Mix&Match: A Resource-Based Complaint Recovery Framework for Tangible Compensation

Abstract

Resource exchange theory suggests service recovery compensation is optimal when it is commensurate with what was lost (e.g., refund for overcharging). However, in practice, companies cannot always follow the theory-driven prescriptions, and the complaint recovery literature remains silent on how to best recover in such suboptimal situations. This study takes a resource-based theory stance to propose Mix&Match, a complaint recovery framework for tangible compensation offers (refunds, redeliveries, or credits) to optimize customer retention and lifetime value in both optimal and suboptimal complaint recovery scenarios. We find that matching tangible compensation with the complaint cause (e.g., redelivery for expired products) is the most effective recovery response for improving customer retention and lifetime value. However, in suboptimal nonmatching scenarios, monetary compensation in the form of store credit proves to be the most effective response.

Keywords

service failure customer complaints tangible compensation customer churn RLV customer base analysis

The “real-time” nature of service delivery has made it almost impossible for service organizations to operate flawlessly (Prasongsukarn and Patterson 2012). The latest Consumer Complaint Survey conducted by the Consumer Federation of America and North American Consumer Protection Investigators (2015) showed that most of the top 10 consumer complaints are service-related issues. While even the most customer-oriented companies experience service failures, how they react to customer complaints and service failures are somewhat under their control. Previous studies suggest that 9 of the 10 customers who experience a failure would return if their issue was properly resolved; this ratio plunges to 22% if the situation is not handled appropriately (Simons and Kraus 2005). Hence, it is vital that when companies face a failure situation, they identify appropriate recovery strategies that can deliver the desired outcome.

Recent academic findings in the service recovery literature prescribe the use of resource exchange theory (Foa and Foa 1974) as the overarching theory to guide recovery choices under different service failure scenarios (Roschk and Gelbrich 2014). By definition, a resource is anything tangible or intangible that can be transmitted from one person to another (Foa and Foa 2012). Resource exchange theory identifies six broad resource categories: love, status, goods, information, service, and money. As Roschk and Gelbrich (2014) argue, for a complaint recovery to be effective, the resource lost due to the failure must match the resource gained from the recovery. For example, Roschk and Gelbrich suggest that in the case of a flawed product (goods-related failure), the recovery choice decision between an exchange (goods-based recovery) and a refund (money-based recovery) should favor the former as it will result in a higher level of satisfaction, loyalty, and positive word of mouth.

In the retail sector, service- and goods-related failures have historically been reported as customers’ main service encounter pain points (Holloway and Beatty 2003; Kelley, Hoffman, and Davis 1993). The collective advice from the latest service failure recovery studies encourages companies to offer tangible compensation in the form of refunds/store credits or exchanges (Gelbrich, Gäthke, and Grégoire 2015; Gelbrich and Roschk 2011) and, at the same time, try to match such recovery offers with the resource that has been lost (Roschk and Gelbrich 2014).

Nevertheless, in practice, the higher costs of a matching recovery policy (Zhu, Sivakumar, and Parasuraman 2004) and the nature of failures can result in companies being unwilling or unable to follow the normative resource-based failure recovery model.¹ To elaborate, an online retailer might see recovering all goods-related complaints with redelivery as an intolerable pressure on its delivery system or simply not cost-effective for low-value complaints. Furthermore, for some types of customer complaints, such as late delivery, it is not possible to make up for what is lost (i.e., time) via a matching offer. In such scenarios, a company might choose or be forced not to follow resource matching, instead adopting a recovery strategy with a mix of matching and nonmatching compensation. This suboptimal but plausible situation has received little attention in the service recovery literature in terms of measuring return and effectiveness.

