Next Article in Journal
Approaching Urban Food Waste in Low- and Middle-Income Countries: A Framework and Evidence from Case Studies in Kibera (Nairobi) and Dhaka
Previous Article in Journal
Spatial Distribution of Freshippo Villages under the Digitalization of New Retail in China
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 15
Issue 4
10.3390/su15043294
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

The Impact of AI’s Response Method on Service Recovery Satisfaction in the Context of Service Failure

by
Zengmao Yang
*,
Jinlai Zhou
and
Hongjun Yang
School of Economics and Management, Beijing University of Posts and Telecommunications, Beijing 100876, China
*
Author to whom correspondence should be addressed.
Sustainability 2023, 15(4), 3294; https://0-doi-org.brum.beds.ac.uk/10.3390/su15043294
Submission received: 19 October 2022 / Revised: 19 November 2022 / Accepted: 9 February 2023 / Published: 10 February 2023
(This article belongs to the Section Health, Well-Being and Sustainability)
Browse Figures
Review Reports Versions Notes

Abstract

:
In order to perpetuate service sustainability and promote sustainable growth in the service sector, it is important to resolve service failures. AI technology is being applied to service jobs in more and more industries, but AI will inevitably fail while providing service. How to carry out service recovery and obtain the understanding and forgiveness of customers is a problem that urgently needs solving in the practice and research of AI services. The purpose of this study was to explore the artificial intelligence remediation mechanism in the context of service failure and to explore the remedial utility of AI’s self-deprecating humor responses. The study conducted data collection through three experiments to test our hypotheses: study 1 verified the main effect of self-deprecating humor responses and the mediating effect of perceived sincerity and perceived intelligence; study 2 verified the moderated effect of the sense of power; and study 3 verified the moderated effect of failure experience. The experimental results show that, in the context of AI for service recovery, self-deprecating humor responses can improve customers’ willingness to tolerate failure, with perceived intelligence and perceived sincerity found to play a mediating role in this. The sense of power also plays a moderating role by affecting perceived sincerity, and failure experience has a moderate effect by affecting perceived intelligence. The theoretical contribution of the article is to introduce the perspective of AI’s self-deprecating humor service recovery, which complements theoretical research in the field of AI services. The management significance of the article is to provide new AI communication strategies and practical suggestions for enterprises and technical personnel.

1. Introduction

Service failure can seriously affect business activities, so service recovery is important to continue service sustainability and to promote sustainable growth in the service industry. Artificial intelligence (AI) refers to programs, algorithms, systems, and machines that embody intelligence and show some similarity to certain characteristics of humans [1]. Due to its advantages in efficiency, precision, and ease of management [2], it is gradually being used in more and more fields, such as customer services, finance, unmanned vehicles, and mobile healthcare. The specific forms it appears in are mostly offline AI robots and online AI service systems. Although AI is more accurate and efficient at work tasks than human employees, it is subject to algorithmic flaws, making it difficult to avoid the possibility of error [3]. The service industry belongs to a wide range of non-habitual environments wherein users’ demands are significantly different, and the frequency of AI mistakes is significantly higher, which poses a challenge to the application of digital technology in enterprises. In this study, in order to demonstrate the effect of AI’s response methods on customers’ service recovery satisfaction, as well as the mechanism behind this, five hypotheses were proposed: “AI self-deprecating humor responses positively affect customers’ service recovery satisfaction, perceived intelligence and perceived sincerity play a mediating role, and failure experience and sense of power moderate the two mediators, respectively”.
Service failure leads to negative customer sentiment, and also causes customers to spread negative word-of-mouth reviews; it can also lead to boycotts, complaints, and other situations, which can seriously affect business operations [4]. Scholars have explored service recovery in terms of recovery methods, types of service failure, and customer traits of failure severity, and have achieved some results [5]. However, the reality is that when a customer asks for something unexpected, service staff usually have the ability to improvise, but AI struggles to respond appropriately [6]. The feelings and attitudes of customers are also very different when they are served by service staff and AI. In addition to making itself appear sincere when recovering, AI’s unique intelligent performance is necessary for customers to have confidence in the outcome of subsequent recoveries [7]. Therefore, the current findings on service failure ignore matters such as the differences between AI and humans and the fact that both make different errors, making it difficult to apply these findings to AI service failure situations. The practice of AI service failure and recovery is in dire need of contextualized theoretical research to guide it.
Self-deprecating humor is a form of communication, and the established literature fully reveals that an expresser can use self-deprecating humor to alleviate an embarrassing situation [8], bringing the expresser closer to the other party and thereby facilitating the completion of relationship recovery. Research confirms the following: internally, when leaders encounter embarrassing events, self-deprecating humor can effectively appease employees’ dissatisfaction [9]; externally, when companies encounter brand and image crises, self-deprecating humor response strategies can effectively recover their corporate image and salvage word-of-mouth [10,11]. Unlike when people use self-deprecating humor, when AI uses self-deprecating humor expressions to communicate with customers, its emotional expressions may amaze customers [12], but they will not make customers equate AI with humans, which leads to different standards of understanding between the two; this in turn amplifies the effect of the self-deprecating humor, alleviating customers’ dissatisfaction and making them more likely to forgive service failure [13]. The psychological mechanism through which AI’s use of self-deprecating humor affects customers, and its effect on service recovery, are the elements that were investigated in this study.
Whether self-deprecating humor applies to AI service remediation scenarios was central to the study, which attempted to interpret this question in light of the cognitive-emotional processing system. The cognitive-emotional processing system, proposed by Mischel and Shoda, asserts that individuals’ cognitive processing and emotional processing of external information affect their behavioral decisions, based on the combined effect of the two processing pathways, which ultimately change individuals’ judgments and decisions about external situations [14,15,16]. In AI service recovery contexts, self-deprecating humor influences customers’ emotional processing through self-deprecation, causes emotional arousal in order to influence perceived sincerity, and influences customers’ cognitive processing through humor in order to convey information to influence perceived intelligence. The cognitive-emotional processing system can clarify the process mechanism by which AI affects service recovery satisfaction through different response styles. This study explored the mediating effect of perceived sincerity (cognitive path) and perceived intelligence (emotional path) between AI response styles and service recovery satisfaction based on this theoretical framework. According to convergence-inhibition theory, customers with different senses of power see things differently, and their perception of sincere performance in service recovery also differs; furthermore, the sense of power has a moderated effect through the path of perceived sincerity. In addition, the use of humor needs to focus on the occasion, and in service recovery scenarios, different failure experiences create very different occasions [17]. Therefore, the role humor plays in process failure and outcome failure scenarios, respectively, should be examined, as it is only when humor is used successfully that customers will affirm the level of AI intelligence they are engaging with. This study determined that failure experiences have a moderated effect through the path of perceived intelligence service recovery satisfaction. In summary, our study is divided into four main parts. Section 1 is the introduction, which leads to the research question; Section 2 is a literature review and a hypothetical deduction that formulates the research hypothesis; Section 3 is the research design, which details how data were collected through experiments and then analyzed to test hypotheses; and Section 4 is the conclusion and discussion, which summarizes the research findings, theoretical contributions, and practical implications, and concludes with the article’s shortcomings and prospects for subsequent research.

