The concept of service recovery was first proposed by Etzel and Silverman (1981) and refers to the measures taken by service providers to compensate for the loss of consumers after the occurrence of service errors in response to negative evaluations by the consumers to change consumers’ attitudes toward enterprises (Groonros, 1988). Service recovery is a management process. Companies need to know that service failures have occurred, identify the reasons for such losses, evaluate their impact, and carry out appropriate management activities to resolve these failures (Tax et al., 1998). Based on this perspective, service recovery is defined as the actions that organizations take in response to a service failure (Steyn et al., 2011) or the process of addressing mistakes (Hu et al., 2013). Implementing effective service recovery after service failures does not necessarily lead to negative results (Hu et al., 2013). Particularly in the restaurant business, service failure is difficult to avoid (Namkung and Jang, 2010; Byun and Jang, 2019), emphasizing the importance of practical service recovery efforts in a restaurant setting. Service failures are effectively managed in a restaurant setting also affects customers’ behavioral intentions, including word-of-mouth (WOM) intention and revisit intention (Ok et al., 2006). During the recovery process, customers usually expect fairness to be involved in compensation for the loss during the service failure. Many researchers have indicated that organizations can use several strategies to recover from service failures, including communicating with customers to provide feedback, offering explanations for their losses (Boshoff and Staude, 2003; La Khanh, 2004). They also apologize for their losses (Mattila and Cranage, 2005; Mostert et al., 2009). Thus, service failures have a positive relationship with service recovery.

This manuscript provides insight into the dynamics of relationship repair by accounting for the impact of time-specific factors on the relationship repair process. Thus, we propose the revised service recovery cycle model (Figure 1), which includes the following three primary stages of service recovery and their attributes: (1) initiation at the early recovery stage; (2) response speed at the immediate recovery stage; and (3) compensation at the follow-up recovery stage.

First, service recovery encompasses a much broader set of activities than addressing complaints because it includes situations in which a service failure occurs the customer lodges but no complaint. Evidence shows that a majority of dissatisfied customers do not bother to complain (Harari, 1992). Prior research has focused solely on failure/recovery situations in which customers have filed a formal complaint with an organization (Tax et al., 1998). Several researchers have suggested that proactive recovery efforts enhance customers’ evaluations of the service provider (Berry, 1995). When an organization initiates recovery during the early stage, the customer is likely to view the proactive effort as an act of courtesy, a demonstration of honesty and forthrightness, and a show of empathic understanding and respect.

Second, we argue that recovery speed is a critical factor during the immediate recovery stage. Recovery speed refers to service providers’ promptness in responding to service failure, which influences consumer satisfaction (Crisafulli and Singh, 2017). Evidence suggests that longer recovery time would lead to additional negative consequences, such as negative WOM and lower levels of recovery satisfaction (Hogreve et al., 2017; Babin et al., 2021). Service providers can reduce these negative impacts by quickly responding to consumers’ complaints and/or negative reviews, and conducting service remediation.

Lastly, consumers perceive inequity following a service failure, and they might be appeased when provided with suitable compensation during the follow-up stage. According to social exchange theory, compensation (e.g., discounts, free merchandise, refunds, and coupons) by an organization could balance the costs and benefits of achieving an equitable exchange in relationship recovery. Walster et al. (1973) showed that compensation is used to restore equity in an exchange relationship when another party has harmed one party. Tax et al. (1998) performed a content analysis of qualitative evaluations of service complaint experiences and showed that compensation is the most essential recovery dimension. Therefore, compensation plays a vital role during the follow-up recovery stage after service failures.

It is essential to differentiate between service recovery and service failure. Service failure is defined as a “situation where a service provider does not meet customer expectations in terms of its service products or engages in service behaviors that customers evaluate as unsatisfactory” (Harrison-Walker, 2012). Service failures can be classified into the following three types: (1) core service failures, such as failures to fulfill basic service needs (Yang and Mattila, 2012); (2) interactional service failures, including the attitudes and behaviors of employees during face-to-face interactions with customers, such as a server treating a customer impassively or impolitely (Yang and Mattila, 2012); and (3) process service failures, which involve how the core service is delivered to the customer, such as slow service or incorrect delivery order (Mohr and Bitner, 1995). Service failure can result in dissatisfied customers and negative WOM. Thus, recovery efforts are critically needed in service failure situations, and service recovery is among the critical antecedents of customer satisfaction and loyalty (Craighead et al., 2004).