When it comes to measuring the effectiveness of service recovery practices, the literature is dominated by perceptual metrics such as satisfaction and repurchase intention (Khamitov, Grégoire, and Suri 2020). In a recent study, Van Vaerenbergh et al. (2019) urge the use of objective behavioral variables such as churn/retention and customer lifetime value (CLV) due to their “stronger impact on practice” and the important role they play in demonstrating the investment return of service recovery. Although few studies investigate the effect of different failure and recovery types and their interactions on customers’ “actual” churn behavior, of these two behavioral metrics, customer churn appears in many previous studies only in the form of perceptual metrics such as return intent (Orsingher, Valentini, and Angelis 2010), loyalty (Roschk and Gelbrich 2014), and repurchase intention (Holloway and Beatty 2003). Nevertheless, financial metrics such as CLV are rarely found in the service failure and recovery literature. This can be attributed to the dominance of survey-based studies and the nature of financial metrics, which are difficult to measure in surveys. From the few studies that examine customer-level financial metrics to evaluate service recovery, Knox and Van Oest (2014) shed no light on the interaction between complaint type and recovery type and its effect on financial metrics. Cambra-Fierro, Melero, and Sese (2015) use the congruence approach (Chandon, Wansink, and Laurent 2000) to match complaints and recovery but focus on the short-term (1-year postfailure recovery) profitability of customers and overlook the impact of the interaction between complaint and recovery types on the lifetime value of those affected. This is a significant gap in the literature given the great emphasis on financial accountability of service-related investments (Rust, Zahorik, and Keiningham 1995; Zhu, Sivakumar, and Parasuraman 2004).

Considering these gaps, this study seeks to address the following research question: What is the most effective tangible compensation for optimizing retention and customers’ lifetime value when deviating/not deviating from the normative model of matching resources in failure recovery? Using resource exchange theory as a roadmap, we specifically focus on tangible compensation in the form of credits, refunds, and redeliveries in addressing goods- or service-type complaints. Thus, this study contributes to the literature on service failure and recovery in two ways. First, the study builds on Roschk and Gelbrich’s (2014) resource exchange framework for complaint recovery to propose Mix&Match, a framework for offering tangible compensation that applies to both matching and mismatching failure and recovery resources. Second, responding to recent calls by Van Vaerenbergh et al. (2019) and Khamitov, Grégoire, and Suri (2020) to employ behavioral and financial metrics to evaluate the effectiveness of service recovery practices, we use customer churn/retention and customer residual lifetime value (RLV) as our outcome variables. Here, customer RLV is defined as the present value of the expected (future) profit generated by an existing customer for the rest of his or her lifetime with the company (Fader, Hardie, and Lee 2005). The use of such objective metrics as the outcome variables of the framework enables us to go beyond behavioral intentions (Gupta and Zeithaml 2006) and derive impactful and relevant recommendations for managers. Table 1 summarizes this study’s contributions.

Table 1.

Contribution Table.

Authors	Outcome Metric	Complaint Type × Recovery Type			Consequences
Authors	Outcome Metric	Framework	Matching	Nonmatching	Churn/Retention	Profitability
Roschk and Gelbrich (2014)	Perceptual	Resource exchange theory	✓	—	✓	—
Cambra-Fierro, Melero, and Sese (2015)	Financial	Congruence approach	✓	—	—	Short/medium-term
Knox and Van Oest (2014)	Behavioral/financial	—	—	—	✓	Long-term (RLV)
This study	Behavioral/financial	Resource exchange theory	✓	✓	✓	Long-term (RLV)

Note. RLV = residual lifetime value.

We extend Knox and Van Oest’s (2014) model to accommodate the type of complaint made, as well as the type of recovery offered, and develop a probability model to seek empirical answers to our research question. We estimate the model using transaction, complaint, and recovery data for customers of a major online fast-moving consumer goods (FMCG) retailer in Australia.

As per the Mix&Match advice, in cases of goods-related complaints, redelivery may be the best recovery strategy to prevent customer churn and increase lifetime value of existing customers. However, when the company addresses goods-related complaints with monetary compensation, store credit should be offered to optimize complainants’ retention and lifetime value. Monetary compensation was found to be the most effective for complainants’ lifetime value when recovery actions deviate from directly matching the resources involved. In this scenario, as our findings suggest, customers respond more positively to monetary compensation in the form of store credit.

Conceptual Framework and Theoretical Background

Overview

Drawing on resource exchange theory (Foa and Foa 1974) and its recent applications in the service recovery literature (Roschk and Gelbrich 2014), this study’s main interest is to empirically investigate the effectiveness of tangible compensation (i.e., refunds, redeliveries, or credits), as a key element of customer satisfaction with service failure recovery (Gelbrich, Gäthke, and Grégoire 2015; Goodwin and Ross 1990), when dealing with customer complaints concerning goods or services. In the retail sector, these two broad complaint categories are reportedly the most common (Holloway and Beatty 2003; Kelley, Hoffman, and Davis 1993).