2. Theoretical Background and Hypothesis Development

2.1. Service Failure and AI Self-Deprecating Humor Response

Service failure means that, in the service process, service personnel provide services that do not meet the minimum acceptable standards, including the needs and expectations of the customer, resulting in customer dissatisfaction and other negative emotions [18]. Most current artificial intelligence is based on internal algorithms that collect and process external information and then react to it; it is difficult for AI algorithms to react properly when situations occur that the algorithm settings have not been built to respond to [19]. Technicians have been working to improve the intelligence of algorithms in pursuit of higher analytics, enabling AI to gain higher intelligence in order to avoid service failures. However, due to the intractable nature of AI, recovery measures in normal human interaction are no longer applicable, and recovery methods suitable for AI service contexts should therefore be explored. This research should not only rely on AI to improve the level of intelligence but also should focus on the emotions of customers in the service process, in order to achieve AI service recovery measures that can produce better results [20]. Among the many current recovery measures, emotional guidance is an important component, and self-deprecating humor is a means of achieving emotional guidance based on the event itself [21]. Self-deprecating humor has been used in interpersonal interactions in companies for service recovery and has been shown to have a significant positive effect on alleviating negative customer sentiment.
Self-deprecating humor as a form of expression is the expression of self-deprecating content through humor [22]. As can be seen, self-deprecating humor is the act of actively pointing out one’s own defects, but there is a big difference between self-criticism and self-reflection; self-hatred is not a serious criticism, and correction behavior [23] is performed in a light-hearted way that cannot help but make others laugh. It has been noted that two characteristics of “self-deprecating humor” can be summarized from this type of behavior in the past: self-deprecation and humor [24]. Self-deprecating humor has been shown to be effective in alleviating the awkwardness of interactions, expressing the faults of a given incident in a pleasant way, and downplaying its negativity [25]. Self-deprecating humor has been applied to marketing, organizational behavior, and news communication, but related research has remained at the level of interpersonal interaction. Current research on AI service recovery has also begun focusing on AI-induced customer empathy [26]. Self-deprecating humor in human–computer interactions, as an efficient means of emotional awakening, is worth exploring further to determine if it can produce significant effects.
This study defines AI self-deprecating humor responses as a specialized recovery measure taken by AI service providers after a service failure; it is AI’s strategy of communicating with customers about the service failure in a humorously self-deprecating manner. When AI makes a humorously self-deprecating response after a service failure, the AI is behaving to a certain extent like a human being, by imitating a human tone of re-ply, which induces customers to think that the AI is criticizing the merchant or the algorithm. This can then make customers think that the AI and themselves are standing in the same position, which greatly closes the distance in the relationship between the two sides and is conducive to alleviating customers’ negative emotions and completing service recovery [27].

2.2. Differences in Customer Attitudes between Manual and AI Services

The provision of services to customers by artificial intelligence is already a future trend in the service industry. As technology has improved, AI has become increasingly adept at mimicking human speech and behavior to provide services; while there are fundamental differences between human and AI service provision, the latter can approximate the former. The experience differences between HCI and HMI have been a hot topic of debate in recent years, and no uniform conclusions have been reached, but these differences certainly lead to very different customer attitudes in their interactions with both.
Especially in the recovery phase of service failure, there are large differences in customer perceptions and attitudes. On the one hand, when service fails, customers have a high standard of understanding with regards to real people and a relatively low standard of understanding with regards to AI. Some studies have confirmed that customers are more likely to forgive AI when they encounter the same types of errors caused by humans and AI [28]. On the other hand, human beings are social creatures, and it is normal for people to have human feelings, so human service providers’ emotional behavior can affect customers in a limited way, but AI having human feelings is abnormal, and therefore the emotional behavior of AI can provide more stimulation to and emotional fluctuations in customers [29]. Research confirms that, when service fails, companies using self-deprecating humor responses can be very effective in influencing customers’ negative emotions. Therefore, when AI uses self-deprecating humor, the anomaly of this human-like behavior should be able to more effectively channel the negative emotions of customers and reduce the negative effects of service failure.

2.3. The Impact of AI Self-Deprecating Humor Responses on Service Recovery Satisfaction

Usually, service failure will cause customer dissatisfaction, which requires some recovery measures to weaken and dissolve the negative emotions of customers, so as to gain their forgiveness [30]. AI service failure also requires recovery measures, and due to the inflexibility of AI, it can only apologize and relieve emotions by replying to some customer responses, and it is difficult to implement recovery measures related to the actual situation other than verbal methods. For example, after service failure, human employees can restore service through verbal, emotional, and real-time remedial actions, but it is difficult for AI to implement non-verbal recovery measures. Therefore, the response content and response method form the core of AI service recovery. Self-deprecating humor as a unique expression has been proven many times to gain the recognition and acceptance of others, and AI can draw on this kind of humor to communicate with customers [31].
“Self-deprecating humor” expressions consist of two main features: humor and self-deprecation, which means that self-deprecating content is expressed in a humorous way. Existing research has confirmed in many ways that a humorous tone can produce emotions such as pleasure and cheerfulness in many situations, and these emotions can cause a series of positive feedback [32]. When customers are affected by humor, they may enter a state of pleasure, alleviating their dissatisfaction to a certain extent, and thus they may be more likely to forgive the service failure suffered. The difference between “self-deprecation” and normal deprecation is that normal deprecation involves both subjects, with one party deprecating the other, while self-deprecation occurs in a single subject, where the individual deprecates himself [33]. When self-depreciation occurs, it causes the onlooker to believe that the other person is standing in one position to depreciate themselves in the other position. When faced with a service failure situation, AI’s use of self-deprecation in service recovery makes the customer perceive that the AI is criticizing and blaming itself for the mistake, which makes the customer feel that the AI is standing in the same position as them and recognizing the problem from the customer’s perspective, which reduces the sense of distance between the positions of the AI and the customer and makes the customer perceive the AI’s sincerity. In general, through self-depreciation, AI can be perceived as standing in the shoes of customers and actively defending their rights, which allows customers to perceive the attitude of the AI or even of a given company towards taking responsibility for and solving problems, while humorous expressions make customers more receptive to the abovementioned perceptions. After customers experience self-deprecating humor, their negative emotions are channeled and defused, and they are more likely to forgive service failures. Thus, we hypothesized the following:
H1. 
AI uses a self-deprecating humor response approach to positively influence customers’ service recovery satisfaction.