The service recovery paradox (SRP) has emerged in the marketing literature as an essential effect of service failures. The SRP is defined as a situation in which post-recovery satisfaction is greater than the satisfaction before the service failure when customers experience high recovery performance (Smith and Bolton, 1998; Maxham, 2001). Effective service recovery may lead to a higher level of satisfaction than the periods when the service was incorrectly performed; recovery encounters offer an opportunity for service providers to increase customer retention (Hart et al., 1990). Based on the disconfirmation framework (McCollough et al., 2000; Oliver, 2010), the SRP is related to secondary satisfaction following a service failure in which customers compare their expectations for recovery their perceptions of the service recovery performance. If positive disconfirmation occurs, i.e., if consumers’ perceptions of the service recovery performance are greater than their expectations, a paradox might emerge (secondary satisfaction becomes more substantial than their pre-failure satisfaction). In contrast, there is a double negative effect in the case of negative disconfirmation as the service failure is followed by a flawed recovery (Smith and Bolton, 1998; McCollough et al., 2000). When customer satisfaction is negatively affected by a service failure, subsequent service recovery reactions may include negative WOM behavior (Hocutt et al., 2006). Positive evaluations occur when the recovery is understood as satisfactory (Matos et al., 2009). Satisfaction with service recovery is defined as positive customer evaluations of the service recovery experience (Bambauer-Sachse and Rabeson, 2015). The degree of success may depend on the type of service involved, the type of failure that occurred, and the type of recovery (Komunda and Osarenkhoe, 2012).

Over the past 20 years, there has been growing interest in the concept of relationship marketing among practitioners and academics (Morgan and Hunt, 1994; Sheth and Parvatiyar, 2000, 2002; Gummesson, 2002). Establishing a long-term relationship with customers generates an outstanding level of customer satisfaction, which, in turn, helps companies gain customers’ trust and loyalty and thus benefits the company overall (Valenzuela et al., 2010). Relationships help participants meet their objectives and depending on the stage of the connection (i.e., beginning stage, cultivating stage, or enhancing stage), the strategic implications might vary. As a result, firms must pay attention to different aspects at different stages, ranging from creating customer knowledge to shaping their perceptions. The relationship between an organization and its customers is strengthened by many marketing actions (Berry, 1983), which leads to further customer retention. Relationship marketing embodies tactical and strategic elements that can positively impact a firm (Lo and Campos, 2018). The implementation of relationship marketing endeavors to enhance the value of an enterprise to its customers and their long-term relationship (Bruhn, 2003). Relationship marketing further explores the marketing concept by focusing on the customer as an important representative figure and examining a promising way to acquire competitive advantages by exchanging information and becoming closer to the customer. This bond is genuinely advantageous to both parties. It allows both buyers and sellers to commit to achieving long-term benefits that offer greater chances for a successful relationship (Ganesan, 1994). Relationship marketing can be a challenging concept to implement, and the development of close bonds with customers is not always possible or alluring (Webster, 1992). Due to advances on the Internet and technology, sellers can utilize technology to process large amounts of data, derive specific information, and gain insight into customer preferences and behaviors. Therefore, companies can design more suitable solutions and products to meet their customers’ needs.

Relationship marketing refers to a customer’s perception of the extent to which the relationship fulfills his/her expectations, predictions, goals, and desires regarding the overall relationship (Wong and Sohal, 2002). A high degree of relationship marketing indicates that customers can rely on the service provider’s integrity and develop confidence in the service provider’s future performance because the past performance level has been consistently satisfactory (Tseng and Seidman, 2007). Kim et al. (2006) indicated that customers’ level of relationship marketing depends on various elements rather than only on the interpersonal relationship between service providers and their customers. Commitment, trust, and relationship satisfaction were considered as focal dimensions of relationship quality (Ulaga and Eggert, 2006; Verma et al., 2016). This study proposes that relationship marketing is composed of satisfaction and trust, while commitment belongs to consumer loyalty. Loyalty refers to a commitment to repurchase a product and/or service, and can be reflected by purchasing behaviors (e.g., frequency, intensity, and proposition) and WOM recommendations (Cossío-Silva et al., 2016).