Taking the resource exchange theory perspective, we propose a framework called Mix&Match that categorizes tangible compensation resources as goods (e.g., redeliveries) and money (e.g., refunds and credits) and endeavor to find the optimal complaint-recovery pair for different scenarios, with goods or service-type complaints. We consider two main scenarios: (1) where the resources involved in the complaint and recovery are matched (goods vs. goods) and, more importantly, (2) where the resources involved in the complaint and recovery are not matched (service vs. goods or money). We empirically test the effectiveness of offering different recovery resources under these two scenarios. We are particularly interested in customer churn/retention and long-term profitability (Knox and Van Oest 2014) as measures of complaint recovery effectiveness.

Complaint and Recovery

Despite companies heavily investing in complaint handling, customer satisfaction with recovery practices has not improved much during the last five decades (Pugh, Brady, and Hopkins 2018). In general, customers acknowledge that service delivery cannot always go as smoothly as planned, and hence, experiencing a failure does not necessarily directly contribute to their dissatisfaction. What can escalate the situation and trigger customer dissatisfaction is the organization’s inability to respond to the situation appropriately (del Río-Lanza, Vázquez-Casielles, and Díaz-Martín 2009). A comprehensive review of services marketing literature identified complaint recovery as one of the critical touch points through which a company can either turn a complaining customer into a delighted and loyal one or worsen the situation by applying inappropriate recovery measures. The latter, in turn, can negatively influence customer satisfaction, word-of-mouth, and repurchase intentions and ultimately induce customers to leave for good (del Río-Lanza, Vázquez-Casielles, and Díaz-Martín 2009; Lopes and da Silva 2015; Maxham 2001; Patterson, Cowley, and Prasongsukarn 2006).

By definition, service failure occurs when a customer perceives a service provider has failed to deliver the core or supplementary service expected in a service exchange (Prasongsukarn and Patterson 2012). This expectation disconfirmation can then lead to a customer complaint. To prevent further damage, service companies attempt to recover the situation by offering the customer some form of compensation for the loss incurred (Patterson, Cowley, and Prasongsukarn 2006). Complaint recovery refers to the steps companies take to neutralize the potentially drastic effects of the failure and restore the customer to a state of satisfaction (Patterson, Cowley, and Prasongsukarn 2006; Simons and Kraus 2005). A recent comprehensive review of journal articles on complaint recovery suggests nearly 97% of the published articles tested the effectiveness of recovery responses while revealing a total of 21 different potential recovery measures (Van Vaerenbergh et al. 2019).

The common core of most studies in this area is the notion of justice and its value in explaining individuals’ evaluations of the rightness of exchanges (Orsingher, Valentini, and Angelis 2010). Perceived through the lenses of social exchange and equity theories, in the event of a service failure, customers can experience a social or economic loss. In such situations, companies’ main focus should be making up for the loss incurred (Gelbrich and Roschk 2011; Smith, Bolton, and Wagner 1999). The degree to which complainants are satisfied with company responses depends on their perception of the fairness of the compensation offered (justice perception; Gelbrich and Roschk 2011). In other words, the explicit or implicit attempt of most scholars in this area is to find the just offer under different scenarios. As a recent development, Roschk and Gelbrich (2014) utilize resource exchange theory (Foa and Foa 1974) to propose a comprehensive complaint recovery guide that best delivers the justice that complainants expect under different complaint-recovery conditions.

Resource-Based Complaint Recovery

According to resource exchange theory (Foa and Foa 1974), in a service failure, customers experience loss of a resource (from the six categories mentioned earlier: love, status, goods, information, service, and money), which leads to an imbalance. This imbalance can be corrected by accepting a resource that resembles the one that was lost as closely as possible. Failure to match resources on the two sides of the exchange in terms of size and type may be perceived as an injustice and lead to dissatisfaction (Dorsch, Törnblom, and Kazemi 2017).

