Double jeopardy: effects of inter-failures and webcare on (un-)committed online complainants’ revenge

Abstract

Purpose

The purpose of this research is the exploration of online complainants' revenge based on their consumer-brand relationship strength and received webcare. The authors introduce inter-failures (i.e. the perceived number of earlier independent service failures that a customer has experienced with the same brand involved in the current service failure) as the central frame condition.

Design/methodology/approach

To test our hypotheses, both a scenario-based online experiment (n = 316) and an online survey (n = 492) were conducted.

Findings

With an increasing number of inter-failures, online complainants with a high-relationship strength move from the “love is blind” effect (no inter-failures) to the “love becomes hate” effect (multiple inter-failures), when they ultimately become more revengeful than their low-relationship strength counterparts. In addition, the authors show that in the case of no or few inter-failures, accommodative webcare has a lasting positive effect over no/defensive webcare for both low- and high-relationship complainants. More importantly, however, when consumers have experienced multiple inter-failures, accommodative webcare becomes ineffective (for low-relationship complainants) or boomerangs by cultivating revenge towards the brand (among high-relationship complainants), but not strategic avoidance.

Research limitations/implications

The findings have pronounced implications for the literature on customer–brand relationships following service failures and the literature, which predominantly emphasizes the unconditionally positive effects of accommodative webcare.

Originality/value

This study is the first that simultaneously considers the prior customer–brand relationship, inter-failures and webcare to explain online complainants' revenge.

Citation

Hutzinger, C. and Weitzl, W.J. (2023), "Double jeopardy: effects of inter-failures and webcare on (un-)committed online complainants’ revenge", Internet Research, Vol. 33 No. 7, pp. 19-45. https://doi.org/10.1108/INTR-02-2022-0115

1. Introduction

When consumers purchase a brand for the first time, they enter a customer life cycle. During this period, positive experiences with the chosen brand help to develop strong relational bonds (Arnould and Thompson, 2005; Robertson et al., 2022), which have favorable effects (e.g. increased customer loyalty) (e.g. Fournier, 1998; Batra et al., 2012). However, during this life cycle, negative experiences – that is, situations when the brand does not meet the customer's expectations (i.e. service failures) (e.g. unfriendly staff, malfunctioning product) – can also occur once or even multiple times (Maxham and Netemeyer, 2002; Weitzl et al., 2018). When failures happen, dissatisfied customers regularly visit a brand's social media channel to voice an online complaint and inform the involved company and other, uninvolved consumers about their problems. This is true for all phases of the customer life cycle (Grégoire et al., 2019; Weitzl and Hutzinger, 2019). Following these complaints, companies have the opportunity to win back their dissatisfied customers by means of reactive “webcare” (i.e. a company's online complaint handling effort; van Noort and Willemsen, 2012). While prior research has provided considerable insights into the general effectiveness of webcare (e.g. Huang and Ha, 2020; Istanbulluoglu, 2017), marketing practitioners are still uncertain about webcare's impact on complainants, who have a varying intensity of relational bonds and different failure histories.

When responding to online complaints, marketers can – on one hand – hope for the so-called “love is blind” effect (Grégoire and Fisher, 2006), which suggests that strong relational ties help them to mitigate negative customer reactions (e.g. revenge-taking). On the other hand, marketers have to fear the “love becomes hate” effect (Grégoire et al., 2009), when strong bonds – rather than weak ones – cultivate revenge and make complainants unreceptive or even more aggressive after recovery efforts. Recent work urges researchers to better explain which conditions lead to these two opposite effects (Khamitov et al., 2020). In this research, we assume that, depending on the number of prior “inter-failures” which complainants share with a specific company, their reactions following webcare gradually shift from the “love is blind” to the “love becomes hate” effect. We define inter-failures as the perceived number of earlier independent service failures that a customer has experienced with the same brand involved in the current service failure. Inter-failures need to be distinguished from “intra-failures” [1]. Table 1 gives an overview of the empirical research on inter-failures.

As Table 1 shows, prior research on inter-failures is rather limited. Yet, no study has simultaneously considered the frequency of inter-failures and the prior complainant-brand relationship in the context of webcare. However, the work of Weitzl et al. (2018), who show that in case of accommodative webcare, complainants that have experienced multiple as compared to few inter-failures make more unfavorable company-related failure attributions, provides some first insights. Nevertheless, the authors neglect the varying stages of complainant-brand relationships and their influence on customer revenge across situations of no, few and multiple inter-failures. In the research at hand, we suggest that the perceived number of inter-failures is a central predictor of the proposed transition from the “love is blind” to the “love becomes hate” effect. This means that online complainants who had a high pre-failure relationship (i.e. brand-committed customers) – as compared to those with a low pre-failure relationship (i.e. brand-uncommitted customers) – only show a more forgiving reaction to webcare if no earlier independent failures have been experienced with this brand. When the perceived number of failures increases over time, forgiveness diminishes and high-relationship online complainants (i.e. customers with a high relationship strength prior to the current failure) become more aggressive than low-relationship complainants. Finally, in the case of numerous inter-failures, high-relationship online complainants feel extreme dissatisfaction and a strong disruption of the customer–brand relationship (Maxham and Netemeyer, 2002). Such severe violations of relational norms (Aggarwal, 2004) lead these complainants to be more revengeful than their low-relationship counterparts (i.e. they have a higher revenge level, which can be summarized as showing a “love becomes hate” effect) – even when controlling for double deviations (i.e. intra-failures) and other factors (e.g. failure severity).

Moreover, we theoretically argue and empirically show that “accommodative webcare” (i.e. a company's complaint response in which it takes full responsibility for a service incident and makes a redress offer; Weitzl, 2019) leads to varying effects depending on complainants' relationship strength and the number of inter-failures before the current incident: Accommodative webcare is more effective than no/defensive webcare to mitigate revenge, but only when no or few failures with the same brand have happened in the past. Further, it can be ineffective for low-relationship complainants or can even intensify hate by sparking a boomerang effect on the company by directly increasing high-relationship complainants' revenge intentions – when they have experienced inter-failures. Table 2 summarizes the proposed transition effects: The transitions of the “love is blind” to the “love becomes hate” effect (i.e. the relationship strength based “revenge level”) and the forgiveness-to-revenge-intensifying effect of accommodative webcare (i.e. the accommodative webcare based “revenge intensity”). Neither of these two processes has been covered by existing research so far.

2. Reactions following online complaints

2.1 Companies' reactions to online complaints

After an online complaint, companies can choose from various public online response strategies to counterbalance complainants' losses. These strategies typically range on a continuum of responsibility from a “no-response” (i.e. the company takes no responsibility), via “defensive webcare” (i.e. limited responsibility), to “accommodative webcare” (i.e. full responsibility) (Weitzl and Hutzinger, 2017). A no-response ignores the complainant, while with defensive webcare, the accused company regularly blames the complainants themselves or a third party for the incident. Accommodative webcare typically includes social benefits, such as an apology or an explanation (Berry, 1995), but often also economic benefits (e.g. a redress offer) which (over-)compensate the online complainant for the inconvenience experienced (van Gils and Horton, 2019). Given extant research (e.g. Lee, 2005), which demonstrates that consumers react very similarly to a company's no- and defensive responses, in this study, we only contrast complainants' reactions to “no/defensive webcare” vs “accommodative webcare”.