Therefore, although service failures occasionally occur, service recovery can substantially impact relationship marketing because, through such marketing, customers gain confidence that long-term service providers will provide benefits. In contrast, when a customer is not satisfied with the service recovery measures taken by the enterprise. The relationship between the customer and the enterprise could be terminated (Zeithaml, 2002). Service recovery belongs to quality management, and its goal is to maintain a good relationship between enterprises and customers (Schweikhart et al., 1993). Good service recovery can improve customer satisfaction and promote customer trust in businesses, and compensatory service recovery can improve the degree of customer satisfaction (Yan and Jia, 2003; Wen, 2004).

Many scholars have shown that relationship marketing plays an intermediary role in the relationship between service recovery and consumer loyalty; thus, service recovery affects relationship marketing and consumer loyalty (Huang, 2006). Crosby et al. (1990) studied the relationship between service recovery and consumer loyalty in the insurance industry and confirmed the intermediary effect of relationship marketing (Chen and Fu, 2015). Huang (2006) used the banking industry as the investigation background and found that relationship marketing between customers and banks is affected by the perception of service recovery and plays an intermediary role in the relationship between service recovery and consumer loyalty. Huang (2006) also investigated the relationship between audio-visual rental customers and companies and further confirmed the intermediary role of relationship marketing. Therefore, based on the above analysis, this paper proposes the following hypotheses.

Consumer loyalty is a desirable marketing indicator, and has been associated with positive organizational outcomes, such as sustainable competition (Kandampully et al., 2015; Huang et al., 2017; Rather et al., 2018; Affran et al., 2019; Rather and Hollebeek, 2019). Thus, a good service provider believes that a growth process usually involves customers, i.e., attracting new customers, retaining existing customers, and motivating customers to spend more and recommend products and services to other people. When customers experience a higher degree of relationship marketing with enterprises or service personnel, customer identification increases, customers are easily satisfied with the services provided, and customers’ willingness to find another service provider decreases (Crosby et al., 1990; Morgan and Hunt, 1994; Ruiz et al., 2008). Liu (2017) investigated the offline retail industry and found that adequately maintaining customer relationships promptly helps retail stores improve their customers’ sense of identity and cultivates dependence on the business. Rahman and Ramli (2016) conducted research that investigated the customers of some banking systems in northern Malaysia and verified the role of customer relationship marketing in the formation of consumer loyalty. In the Internet context, relationship marketing positively impacts consumers’ repeated purchase behavior (Jin et al., 2008; Zheng, 2008).

Relationship marketing includes customer satisfaction and customer trust, which have been shown to play a significant role in promoting consumer loyalty (Zhao, 2005). Specifically, Mayer et al. (2007) defined trust as the vulnerability of one party to the actions of another party based on expectations that the other party is performing in the desired way. The development of trust is considered a critical result of establishing long-term successful relationships among all parties involved. Moreover, many researchers have suggested that customers’ trust plays a significant role in building long-term relationships and achieving consumer loyalty (Feng and Zhang, 2009). For instance, research has shown that when customers consistently receive competent service, their trust levels increase, which results in their maintenance of long-term relationships with the firm (Balaji, 2015).