In a service failure, the lost resource may be money as a result of overcharging, goods when a damaged item is received, or service when the online grocer’s driver is late. Consumers prefer to receive a form of compensation that matches the type of resource they lost due to the failure (Roschk and Gelbrich 2014). In the current study, this means when customers’ complaints are goods-type, resource-exchange theory favors redelivery as the preferred recovery type. In addressing our research question in the “no deviation” (matching) scenario, we empirically investigate the superiority/inferiority of resource matching practices in tangible compensation offers for both customer churn likelihood and long-term profitability (measured via RLV).

Although, in theory, matching resources can be the most effective recovery strategy, in practice and due to limitations stemming from the complaint types and resources involved or for reasons based on companies’ strategic decisions, companies might deviate from the normative (theory-driven) model of matching resources in their complaint recovery endeavors. In the case of our framework, this is when dealing with a service-type complaint, where, due to operational constraints (e.g., costs) or the nature of the failure (e.g., late delivery), the focal company cannot or decides not to offer a matching recovery. In such situations, companies need to use an alternative (nonmatching) resource—for instance, offering a credit as a monetary resource—to compensate the disgruntled customer. As our research question indicates, under the “deviation scenario,” the challenge is to find the nonmatching resource that best recovers the complaint situation—an issue on which the literature is silent.

To speculate about the possible outcomes of this exchange, we tap into the notion of the functional relationship between resources, as presented by Foa and Foa (2012). Each of the resource types identified can be characterized by their level of concreteness (vs. abstractness) and particularism (vs. universalism). Concreteness measures a resource’s tangibility and the degree to which it can take a material form and be touched. Resources range from completely tangible (i.e., high concreteness) to completely intangible (i.e., low concreteness). The particularism of a resource tells us to what level its value is defined by the exchange partners involved, as opposed to being commonly accepted (Foa 1971; Roschk and Gelbrich 2017). Thus, the value of a resource with a high level of particularism (e.g., service) is subject to negotiation and is dependent on the individuals involved in the exchange (a high level of value subjectivity), while the value of universal resources (i.e., resources with a low level of particularism, such as money) is based on a more objective and universal scale (a low level of subjectivity in perceived value; Dorsch, Törnblom, and Kazemi 2017; Foa and Foa 1974).

Mapping the six resource types (including those in this study: service, goods, and money) based on their levels of concreteness and particularism produces the configuration in Figure 1. Here, resources with greater similarity in terms of concreteness and particularism are in close proximity (e.g., money and goods), whereas those with lower similarity levels are far from each other (e.g., money and service). Hence, as Figure 1 suggests, the closeness of resources in the illustrated configuration can be considered a proxy to their similarity. When dealing with complaint recovery cases where resource matching does not take place, Roschk and Gelbrich’s (2014) resourced-based service recovery logic suggests that resources proximally situated to one another are expected to provide a better recovery effect, due to their greater similarity. When dealing with a service-related complaint, this means a goods-related recovery would be perceived as better justice and be more effective than a money-related recovery offer.

Figure 1.

Functional relationships among six resource categories, with the resources involved in this study presented in black (adapted from Dorsch, Törnblom, and Kazemi, 2017). The distance between two resources indicates the level of similarity in terms of concreteness and particularism. The closer the resources are in this map, the more similar they are. For instance, Money and Goods are more similar than Money and Service. In a nonmatching situation, we investigate whether managers should seek more similarity between the complaint and recovery resources or offer recoveries with the lowest particularism which translates to having more objective and commonly accepted value.

Nevertheless, findings from Roschk and Gelbrich’s (2014) experimental study also indicate a potential alternative scenario in such cases. When we deviate from resource matching in complaint recovery, monetary recovery has been found to have the greatest effect on customer repurchase intentions. The characteristics of money as a resource suggest that it has a low particularism level, meaning its value is more objective and commonly accepted than any other resource. This characteristic can help complainants better perceive the compensation’s value and potentially a greater level of justice, leading to more effective complaint recovery. In addressing our research question in deviation (nonmatching) scenarios, we intend to determine whether the first or second mechanism is more effective and what is the best compensation offer in such cases to optimize customer retention and RLV.

