INTRODUCTION:

Artificial Intelligence (AI) is expected to reshape practically all service sectors, as the industrial revolution did to manufacturing (Wirtz et al., 2018). One phenomenon of this transformation is the increasing usage of conversational agents such as chatbots in (online) service encounters, which some expect to be the dominant customer-firm dyad in the next years (Tsai et al., 2021). The benefits and key drivers for companies are clear, since chatbots facilitate time-independent customer support at marginal costs. Considering the importance of immediacy for service recovery actions (e.g. Smith et al, 1999), the implementation of AI-based chatbots promises decisive benefits in this context. Despite the practical relevance, the effect of chatbot initiated service recoveries (Grégoire and Mattila, 2021) and transferability of established concepts in this context (Blut et al., 2021) remain underexplored in current service marketing literature.
Against this background, we aim to gain a better understanding of factors that contribute to service recovery satisfaction in human-chatbot-interactions. We address two major research questions: (1) Are there differences regarding the influence of satisfaction determinants between chatbot and human initiated recoveries like justice perception? (2) To what extent does a chatbot’s social cognition influence customers’ recovery process evaluation?
As theoretical foundation, we use the Computers are Social Actors paradigm (Nass et al., 1994) and Role Theory (Solomon et al., 1985) to explain differences regarding customers’ service recovery evaluation. Solomon et al. (1985) assume that in dyadic service encounter interactions, a customer’s evaluation is determined by his or her expectations of the counterpart and that congruence must prevail. We assume that the higher the degree of congruence between role and actual behavior, the more likely complaint satisfaction occurs and exceeding the expected lowers its effectivity.

METHODOLOGY:

A scenario based between-subject-design was conducted. The participants (sample size: 186 US customers) were confronted with a successful chat-based service recovery in a service context, either initiated by a chatbot or a human.

CONCLUSION:

Our analysis discloses some interesting results that can be used to optimize the effect of chatbot initiated service recoveries. First, the influence of interactional, procedural, and distributive justice significantly differs between human-human and human-chatbot interactions: while interactional and procedural justice show stronger influence for human interactions, customers especially value distributive justice in a chatbot recovery. Additionally, customers perceive interactional and procedural justice as less favorable in chatbot interactions, while service recovery satisfaction stays the same. Perceived competence and warmth moderate the effect of service recovery origin on interactional justice, suggesting that practitioners should design chatbots in a way that maximizes social cognition of chatbots in this vein. Based on our limitations several fruitful avenues for further research can be identified. First, research should examine recovery processes in different service settings and to overcome concerns with a solid intention measurement, field experiments with behavioral outcomes are needed. Second, researcher should analyze the impact of different types of chatbots (e.g. level of anthropomorphism and/or intelligence) on service recovery perception.

References upon request.