Customer satisfaction is customers’ overall attitudes toward a product, service or experience after their purchase (Leninkumar, 2017). Existing studies suggest that satisfaction positively affected consumer loyalty (Ruyter and Wetzels, 2000; Deng et al., 2010). Although satisfaction may not always be a reason that customers remain loyal (Gerpott et al., 2001), satisfied customers are more loyal than unsatisfied ones. This study thus proposes the following hypothesis:

The evidence suggests that appropriate service remedies can reduce or even eliminate consumers’ complaints and promote consumers’ trust in enterprises, a common marketing strategy of enterprises (Hart et al., 1990). Like offline service recovery, online service recovery includes responsiveness to customer complaints, resolving the communication process, and economic compensation for service losses (Parasuraman et al., 2005). Online companies are increasingly focused on reducing service failures, enhancing online service quality, and increasing customer satisfaction to attract more customers to make purchases and earn more profit (Zhao et al., 2014). Kelley et al. (1993) identified the top seven recovery strategies used to retain customers: discounts, correction, management/employee intervention, correction plus, replacement, apology, and refund. Smith et al. (1999) concluded that customers prefer recovery in both the value and form of recovery that matches the failure they experienced. Based on research investigating service failure and recovery (Smith et al., 1999; Mattila and Cranage, 2005; Weber and Sparks, 2009), apology and compensation are two key strategies used in service recovery. Moreover, the four attributes of perceived justice proposed by the extant research are compensation, response speed, apology, and recovery initiation (Chou, 2015). Based on research conducted by Smith et al. (1999), service recovery includes the four dimensions of compensation, response speed, apology, and initiation. Compensation, response speed, and apology are often in the business press, are particularly salient to customers, can easily be acted on by managers, and can be manipulated through written scenarios in an experimental context (Hart et al., 1990). Recovery initiation has received much attention in the business literature but has not been addressed empirically. We also expect interaction effects between the failure context and recovery attributes as subsequently described. According to the context of online takeout services, we selected compensation, response speed, and initiation as the three attributes of service recovery. Service recovery denotes a series of activities performed by companies to respond to a customer complaint regarding a service failure (Zhao et al., 2014). These strategies and activities aim to remedy problems with services and products (Kelley et al., 1993).

The most common definition of consumer loyalty was provided by Oliver (2010), who states that such loyalty occurs consistently in the future, thereby causing repetitive same-brand or same-brand-set purchasing despite any situational factors. Various authors have found that an increase in consumer loyalty increases profits, reduces the costs of acquiring new customers, and decreases the costs of serving current customers (Reichheld and Sasser, 1990). Increasing competition, particularly in the service industry, has caused firms to become very concerned with attracting potential customers and maintaining long-term relationships with their current customers, which is the aim of consumer loyalty (Bojei and Alwie, 2010). The measurement of consumer loyalty includes the following two aspects: attitudinal loyalty and loyalty in behavioral intention (Su et al., 2010; Zhang et al., 2010). Attitudinal loyalty is reflected in the willingness of consumers to recommend a service provider to other consumers or the commitment to re-patronize a preferred service provider (Chou, 2015). Behavioral intention is reflected in the frequency with which a customer chooses the same product or service compared to the total frequency with which this specific product or service is consumed (Chou, 2015).

In the online service context, the quality-of-service remedy has a positive impact on loyalty as expressed through customer behavioral intention. Etzel and Silverman (1981) show that service remedy positively affects loyalty through both customer attitude and behavioral intention. Further studies have shown that the response speed of service remediation, tangible compensation, and the initiation of remediation can promote the occurrence of repurchasing behavior and good WOM publicity intentions after the occurrence of service failures (Wang, 2014).

Service recovery is positively related to consumer loyalty, while service failure negatively relates to customer loyalty (Wang et al., 2011; Balaji et al., 2017). When loyal customers face service failure, they are more likely to re-evaluate their experience and service quality and then modify their decisions when competition is high (Fox et al., 2018). Therefore, more service providers have used service recovery as a means to improve customer retention (Vázquez-Casielles et al., 2012). Buttle and Burton (2002) asserted that most customers whose problems are resolved would repurchase, when organizations use an appropriate service recovery strategy. Recovery strategies thus aim to offset the dissatisfaction caused by service failure and reinforce positive WOM (Spreng et al., 1995). For example, travelers would recommend their airlines to others with service recovery efforts and/or incentives (Steyn et al., 2011). Additional evidence indicates that well-executed service recovery efforts can enhance customer satisfaction and loyalty (Buttle and Burton, 2002; Mostert et al., 2009; Steyn et al., 2011; Hu et al., 2013). Effective service recovery can increase customers’ trust, enhance consumer loyalty and increase customers’ willingness to repurchase in the future (Hu et al., 2013). Moreover, suppose merchants take effective service remedy measures. In that case, customers’ repurchase behavior increases, and customer satisfaction is significantly affected by the types of service remedy measures (oral apology, economic compensation, initiation, and response speed) (Zhang, 2013).