2.2 Complainants' post-webcare reaction

It is assumed that, after obtaining webcare, online complainants regularly adapt their thoughts, feelings, intentions and habits towards the involved brand (Grégoire et al., 2019; Weitzl et al., 2018). One of the most critical intentions is their propensity to take further revenge because some of their expectations remained unmet. Grégoire et al. (2009, p. 19) define revenge as “customers' need to punish and cause harm to firms for the damages they have caused.” Consequently, we conceptualize post-webcare “revenge intention” as online complainants' inclination to punish and cause inconvenience to the involved company by spreading negative word-of-mouth about the brand to fellow consumers after receiving webcare. Following Grégoire et al. (2010), we suggest that post-webcare revenge arises due to a “cognitions-emotions-actions” sequence: Accordingly, when complainants appraise the received webcare strategy as insufficient or unjust, they feel anger, which in turn triggers revenge intentions and guides revenge actions (e.g. boycotting). While dissatisfied complainants can exert different forms of retaliation (e.g. deliberately avoiding the brand in future), we focus on revengeful negative word-of-mouth as one of the most active, direct and aggressive forms which detrimentally impacts the criticized company (Grégoire and Fisher, 2008; Grégoire et al., 2018).

The main section of this paper is structured as follows: We first focus on the case of no inter-failures by discussing the conceptual background, developing the respective hypotheses and testing them. Afterwards, we move to the case where complainants have experienced inter-failures and proceed in a similar manner. This procedure enables us to stepwise assess our conceptual framework predicting the interplay of the three focal drivers of online revenge: relationship strength, inter-failure frequency and webcare type (see Figure 1).

3. Relationship's effect on post-webcare revenge in the case of no inter-failures

In this research, we distinguish between two types of online complainants: “low-relationship online complainants”, who were in an exchange relationship and “high-relationship online complainants”, who were in a communal relationship with the brand prior to the current failure (Aggarwal, 2004). In exchange relationships, interaction partners follow a quid pro quo approach. That is, one partner gives benefits to the other partner (e.g. money) in order to get something in return (e.g. a service). The partners typically interact for purely transactional purposes (e.g. granting access to a service). In contrast, communal relationships are particularly based on social factors, where the transaction also conveys symbolic meaning (e.g. expression of dedication) for the consuming party (Aggarwal, 2004; Clark and Mills, 1993). Customers in communal relationships have a strong sense of belonging and commitment towards the brand as the interaction partner, as well as an intense drive to maintain emotional ties with it (Ringberg et al., 2007).

We predict that when no inter-failures have occurred, high-relationship complainants react with lower levels of revenge after receiving webcare than low-relationship complainants. This argument is mainly supported by an evolving stream of literature (Hess et al., 2003; Tax et al., 1998), finding that strong relationships make consumers respond more favorably to failures and companies' attempts to recover them afterwards. This pattern corresponds to the moderating role of commitment identified in the information-processing literature, according to which uncommitted consumers show more attitude change with negative than positive information, while highly committed consumers do not (Ahluwalia, 2002; Ahluwalia et al., 2000). Following the social cognition literature, individuals who are in an ambiguous situation tend to overlook or downplay information that is inconsistent with their earlier positive cognitions when they feel a strong connection with an object (e.g. a specific company) (Herr et al., 1983). In the service recovery literature, this pattern is described by Grégoire and Fisher (2006) as the “love is blind” effect. The authors demonstrate that strong relationships can mitigate detrimental effects of failures, such as decreasing loyalty. Among the research identifying such a benevolent reaction, Mattila's (2004) study shows that after a failure, customers with high – as compared to low – affective commitment (defined as a sense of belonging and involvement with the company) are less likely to share negative word-of-mouth and more likely to remain loyal.

We argue, however, that the above-mentioned effects only unfold when high-relationship complainants have not experienced any negative prior incidents with the brand involved in the current failure (i.e. no inter-failures). That is, we propose one of the “love is blind” effect's boundary conditions. This claim is supported by the cognitive dissonance theory (Festinger, 1957), which suggests that individuals have a strong drive to dissolve mental imbalances. A failure by a beloved brand triggers such an imbalance, but the company's attempt to restore the relationship is interpreted as positive information consistent with the earlier favorable stance. When the committed complainant is a “tabula rasa” in terms of inter-failures, the brand-obliging cognitive biases (Ahluwalia, 2002; Ahluwalia et al., 2000; Herr et al., 1983), such as overemphasizing the positive aspects of any form of webcare and downplaying the negative consequences of the current failure, become effective. Thus, we hypothesize:

H1.

(“Love is blind” hypothesis). Only in case of no inter-failures, high-relationship complainants have a lower revenge intention than low-relationship complainants when they receive (a) no/defensive webcare or (b) accommodative webcare.

We further theorize that when no inter-failures exist, for both low-relationship and high-relationship complainants, accommodative webcare leads to a lower revenge intention than no/defensive webcare. We develop H2 – in line with Weitzl et al. (2018) – by referring to the equity theory (Adams, 1965), which claims that individuals who experience unfairness (e.g. caused by a failure) try to restore their psychological equity. Online complainants can at least partially achieve this rebalance by means of social benefits (e.g. apology) and economic benefits (e.g. refunds) conveyed in accommodative webcare (del Rio-Lanza et al., 2009). In the no-prior-failure case, the positive effects of accommodative responses on complainants are well documented for online complaint-handling (Chang et al., 2015). Accommodative webcare typically includes monetary benefits (e.g. price reduction), which have been shown to positively affect the perceived fairness of the problem resolution process (Tax et al., 1998). Consequently, we argue that compensation leads to lower levels of revenge.

In contrast, defensive webcare lets consumers believe that the failure was the company's fault (Weitzl, 2019), which triggers anger and negative cognitions (Lee, 2005). No-responses ignoring customers' problems provide a comparable level of benefit (van Noort and Willemsen, 2012). When online complainants have no inter-failures with the company, we assume that the effectiveness of webcare is not conditional on their relationship strength. Therefore:

H2.

(“Generosity stimulates forgiveness” hypothesis). In the case of no inter-failures, accommodative webcare, as compared to no/defensive webcare, leads to lower revenge intention of (1) low-relationship complainants and (2) high-relationship complainants.

4. Study 1: no inter-failures

4.1 Study design and method

To test our first set of hypotheses, we conducted a scenario-based online experiment with a 2 (relationship strength: low vs high) × 3 (webcare: no vs defensive vs accommodative) between-subjects design. Data was collected from 316 adult consumers [2] (female = 62.22%, M_age = 25.56 years, SD_age = 6.00) via convenience sampling. The scenario – inspired by Grégoire et al. (2009) – asked participants to imagine inviting two friends to an Italian restaurant (“Da Marco”) to celebrate a birthday. In the low-relationship condition, participants were told that they were regular customers of Italian restaurants, although they were first-time visitors to this specific restaurant. Therefore, they did not know what to expect regarding food, service and atmosphere. In the high-relationship condition, participants were told that they regularly went to Italian restaurants and that they had already visited “Da Marco” several times. Consequently, they did know that they could expect excellent food, great service and a pleasant atmosphere. Furthermore, they were told that they were committed customers of the restaurant and considered the owner a friend.