2.4. Mediating Effects of Perceived Intelligence and Moderated Effects of Failure Experience

Perceptual intelligence is a probability indicator customers can use to evaluate AI’s ability to achieve a certain goal, characterizing the level of intelligence, capability, and efficiency of the AI. The successful use of humorous expressions requires that the person expressing them has sufficient ability. First, people who are good at humor should have a number of skills, such as good timing, assessing the likelihood of failure, and using tactful wit. Secondly, humor has certain risks; different occasions, timing, the other party’s personality, and other factors may lead to the failure of humor, making the other party feel offended and disgusted with the humor expresser. Finally, many scholars have demonstrated that a strong link exists between humor and competence, and it can be inferred that people generally perceive humorous people as having a higher level of competence.
When coming to an AI service scenario, if the AI successfully uses humor, it may make the customer think that the AI has the appropriate capabilities [34]. AI’s use of humorous expressions also requires an assessment of the risk of use. Furthermore, customers associate humor with perceived intelligence. Firstly, AI that uses humor may make customers feel that the AI is fresh and innovative, and this feeling can trigger positive emotions such as fondness and pleasure, which can help to alleviate feelings of dissatisfaction caused by service failure. Secondly, AI’s use of humor shows that the AI has a high level of intelligence, which makes customers believe that the AI can still take control of the situation and solve the given problem sufficiently, which reinforces the customer’s perception of the AI’s intelligence, and increases their expectations that the AI will do better in future situations [35].
In summary, AI’s use of a self-deprecating humor response allows customers to perceive AI’s intelligence and have confidence in AI’s ability to effectively solve problems. In addition, the human touch suggested by humorous expressions makes customers feel positive emotions, causing them to look forward to the AI’s subsequent behavior and to increase their acceptance of previous service failure behaviors, ultimately improving their attitudes toward the AI. Thus, we hypothesized the following:
H2. 
Perceived intelligence mediates the impact of AI self-deprecating humor responses on customers’ service recovery satisfaction.
Since the use of humor needs to distinguish between different occasions, and the di-vision of occasions in service failure contexts is generally based on the type of failure, this study hypothesized that the effect of AI self-deprecating humor responses on customer perceived intelligence is influenced by the contextual factor of the type of service failure. Many scholars classify service failure into two types: process failure and outcome failure. Process failure usually refers to mistakes caused by or deficiencies of service personnel in the service process, such as poor attitudes. Outcome failure usually refers to the failure of service staff to fulfill the core requirements of the customer, resulting in the failure of the customer to achieve their purpose of consumption, such as delivering the wrong item to the customer, or giving them the wrong information [36].
The two types of service failure have different impacts on customers, with there being major differences in customers’ recovery claims. Process failure leads to the loss of customers on the emotional and experience levels; customers are more dissatisfied with their customer experience and feel a lack of emotion, so in this context, we should pay attention to the needs of customers on the emotional level, make customers feel warm and happy through caring greetings or humorous and witty words, and eliminate customers’ negative emotions with appropriate recovery measures [37]. Outcome failure leads to the customer’s core demands or even basic needs not being met; when this occurs, the customer is more inclined to seek the given enterprise’s explanation and to come up with a specific remedial plan, so in this context, we should focus on the actual result-oriented needs of the customer and provide them with reasonable explanations to make them perceive the determination of the enterprise and service personnel to remedy the failure, so as to alleviate the negative emotions of customers.
With regards to AI service failure, this study hypothesized that the type of service failure moderates the impact of AI self-deprecating humor responses on perceived intelligence. As mentioned earlier, successful humor requires attention to the occasion and timing of its use; if used in the wrong context or at the wrong time, it may even have the opposite effect to that which it intended to have. When AI causes process failure, the customer has a deficit in the emotional dimension. In these cases, self-deprecating humor responses can make the customer feel pleasant, while the AI’s successful use of humor can also make the customer affirm the AI’s ability and intelligence level, which makes the customer expect the AI’s subsequent recovery behavior, thus achieving the effect of alleviating their dissatisfaction [38]. When AI causes outcome failure, the customer’s core demands are not met; when this happens, self-deprecating humor responses may not have a positive impact, and may even cause the customer to feel that the AI is not taking the situation seriously. In these cases, the failure of the humor may not make the customer perceive the intelligence of the AI, but instead may deepen the customer’s dissatisfaction with the given company and AI. As can be seen, the effectiveness of self-deprecating humor responses varies widely among different types of failures. Thus, we hypothesized the following:
H3. 
Failure experience moderates the mediating role of perceived intelligence in the impact of AI’s self-deprecating humor responses on customers’ service recovery satisfaction.

2.5. The Mediating Effect of Perceived Sincerity and the Moderated Effect of the Sense of Power

Perceived sincerity refers to people’s perception of others’ perceivable sincere information that conveys authenticity. The AI in this study caused customers to perceive sincerity through the self-deprecation involved in self-deprecating humor [39]. Self-deprecation is one element of self-deprecating humor responses, a critique and mockery of oneself that can make bystanders feel that the individual is in a different position. Unlike people’s natural tendency to receive compliments and praise for themselves, self-depreciation can reflect an individual’s humility. In addition, self-depreciation shows an individual’s awareness of their own flaws and the courage to admit their shortcomings, ultimately reflecting the individual’s sincerity. When coming to an AI service failure situation, a self-deprecating response by AI to the service failure allows the customer to perceive that the AI is on the same page as the customer and that it is critical of its own mistake, and, furthermore, that it intends to actively recover the customer’s loss. Firstly, self-depreciation makes the customer think that the AI is looking at the problem from the customer’s perspective, which reduces the sense of distance between the AI and the customer, making the customer have some trust in the AI. Secondly, the gesture of humility expressed by self-deprecation makes the customer feel that the AI is daring to admit its mistake and that it is determined to restore service. In summary, self-depreciation affects the customer’s perceived sincerity of AI.
Although existing research has confirmed that apologies are one of the methods that can be used to gain forgiveness, their effectiveness is determined by the sincerity of the apology and the perceived sincerity of the other party. Sincerity is the core element in-volved in two parties’ gaining each other’s trust, and people usually have more recognition and tolerance for the people that they trust [40]. Perceived sincerity has a very critical impact on whether an apologizer can gain the forgiveness of the other party. Apologies are also one of the important methods AI can use to restore service in the case of AI service failure. However, an effective apology needs to make the customer perceive the sincerity of the AI, thus influencing the customer’s dissatisfaction and causing them to have a positive perception of the content of the AI’s response, which in turn alleviates the customer’s negative emotions caused by the service failure, and ultimately generates positive feedback on the customer’s part regarding the AI and the given company. Thus, we hypothesized the following:
H4. 
Perceived sincerity mediates the effect of AI self-deprecating humor responses on customers’ service recovery satisfaction.
As subjects in a given consumption scenario, customers and service providers are in different positions, with customers usually in the superior one; consequently, customers often generate a sense of power to judge the behavior of service providers. Power is an important driver in social interaction and is generally seen as being involved in influencing outsiders, as well as in control over limited resources at the level of social relations. A sense of power is a subjective perception of people’s own influence, a concept of social relations, and the level of an individual’s influence on the will, behavior, and effects of other social subjects [41]. The sense of power as an important social concept has been widely used in organization, marketing, psychology, and political science. According to the approach-inhibition theory, it is possible to analyze the differences between individuals with different levels of the sense of power. Compared to individuals with a low sense of power, individuals with a high sense of power activate the “behavioral approach system” when making perceptions and decisions in social activities, which leads individuals to pay more attention to positive, favorable information and gives them greater confidence and a more positive emotional state. Compared to individuals with a high sense of power, individuals with a low sense of power activate the “behavioral inhibition system” during their perceptions and decision-making processes, directing them to focus more on risk factors and hazards, which results in a lower level of confidence and a negative bias in behavior and decision making [42].
With regards to situations where an AI service fails, this study considered that the sense of power moderates the effect of AI’s self-deprecating humor responses on perceived sincerity. Firstly, as mentioned above, high-power customers are influenced by the behavioral approach system, which will make them pay more attention to positive information rather than risk factors during an AI service failure; here, the sincerity of the self-deprecating response of the AI will be noticed by these customers, and they will have a higher perceived sincerity for AI who make self-deprecating responses. Secondly, again, as mentioned above, low-power customers are influenced by the behavioral inhibition system, which makes them focus more on negative information such as risk factors. During an AI service failure, these types of customers will pay more attention to the causes and adverse consequences of the service failure, and due to such customers having a low-er confidence level, it will be harder for them to agree with the AI’s self-deprecating humor response; they will therefore feel that the AI is not exhibiting sincerity. Thus, we hypothesized the following:
H5. 
Sense of power states moderates the mediating role of perceived sincerity in the effect of AI self-deprecating humor responses on customers’ service recovery satisfaction.
In summary, the theoretical framework of this study is constructed based on the above hypotheses, as shown in Figure 1.