Based on the above reasoning, this study proposes the following hypotheses.

The revised service recovery cycle model elaborated in Figure 1, this study proposes the conceptual stage model of service recovery (Figure 2), which not only includes the three stages of service recovery and their recovery attributes in these stages. Notably, it examines how these recovery attributes influence relationship quality and customer loyalty.

The online survey was launched in March 2019. Our sample was purchased from a contracted marketing agency, who emailed our survey link to their participants. The participants in our survey were individuals who (1) had purchased via online food delivery apps, and (2) had reviewed and commented on service providers in the past 3 months (Table 1). When participants clicked on the survey link, they first accessed a page that described the study purpose, confidentiality, and privacy protocols. According to our contract with the marketing agency, we eliminated nine incomplete questionnaires and led to our final sample size of 301. Our response rate was 97.1%.

All items in the online survey instrument used the five-point Likert scale measurement method (1 = strongly disagree to 5 = strongly agree). First, we modified and developed the service recovery scale based on Mattila (2001); Mattila and Cranage (2005), Fan and Du (2006), and Ma et al. (2009), including nine items in the three attributes of initiation, response speed, and compensation. In Table 2, our factor analysis results showed that the internal consistency coefficients of initiation, response speed and compensation are respectively 0.769, 0.896, and 0.790. The Kaiser–Meyer–Olkin (KMO) value is 0.877. Second, relationship marketing in our study had two dimensions of customer satisfaction and customer trust. We developed the customer satisfaction scale based on Oliver (1980) and Leuthesser et al. (2003), which included a total of four items. We used Morgan and Hunt (1994)’s scale of customer trust with a total of four items. As shown in Table 3, the internal consistency coefficients of the customer satisfaction and customer trust scales are respectively 0.888 and 0.887. The KMO value of the scale is 0.831. Third, we examined consumer loyalty from attitudinal loyalty and behavioral intention (Table 4), according to Zeithaml et al. (1996) and Zhou (2005) using four items. The internal consistency coefficients of the attitudinal loyalty and behavioral intention scales are respectively 0.814 and 0.778. The KMO value of the scale is 0.822. In Table 5, the correlation coefficients among the three factors ranged from 0.26 to 0.61 (P < 0.01). Our survey instrument overall has a good level of reliability and validity.

The data analysis in this study was divided into five steps. First, the descriptive statistics were analyzed to determine the characteristics and distribution of the measured variables, and the reliability of the scale was examined. Second, a measurement model with all dimensions was established by using confirmatory factor analysis (CFA) to test the fit of the measures. In the next step, a baseline path model was developed with structural equation modeling (SEM) to test H1 to H5. SEM was used to test the relationship between the independent variable and the dependent variable. For mediation to occur, the following four criteria must be met: (1) the independent variable should be significantly associated with the dependent variable; (2) the independent variable should be related to the mediator; (3) the mediator should be related to the dependent variable; and (4) the association between the independent and dependent variables must be reduced when the mediator is partially omitted. All variables (initiation, response speed, and compensation) were included in the measurement model (Figure 1). Fourth, the goodness of fit was tested under SEM to explore paths of influence and the magnitude of explained dependent variable. Finally, a structural model was developed, and the regression weights were compared to test the hypotheses.

Around half of the participants (49.5%) were female, and 50.5% were male (Table 1). The mean age of the respondents was 31.9 years (SD = 7.9). Most of the respondents (35.9%) had a Bachelor’s degree or below, and the rest had a Master’s degree (28.6), a doctoral degree (15.3%), or high school and below (20.3%). The mean annual income of the respondents was $11,480 (SD = 9.49). Approximately one-quarter (25.6%) of the respondents had an annual income of $7,200 or below, 27.2% had an income between $7,200 and $14,400, 28.6% had an income between $14,400 and $28,800, and 18.6% had an income above $28,800.

The results show that in service recovery, consumers have the highest positive perception of compensation (M = 3.86, SD = 0.88), followed by timeliness (M = 3.85, SD = 0.85) and loyalty (M = 3.83, SD = 0.83; Table 4). In general, consumers agree with service recovery overall. Regarding relationship marketing, consumers have the same perception of satisfaction and trust (M = 3.67, SD = 0.89); in consumer loyalty, consumers’ attitudinal loyalty (M = 3.66, SD = 0.87) and behavioral intention loyalty (M = 3.66, SD = 0.85) are also similar. Here, 1 indicates initiation, 2 indicates response speed, 3 indicates compensation, 4 indicates satisfaction, 5 indicates trust, 6 indicates attitudinal loyalty, and 7 indicates behavioral intention.

In the second step, the fits of the measures were assessed by using CFA. In this research, the model fits were evaluated through the comparative fit index (CFI), the goodness-of-fit index (GFI), the normed fit index (NFI), and the root mean square error of approximation (RMSEA). According to Bollen (1989) and Byrne (1998), a model is regarded as acceptable if the CFI exceeds 0.93, the NFI and the GFI exceed 0.90, and the RMSEA is less than 0.80.

Our CFA results show that the initial measurement model, which consists of nine items for three factors (i.e., initiation, response speed, and compensation), eight items for one factor (i.e., relationship quality), and four items for one factor (i.e., consumer loyalty), had acceptable fit indices (n = 301, χ² = 594.995, DF = 344, CFI = 0.968, GFI = 0.875, NFI = 0.927, RMSEA = 0.047).

As all regression weights were significant (p < 0.01), the measurement model was further refined, as standardized residuals greater than 2.57 are considered to be statistically significant (Bollen, 1989), and large modification indices (those greater than 3.84) are considered to be statistically significant (Hayes, 2018). Furthermore, the discriminant validity of the measures was assessed by comparing the squares of the correlations between each pair of factors with their average variance extracted values (AVEs). The correlation coefficients among the three factors ranged from 0.26 to 0.61. Since the AVEs for all seven latent factors were higher than the squares of all correlation coefficients, the discriminant validity was acceptable. As shown in Tables 2, 3, the composite reliability values for initiation, response speed, compensation, satisfaction, trust, attitudinal loyalty and behavioral intention loyalty were 0.81, 0.90, 0.83, 0.90, 0.89, 0.89, 0.81, and 0.81, respectively. All values were higher than the suggested value of 0.80. The AVEs for all seven factors were equal to (for employee characteristics) or higher than the suggested value of 0.50. Thus, the convergent validity of the scale is acceptable.

In the next step, H1–H5 were tested by establishing a baseline structural model (n = 301) that included the initiation, response speed and compensation of service recovery as an exogenous variable and relationship marketing and service loyalty as endogenous variables. As shown in Table 6 and Figure 3, the baseline structural model has acceptable fit indices (n = 301, χ2 = 594.995, DF = 344, CFI = 0.968, GFI = 0.875, NFI = 0.927, RMSEA = 0.047). The SEM results reveal that response speed had a significant direct effect on relationship quality (β = 0.27; p < 0.001) and customer loyalty (β = 0.23; p < 0.001); therefore, H2 and H6 are supported. Compensation had a significant direct effect on customer loyalty (β = 0.34; p < 0.001), which supports H7, and initiation had a significant direct effect on relationship quality (β = 0.29; p < 0.001) and customer loyalty (β = 0.23; p < 0.001), which supports H6 and H2. Relationship quality had a significant direct effect on customer loyalty (β = 0.21; p < 0.001), which supports H4. However, compensation had no significant direct effect on relationship quality (β = 0.08). These findings thus support H1, H2, H4, H6 and H7.