Afterwards, all participants were instructed to imagine a service failure in which the served food had several deficiencies (e.g. low quality, wrong preparation) and that they were billed for several unconsumed beverages (30€) in their final check of 105€. While the restaurant owner listened to the participants' in-house complaint, he neither apologized nor corrected the incorrect check (i.e. double deviation). Because of this negative experience, participants were further told that they consequently posted an online complaint on the restaurant's Facebook page. In the no-webcare condition, the restaurant did not react to this complaint. In the defensive webcare condition, the restaurant doubted the reasons for complaining and considered itself not responsible for the incident. In the accommodative webcare condition, the restaurant apologized, provided an explanation, took full responsibility and gifted a 40€ voucher for the next visit. Participants were randomly assigned to one of the six experimental groups.

To check the manipulation of relationship strength, we measured customer commitment (3 items adapted from Grégoire et al. (2009), e.g. “I intended to maintain the relationship with the restaurant Da Marco for a long time.”; α = 0.86) on a 5-point Likert-scale ranging from 1 (=totally disagree) to 5 (= totally agree). An independent-samples t-test revealed that customer commitment was significantly higher in the high-relationship-strength group than in the low-relationship-strength group (M_{high RS} = 3.62, SD_{high RS} = 0.88; M_{low RS} = 2.52, SD_{low RS} = 1.04, t(314) = 10.16, p < 0.001). Thus, our manipulation of relationship strength was successful. To check the manipulation of webcare, we measured the responsibility taken by the company (4 items, e.g. “The restaurant ‘Da Marco’ explained what caused the incident.”; α = 0.97) on a 5-point scale ranging from 1 (= totally disagree) to 5 (= totally agree). Independent-samples t-tests revealed that the responsibility taken was significantly higher in the accommodative-webcare group (M = 4.24, SD = 0.77) than in the defensive-webcare group (M = 1.21, SD = 0.56, t(215) = 33.17, p < 0.001) and in the no-webcare group (M = 1.17, SD = 0.58, t(203) = 31.99, p < 0.001). Since the responsibility taken did not differ between the defensive- and the no-webcare group (t(208) = 0.57, p = 0.572) and additional analyses [3] did not show any significant differences in their characteristics, the two groups were collapsed into one. Given this, our manipulation of no/defensive vs accommodative webcare was successful.

4.2 Measures

The online questionnaire included standard socio-demographics and a series of multi-item scales taken from established academic literature to capture the study's key constructs and covariates. The measures – scored from 1 to 5, with higher values indicating higher agreement etc. – captured, amongst others, post-webcare revenge intention [4 items adapted from Joireman et al. (2013) and Kaltcheva et al. (2013), e.g. “After this incident I will spread negative word-of-mouth about the company.”; α = 0.91]. Negative word-of-mouth has been operationalized as an essential component of revenge (Grégoire and Fischer, 2008). Furthermore, several constructs were included that were shown to be important determinants for consumers' reactions in a failure episode, like failure severity (3 items adapted from Hess et al. (2003), e.g. “I consider the incident a major problem.”; α = 0.83), anger (3 items adapted from Zarantonello et al. (2016), e.g. “I feel enraged.”; α = 0.91) and attribution of stability (2 items adapted from Wirtz and Mattila (2004), e.g. “The cause of the incident was something permanent.”; α = 0.90). Confirmatory factor analysis (CFA) pointed towards a satisfactory fit of the measurement model and acceptable psychometric properties of the measures used, such as adequate levels of convergent and discriminant validity [4].

Cross-sectional studies with self-reported measures are vulnerable to the “common method bias” (Podsakoff et al., 2003). To prevent this, we employed several procedural remedies at the data-collection stage. Specifically, we ensured the participants' anonymity, informed all participants that there were no right or wrong answers, implemented a counterbalancing question order and tried to improve the psychometric quality of the scale items with a pretest to avoid item ambiguity. Furthermore, regarding statistical remedies, we first applied Harman's single-factor test on the involved constructs, which revealed that less than 50% of the variance was explained by a common factor. Second, Liang et al.’s (2007) latent method construct approach was applied by means of CFA. Here, we specified a method factor together with the original latent variables, in order to check whether the AVE by the method factor was substantially less than the AVE by the substantive original factors. The average substantive variance was significantly higher than the method variance, which was less than 12% of the former in our model. Finally, we also utilized Lindell and Whitney's (2001) marker variable assessment technique to detect and control the bias. This analysis showed that the common factor's path loadings on the involved scale items were reasonably low (<1%). Taken together, the three tests consistently indicated that common method bias was less of a concern in study 1.

4.3 Results

Hypotheses Testing. To test H1 and H2, we ran an ANCOVA with revenge intention as the dependent variable and relationship strength and webcare as factors. Gender, age and failure severity were included as covariates. Due to unequal cell sizes, the sample was bootstrapped with 1,000 replications (Hayes, 2018). The results for revenge intention showed a significant effect of relationship strength ((F (1, 308) = 23.34, p < 0.001, partial eta² = 0.07) and webcare ((F (1, 308) = 51.70, p < 0.001, partial eta² = 0.14). Furthermore, the interaction effect between relationship strength and webcare was – as expected – not significant ((F (1, 308) = 0.00, p = 0.967, partial eta² = 0.00). Independent-samples t-tests revealed that, in the case of no inter-failures and no/defensive webcare, high-relationship complainants (M = 3.40, SD = 1.17) had lower revenge intention than low-relationship complainants (M = 4.04, SD = 0.94, t(208) = −4.26, p < 0.001). Thus, H1a was supported. The same held when low- and high-relationship complainants received accommodative webcare (M_{high RS} = 2.45, SD_{high RS} = 1.16; M_{low RS} = 3.11, SD_{low RS} = 1.18, t(104) = −2.79, p < 0.01). Thus, H1b was also supported (see Table 3 for results).

Furthermore, in line with H2a, low-relationship complainants had lower revenge intention when they had received accommodative webcare (M = 3.11, SD = 1.18), as compared to no/defensive webcare (M = 4.04, SD = 0.94, t(126) = −4.79, p < 0.001). In support of H2b, the same held for high-relationship complainants (M_{acc. webcare} = 2.45, SD_{acc. webcare} = 1.16; M_{no/def. webcare} = 3.40, SD_{no/def. webcare} = 1.17, t(186) = −5.31, p < 0.001). Table 3 summarizes the results.

4.4 Discussion

In study 1, we identify that the “love is blind” effect (Grégoire and Fisher, 2006) is apparent when high-relationship online complainants have faced no inter-failures with the same brand in the past. In this situation, they have a lower inclination to take revenge after receiving webcare than low-relationship complainants. Our results testify that building strong bonds with customers pays off in the case of service failures – but only when earlier service experiences with this company have been error-free. In addition, we also demonstrate that in the case of no inter-failures, accommodative webcare – as compared to less-beneficial responses (i.e. no/defensive webcare) – leads to a lower revenge intention for both low- and high-relationship complainants. Since the literature remained silent on the possible increasing detrimental influences of few and multiple inter-failures, which are likely to destroy the “love is blind” effect, we investigate these situations in study 2.