3. Research Design

3.1. Pre-Experiments

The purpose of the pre-experiment was to examine the differences between the AI‘s self-deprecating humor responses and normal responses, and whether the subjects could perceive the “self-deprecating” and “humorous” content and compare this with the normal responses, so as to allow us to control the “self-deprecating humor“ in the formal experiment. Fifty subjects were recruited for the pre-experiment, and the subjects viewed the content of each of the two responses and then rated the characteristics of these two responses.
The content was introduced to the subjects in the following way: they were told that they were receiving services from the AI and that it could not respond to their request accurately, so they had expressed their dissatisfaction with it. The AI in the self-deprecating humor response group responded: “I was just bragging about being able to solve a lot of problems, and I was immediately smacked in the face, so I’m really not smart at all, I’m really very sorry.” The AI in the normal response group responded: “I’m sorry I didn’t solve your problem, and I hope I can help you next time.”
After the subjects read the material, the subjects were asked to respond on a scale. First of all, referring to the work of Xu Lan and others, for the material of the self-hatred response, eight options were presented for the question items, including “humorous”, “self-deprecating”, “denying”, “sincere”, “defending”, etc. Second, following Marcus’ re-search, the subjects were asked, “Do you think the AI acknowledges its own shortcomings through self-deprecating humor [43]?” All scales were measured on a seven-point Likert scale, with one meaning strongly disagree and seven meaning strongly agree.
All subjects completed the pre-experiment, and 50 valid samples were obtained. The analysis of the data shows that the most frequently selected characteristics of the self-deprecating humor responses were “humor” and “self-deprecation”, with these selected 38 and 37 times, respectively. The other six characteristics were selected less than 25% of the time. It can thus be seen that the response content of the pre-experiment was successful in controlling for self-deprecating humor and was therefore able to be used in the subsequent formal experiment. The scale data were then analyzed, yielding M(normal response) = 1.46, SD = 0.503; M(self-deprecating humor response) = 5.30, SD = 0.614, p < 0.01. Once again, this proved the success of the content control for self-deprecating humor responses.

3.2. Study 1

The purpose of study 1 was to verify the impact of AI self-deprecating humor responses on service recovery satisfaction when AI service fails. To test this, study 1 tested hypothesis 1 by comparing the differences in customer perceptions between the group receiving self-deprecating humor responses and the group receiving normal responses.

3.2.1. Experimental Design

The subjects were selected before the experiment started, and only those who had used the AI service before were allowed to participate in the experiment; a total of 117 subjects were finally recruited for study 1. The subjects were informed that this experiment was a study on customer experience in order to avoid the subjects’ guessing the real intention of the experiment. All subjects were randomly assigned to two groups: the self-deprecating humor response group and the normal response group. The experiment set the AI to be a chatbot for VR glasses. The subjects were placed in the following scenario: they were about to buy VR glasses at a store and were interested in learning more about this product, so they asked the service agent for specific information about the product; the service agent then indicated to the subjects that she was an AI chatbot named Xiao Mei. Next, the communication process between the customer and the AI chatbot was shown to the subjects in the form of chat transcripts, and the AI answered the customer’s questions incorrectly, causing dissatisfaction.
At this point, the AI of the self-deprecating humor response group responded: “Xiao Mei just now also bragged about being able to solve a lot of problems, the result was im-mediately hit in the face, it is not intelligent at all, I am very sorry”. The AI in the normal response group responded: “I’m sorry I didn’t solve your problem, I hope I can help you next time.”
Then, subjects were asked to respond on the scales provided. The scales were all de-rived by contextualizing well-established scales from existing studies, with the scale for perceived intelligence drawing on Warner’s research; this had a total of five questions: I think this AI is competent, I think this AI is intelligent, and so on [44]. The scale for perceived sincerity drew on Arran’s study, and this had a total of four questions: I felt the AI’s apology was sincere, I thought the AI gave a reasonable explanation for the matter, and so on [9]. The scale for service recovery satisfaction drew on Paul’s research and included a total of four questions: I feel that the AI’s apology was acceptable, I feel that the AI’s apology met my expectations, and so on [45]. All scales were measured on a seven-point Likert scale. At the end of the scale, demographic information was collected.

3.2.2. Results and Discussion

(1) Reliability test. The reliability of the scales was tested, and Cronbach’s α for perceived intelligence was 0.924, Cronbach’s α for perceived sincerity was 0.848, and Cronbach’s α for service recovery satisfaction was 0.906. This shows that the experimental scales all had good internal consistency.
(2) Main effect test. An independent samples t-test was conducted to verify whether there was a significant difference in service recovery satisfaction between the self-deprecating humor response group and the normal response group. The results showed that service recovery satisfaction was significantly higher in the self-deprecating response group (M(self-deprecating humor) = 3.853, SD = 1.517) than in the normal group (M(normal) = 3.122, SD = 1.336), t(115) = 2.772, p < 0.01), and therefore that hypothesis 1 was supported.
(3) Mediated effect.
To test the mediating effect of perceived sincerity, the process 3.4 plug-in in Spss was used and model 4 was selected with a confidence level of 95%. In the model, with the response method as the independent variable, perceived sincerity as the mediating variable, and service recovery satisfaction as the dependent variable, we tested whether there was a mediating effect of perceived sincerity between response method and service recovery satisfaction. By analyzing the indirect effect of response style on service recovery satisfaction, the results showed a confidence interval of [0.027, 0.813], which did not contain 0, indicating there was a significant mediating effect of perceived sincerity with a mediating effect value of 0.421.
To test the mediating effect of perceived intelligence, the process plug-in was used, and the response method as the independent variable, perceived intelligence as the mediating variable, and service recovery satisfaction as the dependent variable were entered into the model. The results showed a confidence interval of [0.162, 0.946], which did not contain 0, indicating that the mediating effect of perceived intelligence was also significant; this had a mediating effect value of 0.557. Table 1 shows the results of the tests for the mediating effects of perceived sincerity and perceived intelligence. Hypothesis 2 and hypothesis 4 were supported.

3.3. Study 2

The purpose of study 2 was to test hypothesis 4 by assessing the moderated effect of the sense of power on perceived sincerity, and also to repeat the verification of hypotheses 1 and 3. Study 2 used a between-group design with 2 (self-deprecating humor response vs. normal response) × 2 (sense of power: high vs. low) groups, for a total of 4 groups. The control for the response method was essentially the same as in study 1.

3.3.1. Experimental Design

The subjects were informed that this was an experiment about consumption experiences before the experiment started to avoid their guessing the actual intention of the experiment. Only subjects who had used the AI service before were chosen to participate in the experiment. A total of 196 subjects were recruited for study 2, and all subjects were randomly assigned to the 4 groups. Subjects were provided with experimental materials and allowed to substitute themselves into the experimental situation. At the same time, to prevent the commodity from influencing the subjects and to enhance the generalizability of the experimental findings, the commodity for study 2 was set as chocolate, a product which differed greatly from that of study 1 (VR glasses). The experiment set up the AI as a chatbot for chocolate products. The scenario was the following: the subjects were about to buy chocolate in a store and were interested in learning more about the product, so they asked the service staff for specific information about the chocolate, and the service staff indicated to the subjects that it was an AI chatbot. Subsequently, a picture of the chat transcript was used to show the subjects how the customer and the AI chatbot communicated, and the AI chatbot answered the customer’s questions incorrectly, resulting in customer dissatisfaction. The content of the two responses of the AI was essentially the same as in study 1.
After reading the experimental material, the subjects were asked to respond to the scales provided. The scales for perceived intelligence, perceived sincerity, and service recovery satisfaction were the same as those used in study 1.
The sense of power scale drew on Cameron’s research and included a total of eight questions: I can make people listen to what I say, my wishes don’t carry much weight, and so on [46]. All scales were measured on a seven-point Likert scale. At the end of the scale, demographic information was collected.