Behavioral and Financial Implications of Complaint Recovery

Despite the importance of the real behavior of customers and its financial impact on practice (Van Vaerenbergh et al. 2019), the complaint recovery literature is dominated by survey-based studies with intentions data (Khamitov, Grégoire, and Suri 2020). Considering the limited managerial relevance of intentions data (Knox and Van Oest 2014; Mittal and Kamakura 2001) and the potentially negative financial consequences of offering suboptimal compensation (Zhu, Sivakumar, and Parasuraman 2004), scholars have recently called for the use of behavioral and financial metrics in evaluating complaint recovery (Khamitov, Grégoire, and Suri 2020; Van Vaerenbergh et al. 2019). To measure the effectiveness of complaint recovery practices under the Mix&Match framework, we consider churn/retention and long-term profitability as our outcome variables and investigate the impact of different tangible compensation scenarios.

Churn/retention (captured using repurchase intention in survey-based studies) has long been an established measure of effective complaint recovery (e.g., Holloway and Beatty 2003; Maxham 2001). In general, customers expect a certain level of service quality when dealing with a company. Moreover, those who experience a failure expect the company to recover the situation. If the recovery offered does not meet customer expectations, there is a double deviation from both service expectations and complaint recovery (Bitner, Booms, and Tetreault 1990; Knox and Van Oest 2014), leading to a perception of poor justice and a decrease in repurchase intention (increase in churn). On the other hand, if recovery expectations are met, repurchase intention might remain stable or even increase (lowering churn intention) as a result of the complaint recovery paradox (i.e., where postrecovery satisfaction is higher than prefailure satisfaction; Maxham 2001; Smith and Bolton 1998).

Furthermore, we look through the congruence approach lenses for the financial effect of recovery (Cambra-Fierro, Melero, and Sese 2015; Mahajan and Churchill 1990). We consider failure and recovery as our two factors and customer profitability as the quantitative outcome. Accordingly, in the complaint recovery context, alignment between the factors (i.e., the loss incurred due to the failure and the perceived gain from the recovery) can lead to enhancement of the outcome (i.e., profitability; Cambra-Fierro, Melero, and Sese 2015). The effect of effective complaint recovery on customers’ short-term profitability has been empirically tested and confirmed by Cambra-Fierro, Melero, and Sese (2015). Here, we empirically examine, under different scenarios, the impact that complaint recovery can have on RLV of customers.

Empirical Context and Model Formulation

Data

The data were obtained from a major online FMCG retailer in Australia. We examine the complete history of purchases, complaints, and their recoveries for all customers with at least one repeat purchase from the company between August 4, 2011, and September 30, 2012. For existing customers, the first 2-month data were used as a warm-up period to initialize their prior transactional behavior. New customers, who registered with the company after the start of the observation period, do not require a warm-up period. Of the 100,896 customers, 23,132 were new. Customers placed a total of 1,211,821 orders during the observation period, which translates to an average of about 12 transactions per customer.

The customer base used for this study comprises business and regular customers. Of the 100,896 customers in the data set, 10,953 were business and 89,943 were regular (nonbusiness) customers. Also, of all the customers in the data set, 53,694 had signed up for the company’s loyalty program (members), while 47,202 had not (nonmembers). Of the 1,211,821 transactions, 62,198 were complaint incidences (5.13%). The average transaction value was $193.71 (Australian dollars) and the average interpurchase time was 16.87 days.

Every transaction can either go smoothly or incur a problem (customer complaint). We classified the reasons for problems as goods-related or service-related. Examples of goods-related complaints include missing items, damaged goods, poor quality, use-by-date issues, warehouse picking errors, and inappropriate substitutions for out-of-stock items. Service-related complaints relate to packaging, delivery, and/or ordering issues. Every complaint raised by a customer is resolved by offering: store credit, refund, or redelivery of goods. The frequencies of the different types of complaints and recoveries are presented in Table 2. In the case of goods-related complaints, while theoretically a matching resource could have been offered through redelivery, the focal company offered a nonmatching resource in 97% of the cases. In the case of service-related complaints, where no matching resource could be offered, the company chose monetary recovery (a resource with a commonly accepted value and low particularism) in 98.7% of the incidents. The figures presented in Table 2 suggest that, when dealing with complaints, the company used a mix of matching and nonmatching tangible compensation, which contradicts the theory-based advice and recommendations made in relevant academic literature.

Table 2.

Frequencies of Types of Complaints and Recoveries.