5. Relationship's effect on post-webcare revenge in the case of few inter-failures

Prior work on the role of inter-failures in the context of webcare is scarce and has never investigated the role of the complainant-brand relationship (see Table 1). Furthermore, existing research only focuses on a single inter-failure (see Weitzl et al. (2018) for an exception) and has never jointly considered both the kind of relationship and the frequency of inter-failures. At the same time, it can be considered as a norm that online complainants have been confronted with inter-failures with the company involved in the current failure. This means that they have experienced at least a few earlier incidents over several independent transactions, which is consistent with Sivakumar et al.'s (2014) dynamic perspective on service delivery. Consumers' history of positive/negative experiences has been noted to determine their judgment and behavioral tendencies towards a brand (Magnini et al., 2007). Even the occurrence of one earlier service failure leads to considerably fewer favorable consumer reactions as compared to no prior incidents. Consumers who have experienced two independent failures with the same brand attribute more blame to the company after the second failure than after the first (Maxham and Netemeyer, 2002). Given this, we theorize that the more inter-failures occur, the more consumers experience a violation of relational norms (Aggarwal, 2004) and the higher their unfavorable failure attributions (e.g. problem stability) (Wirtz and Mattila, 2004). More specifically, we argue that the more the perceived number of independent failures of a company increases, the less likely committed consumers are to show the “love is blind” effect. With a rising number of inter-failures, these complainants learn that poor service quality is a long-lasting (i.e. stable) problem. Consequently, high-relationship complainants start doubting their personal affection towards the brand. Their post-webcare revenge inclination rises and aligns to that of uncommitted online complainants. Given this, we hypothesize the following:

H3.

(“Diminishing love is blind” hypothesis). In the case of few inter-failures, high-relationship complainants have a comparable revenge intention as low-relationship complainants when they receive (a) no/defensive webcare or (b) accommodative webcare.

It is further suggested that – irrespective of complainants' relationship strength – webcare types trigger similar consumer reactions in the no and few failures cases. This means that, following a limited number of earlier independent failures, complainants are still expected to react more positively to accommodative responses (including an apology, an explanation and/or a form of monetary compensation) than to no/defensive webcare responses. While the stress on consumers certainly increases in this situation, it does not exceed a dissonance level when the problem of the current failure cannot be restored with a benevolent response anymore. Thus, committed and uncommitted complainants are receptive to sincere online recovery attempts. It follows:

H4.

(“Enduring forgiveness” hypothesis). In the case of few inter-failures, accommodative webcare, as compared to no/defensive webcare, leads to a lower revenge intention of (a) low-relationship complainants and (b) high-relationship complainants.

6. Relationship's effect on post-webcare revenge in the case of multiple inter-failures

Furthermore, it is assumed that when high-relationship complainants have experienced multiple inter-failures with the same brand in the past, the “love becomes hate” effect (Grégoire et al., 2009) is likely to occur. Research revealed that high-relationship complainants are sometimes more willing to retaliate against the company involved in the current failure than low-relationship complainants (Grégoire and Fisher, 2006, 2008). Committed customers, who strongly identify with a company, typically have higher expectations about the service they believe they deserve. However, having experienced many inter-failures contrasts sharply with these expectations, which cultivates revenge more strongly than among uncommitted customers (Brockner et al., 1992). Grégoire and Fisher (2008) conclude that perceived betrayal (i.e. customers' perception that a company has violated normative aspects of their relationship on purpose) is the key driver of the effect: Specifically, high-relationship complainants are more inclined to feel betrayed, as they typically place more confidence in the company than low-relationship complainants. In the case of a failure, these consumers think that the company has violated the social norm required to maintain the communal relationship (Koehler and Gershoff, 2003). This results in a feeling of broken trust and complainants' perception that the failure is an act of betrayal.

We acknowledge that perceived betrayal creates a critical momentum that energizes dissatisfied customers to restore fairness by taking revenge. More specifically, a perceived infringement of the social normative standard is generally regarded as a justification for rebalancing their relationship and re-establishing social order by exerting justified retaliation. This is particularly true if complainants have experienced multiple independent failures with the brand in the past. This accumulation of negative experiences increases high-relationship complainants' certainty that increased retaliation is appropriate. In contrast, for low-relationship complainants, several negative experiences with the brand lead to a relatively weaker perceived infringement and consequently to a reduced willingness to take revenge after receiving webcare. Based on these arguments, we hypothesize:

H5.

(“Love becomes hate” hypothesis). In the case of multiple inter-failures, high-relationship complainants have a higher revenge intention than low-relationship complainants when they receive (a) no/defensive webcare or (b) accommodative webcare.

Further, we claim that, in the case of multiple inter-failures, the advantage of accommodative webcare over no/defensive webcare for low-relationship complainants visible for a limited number of inter-failures is likely to diminish. The development of H6 is in line with the argumentation of Weitzl et al. (2018). According to prospect theory (Kahneman and Tversky, 1979) and the disconfirmation framework (Mittal et al., 1998), a considerably greater number of positive experiences is required to overcome a single or multiple negative events. Therefore, a person who has suffered a great number of problems is likely to hold attitudes that are more negative and to have a more suspicious stance towards a brand. This holds even when a company offers social and economic benefits by means of accommodative webcare (i.e. a single positive experience). Furthermore, it has been shown that each additional unit of a positive service performance, such as an accommodative response, has a diminishing value for the consumer (Mittal et al., 1998). In the case of multiple prior service failures, low-relationship complainants may become desensitized to recovery attempts and focus more on the current negative outcomes. Consequently, following multiple negative prior experiences, webcare may be unable to reduce unfavorable failure attributions (de Matos et al., 2007). Attribution theory suggests that consumers are more inclined to make negative attributions when a series of failures has happened (Maxham and Netemeyer, 2002; Magnini et al., 2007). Over time, low-relationship complainants learn from their various negative interactions and develop relatively stable attitudes and attributions that make them unreceptive to webcare. Thus, we propose:

H6.

(“Ineffective webcare” hypothesis). In the case of multiple inter-failures, for low-relationship complainants, accommodative webcare, as compared to no/defensive webcare, leads to a comparable revenge intention.

In contrast, we theorize that a different mechanism guides the varying reactions to webcare shown by high-relationship complainants: More specifically, given their extensive number of inter-failures experienced, these complainants feel strongly betrayed and consider themselves to be unrewarded for their commitment and past loyalty. We argue that committed complainants' unfavorable stance can even be intensified when they receive inappropriate webcare. In this regard, the potential negative effect of apologies is noteworthy: By apologizing, the sender expresses responsibility and remorse for the violation (Skarlicki et al., 2004; Tomlinson, 2012). Offering apologies has been shown to reduce interactional fairness in a bargaining situation (Skarlicki et al., 2004) and can lead to lower perceived competence of the sender (Chaudhry and Loewenstein, 2019). After a severe violation of trust, apologies might be ineffective in improving the situation given that they acknowledge guilt (Kim et al., 2006; Riordan et al., 1983). Most importantly, apologies which are not associated with honest feelings of guilt have the potential to severely boomerang to the sender (Kim et al., 2004). Therefore, companies have been occasionally advised not to use apologies in their communication (Skarlicki et al., 2004). Social psychologists have shown that victims of a problem are more inclined to retaliate when another party validates their interpretation of the event and justifies their negative emotions (Eaton, 2013). Consequently, we theorize that high-relationship complainants regard accommodative webcare as a suitable validation mechanism. Their intense frustration leads them to interpret this reaction consistently with their already cultivated strong desire for retaliation. As outlined before, with an accommodative response, companies take full responsibility for the current failure. We argue that high-relationship complainants who are dissatisfied by multiple inter-failures interpret this well-meant webcare negatively as evidence of the company's own responsibility. The company's confession allows consumers to rationalize their negative emotions and post-webcare revenge. Hence:

H7.