3.3.2. Results and Discussion

(1) Reliability test. The reliability of the scales was tested, and Cronbach’s α value for perceived intelligence was 0.921, Cronbach’s α value for perceived sincerity was 0.85, Cronbach’s α value for sense of power was 0.860, and Cronbach’s α value for service recovery satisfaction was 0.88. This shows that the experimental scales all had good internal consistency, and therefore the results of the scales were able to be used for further analysis.
(2) Main effects test. To analyze the subjects’ sense of power, the higher their score, the higher the level of their sense of power, and vice versa. The median sense of power scores was taken, and those above the median were considered a high sense of power, and vice versa.
First, an independent samples t-test was conducted on service recovery satisfaction in both groups, and the results showed that service recovery satisfaction was significantly higher in the self-deprecating humor response group than in the normal group: (p < 0.01) M(self-deprecating humor) = 3.343, SD = 1.491; M(normal) = 2.419, SD = 1.137. Hypothesis 1 was repeatedly tested. Second, a two-factor ANOVA of 2 (self-deprecating humor response vs. normal response) × 2 (sense of power: high vs. low) was conducted. The results of the analysis showed a significant interaction effect between the response method and sense of power (F(3,192) = 3.837, p < 0.05): M(self-deprecating humor, low) = 3.012, SD = 1.366; M(self-deprecating humor, high) = 3.757, SD = 1.552; M(normal, low) = 2.426, SD = 1.069; M(normal, high) = 2.413, SD = 1.199.
(3) Test for moderated effect of sense of power
The sense of power was set as a categorical variable through the process V3.4 plug-in of SPSS, using model 7, to test the moderated effect, and the results are shown in Table 2. As seen in Table 2, the interaction between the response style and sense of power had a significant effect on service recovery satisfaction (β = 1.302, p < 0.01, R2 = 0.154). Thus, the sense of power moderated the relationship between the response method and service recovery satisfaction.
The results of the regression analysis showed the influence of the moderated effect of the sense of power, demonstrating how the influence of the response method on service recovery satisfaction differed between different levels of sense of power. Furthermore, the graph shown in Figure 2 was plotted based on the results of data analysis. As seen in Figure 2, when the level of their sense of power was high, customers’ willingness to forgive was stronger for self-deprecating humor responses compared to normal responses (β = 1.423, p < 0.01,95% confidence interval [0.938, 1.907], which did not contain 0). When the level of their sense of power was low, customers’ willingness to forgive was stronger for normal responses compared to self-deprecating humor responses (β = 0.120, p = 0.62,95% confidence interval [−0.365, 0.606], inclusive of 0). Thus, the level of sense of power affected the relationship between the response method and service recovery satisfaction, and therefore hypothesis 5 was supported.
Table
Table 2. Results of the test for moderated effect of sense of power.
Table 2. Results of the test for moderated effect of sense of power.
VariableService Recovery Satisfaction
βSET
Constant4.209 ***0.8904.726
Response method−1.182 **0.550−2.148
Sense of power−1.672 ***0.551−3.031
Rseponse method × Sense of power1.302 ***0.3473.744
R20.154
F14.021
(** p < 0.05, *** p < 0.01).

3.4. Study 3

The purpose of study 3 was to verify the moderated effect of failure experience on service recovery satisfaction through perceived intelligence, in order to support hypothesis 5 and to test hypotheses 1 and 2 again. Study 3 used a between-group design with 2 (self-deprecating humor response vs. normal response) × 2 (process failure vs. outcome failure) groups, for a total of 4 groups. The control for the response method was essentially the same as that used in study 1.

3.4.1. Experimental Design

Study 3 also only allowed participants who had used the AI service before to participate in the experiment. A total of 209 subjects were eventually recruited for the study. Subjects were informed that this experiment was a study on customer experience in order to avoid subjects’ guessing the true intention of the experiment. All the subjects were randomly assigned to one of the four groups. To enhance the generalizability of the experimental findings, study 3 set the experimental context in an offline AI service scenario, where the AI was an intelligent robot placed in a shopping mall. The intelligent robot was shown to the subject through a video, and the subject was asked to imagine that they were looking for a store in the mall, and the intelligent robot in the mall had come to the subject and told them that it could take them to the store they wanted to go to. Process failure: The subject told the robot the store they wanted to go to, but the robot was unable to identify the problem until after five or six times; after understanding the request, the robot took the subject to the target store. The subject would then complain that even though the location had been given, the robot had limited intelligence, as it had made them say the name of the store so many times. Outcome failure: The subject indicated the store they wanted to visit to the robot, and the robot told them the target store location, but the subject would arrive and find that the robot told them the wrong location. The subject would then complain: “You made me go around and come to the wrong position, is this your level of intelligence, what a waste of my time!” The control of the response method was basically the same as the pre-experiment.
The subjects were then asked to respond to the scales. The scales were the same as those in study 1, and all scales continued to use the seven-point Likert scale. At the end of the scale, demographic information was collected.

3.4.2. Results and Discussion

(1) Reliability test. The reliability of the scales was tested, and Cronbach’s α value for perceived intelligence was 0.892, Cronbach’s α value for perceived sincerity was 0.837, and Cronbach’s α value for service recovery satisfaction was 0.832. This shows that the experimental scales all had good internal consistency, and that the results of the scales were able to be used for further analysis.
(2) Main effects test. An independent samples t-test was conducted on service recovery satisfaction for the different response methods, and the results showed that service recovery satisfaction was significantly higher in the self-deprecating humor response group than in the normal response group (p < 0.01, t(207) = −2.847): M(self-deprecating humor) = 4.542, SD = 1.156; M(normal) = 4.081, SD = 1.185. Hypothesis 1 was again supported in this experiment.
Subsequently, a two-factor ANOVA of 2 (self-deprecating humor response vs. normal response) × 2 (process failure vs. outcome failure) was conducted. The analysis showed a significant interaction effect between the response method and type of failure (F(3,205) = 4.175, p < 0.05): M(process, normal) = 4.121, SD = 1.267; M(process, self-deprecating humor) = 4.908, SD = 1.056; M(outcome, normal) = 4.038, SD = 1.101; M(outcome, self-deprecating humor) = 4.176, SD = 1.145.
(3) Test for moderated effects of failure experience. Based on the process V3.4 plug-in of the Spss program, using model 7, the failure experience was set as a categorical variable to test the moderated effect of failure experience on service recovery satisfaction through perceived intelligence, and the test results are given in Table 3. As seen in Table 3, the interaction effect of the response method and failure experience had a significant effect on service recovery satisfaction (β = −0.886, p < 0.01, R2 = 0.059). Thus, failure experience was found to moderate the relationship between the response method and service recovery satisfaction.
The results of the regression analysis showed the influence of the moderated effect of failure experiences, which can be seen in the differences in service recovery satisfaction under the influence of different failure experiences. The graph shown in Figure 3 was drawn based on the results of the data analysis. As seen in Figure 3, when in a process failure scenario, customers’ willingness to forgive was stronger for self-deprecating humor responses compared to normal responses (β = 0.823, p < 0.01, 95% confidence interval [0.365,1.287], which did not contain 0). When in an outcome failure scenario, customers’ willingness to forgive was stronger for self-deprecating humor responses compared to normal responses (β = −0.059, p = 0.801, 95% confidence interval [−0.527,0.407], inclusive of 0). It can be seen that the effect of the response method was no longer significant in the outcome failure scenario. As a result, failure experience moderated service recovery satisfaction through perceived intelligence, and hypothesis 3 was therefore supported.
Table
Table 3. Results of the test for moderated effects of failure experience.
Table 3. Results of the test for moderated effects of failure experience.
VariableService Recovery Satisfaction
βSEt
Constant5.424 ***0.8326.518
Response method−0.945 *0.528−1.788
Failure experience−1.207 **0.522−2.309
Response method × Failure experience0.89 ***0.333−2.660
R20.059 ***
F4.317
(* p < 0.1, ** p < 0.05, *** p < 0.01).

