The deployment of AI chatbots is driving significant changes in the way that service is delivered and experienced (Belanche, Casaló and Flavián, 2021; Ostrom et al., 2021). The co-creation literature emphasises actor-to-actor networks that create value for each other (Vargo and Lusch, 2004); however, the introduction of AI chatbots in the frontline adds an uncharted dimension to the study of co-creation, as AI developments promise value co-creation that changes as the AI adapts to other actors (such as customers), and these actors then adapt to the AI. AI chatbots take an active part in the service encounter, and as pseudo frontline employees, become important actors in the value co-creation process (Verhagen et al., 2014).

However, chatbots often fall short of customer expectations, resulting in service failure (Sheehan, Jin and Gottlieb, 2020). In this manner, the failure can also be perceived as co-created, since customers invest considerable time and energy (resources) in order to interact with the chatbot and co-create the interaction (Heidenreich et al., 2015). The negative effects of such service failures may be further compounded if customers are not allowed to choose their preferred customer representative: a human agent or a chatbot; or if customers perceive to be interacting with a human, but would be interacting with a chatbot instead (Robinson et al., 2020). Although previous literature extensively examined the determinants of consumer use and adoption of novel technologies, extant research has largely been conducted in voluntary co-creation contexts, and has neglected the investigation of less harmonious forms of co-creation, such as forced co-creation contexts.

Different co-creation contexts are also likely to be influential in forming customer expectations regarding the forthcoming co-creation experience, the chatbot’s performance, as well as possible chatbot limitations (Crolic et al., 2022). The nature of such expectations can have a significant impact in shaping customer evaluations about their participation in co-creation (Nijssen, Schepers and Belanche, 2016) and eventually in assigning attributions for failure (Belanche et al., 2020). However, the literature only offers contradictory findings regarding the possible impact of different co-creation contexts on customer expectations, and responsibility attributions.

Against this background, the aim of this study is to investigate not only the customer perception of chatbot interactions in different co-creation contexts, but also to understand how such contexts may influence customer expectations and the resulting responsibility attributions when chatbot failure occurs.

Guided by a Pragmatism research philosophy, a mixed methods approach was adopted in order to be able to effectively address the research aim. Specifically, an exploratory sequential design was employed, consisting of the collection and analysis of qualitative data in the first stage, followed by a quantitative phase which built on the findings obtained from the qualitative study.

The qualitative study comprised a total of 39 semi-structured interviews conducted in a face-to-face format. This study demonstrated how customers perceive three distinct co-creation contexts when interacting with chatbots. One context, volitional co-creation, is more aligned with the concept of voluntary participation and collaboration that is prevalent in the co-creation literature. Yet, the other two contexts, coercive and deceptive co-creation, are novel to the co-creation literature and as such, are worthy of further investigation. As a result, the qualitative study advanced a research framework and accompanying hypotheses regarding the impact of distinct co-creation contexts on expectations and attribution of responsibility.

Focusing on the investigation of coercive co-creation contexts, the hypotheses were tested through two experimental research studies conducted in a customer service setting. Data was collected from US participants (N = 315; N = 324), who, as part of the experimental research, were asked to interact with a chatbot that was specifically programmed for this study. Half of the participants were forced to interact, whereas the other half were given the illusion of choice between interacting with a chatbot and an alternative customer service channel. Structural Equation Modelling (SEM) was used to evaluate the experimental data.

The experimental studies found that in situations which result in service failure, customers who were forced to interact with chatbots, attribute more negative responsibility towards the company, than customers who were given a choice among several contact options. In such cases, customers blame the company directly for the failure. When compared to those customers who were offered a choice, customers who were forced to interact with the chatbot showed stronger controllability and stability attributions towards the company; in other words, they perceived the company as having had the ability to prevent the failure, whilst also perceiving the failure as more permanent and likely to happen again. These results were replicated within the two different service settings that were included in the experimental research. These results show that despite all the benefits associated with chatbots, it is important to understand why customers may feel forced into interacting with a chatbot, and the resulting customer evaluations following such interactions. Although the co-creation literature has shown a strong reliance on the notion of co-creation as a harmonious, voluntary collaborative activity, this study challenged this view and contributed to the co-creation literature by demonstrating the possibility of discordant forms of co-creation, which may especially arise in human-to-non-human interactions. Additionally, this study also contributes to the service failure literature by assessing the impact of service failures in an AI context and uncovering the underlying psychological processes following a co-created service failure between a customer and an AI chatbot.

The experimental research also found support for the mediating effect of disconfirmation of expectations. Forcing customers to interact with chatbots results in a more negative disconfirmation of expectations (the experience rated as worse than expected) than those customers who were given a choice. This, in turn, leads to stronger attributions of controllability, stability and causality towards the company. It is likely that forced involvement represents a more comprehensive level of mental involvement for customers (Reinders, Dabholkar and Frambach, 2008), in the process expecting more from the interaction, and ultimately being more disappointed when the service results in failure. This result confirms the role of customer expectations as a reference point when judging the actual service (Oliver, 2015) and contributes to the co-creation literature by establishing disconfirmation of expectations as an important link between the perceived freedom of choice when co-creating and the resulting attributions of responsibility.

The findings also suggest a number of strategic implications for managers who are considering the partial or complete replacement of human staff with chatbots in customer service settings.