(“Boomeranging generosity” hypothesis). In the case of multiple inter-failures, for high-relationship complainants, accommodative webcare, as compared to no/defensive webcare, leads to a higher revenge intention.

7. Explaining the transition from the “love-is-blind” to the “love-becomes-hate” effect

Prior research consistently shows that consumers' revenge is predominantly caused by anger (e.g. Grégoire et al., 2009). This negative emotion is – according to appraisal theory (Scherer, 1999) – triggered by unfavorable blame attributions. Recent research demonstrates that particularly the attribution of problem stability (i.e. the perceived relative permanence of a problem) is essential for consumers (Weitzl et al., 2018). This research proposes that the relationship strength and the number of inter-failures dictate the complainants' causal inferences of the problem. More specifically, we claim that in case of no inter-failures, stronger relational bonds lead to a less unfavorable stability attribution than weak bonds. This assumption is consistent with consumer–brand identification theory (Stokburger-Sauer et al., 2012) claiming that brand lovers with no prior negative experiences are more forgiving. However, when complainants have experienced multiple inter-failures before, this effect is reversed and an increased relationship strength stimulates more unfavorable attributions of stability, which ultimately leads to increased revenge via anger. Therefore:

H8.

(“Transition” hypothesis). The effect of relationship strength on revenge is mediated by the attribution of stability and anger depending on the number of inter-failures.

8. Study 2: few and multiple inter-failures

8.1 Study design and method

An online survey with 492 adult consumers from a consumer panel (female = 42.90%, M_age = 35.10 years, SD_age = 11.06) that had experienced a real service failure (e.g. an incompetent salesperson) during a personal consumption process within the last 6 months was conducted. All respondents had experienced a recovery failure (i.e. a “double deviation”) following the current service failure. Because of this incident, they decided to post a public online complaint on the company's social media page. This approach of studying real negative experiences of dissatisfied consumers is in line with comparable research (Bougie et al., 2003). The fully standardized online questionnaire measured participants' pre-failure relationship strength (6 items adapted from Harrison-Walker (2001), e.g. “I had a special relationship with this company.”; α = 0.89). We consequently classified them into groups with low (n_{low RS} = 238) and high relationship strength (n_{high RS} = 254) by a median-split. Furthermore, participants indicated the frequency of inter-failures (i.e. the amount of earlier independent failures with the same brand) on a 5-point rating scale ranging from 1 = “no failures” to 5 = “multiple failures”. Based on inter-failures, we selected the participants with few (n = 239) and multiple inter-failures (n = 253) [5]. To measure webcare, we asked participants how the company reacted to their online complaint. The respective items comprised no/defensive webcare (adapted from Davidow (2003), e.g. “The company denied responsibility.”) and accommodative webcare (adapted from Cambra-Fierro et al. (2015), e.g. “Apologized to me.”; “Offered me a discount.”). Based on these characteristics, participants were classified into four groups (2 (relationship strength: low vs high) × 2 (webcare: no/defensive [6] vs accommodative)) for both the few and the multiple inter-failure conditions.

8.2 Measures

The online questionnaire further included measures of the key constructs (scored from 1 to 5, with higher values indicating higher agreement etc.), the standard socio-demographics as well as covariates: This included failure severity (3 items adapted from Hess et al. (2003), e.g. “I consider the incident a major problem.”; α = 0.78), attribution of stability (2 items adapted from Wirtz and Mattila (2004), e.g. “The cause of the incident was something permanent.”; α = 0.60), anger (3 items adapted from Zarantonello et al. (2016), e.g. “I felt enraged.”; α = 0.83) and revenge intention (4 items adapted from Joireman et al. (2013) and Kaltcheva et al. (2013), e.g. “After this incident I will spread negative word-of-mouth about the company.”; α = 0.87). Again, CFA results indicated satisfactory psychometric measure properties, including convergent and discriminant validity [7]. Furthermore, by using the same three-step approach as in study 1, no signs of a severe common method bias were identifiable, which further supported the validity of the measures.

8.3 Results

8.3.1 Hypotheses testing: few inter-failures

To test H3 and H4 – consistent with our analyses on no inter-failures – we ran an ANCOVA with revenge intention as dependent variable and relationship strength and webcare as factors. To control for possible influences, we again included gender, age and failure severity as covariates. The bootstrapped results for revenge intention showed a non-significant effect of relationship strength (F (1, 232) = 0.35, p = 0.558, partial eta² = 0.00) and a significant webcare effect (F (1, 232) = 13.97, p < 0.001, partial eta² = 0.06). At the same time, the interaction effect between relationship strength and webcare was – as expected – not significant ((F (1, 232) = 0.13, p = 0.719, partial eta² = 0.00). Independent-samples t-tests revealed that, in the case of few inter-failures and no/defensive webcare, high-relationship complainants (M = 3.30, SD = 1.05) had a revenge intention comparable to low-relationship complainants (M = 3.25, SD = 1.09, t(169) = 0.29, p ≥ 0.05). Hence, H3a was supported. The same pattern emerged when they received accommodative webcare (M_{high RS} = 2.58, SD_{high RS} = 1.07; M_{low RS} = 2.62, SD_{low RS} = 1.01, t(66) = −0.14, p ≥ 0.05). Thus, H3b was supported (see Table 4 for findings).

Furthermore, in line with H4a, low-relationship complainants had a lower revenge intention when they received accommodative webcare (M = 2.62, SD = 1.01), as compared to no/defensive webcare (M = 3.25, SD = 1.09, t(164) = −3.48, p < 0.001). The same held for high-relationship complainants (M_{acc. webcare} = 2.58, SD_{acc. webcare} = 1.07; M_{no/def. webcare} = 3.30, SD_{no/def. webcare} = 1.05, t(71) = −2.57, p < 0.05). Thus, H4b was supported.

8.3.2 Hypotheses testing: multiple inter-failures

To assess H5–H7, we used an ANCOVA with the same variables as in our earlier analyses and applied bootstrapping (Hayes, 2018). The findings indicated a significant effect of relationship strength (F (1, 243) = 20.11, p < 0.001, partial eta² = 0.08) and a weakly significant effect of webcare (F (1, 243) = 3.53, p = 0.062, partial eta² = 0.01) on revenge intention. The interaction effect between relationship strength and webcare was not significant ((F (1, 243) = 0.86, p = 0.354, partial eta² = 0.00). Follow-up independent-samples t-tests revealed that, in the case of multiple inter-failures and no/defensive webcare, high-relationship complainants (M = 3.96, SD = 0.76) had a higher revenge intention than low-relationship complainants (M = 3.56, SD = 0.80, t(141) = 3.01, p < 0.01). Thus, H5a was supported. The same held when complainants received accommodative webcare (M_{high RS} = 4.31, SD_{high RS} = 0.62; M_{low RS} = 3.71, SD_{low RS} = 0.92, t(108) = 3.40, p < 0.001). Consequently, H5b was supported.