4. Conclusions and General Discussion

4.1. Conclusions

This study explores the mechanism of AI’s self-deprecating humor responses and its effect on service recovery satisfaction in the context of AI service failure, introducing perceived intelligence and perceived sincerity as mediating variables. On this basis, boundary conditions for the effect of AI’s self-deprecating humor responses on service recovery satisfaction were explored, and the sense of power and the type of failure were introduced as moderating variables to verify differences in customer attitudes toward AI’s self-deprecating humor responses under different circumstances. From the results of the experiment, it can be concluded that, firstly, AI’s self-deprecating humor responses positively influenced service recovery satisfaction, and these responses were more likely to gain customers’ forgiveness compared with the traditional apology response. Secondly, perceived intelligence and perceived sincerity were influenced by humor and self-depreciation, respectively, and played a mediating role in the impact of service recovery satisfaction. Finally, the sense of power and type of failure played a moderating role in the effect of AI’s self-deprecating humor responses on service recovery satisfaction. Compared to customers with a low sense of power, customers with a high sense of power were influenced by self-depreciation and experienced higher service recovery satisfaction; humor in process failure can better enhance service recovery satisfaction through perceived intelligence compared to outcome failure.

4.2. Theoretical Contributions

First, this study explored for the first time the effect of AI’s self-depreciation humor on service recovery, as well as the mechanism behind this. Previous studies on AI service recovery have explored the ability of AI’s intelligence to provide services more accurately so as to avoid mistakes [47], or to use various normal recovery methods such as apology, explanation, and compensation [20]. Some scholars have also begun to focus on the emotional resonance of AI and customers. The present study’s findings are consistent with those of existing research, which have found that empathic responses are better at gaining service recovery satisfaction. The study then deepens existing research by exploring more efficient self-depreciation humor on this basis. The study uses self-depreciation humor responses as a service recovery approach for AI, examines the impact of this approach on service recovery satisfaction, provides a new perspective on AI service recovery strategies, fills the current research gap of using AI for emotional intelligence-based remediation, and plays a role in promoting the subsequent expansion of AI service recovery-related research.
Second, this study analyzed the impact paths of AI self-depreciation humor as perceived intelligence and perceived sincerity. Humans use self-depreciation humor to in-fluence customer attitudes by creating perceptions of competence and warmth [48]. As inanimate objects, computers’ interaction with humans is also very different from interpersonal interaction, and the effect paths relied on by AI to produce effects using self-depreciation humor are also different from those of human self-depreciation humor [49]. This study explored how customers perceive AI’s ability to engage in the two most important elements of self-depreciation humor, and ultimately identified perceived intelligence and perceived sincerity as the perceptions generated by customers, which are influenced to produce the effect of changing emotions and attitudes. Firstly, this conclusion echoes the influence path of self-depreciation humor discussed in existing studies; secondly, it provides a new interpretation of the self-depreciation humor mechanism, through the AI perspective, which emphasizes sincerity in addition to the level of intelligence. This study builds a more refined theoretical framework for research into AI service recovery and also provides a more precise theoretical chain for explaining customer attitudes during AI service recovery.
Finally, this study identified the boundary conditions and mechanisms of influence of the sense of power and type of failure, and their effects on service recovery satisfaction. It was found that the sense of power influenced perceived sincerity through perceptions of self-deprecation and the type of failure influenced perceived intelligence through perceptions of humor, revealing a theoretical black box in which these two conditions differentially affected service recovery satisfaction, enabling better analysis and exploration of the impact effects of AI’s use of self-depreciation humor expressions. The findings of the study provide theoretical support for the positive effects of AI’s use of self-depreciation humor, and also provide a theoretical basis for further delving into the strategies adopted by, and the effects and mechanisms of action produced by AI in service recovery.

4.3. Practical Contributions and Implications

First, this study provides a new recovery strategy for AI encountering service failure. Companies should focus on AI and human empathy when pursuing the development of higher AI intelligence levels. With the increasingly widespread use of AI in service scenarios, the communication methods between customers and AI are also showing various trends, such as bi-directional, diversified, and functional trends. While these changes have made it more difficult for AI to recover from mistakes, they have also created the conditions for new recovery methods to emerge. The findings of the study confirm that the use of self-deprecating humor responses is a measure that can defuse negative customer attitudes after service failure. Thus, this study suggests that companies should not only focus on pursuing the development of AI with a high intelligence level but should also make AI adopt appropriately a self-deprecating humor response strategy after it makes mistakes in order to close the distance of their relationships with customers, thereby alleviating customers’ negative emotions, resolving their negative attitudes, and gaining their forgiveness.
Second, companies should focus on the right occasions for AI to use self-deprecating humor responses. Research findings have confirmed that the use of self-deprecating humor in different contexts produces a wide range of effects, and that mistakes made by AI in the service process and mistakes made in the service outcome trigger different customer emotions and attitudes and lead to different customer perceptions of self-deprecating humor. Using self-deprecating humor in inappropriate situations can lead to the illusion that the customer is not being respected and that the AI is not taking the given mistake seriously, which not only fails to alleviate customers’ negative emotions, but may even have the opposite effect. Therefore, companies should be cautious about the occasions and effects of AI’s use of self-deprecating humor, especially those of the humor factor involved. Although humor can make users appear more intelligent, it is difficult to produce higher perceived intelligence and complete service recovery if the specific context is ignored.
Finally, companies should base the content of self-deprecating humor responses on specific service matters and scenarios and types of failure to make customers perceive not only the intelligence of the AI, but also the sincerity of the AI with regards to service recovery. This study confirms that AI’s self-deprecating humor makes customers more willing to forgive AI through the perceived sincerity generated by the self-deprecation and the perceived intelligence generated by the humor, which is a good complement to the practical application of AI. Both companies and technology parties should focus on the fact that the response content should take into account both characteristics of self-deprecating humor and should obtain customers’ forgiveness using the joint effect of these two characteristics.

4.4. Limitations and Future Research

First, this study chose the sense of power and type of failure as boundary conditions, but there may be other factors that are involved in the effect of self-depreciation humor, so subsequent studies should explore these other potential factors. In the service failure context, the severity of a given failure may also affect the effect of the self-depreciation humor; in addition, the consumption context may also be one of the influencing factors, as customers pursue different values in different consumption contexts. Second, this study only used the online scenario experiment for data collection, which has the characteristics of high efficiency and convenience and wider experimental scope; at the same time, if the experimental procedure is properly designed, it can also make the experimental results achieve high internal and external validity. In the future, if conditions permit, on-site experiments can be conducted in the laboratory to make subjects perceive the behavior of artificial intelligence more directly in order to enhance the reliability of the experimental results. Finally, this study only considered the recovery strategy of self-deprecation; in fact, there are many other methods that can be used to recover from a service failure, so future research should continue to explore other proven AI service recovery measures in order to further improve AI service theory research.