		Type of Recovery			Total
		Credit	Redelivery	Refund	Total
Type of Complaint	Goods-related	58,187 (NMR)	1,768 (MR)	183 (NMR)	60,138
	Service-related	2,028 (NMR)	26 (NMR)	6 (NMR)	2,060
	Total	60,215	1,794	189	62,198

Note. MR = matched recovery; NMR = nonmatched recovery.

Model

We used the probability modeling approach (Fader and Hardie 2009) to capture customers’ probability to churn as in Knox and Van Oest (2014). Customer $i = 1 . . . I$ has a history of purchases with indexing $j = 1 . . . J_{i}$ . Every transaction can either go smoothly or a complaint may be made with probability $π_{c o m, i, j}$ . We distinguish two types of complaints, goods-related (occurring with probability $π_{c o m_g o o d s, i, j}$ ) and service-related (occurring with probability $π_{c o m_s e r v i c e, i, j}$ ). These probabilities are defined such that $π_{c o m_g o o d s, i, j}$ + $π_{c o m_s e r v i c e, i, j}$ + $π_{c o m_n o n e, i, j}$ = 1. If a complaint is made, it is resolved with probability 1 by providing a refund or store credit or by redelivering the goods. After every transaction j for customer i, there is a probability $π_{c h u r n, i, j}$ that the customer will defect. If the customer does not churn, she places another $j + 1$ order with probability $1 - π_{c h u r n, i, j}$ and interpurchase time $t_{i, j + 1}$ . Omitting subscript i, this flow of events is depicted in Figure 2. The detailed derivation of the likelihood is presented in Appendix A.

Figure 2.

Flowchart of events for a customer.

Customer Churn

The probability of churn after the current transaction j is modeled using a logistic function:

π_{c h u r n, i, j} = \frac{exp (X_{c h u r n, i, j} β_{c h u r n})}{1 + exp (X_{c h u r n, i, j} β_{c h u r n})},

where $X_{c h u r n}$ is a set of covariates influencing the churn decision and $β_{c h u r n}$ are estimated parameters (see Table 3 for a complete list of covariates). The probability of churning varies not only across customers (subscript i) but also across transactions for each customer (subscript j), allowing us to capture observed heterogeneity in churn.

Table 3.

Model Parameter Estimates for $π_{c h u r n}$ and $f_{p u r c h} .$

Parameter	Description	Estimate	SE
$π_{c h u r n}$	Probability to churn^a	0.029
$β_{c h u r n, 0}$	Intercept	−3.636***	0.016
$β_{c h u r n, 1}$	Member	−0.470***	0.012
$β_{c h u r n, 2}$	Business	−1.176***	0.031
$β_{c h u r n, 3}$	Member × Business	0.095**	0.048
$β_{c h u r n, 4}$	Goods × Credit	−0.279***	0.031
$β_{c h u r n, 5}$	Goods × Redelivery	−2.287***	0.446
$β_{c h u r n, 6}$	Goods × Refund	0.708*	0.396
$β_{c h u r n, 7}$	Service × Credit	−0.277*	0.162
$β_{c h u r n, 8}$	Service × Redelivery	−0.056	1.515
$β_{c h u r n, 9}$	Service × Refund	−0.008	4.026
$β_{c h u r n, 10}$	Quarter 1	0.563***	0.018
$β_{c h u r n, 11}$	Quarter 2	0.936***	0.017
$β_{c h u r n, 12}$	Quarter 3	−0.009	0.032
$β_{c h u r n, 13}$	Ratio of prior credits	0.618***	0.048
$β_{c h u r n, 14}$	Ratio of prior redeliveries	2.916***	0.148
$β_{c h u r n, 15}$	Ratio of prior refunds	−0.167	0.587
$κ$	Shape parameter	1.094***	0.001
$λ$	Scale parameter^a	18.640
$δ_{p u r c h, 0}$	Intercept	2.558***	0.002
$δ_{p u r c h, 1}$	Member	−0.072***	0.002
$δ_{p u r c h, 2}$	Business	−0.336***	0.003
$δ_{p u r c h, 3}$	Member × Business	0.012***	0.005
$δ_{p u r c h, 4}$	Lag Goods × Lag Credit	−0.081***	0.006
$δ_{p u r c h, 5}$	Lag Goods × Lag Redelivery	−2.078***	0.030
$δ_{p u r c h, 6}$	Lag Goods × Lag Refund	−0.010	0.116
$δ_{p u r c h, 7}$	Lag Service × Lag Credit	−0.094***	0.022
$δ_{p u r c h, 8}$	Lag Service × Lag Redelivery	−1.224***	0.179
$δ_{p u r c h, 9}$	Lag Service × Lag Refund	0.468	0.419
$δ_{p u r c h, 10}$	Quarter 1	0.383***	0.003
$δ_{p u r c h, 11}$	Quarter 2	0.482***	0.003
$δ_{p u r c h, 12}$	Quarter 3	0.721***	0.003
$δ_{p u r c h, 13}$	Ratio of prior credits	0.025***	0.010
$δ_{p u r c h, 14}$	Ratio of prior redeliveries	0.852***	0.060
$δ_{p u r c h, 15}$	Ratio of prior refunds	0.672***	0.207
$ρ$	Decay rate	0.983***