As predicted in H6, low-relationship complainants had comparable revenge intentions when they received accommodative webcare (M = 3.71, SD = 0.92), as compared to no/defensive webcare (M = 3.56, SD = 0.80, t(70) = 0.62, p ≥ 0.05). In contrast, high-relationship complainants had a higher revenge intention when they received accommodative webcare (M = 4.31, SD = 0.62), as compared to no/defensive webcare (M = 3.96, SD = 0.76, t(179) = 3.36, p < 0.001). Therefore, H7 was supported. Table 5 summarizes the results for H5–H7, while Figure 2 illustrates the transition from the “love is blind” effect to the “love becomes hate” effect for revenge intentions and the varying effectiveness of webcare.

8.3.3 Explaining the transition from “love is blind” to “love becomes hate”

To test our prediction about the changing effect of relationship strength on attribution of stability for no inter-failures versus multiple inter-failures, we ran Model 6 of the SPSS PROCESS macro (version 3.0; Hayes, 2018). For analyses, we used 5,000 bootstrap samples to estimate the 95% bias-corrected bootstrap confidence intervals to arrive at conclusions about the effects of relationship strength. We evaluated a model with relationship strength as the independent variable, attribution of stability and anger were the two sequential mediators, and revenge intention was the dependent variable. We ran the model for both no inter-failures and multiple inter-failures.

The results for no inter-failures showed a significant negative effect of relationship strength on complainants' attribution of stability (b = −0.75, p < 0.001). Furthermore, we identified a positive effect of attribution of stability on anger (b = 0.34, p < 0.001) and a positive effect of anger on revenge intention (b = 0.55, p < 0.001). The indirect effect of relationship strength on revenge intention via complainants' attribution of stability and anger was negative and significant (b = −0.14, p < 0.05). Its direct effect (b = −0.44, p < 0.001) was also negative, as was the total effect (b = −0.68, p < 0.001). These findings show that, in the case of no inter-failures, high-relationship complainants are more forgiving and less revengeful than low-relationship complainants. This corresponds to the “love is blind” effect.

The results for multiple inter-failures show a different picture: Here, relationship strength had a significant and positive effect on attribution of stability (b = 0.67, p < 0.001). The latter triggered anger (b = 0.27, p < 0.001), which, in turn, positively affected revenge intention (b = 0.11, p < 0.05). The indirect effect of relationship strength on revenge intention was positive and significant (b = 0.02, p < 0.05). Its direct effect (b = 0.25, p < 0.05) was positive, as was its total effect (b = 0.54, p < 0.001). These results indicate that, when facing multiple inter-failures, high-relationship complainants are less forgiving and take more revenge than low-relationship complainants. This is consistent with the “love becomes hate” effect (Figure 3 summarizes the results). Therefore, H8 was supported.

8.4 Discussion

While study 1 identified that the “love is blind effect” occurs when complainants have not experienced inter-failures with the same brand, with study 2 we provide evidence that this effect vanishes when few inter-failures have occurred. More specifically, we show that, with few inter-failures, the revenge levels of low-relationship and high-relationship complainants assimilate. Most notably, study 2 pioneers in demonstrating that the stronger retaliatory reactions of high-relationship complainants – typically denoted in literature as the “love becomes hate” effect (i.e. committed complainants' revenge level is above the level of uncommitted complainants; Grégoire et al., 2009) – occur when the failure history includes multiple inter-failures with the company.

Furthermore, in the case of few inter-failures, accommodative webcare is still more effective than no/defensive webcare in mitigating negative reactions of both groups of complainants. In the case of multiple inter-failures, the two webcare types are equally ineffective for low-relationship complainants. However, accommodative webcare provided to high-relationship complainants boomerangs on the brand by increasing post-webcare revenge intention. This suggests that well-meant webcare is here interpreted by recipients as the company's confession of guilt and evidence of responsibility, which triggers aggressive consumer behavior.

9. General discussion

Companies put a lot of time and effort into building and maintaining strong relationships with their customers. Prior research (e.g. Batra et al., 2012; Brown et al., 2005) consistently shows that these endeavors lead to positive customer reactions (e.g. purchases, positive word-of-mouth) in “good times”, when customers' expectations are met. However, literature is inconsistent concerning the reactions of high-relationship consumers when a service failure (e.g. unfriendly staff, malfunctioning product) occurs, showing that they can react either more negatively (Grégoire and Fisher, 2006) or more positively (Mattila, 2004) than uncommitted customers.

In this research, we provide consistent evidence that the favorability of reactions is conditional on complainants' “inter-failures”: These are the perceived number of earlier independent service failures that a customer has experienced with the same brand involved in the current service failure. Specifically, we show that in the case of no inter-failures, high-relationship complainants have a lower revenge level than low-relationship complaints, which is denoted as the “love is blind” effect (e.g. Grégoire and Fisher, 2006). However, this positive effect vanishes when the perceived number of inter-failures increases. When high-relationship complainants have experienced few failures with the same brand in the past, their revenge level aligns with their low-relationship counterparts – and clearly exceeds it after having faced multiple inter-failures. This is referred to as the “love becomes hate” effect (Grégoire et al., 2009). Hence, our work shows for the first time the diminishing value of strong customer–brand commitment, which explains the transition from one to the other effect due to a simple mechanism. We explain this process by showing that the attribution of stability and anger are important mediators between relationship strength and revenge intention.

Another critical finding of this research is that accommodative webcare (i.e. when the company takes full responsibility for a failure) can lead to lower revenge intensity among committed and uncommitted online complainants. However, the perceived number of inter-failures is the boundary condition for this effect: Only if complainants have experienced no or few inter-failures, accommodative webcare is superior to no/defensive webcare in reducing consumer revenge. However, when high-relationship complainants have been exposed to multiple inter-failures, accommodative webcare leads to higher revenge than no/defensive webcare. For their uncommitted counterpart, webcare becomes ineffective in the case of multiple inter-failures (Table 6 summarizes the research findings). With these findings, this research furnishes academics and practitioners with several critical implications.

9.1 Theoretical contributions

First, our research contributes to the literature on customer-company relationships following service failures. Prior research focused on either the demonstration of the “love is blind” effect (i.e. a situation when committed consumers have a lower revenge level than uncommitted consumers; Mattila, 2004) or the “love becomes hate” effect (i.e. a situation when committed consumers have a higher revenge level than uncommitted consumers; Grégoire et al., 2009). In contrast to these valuable contributions, this research explains the rise and the transition of consumer love into hate by considering a simple, but prevailing mechanism. Specifically, earlier research has neglected to show that consumers' experiences of the number of earlier independent service failures with the same brand (i.e. inter-failures) may have a pronounced impact on online complainants. This influence is present even when keeping intra-failures (i.e. service failures that are followed by recovery failures such as “double deviations”) constant, which are a typical circumstance of online complaint handling (Weitzl, 2019). Earlier research has largely focused on the role of double deviations and its variations for complaint-handling (e.g. Bitner et al., 1990). However, the role of the perceived number of earlier independent failures (i.e. “inter-failures”) for the love-to-hate transition process has remained widely ignored (see Table 1). One notable exception is the contribution of Weitzl et al. (2018). Here, the authors demonstrate that online complainants who have experienced multiple inter-failures make more unfavorable brand-related failure attributions following accommodative webcare as compared to individuals experiencing no inter-failures. However, the possible impact of complainant-brand relationships on revenge taking remained neglected.