Author Contributions

Z.Y. was responsible for data curation, formal analysis, and writing—original draft; J.Z. was responsible for supervision and software; H.Y. was responsible for validation and visualization. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Informed consent was obtained from each respondent.

Data Availability Statement

Not applicable.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Zhang, C.; Lu, Y. Study on Artificial Intelligence: The State of the Art and Future Prospects. J. Ind. Inf. Integr. 2021, 23, 100224. [Google Scholar] [CrossRef]
  2. Guo, K.; Lu, Y.; Gao, H.; Cao, R. Artificial Intelligence-Based Semantic Internet of Things in a User-Centric Smart City. Sensors 2018, 18, 1341. [Google Scholar] [CrossRef] [PubMed]
  3. Bigman, Y.E.; Gray, K. People Are Averse to Machines Making Moral Decisions. Cognition 2018, 181, 21–34. [Google Scholar] [CrossRef] [PubMed]
  4. Manu, M.; Sreejesh, S. Addressing Service Failure and Initiating Service Recovery in Online Platforms: Literature Review and Research Agenda. J. Strateg. Mark. 2021, 29, 658–689. [Google Scholar] [CrossRef]
  5. Borah, S.B.; Prakhya, S.; Sharma, A. Leveraging Service Recovery Strategies to Reduce Customer Churn in an Emerging Market. J. Acad. Mark. Sci. 2020, 48, 848–868. [Google Scholar] [CrossRef]
  6. Bag, S.; Gupta, S.; Kumar, A.; Sivarajah, U. An Integrated Artificial Intelligence Framework for Knowledge Creation and B2B Marketing Rational Decision Making for Improving Firm Performance. Ind. Mark. Manag. 2021, 92, 178–189. [Google Scholar] [CrossRef]
  7. Choi, S.; Mattila, A.S.; Bolton, L.E. To Err Is Human(-Oid): How Do Consumers React to Robot Service Failure and Recovery? J. Serv. Res. 2021, 24, 354–371. [Google Scholar] [CrossRef]
  8. Janes, L.; Olson, J. Humor as an Abrasive or a Lubricant in Social Situations: Martineau Revisited. HUMOR 2015, 28. [Google Scholar] [CrossRef]
  9. Caza, A.; Zhang, G.; Wang, L.; Bai, Y. How Do You Really Feel? Effect of Leaders’ Perceived Emotional Sincerity on Followers’ Trust. Leadersh. Q. 2015, 26, 518–531. [Google Scholar] [CrossRef]
  10. Hassey, R.V. How Brand Personality and Failure-Type Shape Consumer Forgiveness. J. Prod. Brand Manag. 2019, 28, 300–315. [Google Scholar] [CrossRef]
  11. Shuqair, S.; Pinto, D.C.; So, K.K.F.; Rita, P.; Mattila, A.S. A Pathway to Consumer Forgiveness in the Sharing Economy: The Role of Relationship Norms. Int. J. Hosp. Manag. 2021, 98, 103041. [Google Scholar] [CrossRef]
  12. Benenson, J.F.; Maiese, R.; Dolenszky, E.; Dolensky, N.; Sinclair, N.; Simpson, A. Group Size Regulates Self-Assertive versus Self-Deprecating Responses to Interpersonal Competition. Child Dev. 2002, 73, 1818–1829. [Google Scholar] [CrossRef]
  13. Epley, N.; Waytz, A.; Cacioppo, J.T. On Seeing Human: A Three-Factor Theory of Anthropomorphism. Psychol. Rev. 2007, 114, 864–886. [Google Scholar] [CrossRef]
  14. Zourrig, H.; Chebat, J.; Toffoli, R. Exploring Cultural Differences in Customer Forgiveness Behavior. J. Serv. Manag. 2009, 20, 404–419. [Google Scholar] [CrossRef]
  15. Christodoulides, G.; Gerrath, M.H.E.E.; Siamagka, N.T. Don’t Be Rude! The Effect of Content Moderation on Consumer-brand Forgiveness. Psychol. Mark. 2021, 38, 1686–1699. [Google Scholar] [CrossRef]
  16. Mischel, W.; Shoda, Y. A Cognitive-Affective System Theory of Personality: Reconceptualizing Situations, Dispositions, Dynamics, and Invariance in Personality Structure. Psychol. Rev. 1995, 102, 246–268. [Google Scholar] [CrossRef]
  17. Amir, O. The Frog Test: A Tool for Measuring Humor Theories’ Validity and Humor Preferences. Front. Hum. Neurosci. 2016, 10, 40. [Google Scholar] [CrossRef]
  18. Iii, J.G.M.; Netemeyer, R.G. A Longitudinal Study of Complaining Custorners’ Evaluations of Multiple Service Faiiures and Recovery Efforts. SAGE 2018, 66, 18512. [Google Scholar]
  19. Huang, B.; Philp, M. When AI-Based Services Fail: Examining the Effect of the Self-AI Connection on Willingness to Share Negative Word-of-Mouth after Service Failures. Serv. Ind. J. 2021, 41, 877–899. [Google Scholar] [CrossRef]
  20. Lv, X.; Yang, Y.; Qin, D.; Cao, X.; Xu, H. Artificial Intelligence Service Recovery: The Role of Empathic Response in Hospitality Customers’ Continuous Usage Intention. Comput. Hum. Behav. 2022, 126, 106993. [Google Scholar] [CrossRef]
  21. González-Gómez, H.V.; Hudson, S.; Rychalski, A. The Psychology of Frustration: Appraisals, Outcomes, and Service Recovery. Psychol. Mark. 2021, 38, 1550–1575. [Google Scholar] [CrossRef]
  22. Abulaish, M.; Kamal, A. Self-Deprecating Sarcasm Detection: An Amalgamation of Rule-Based and Machine Learning Approach. In Proceedings of the 2018 IEEE/WIC/ACM International Conference on Web Intelligence (WI), Santiago, Chile, 3–6 December 2018; pp. 574–579. [Google Scholar]
  23. Kim, M.-H. Why Self-Deprecating? Achieving ‘Oneness’ in Conversation. J. Pragmat. 2014, 69, 82–98. [Google Scholar] [CrossRef]
  24. Tang, L.; Sun, S. How Does Leader Self-Deprecating Humor Affect Creative Performance? The Role of Creative Self-Efficacy and Power Distance. Finance Res. Lett. 2021, 42, 102344. [Google Scholar] [CrossRef]
  25. Yu, C. Two Interactional Functions of Self-Mockery in Everyday English Conversations: A Multimodal Analysis. J. Pragmat. 2013, 50, 1–22. [Google Scholar] [CrossRef]
  26. Tsarenko, Y.; Rooslani Tojib, D. A Transactional Model of Forgiveness in the Service Failure Context: A Customer-driven Approach. J. Serv. Mark. 2011, 25, 381–392. [Google Scholar] [CrossRef]
  27. Tsarenko, Y.; Tojib, D. Consumers’ Forgiveness after Brand Transgression: The Effect of the Firm’s Corporate Social Responsibility and Response. J. Mark. Manag. 2015, 31, 1851–1877. [Google Scholar] [CrossRef]
  28. Murphy, J.; Gretzel, U.; Pesonen, J. Marketing Robot Services in Hospitality and Tourism: The Role of Anthropomorphism. J. Travel Tour. Mark. 2019, 36, 784–795. [Google Scholar] [CrossRef]
