicitly covers these gaps, for example, mapping antecedents, manifestations, elements and consequences of value cocreation and codestruction. These searches will also be useful in cases where the AI represents more than one actor. Considering some sensitive issues in the use of AI that can minimize the well-being of actors, there is a need for more research on privacy, ethics, governance, security, transparency and algorithmic bias. In addition, some research questions are pointed out:

Methods – From a methodological standpoint, the development of empirical studies, such as qualitative research, is recommended, which allows for an in-depth understanding of the phenomenon, especially regarding research associated with technologies still in the initial phase of diffusion and adoption. On the other hand, quantitative studies are needed to generalize findings on human−AI interactions, and experimental research helps obtain greater explanatory power. In this aspect, it is advisable to develop and validate scales. Moreover, longitudinal studies help track value formation over time and can help answer the following research questions:

This review revealed that interactions between human actors and AI-enabled autonomous nonhuman actors (human-AI interaction) can result in value cocreation or value codestruction, or both. Furthermore, these phenomena are also likely to occur in service encounters where more than one human actor is replaced by AI (AI−AI interaction). AI is seen as an increasingly independent nonhuman actor that integrates resources and interacts with other actors in service ecosystems. However, caution and balance are needed in its adoption, given the potential for mixed and multifaceted value formation.

Research on value formation in interactions with AI is in the introductory phase, with a low volume of publications. The greatest research emphasis is associated with value cocreation, and few articles have directly addressed the concept of value codestruction. In short, these are “hot” topics for research and theoretical-empirical deepening, given the progressive adoption of AI by organizations. As a way of contributing to the field, this work presents a detailed research agenda.

The review followed a rigorous and transparent process. Nonetheless, we point out some methodological limitations that may have limited the scope of the analysis, such as the selection of only two databases for retrieving articles, Web of Science and Scopus. Also, the journal selection criteria were based on scientific indicators, and although they ensured the selection of articles of recognized quality, they may have eliminated relevant works. Furthermore, only scientific articles published in English were retrieved. In future reviews, it is recommended to explore the topic with a broader scope, including research on AI and other associated technologies, such as the IoT, big data, and virtual reality, and to expand the databases researched.