Services are prone to failures, therefore, ensuring effective recoveries is essential for service providers. When engaging in service recovery, companies "are essentially hoping for forgiveness" (Harrison-Walker, 2019, p. 2). Yet, customers' reactions to service failures and recovery experiences (SFR) are predominantly studied in terms of justice perceptions or service evaluations, while forgiveness remains unexplored. Typically, research takes the perspective of victims of the SFR, however, witnesses are also affected.
Early studies confirm that forgiveness is critical for positive service evaluations and favorable behavioral intentions (Babin et al., 2021; Tsarenko et al., 2019). Individuals' decision to forgive (decisional forgiveness) doesn’t necessarily coincide with the ability to let go of the grudge (emotional forgiveness). Companies can facilitate forgiveness by employing psychological (i.e., apology) and tangible (i.e., compensation) recovery strategies (Gelbrich and Roschk, 2011), with the varying effects of these strategies. Little evidence exists about the consequences of combined service recovery strategies. Moreover, it is essential to study these effects from the perspective of victims and observers of SFR since 1) services often happen publicly; 2) observers outnumber victims; 3) following the deontic justice principle, individuals react to injustice towards themselves and their surroundings; 4) literature reports inconclusive findings as to how customers react to observing service recoveries (Sharifi et al., 2017; van Vaerenbergh et al., 2013).
The 2x3 experiment aims at shedding light on the identified research gaps by answering the central research question: What is the relative effect of single versus combined service recovery actions on emotional and decisional forgiveness for victims and observers of SFR?
471 respondents, recruited via the online panel Prolific (www.prolific.io) were randomly assigned to one of the six scenarios describing harm direction: victim versus observer, and service recovery strategies: apology, apology and excuse, apology and compensation.
The results of MANOVA analyses reveal that service providers' efforts to recover the failed experiences facilitate forgiveness for both victims and observers. Notably, the effects are asymmetric. While offering compensation with apology increases the likelihood of decisional forgiveness among victims and observers, only for the latter group such decisions translated into emotional forgiveness. Consumed by negative emotions, victims might find it harder to come to peace with the service failure. Hence, they are likely to hold a grudge against the transgressor despite declaring forgiveness (Tsarenko et al., 2019; VanOyen Witvliet et al., 2001). Conversely, observers acknowledge the additional efforts and report a higher emotional forgiveness level, with effects being stronger for psychological than tangible compensation. The latter reflects the deontic justice principle (Cropanzano et al., 2003). The revealed asymmetry of the recovery effects on forgiveness is new to service and hospitality research, which either examined forgiveness as a unidimensional construct (Casidy and Shin, 2015) or reported symmetric effects for both emotional and decisional forgiveness (Shin et al., 2018). Surprisingly, the identified effects of recovery strategies on forgiveness were stronger for observers than for victims. Consequently, service recovery strategies are capable of affecting service evaluations not just for the person experiencing the failure but also for bystanders.

Organizations use different recovery tools such as monetary or psychological compensation to address customers’ complaints following service failures (Roschk & Gelbrich, 2014). Prior research shows that such measures can improve customer fairness perception and satisfaction (Tax et al., 1998), and also reduce customer anger and negative word of mouth (Gelbrich, 2010). However, there is limited research on the role of technology in service recovery. Specifically, while gamification has gained increased popularity in marketing over the past few years (Larivière et al., 2017), there is still a lack of research on the role of gamification in service recovery. Accordingly, the aim of this study is to examine the effectiveness of gamification as an alternative recovery tool. Further, the majority of studies in service recovery have used justice as the “dominant theoretical framework” to explain customers’ reactions to failures (Mattila, 2001, p. 584). However, prior research in gamification suggests that perceived enjoyment is better suited to predict customers’ usage intentions (Ryan and Deci, 2000). Hence, this study aims to shed light on the mechanism that explains customers’ reactions to gamified recovery. In addition, prior research shows that failure severity and the type and level of compensation can influence compensation effectiveness (Grewal et al., 2008). Therefore, the moderating roles of failure severity as well as reward type and level will also be examined.
This will be a quantitative study with three experiments. In the first study, a scenario-based experiment will be employed to examine the direct effects of gamified recovery. This will be a single factor design with three conditions: gamified recovery, non-gamified recovery, and no recovery. The scenario depicts a service failure related to an overcooked meal using a video clip to enhance ecological validity (Gelbrich et al., 2015). In the non-gamified recovery condition, participants are informed that they will receive a 20% refund. In the gamified recovery condition, participants are informed that they can play a game to win a reward and they receive a 20% refund. Finally, in the no recovery condition, participants will not receive any compensation. The second experiment will be conducted in the same restaurant setting, to examine the moderating effects of perceived severity. This will be a 2 (failure severity: low vs high) * 2 (recovery type: gamified vs. non-gamified) between subject design. The third experiment will be a 3 (recovery type: gamified visible reward vs. gamified hidden rewards vs. non-gamified) * 2 (compensation level: low vs. high) between subject design in a hotel context to examine the interaction effects of compensation type and compensation level.
This study will contribute to the gamification literature by directly addressing Van Vaerenbergh et al. (2018)’s call for research on the effectiveness of gamification in a service recovery context. It will also examine the mediating effects of perceive enjoyment (vs. perceived justice) as well as the moderating effects of failure severity and reward visibility as theoretically and managerially relevant boundary conditions. Ultimately, it offers insights on how to capitalize on modern technologies to better handle service failures.