Second, while accommodative webcare is typically considered the most appropriate response strategy to mitigate negative consumer reactions (e.g. revenge-taking) (e.g. van Noort and Willemsen, 2012; Weitzl and Hutzinger, 2017), we identify complainants' inter-failures as an important boundary condition. Only if complainants have experienced no or a few inter-failures, accommodative webcare can realize its potential to more effectively reduce revenge intentions than no/defensive webcare. However, when inter-failures continue to increase, accommodative webcare becomes ineffective for uncommitted complainants. These complainants' reactions can be explained by vindictive motives aimed at harming the brand (Weitzl, 2019). These complainants do not strive for redress and solving the service problem (Istanbulluoglu, 2017), instead they try to bring down the involved brand by sharing negative sentiments with other consumers – a form of revenge.

Third, this research demonstrates that accommodative webcare can also boomerang and cause significant reputational damage for the brand involved in a specific situation: This counter-intuitive pattern arises when companies respond to online complaints of high-relationship complainants with a well-meant but negatively interpreted accommodative comment. We thus contribute to the evolving literature on company-harming effects of online recoveries (e.g. Weitzl and Hutzinger, 2019). However, earlier research has never investigated online complainants' rationale that, when they have experienced multiple inter-failures, they regard accommodative responses as evidence that the brand is responsible for disturbing the relational bonds. These complainants are convinced that active retaliation by means of negative word-of-mouth is justified here. In an additional analysis, we also examined complainants' avoidance intention (4 items adapted from McCullough et al. (1998) and Zeithaml et al. (1996), e.g. “I will avoid the company.”; α = 0.87). Our additional results for high-relationship complainants that had faced multiple inter-failures in the past revealed that accommodative webcare did not elicit a greater avoidance intention (M = 4.23, SD = 0.81) than no/defensive webcare (M = 4.04, SD = 0.73, p ≥ 0.05). According to the literature, revenge intention is associated with expressions of aggressiveness like punishments directed at companies, while avoidance intention is more passive and leads to escape (McCullough et al., 1998). Based on prior research (Grégoire et al., 2009), we argue that complainants' post-webcare revenge intention is a short-term inclination fueled by immediate negative emotions (i.e. anger) aimed at venting these negative feelings. After a serious problem, high-relationship consumers can experience extreme levels of anger (Grégoire and Fisher, 2008), which triggers immediate retaliation (i.e. negative word-of-mouth) that can be short-dated. In contrast, consumer avoidance of a specific firm is a more strategic decision that is not seriously affected by a single experience (e.g. an accommodative response that is processed unfavorably), but is a result of various strategic considerations (e.g. perceived switching barriers). Therefore, a single accommodative response is not enough to alter committed complainants' exit desires either positively or negatively when they have experienced multiple independent failures before. These insights are consistent with prospect theory (Kahneman and Tversky, 1979) and enable us to estimate the extent of the possibly detrimental effects of accommodative webcare in common online complaining situations which had remained yet uninvestigated (see Table 6).

9.2 Managerial implications

Most marketers are aware that their social media channels have become a vital platform for dissatisfied customers to voice their discontent about a failure. Nevertheless, they should also realize that the way they respond to online complaints is crucially important in attenuating further negative reactions (e.g. revenge). This research suggests that, when selecting their webcare strategy, marketers are well advised to consider both the complainant type (i.e. the level of relationship strength) and the individual prior failure history (i.e. the perceived number of service failures that complainants have experienced with the company in the past). Segmenting complainants along these two dimensions enables marketers to increase the effectiveness of their webcare. For example: Accommodative webcare that includes both social benefits (e.g. sincere apology, detailed explanation) and economic benefits (e.g. vouchers, gifts) is resource-intensive and should thus only be provided to receptive complainants. This study indicates that customers who have experienced either no or few failures before the focal incident belong to this group. Here, taking responsibility for the failure is more beneficial than whitewashing the failure by defending the company against accusations (e.g. by blaming a third party) or remaining silent. Such accommodative webcare is equally effective for committed and uncommitted complainants, but the revenge level depends on their relationship strength. Complainants have a certain zone of tolerance in which they value accommodative webcare. When they have experienced multiple prior failures with the same brand, well-meant webcare becomes ineffective for low-relationship complainants and can even cause severe damage when provided to high-relationship complainants. Specifically, accommodative webcare here intensifies the willingness of former brand lovers to spread high levels of negative word-of-mouth to others. Therefore, identifying the right webcare audience is key. A decision tree helping service managers to better handle online complaints is provided in Figure 4.

Besides segmenting online complainants on their failure history and relationship strength, management can influence these critical aspects. Obviously, perceived failure history length can be altered by reducing the number of actual service failures through total quality management (Spencer, 1994). However, besides expensively reducing the number of actual failures (which can be collected in CRM systems), marketers can particularly benefit from a reduction of perceived service failures. Earlier research shows that consumers who feel strong, self-relevant negative emotions may try to forget the underlying negative events (e.g. failures; Yi and Baumgartner, 2004). We have learned from research that individuals apply different strategies to cope with unpleasant situations, including mental disengagement (i.e. doing other things to distract the mind from the problem), distancing (i.e. refusing to think about the problem) and denial (i.e. refusing to believe that a problem has happened) (e.g. Sengupta et al., 2015). Individuals can manage their negative feelings, for instance, by applying denial when experiencing heightened emotional distress, such as when facing negative feedback or some kind of self-relevant unflattering information that challenges the person's positive self-concept (Strizhakova et al., 2012). Service failures particularly challenge the self-concept of consumers with a strong consumer-brand identification (CBI), which triggers a very inconvenient emotional situation that can be resolved by denial of the problem. Hence, marketers can benefit from building denial-inducing CBI (but only until negative experiences begin to prevail). It is likely that in this “zone of tolerance”, committed consumers are inclined to forget minor service problems.

On the other hand, marketers can strive to reduce the relational ties with already committed customers by minimizing their relationship marketing efforts when the perception of recurring problems rises. Here, the company would benefit from a significant decline in online complainants' revenge level. However, such a radical strategy may only help to mitigate their short-term revenge intentions and may – on the other hand – induce other detrimental consumer reactions (e.g. decreasing brand loyalty, lower repurchases) at the same time.

9.3 Limitations and future research

This work should be interpreted in the light of some limitations. First, while we consider several covariates in the analyses, other factors might affect respondents' reactions. For instance, past research has shown that personal characteristics such as propensity to complain (Davidow and Dacin, 1997) have an impact on revenge-taking. The same holds for contextual factors, such as the number of people affected by the failure (Zhou et al., 2013). Therefore, future research should introduce additional control variables to evaluate the stability of the identified effects. Second, while this research provides important insights, future research should cross-validate our findings with additional experiments and independent samples. Finally, while this research investigates marketer-initiated webcare (Weitzl and Hutzinger, 2017), the responses of other consumers witnessing the online complaint have been shown to also affect complainants' reactions (Schaefers and Schamari, 2016). Therefore, further work that isolates the effects of these reactions in various situations (e.g. after multiple inter-failures) is needed.