  29. Lv, X.; Liu, Y.; Luo, J.; Liu, Y.; Li, C. Does a Cute Artificial Intelligence Assistant Soften the Blow? The Impact of Cuteness on Customer Tolerance of Assistant Service Failure. Ann. Tour. Res. 2021, 87, 103114. [Google Scholar] [CrossRef]
  30. Ma, K.; Zhong, X.; Hou, G. Gaining Satisfaction: The Role of Brand Equity Orientation and Failure Type in Service Recovery. Eur. J. Mark. 2020, 54, 2317–2342. [Google Scholar] [CrossRef]
  31. Xie, Y.; Peng, S. How to Repair Customer Trust after Negative Publicity: The Roles of Competence, Integrity, Benevolence, and Forgiveness. Psychol. Mark. 2009, 26, 572–589. [Google Scholar] [CrossRef]
  32. Warren, C.; Barsky, A.; Mcgraw, A.P. Humor, Comedy, and Consumer Behavior. J. Consum. Res. 2018. [Google Scholar] [CrossRef]
  33. Gkorezis, P.; Bellou, V. The Relationship between Leader Self-Deprecating Humor and Perceived Effectiveness: Trust in Leader as a Mediator. Leadersh. Organ. Dev. J. 2016, 37, 882–898. [Google Scholar] [CrossRef]
  34. Bartneck, C.; Kulić, D.; Croft, E.; Zoghbi, S. Measurement Instruments for the Anthropomorphism, Animacy, Likeability, Perceived Intelligence, and Perceived Safety of Robots. Int. J. Soc. Robot. 2009, 1, 71–81. [Google Scholar] [CrossRef]
  35. Huang, R.; Ha, S. The Effects of Warmth-Oriented and Competence-Oriented Service Recovery Messages on Observers on Online Platforms. J. Bus. Res. 2020, 121, 616–627. [Google Scholar] [CrossRef]
  36. Chuang, S.-C.; Cheng, Y.-H.; Chang, C.-J.; Yang, S.-W. The Effect of Service Failure Types and Service Recovery on Customer Satisfaction: A Mental Accounting Perspective. Serv. Ind. J. 2012, 32, 257–271. [Google Scholar] [CrossRef]
  37. Sinha, J.; Lu, F.-C. “I” Value Justice, but “We” Value Relationships: Self-Construal Effects on Post-Transgression Consumer Forgiveness. J. Consum. Psychol. 2016, 26, 265–274. [Google Scholar] [CrossRef]
  38. Babin, B.J.; Zhuang, W.; Borges, A. Managing Service Recovery Experience: Effects of the Forgiveness for Older Consumers. J. Retail. Consum. Serv. 2021, 58, 102222. [Google Scholar] [CrossRef]
  39. Wenzel, M.; Okimoto, T.G.; Hornsey, M.J.; Lawrence-Wood, E.; Coughlin, A.-M. The Mandate of the Collective: Apology Representativeness Determines Perceived Sincerity and Forgiveness in Intergroup Contexts. Pers. Soc. Psychol. Bull. 2017, 43, 758–771. [Google Scholar] [CrossRef]
  40. Hu, Y.; (Kelly) Min, H.; Su, N. How Sincere Is an Apology? Recovery Satisfaction in A Robot Service Failure Context. J. Hosp. Tour. Res. 2021, 45, 1022–1043. [Google Scholar] [CrossRef]
  41. Wei, C.; Liu, M.W.; Keh, H.T. The Road to Consumer Forgiveness Is Paved with Money or Apology? The Roles of Empathy and Power in Service Recovery. J. Bus. Res. 2020, 118, 321–334. [Google Scholar] [CrossRef]
  42. Kim, S.; McGill, A.L. Gaming with Mr. Slot or Gaming the Slot Machine? Power, Anthropomorphism, and Risk Perception. J. Consum. Res. 2011, 38, 94–107. [Google Scholar] [CrossRef] [Green Version]
  43. Marcus, A.A.; Goodman, R.S. Victims and Shareholders: The Dilemmas Of Presenting Corporate Policy During A Crisis. Acad. Manage. J. 1991, 34, 281–305. [Google Scholar] [CrossRef]
  44. Warner, R.M.; Sugarman, D.B. Attributions of Personality Based on Physical Appearance, Speech, and Handwriting. J. Personal. Soc. Psychol. 1986, 50, 792–799. [Google Scholar] [CrossRef]
  45. Trianasari, N.; Butcher, K.; Sparks, B. Understanding Guest Tolerance and the Role of Cultural Familiarity in Hotel Service Failures. J. Hosp. Mark. Manag. 2018, 27, 21–40. [Google Scholar] [CrossRef]
  46. Anderson, C.; Galinsky, A.D. Power, Optimism, and Risk-Taking. Eur. J. Soc. Psychol. 2006, 36, 511–536. [Google Scholar] [CrossRef]
  47. Belanche, D.; Casaló, L.V.; Flavián, C.; Schepers, J. Service Robot Implementation: A Theoretical Framework and Research Agenda. Serv. Ind. J. 2020, 40, 203–225. [Google Scholar] [CrossRef]
  48. Jiang, T.; Li, H.; Hou, Y. Cultural Differences in Humor Perception, Usage, and Implications. Front. Psychol. 2019, 10, 123. [Google Scholar] [CrossRef]
  49. Huang, M.-H.; Rust, R.T. Engaged to a Robot? The Role of AI in Service. J. Serv. Res. 2021, 24, 30–41. [Google Scholar] [CrossRef]
Sustainability 15 03294 g001 550
Figure 1. Theoretical framework.
Figure 1. Theoretical framework.
Sustainability 15 03294 g001
Sustainability 15 03294 g002 550
Figure 2. Results of the simple slope test of response style and service recovery satisfaction at the level of sense of power.
Figure 2. Results of the simple slope test of response style and service recovery satisfaction at the level of sense of power.
Sustainability 15 03294 g002
Sustainability 15 03294 g003 550
Figure 3. Results of study 3 with significance values.
Figure 3. Results of study 3 with significance values.
Sustainability 15 03294 g003
Table
Table 1. Results of perceived sincerity and perceived intelligence mediating effects test.
Table 1. Results of perceived sincerity and perceived intelligence mediating effects test.
VariablePerceived SincerityPerceived IntelligenceService Recovery
Satisfaction
βSEtβSEtβSEt
Constant2.428 ***0.3906.2212.212 ***0.4035.4810.3080.2951.043
Response method0.542 **0.2482.1790.737 **0.2572.8670.1560.1670.937
Perceived sincerity 0.450 ***0.0825.437
Perceived intelligence 0.447 ***0.0855.222
R20.0390.0660.655
F4.7518.22471.525
(** p < 0.05, *** p < 0.01).
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yang, Z.; Zhou, J.; Yang, H. The Impact of AI’s Response Method on Service Recovery Satisfaction in the Context of Service Failure. Sustainability 2023, 15, 3294. https://0-doi-org.brum.beds.ac.uk/10.3390/su15043294

AMA Style

Yang Z, Zhou J, Yang H. The Impact of AI’s Response Method on Service Recovery Satisfaction in the Context of Service Failure. Sustainability. 2023; 15(4):3294. https://0-doi-org.brum.beds.ac.uk/10.3390/su15043294

Chicago/Turabian Style

Yang, Zengmao, Jinlai Zhou, and Hongjun Yang. 2023. "The Impact of AI’s Response Method on Service Recovery Satisfaction in the Context of Service Failure" Sustainability 15, no. 4: 3294. https://0-doi-org.brum.beds.ac.uk/10.3390/su15043294

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop