COVID-19 outbreak forced the customers to purchase necessities online rather than offline. Malaysia announced its first case of Corona on January 25, 2020. After a rapid increase in Coronavirus cases in March, the government imposed movement control order (MCO) on March 18, 2020 (Isa et al., 2020). For minimizing the chance of spreading coronavirus, government ordered many businesses to be closed (Martin-Neuninger and Ruby, 2020). Only specific businesses like grocery stores can remain open so that minimum customers come out for purchasing their necessities. During the MCO, many businesses were closed. During that time, there was a drastic switch to webstores. E-commerce is considered as an essential service in Malaysia (Isa et al., 2020). Pandemic altered all the traditional shopping behaviors (Hasanat et al., 2020).

During the MCO, online shopping became popular medium of purchasing not only for customers but also for traders (Isa et al., 2020). Social distancing also forced customer to purchase online so that they can save their time (Hasanat et al., 2020; Isa et al., 2020). During MCO, different webstores like happy fresh, Lazada, and Shopee experienced increase in orders. A 10–15% increase in orders has been reported by webstores, which created troubles for suppliers to store and supply demanded products (Mymetro, 2020). As online shopping percentage increases, service failures are also increased in different ways. For example, required products were out of stock, ordered items were delivered late, wrong items were delivered, damaged items were delivered, and there were website connectivity issues, online payment transaction issues, and personal information privacy issues. Statistics shows that there were 11.9 million webstore users in 2019. The number of users has been increased to 14.4 million until now. The revenue for online shopping has been increased by 89% in 2020. Lazada had the highest number of active users in the third quarter of 2019. Like other companies, Lazada is also contributing in economy of Malaysia to sustain in COVID-19 (Müller, 2020). For example, Lazada also helped in collecting donations for homeless people. As well as offered SMEs to use Lazada’s platform for selling their products. Lazada is a big online shopping webstore and has a majority of active online customers. However, Lazada itself was not well equipped to cope with the current situation of COVID-19. Therefore, Lazada also faced different problems in delivering satisfying services to its customers, such as late in the delivery of orders. Therefore, Lazada was chosen for the current study.

Many researchers have contributed to consumer behavior during COVID-19. They discussed the problems that are faced by the customers while using online shopping during COVID-19. For instance, customers faced trouble in getting fresh products and vegetables (Li J. et al., 2020), consumers’ grocery purchasing behavior during COVID-19 (Grashuis et al., 2020; Hall et al., 2020; Li J. et al., 2020), spending pattern (Andersen et al., 2020), food purchasing habits (Richards and Rickard, 2020), and online shopping behavior (Hasanat et al., 2020; Kim, 2020). In the era of pandemic, while customers are already facing troubles to get their required products, issue of service failure ruins the customer’s relationship with service provider. Companies must consider the issue of service failure during COVID and think differently to retain their customers. The following table provides an overview of research conducted on consumer behavior during COVID-19. In recent studies conducted during COVID-19, researchers mainly focused on purchasing trends of customers, shifting from offline shopping to online shopping, online grocery shopping, and technology adoption. However, the concept of service failure and service recovery is oversighted.

Service failure occurs when the customers are dissatisfied with service/product or its delivery process (Maxham, 2001). When a customer received a wrong service, it goes and leaves customers feeling negatively about the service experience; as a result, a service failure has happened (Gelbrich and Roschk, 2011). Service failures increase the switching intention of customers as well as decrease the loyalty toward service provider (Pieters et al., 2019). When a customer faces a service failure during his purchase, their emotions hurt, and the customer will try to avoid purchase due to the fear of the repetition of service failure (Kamble and Walvekar, 2019). The customer induced to seek services from other available service providers. The negative response of service providers toward the service failure produces negative outcomes, such as negative WOM and decrease in profit (Tax et al., 1998; Bitner et al., 2000; Tronvoll, 2011), and spreads in market like a virus (Zhu et al., 2021). If service provider cannot redress the dissatisfied customer properly, he will not only switch the services of service provider but also share his bad experience with his social circle (Cai and Qu, 2018). Service failures are inevitable because of the integral inconsistency of service performance (Zeithaml et al., 1990). In response to the service failure, the customers might complain to the service provider. Poorly handled complaints are certainly not forgotten, and these customers are vulnerable to defection (Rotte et al., 2006). Therefore, service providers need to handle the complaining behavior properly to retain their customers and ultimately maximize their profit. A defensive strategy to keep the existing customer is less expensive compared to the offensive strategy that attracts new customers. Attracting new customers is five times more expensive as keeping an existing one (Timm, 2001). It has been observed that if a company brings 5% of its angry customers, then the profit will boost from 25 to 95% (Kotler, 2003; Manzano-Machob, 2013).

A number of existing studies have investigated service failure, including delivery failure (delivery later than promised, wrong item delivered or damaged items delivered), system failure (navigational problem, insufficient product information), product quality failure (poor product quality), website security failure (credit card fraud, sharing personal information to e-retailers), payment problems (payment overcharged, confusing purchasing process), and customer support failure (poor communication, unfair return policies) (Holloway et al., 2005; Holloway and Beatty, 2008).

Though, different studies reported different types of service failure in their findings. The service failure at any stage of complete service obtaining process, starting from ordering till consumption, is overlooked. In this study, we have tried to consider the complete service-acquiring process and the response of customers at different stages of this process. Based on the above discussion, the following hypothesis is proposed:

Complaining can be behavioral or non-behavioral (Singh, 1988). In behavioral study, the customer complain to company (seller, retailer, website, or service provider), third-party (legal and consumer protection organizations), or friends and family (Singh, 1988). On the other hand, customers face service failure but do not launch a visible complaint, which is non-behavioral complaining response. When customers have encountered a service failure, they want redress to vent their frustration and anger. Now in the pandemic situation, companies are still focusing on the same strategies of service recovery (compensation, explanation, and apology). But customers’ behaviors and expectations are changed due to current situation. It is very hard to encourage customers to complaint to service provider as they consider it more difficult in online setting. Comparatively in offline store, the customer can elaborate his/her complaint in detail. In online complaints, more psychological effort is required as compared to offline complaints. Customers have to wait for long to get resolution of their complaints. In offline setting, customers can go to service provider and can confirm the status of their complaint, but in online setting, 50% of complaints are ignored by the sellers (Rosenmayer et al., 2018).

In online complaining, non-verbal communication and face-to-face interaction are not available, which have a psychological impact on the customer (Rosenmayer et al., 2018). Criticality in launching a complaint is a big issue in complaining to webstore. In the online complaining system, many customers do not know how to launch and follow up a complaint (Järvenpää, 2017). Another factor in non-complaining behavior is an alternative to same services that are available in market. Customers silently switched to other webstores and do not bother to the complaint. Customers who complaint to webstores do not get proper response from the webstores that might be due to less staff available to respond to customers timely. Customers complained to the webstores, but they did not get service recovery as per their expectations. The above-discussed issues might be the factors, which leads the customers to switch the webstore to fulfill their needs.

Complaints are the opportunities for the companies to improve their services or redesign their modes of providing services (Turner, 2018). Complaining helps in many ways to service providers. If customer does not complain to the service provider, it causes loss not only for the customers but also for the service provider (Sands et al., 2020). In previous studies, it is mentioned that after service failure, customers complain publicly or privately (Istanbulluoglu et al., 2017). It is still needed to study the complaining behavior of customers in online shopping in special circumstances like COVID-19. In normal situation, the customer can purchase from offline store to fulfill their needs temporarily until they acquire service/product from webstore. In COVID situation, customers are more careful about the health of their families and community so they prefer to purchase online. Though it is easy to purchase through webstores, it also has its own shortcomings. If customers do not get proper response from the webstore, they do not bother to complain again they exit silently. Customers have more choices in current competitive era (Liu and Atuahene-Gima, 2018). So the complaining method must be very easy so that a common. Customer who does not have much grip on technology cannot launch complaint easily. Sometimes customers only do not complaint due to the lengthy procedure of launching a complaint (Järvenpää, 2017). COVID-19 has forced the companies to decrease their employees to minimize the risk of pandemic (Blustein et al., 2020). Further, the government has implemented the lockdown and restricted the movement of people (Isa et al., 2020). These restrictions have created problems in transportation. On the contrary, the increase in orders has created a severe problem for companies to fulfill the customer’s requirement on time. The fear of COVID-19 has forced the people to order consumer products through online channels (Grashuis et al., 2020). If these items were not delivered on time, the customer found the alternatives and switched to other service providers. The complaining behavior of customer provides a chance to webstore to make their services better and retain their customers by providing a better service recovery. Webstores encourage customers to report their complaints to webstores so that they can make error-free services. Complaining behavior has a direct link with service recovery; hence, the following hypothesis is proposed:

In a highly competitive service environment, it has become very difficult for organizations to attract their dis-satisfied customers and develop an effective strategy for providing service recovery to retain their customers (Migacz et al., 2017). In a competitive environment, customers have more power to select alternative product/service. Previous studies show that an interest in service recovery has increased because service failure experience often leads to customer switching. Although the first rule for providing services should be to do things in their right manners, Zeithaml et al. (2006) have developed different strategies for satisfying customers to help the marketers: act quickly, encourage and track complaints, treat customers fairly, cultivate relations with customers, and provide explanation.

When service failure occurs, it becomes essential for webstore to reacquire dissatisfied customers so that financial and reputational losses can be minimized. Providing service recovery to dissatisfied/complaining customers is an opportunity for retailers to model the perception of customers about their brand (Turner, 2018). In the literature, it is revealed that poor service recovery was offered to the customers. For example, some online retailers respond only to a half of the complaints they received, and in response to those complaints, they just offered apology or empathy (Rosenmayer et al., 2018). Webstores must be careful about the complaints of customers as well as their expectation regarding service recovery. The following table provides a review on studies of service recovery in online shopping context. If a customer is dissatisfied with service failure, she/he might be dissatisfied/satisfied with service recovery (Tarofder et al., 2016). 50% of complains are dissatisfied with the service recovery provided by service provider (Bradley and Sparks, 2012). It is also reported in a study that only 30% customers are recovered by the service provider after service failure (Michel and Meuter, 2008). A customer’s future intention is based on type of service failure and recovery strategy (Riscinto Kozub et al., 2014). It is very important for service providers to understand the expectations of customers regarding service recovery (Zhu et al., 2021). In another study, it is reported that 70% of service recovery efforts were wasted due to misunderstanding the expectations of customers (Maher and Sobh, 2014). Previous studies investigated the issue of service failure and repurchase behavior. They posit that after a service failure, the customer tends to spread NWOM and have less intention to repurchase (Petitjean, 2013). It is not necessary that in all cases, the customer will continue purchasing from the same service provider in the future (Harrison-Walker, 2012).

The service recovery strategies are not same for online customers and offline customers (Baron et al., 2005). The environment of online/offline shopping plays an important role in providing service recovery (Baron et al., 2005). If choice of recovery were provided to customers, it would have a significant impact on overall satisfaction with webstore (Chang, 2008).

Switching is defined as “replacing or exchanging the current service provider with another service provider” (Bansal and Taylor, 1999). Switching is considered as opposite to loyalty (Singh and Rosengren, 2020). Loyalty focuses on positive outcome of purchase experience with webstore, and a loyal customer must purchase from the same seller in the future. Switching behavior shows the negative outcome of customer that leads customer to leave. Technological advancements changed the business trend nowadays. Companies are shifting their businesses toward online channels to achieve competitive advantage in the market. Currently, customers have many options to fulfill their necessities but switching behavior is harmful action (Istanbulluoglu et al., 2017). In the literature, it is argued that attracting new customers is five to eight times more expensive than retaining new customers (Wu and Huang, 2015). Service failure is one of the major factors that affects switching intentions (Pieters et al., 2019). Literature is evident that after service failure if service provider offers service recovery to the customers, it will remove dissatisfaction and make the customer loyal, but due to low switching cost, customers shift toward other webstores with a single click (Zhu et al., 2021).

In online shopping, less or no human interaction creates difficulties for customers to express their problems to the service provider, which leads to high switching behavior (Holloway and Beatty, 2003; Javed et al., 2020). In online shopping, if service provider provides service recovery to the customer but the customer was expecting more than offered, in this situation the customer will switch even after getting service recovery (Maher and Sobh, 2014). Effective service recovery provided by webstores can reduce the switching rate of customers. It will be only possible if webstores keep in touch continuously during whole process of taking service. Timely and continuous response to customer queries might reduce the switching intention. Based on the discussion, we proposed the following hypothesis:

Profit maximizing is the main objective of all organizations. To attain this objective, they try to make their customer loyal and retain them. Currently, researchers are focusing more on customers’ anti-consumption behaviors (Curina et al., 2020). Anti-consumption behavior explains the impact of negative emotions evoked by service failure and their influence on loyalty, repurchase intention, and frequency of use (Jayasimha et al., 2017; Zarantonello et al., 2018). Risk of service failure also affects the repurchase behavior of customer (Lăzăroiu et al., 2020). From a managerial perspective, it is necessary for service providers to deal with customers’ service failure issues effectively. It directly or indirectly affects customer’s repurchase intention. In traditional markets, different service recovery strategies are using to improve customers’ repurchase intention. However, in online shopping, customer’s behavior toward webstores is different. If customers did not get service recovery as per his/her expectations, he will not repurchase from that webstore.

Currently, due to the COVID-19, customers are remaining at their homes and they have more time to search their required products available on different shopping sites. So, in this situation, when a customer has faced service failure, there is higher chance of his switching. For example, if a customer purchases from an online shopping during a promotion campaign but the received product is not as per his expectations, he contacts customer care and asks for compensation. The service provider promises a refund. When the customer gets the compensation, he is likely to revisit the same webstore for repurchase. However, once the promotion campaign has ended, he observes that the same product is available at a higher price. In that case, he will not repurchase and switch to other available options. Repurchase intention is directly linked with efficient service recovery. In current situation of COVID, webstores might focus on some additional strategies so that customer’s repurchase intention can be enhanced. Based on the above discussion, the following hypothesis is proposed:

Customers always try to get the maximum value of service/product (Lu et al., 2020). They expect more from the service provider (Lu et al., 2020). If a customer gets as per his expectations, it makes him satisfied, which in turn leads to future purchase, and vice versa (Michel and Meuter, 2008). The expectation–disconfirmation theory has been used widely in different contexts to investigate the customer’s post-purchase behavior (Ayanso et al., 2015). The expectancy confirmation theory (ECT) is developed to measure the satisfaction of customer and to check the impact of satisfaction on the willingness of customer to repurchase (Ayanso et al., 2015). As per expectancy disconfirmation theory presented by Oliver (1977), confirmation happens when performance of service/product matches the expectation. When a customer has a bad experience with a service/product, it causes negative disconfirmation, and positive disconfirmation happens when a customer gets better service/product performance than expected. Expectancy–disconfirmation model explains the comparison between expected and actual service/product performance, further resulting in dissatisfaction/satisfaction (Oliver and DeSarbo, 1988). Customers make the evaluation of actual performance of service/product on the basis of his/her expectations and results could be any one of the following three outcomes: (1) Positive disconfirmation (if actual performance of service/product is higher than expected), (2) confirmation (if actual performance of service/product is equal to expectations), and (3) negative disconfirmation (if actual performance of service/product is lower than expected). Based on the three outcomes, the customer further decides its future intentions. If the customer is highly satisfied, he becomes loyal customer, and as the result of dissatisfaction, the customer will switch the service provider. Many scholars have identified that expectancy–disconfirmation theory has an effect on customer’s satisfaction (James et al., 2015; Sarkar et al., 2015), which leads to repurchase or switching behavior. This paper studies the probable future intentions of customers after facing service failure. Customers complain to service provider or their friend and family to vent their frustration. Service provider will provide service recovery to customers who faced service failure. So, if the service recovery will be as per expectations of customer, it will be the confirmation stage. In this situation, a customer might repurchase or switch. If the customer receives less service recovery with respect to his expectations, it will be the negative disconfirmation. In this situation, more chances are toward switching webstores. More than expected, service recovery is positively disconfirmed. Positive disconfirmation might lead to repurchase intention. A customer decides his future purchase from the same seller or switching to other webstores based on the evaluation of the expected and actual service recovery.

A questionnaire was constructed to collect data from the targeted customers. The questionnaire consisted of multiple items based on the previous literature and was divided into two parts. The first part of the questionnaire was about the basic information of respondents. Basic information like gender, education, occupation, and age was gathered to understand the characteristics of respondents. All questions related to basic information formalized on a nominal scale were used to measure the respondents’ characteristics. The second part of questionnaire includes the questions related to variables of the research. For measuring the customers’ response, a 5-point Likert scale was employed ranging from 1 (strongly agree) to 5 (strongly disagree). Items of different variables, service failure (Li et al., 2016; Das et al., 2019), complaining behavior (Singh, 1988), service recovery (Parasuraman et al., 2005), repurchase intention (Jeon et al., 2017), and switching intention (Nikbin et al., 2012) were adapted from the literature to compose the questionnaire.

All customers who purchased services or products through online shopping and faced any kind of service failure are the respondents of our study. Because we do not have the exact customers’ list who might be our respondents, we used snowball sampling technique, which includes purposive sampling. For data collection, snowball sampling is widely used by researchers (Browne, 2005; Baltar and Brunet, 2012), especially in those cases when we want to target as maximum as possible respondents with same characteristics, and it seems hard to reach (Sadler et al., 2010). The questionnaire was created on google forms (Sekaran and Bougie, 2019) and distributed in target respondents through using various online channels like WhatsApp groups, Facebook groups, and emails. It was crucial for us to collect data through online channels because limited opportunities for physical data collection were available during pandemic and lockdown. Collecting data using online channels allows us to ensure the safety of respondents as well as timely data collection.

Common method variance usually occurs when data are collected from a single source in a single sitting (Yüksel, 2017). It may affect the structural relationships (Kline, 2015) and undermines validity (MacKenzie and Podsakoff, 2012). Two statistical controls designed were used to minimize the risk of CMV. First, Harman’s single factor method was implemented to detect the CMV in the data. The results indicate that 19.28% of the total variance by the single factor was the highest variance explained which is very less than the norm of 50% (Fuller et al., 2016). Second, the full multicollinearity test was done as per recommendation of Kock (2015). It reveals that pathological VIF values for all latent variables ranged from 1.000 to 2.850, which is well below the 3.3 threshold, validating that the data are free from CMV problem.

Factor loading, Cronbach’s alpha, composite reliability (CR), average variance extracted (AVE), and discriminant validity were assessed to test the measurement model for reflective constructs. The findings are presented in Table 4. The values of factor loadings of each first-order reflective construct were higher than 0.5 for retaining the items (Hair et al., 2020). However, COMB2, COMB3, COMB5, INFF2, SYSF2, SYSF4, PRDF1, and PROSF1 were dropped due to low factor loadings. Cronbach alpha (α) > 0.70 and CR > 0.70 show a high degree of internal consistency, while AVE > 0.50 shows a high degree of convergent validity (Hair et al., 2019). Next to this, indicator multicollinearity was also evaluated through VIF test. The results tabulated in Table 4 reveal that VIF of each item is well below the limiting value of 5, suggesting that multicollinearity is not a problem in this study (Hair et al., 2020). Furthermore, heterotrait–monotrait ratio (HTMT) criteria were employed as per the recommendation of Hair et al. (2019) to determine the discriminant validity due to its superiority over other methods. Table 5 shows that HTMT values of each construct are less than the cutoff score of 0.90, indicating that all constructs are distinct. Hence, it can be concluded that all reflective constructs established a convergent and discriminant validity in this study.

This study proposed service failure and service recovery as a type two higher-order (reflective-formative) constructs. Therefore, a disjoint two-stage approach, as suggested by Shmueli et al. (2019), was adopted, which was employed in three steps. In first step, convergent validity was checked by means of redundancy analysis. The findings show that service failure and service recovery have a correlation of 0.792 and 0.828 with its global item, respectively, which is sufficiently above 0.70. It indicates convergent validity established for higher-order (reflective-formative) constructs. In the second step, the multicollinearity of the indicators (VIF) was used to determine the formative measure. The results are tabulated in Table 4. The VIF value is well below the cutoff value of three for all measures (Hair et al., 2019), meaning that collinearity is not a serious concern in this study. In the last, a bootstrapping procedure with 5,000 subsamples was used to assess the significance of weights. The results in Table 4 indicate that the weights of all indicators of both higher-order reflective formative constructs were significant. As such, it can be concluded that the measurement model was validated.

Following the measurement model, structural model was assessed to analyze the statistical significance of path coefficients, explanatory power, predictive relevance, and their effect sizes. A bootstrapping with 5,000 subsamples was carried out to test significance of proposed relationships. As shown in Table 6, all four hypothesized relationships are statistically significant. Service failure is positively related to complaining behavior (H1: β = 0.363, p = 0.000). Similarly, complaining behavior is positively related to service recovery (H2: β = 0.294, p = 0.000). Moreover, the relationships between service recovery and switching intention (H3: β = 0.300, p = 0.000) is positively significant. Besides, the effect of service recovery on repurchase intention (H4: β = 0.3245, p = 0.000) is also positively significant. However, this effect is less significant as compared to the effect of service recovery on switching intention. Furthermore, effect sizes (f²) are also tabulated in Table 6, which indicates the effect sizes of weak to medium range.

Further, coefficient of determination (R2) was used to evaluate the explanatory power of dependent variables by the independent variables. The R2 values are given in Table 7, which show moderate to weak explanatory power of the model (Cohen, 2013). Similarly, blindfolding procedure was used to assess the predictive relevance of this study. Findings in Table 7 reveal that Q2 values are less than 0.25, which is an indicative of low predictive relevance in this study (Hair et al., 2019).