Figures

Figure 1

Conceptual framework of revenge-inducing factors

Figure 2

From the “love is blind” effect to the “love becomes hate” effect: Revenge intentions following no, few and multiple inter-failures

Figure 3

Mediation model: Effect of relationship strength on revenge intention via attribution of stability and anger

Figure 4

Webcare choice to reduce online complainants' revenge intention based on number of inter-failures and relationship strength

Table 1

Contribution table of the empirical research on inter-failures

Study	Inter-failures	Prior customer-company relationship	Webcare type	Takeaway
Betts et al. (2011)	Single inter-failure	Not considered	General: company control of the service failure	Post-recovery satisfaction and negative word-of-mouth do not differ between a service failure without a prior failure and with a prior failure. High company control leads to lower post-recovery satisfaction and higher negative word-of-mouth than low company control. However, these effects are not moderated by prior failure
Chuang et al. (2012)	Indirect: prior negative experience with the service	Not considered	Specific: psychological; tangible	In case of a negative prior experience, a current outcome failure leads to higher satisfaction with tangible recovery as compared to psychological recovery, while the opposite is true for a current process failure. In case of a positive prior experience, satisfaction with recovery is independent of type of failure and type of recovery
Magnini et al. (2007)	Single inter-failure	Not considered	General: recovery excellence	Given excellent webcare, there is a greater chance that a customer's post-failure satisfaction exceeds pre-failure satisfaction, if it is the company's first – as compared to second – service failure with the customer
Maxham and Netemeyer (2002)	Single inter-failure	Not considered	General: recovery satisfaction	Customers reporting two service failures have higher failure severity, higher blame attribution towards the company and higher recovery expectations for the second failure than for the first failure. Customers reporting two service failures and perceiving either two satisfactory or two unsatisfactory service recoveries, have lower overall satisfaction with the firm, repurchase intent, and favorable WOM after the recovery of the second failure than after the second failure
Tax et al. (1998)	Indirect: prior negative experience with the company	Not considered	General: recovery satisfaction	The positive effect satisfaction with current service failure handling has on customers' commitment towards the company is reduced the more negative the prior experience with the company is
Watson (2012)	Single inter-failure	Not considered	Specific: assistance; compensation; assistance and compensation	One – as compared to no – prior service failure leads to lower customer satisfaction, lower customer loyalty and higher complaint behavior when confronted with a current service failure. However, these associations are not moderated by webcare type
Weitzl et al. (2018)	Few inter-failures vs Multiple inter-failures	Not considered	Specific: no; defensive; accommodative	Complainants that experienced multiple – as compared to few – prior service failures with the company involved in the current service failure, perceive higher attributions of controllability, stability and locus when they receive an accommodative recovery
Wu and Lo (2012)	Single inter-failure	Not considered	Specific: defensive	After the second service failure, customers have comparably high expectations towards the company than after the first service failure. After the second service failure, customers have lower negative emotions than after the first service failure
This research	No inter-failures vs Few inter-failures vs Multiple inter-failures	Relationship strength (commitment)	Specific: no; defensive; accommodative	Depending on inter-failures, relationship strength has a macro effect on revenge level and accommodative webcare has a micro effect on revenge intensity (see Table 6)

Note(s): We only included studies that have empirically examined inter-failures in Table 1. Therefore, studies that focused on firm failures (e.g. firm bankruptcy) (e.g. Jenkins et al., 2014), which examined one prior service failure – but no current service failure (e.g. Allen et al., 2015), which included inter-failures as covariate – but did not report any respective results (e.g. Smith et al., 1999) or which have been declared as inter-failure studies – but are in fact intra-failure studies (e.g. Bunker and Bradley, 2007) – were not included in Table 1

Table 2

Overview of the proposed effects

Table 3

No inter-failures: effect of relationship strength and webcare on revenge

Dependent variable: Revenge intention
		Webcare
		No/defensive	Accommodative	Δ
Relationship strength	High	3.40	2.45	−0.94***
	Low	4.04	3.11	−0.93***
	∆	0.64***	0.65**

Note(s): Table shows group means. **p < 0.01, ***p < 0.001

Table 4

Few inter-failures: effect of relationship strength and webcare on revenge

Dependent variable: Revenge intention
		Webcare
		No/defensive	Accommodative	∆
Relationship strength	High	3.30	2.58	−0.72*
	Low	3.25	2.62	−0.63***
	∆	−0.05n.s	0.04n.s

Note(s): Table shows group means. n.s. = not significant (p ≥ 0.05), *p < 0.05, ***p < 0.001

Table 5

Multiple inter-failures: Effect of relationship strength and webcare on revenge

Dependent variable: Revenge intention
		Webcare
		No/defensive	Accommodative	∆
Relationship strength	High	3.96	4.31	0.35***
	Low	3.56	3.71	0.14n.s
	∆	−0.40**	−0.60***

Note(s): Table shows group means. n.s. = not significant (p ≥ 0.05), **p < 0.01, ***p < 0.001

Table 6

Summary of research findings

Notes

1. An intra-failure describes a situation when a service failure is followed by a single (i.e. a “double deviation”) or several unsuccessful recovery attempt(s) (Bitner et al., 1990). This failure type tends to increase consumers' discontent (Joireman et al., 2013) and it is the norm that many online complainants have experienced them (Weitzl, 2019). In academic research, intra-failures have been extensively studied (e.g. Grégoire et al., 2010; Pacheco et al., 2019).

2. We calculated the required sample size, given that we expect a medium effect size f of 0.25, in order to achieve a high level of power, i.e. 0.80 (Cohen, 1992). The results of G*Power (version 3.1.9.7; Faul et al., 2007) revealed that the required sample size is 179, which is clearly exceeded by the actual sample size in our study. Thus, our study has sufficient statistical power.

3. The additional analyses revealed no significant differences between the two groups for gender (X²(1, N = 210) = 0.21, p = 0.645), age (t(208) = 0.25, p = 0.803), dissatisfaction (t(208) = 0.92, p = 0.360), distributive justice (t(208) = 1.04, p = 0.299), failure severity (t(208) = 1.95, p = 0.053), attribution of blame (t(208) = 0.45, p = 0.652), attribution of controllability (t(208) = 1.06, p = 0.289) and attribution of stability (t(208) = 0.33, p = 0.746).

4. Details on the descriptive characteristics, correlations and psychometric properties of the measures used in this research's studies are provided in Web Appendix 1.

5. Again, as explicated in study 1, our actual sample sizes for few and multiple inter-failures both provide sufficient statistical power.

6. Complainants who have received no and defensive webcare were collapsed into one group. Consistent with study 1, a series of analyses confirmed that the two groups did not significantly differ in various characteristics.

7. Again, details on the descriptive characteristics and psychometric properties of the measures used in study 2 are provided in Web Appendix 1.

Appendix

The supplementary material for this article can be found online.