^a Average across i and j.

*p < .10. **p < .05. ***p < .01.

Note. SE = standard error.

Event Timing

The interpurchase time between two consecutive orders is captured by Weibull density:

f_{p u r c h} (t_{i, j}; λ_{i, j}, κ) = \frac{κ}{λ_{i, j}} {(\frac{t_{i, j}}{λ_{i, j}})}^{κ - 1} exp {- {(\frac{t_{i, j}}{λ_{i, j}})}^{κ}},

where $κ > 0$ is the shape parameter and $λ > 0$ is the scale parameter. For $κ = 1$ , the Weibull distribution collapses to a more restrictive exponential distribution.

We allow the scale parameter $λ_{i, j}$ to depend on a number of covariates such as customer type (business vs. nonbusiness), loyalty program membership (members vs. nonmembers), previous exposure to complaint recovery actions, and seasonality:

λ_{i, j} = exp (X_{p u r c h, i, j} δ_{p u r c h}) .

The complete list of covariates can be found in Table 3.

Customer Complaints

Because we distinguish two types of complaints, there are three total outcomes for any transaction: a goods-related complaint, service-related complaint, or no complaint. Therefore, we use multinomial logit specification for $π_{c o m}$ . By setting the no complaint category as the reference, the probability of complaints can be specified as follows:

π_{c o m, i, j} = {\begin{cases} \frac{exp (X_{c o m, i, j} β_{c o m_g o o d s})}{1 + exp (X_{c o m, i, j} β_{c o m_g o o d s}) + exp (X_{c o m, i, j} β_{c o m_s e r v i c e})}, i f goods \\ \frac{exp (X_{c o m, i, j} β_{c o m_s e r v i c e})}{1 + exp (X_{c o m, i, j} β_{c o m_g o o d s}) + exp (X_{c o m, i, j} β_{c o m_s e r v i c e})}, i f service \\ \frac{1}{1 + exp (X_{c o m, i, j} β_{c o m_g o o d s}) + exp (X_{c o m, i, j} β_{c o m_s e r v i c e})}, i f none \end{cases},

where $X_{c o m}$ is a set of covariates influencing the occurrence of complaints and $β_{c o m_g o o d s}$ and $β_{c o m_s e r v i c e}$ are estimated parameters for goods- and service-related complaints relative to no complaint.

Addressing Potential Endogeneity

Since the company reacts to all complaints, there is no endogeneity in its decision with regard to providing recoveries. However, there may be endogeneity in the type of recovery decision. We investigated the source of this potential endogeneity using the two-stage residual inclusion (2SRI) technique but did not find any evidence to support its presence. The details of this investigation are presented in Appendix B.

Covariates

The complete list of covariates used in the model is presented in Table 3. The operationalization of key variables together with expected effects are discussed below.

Goods and Service are dummy variables indicating whether the complaint relates to goods or to services. If both Goods and Service are zeros, then no complaint is received.

Credit, Redelivery, and Refund are also dummy variables related to the type of recovery offered. These dummy variables are applicable only to cases where a complaint is made.

Prior Purchases is the recency weighted number of prior purchases. We adopt the approach used by Ansari, Mela, and Neslin (2008) and Van Diepen, Donkers, and Franses (2009) to incorporate recency weighting using an exponential decay function:

P r i o r P u r c h a s e s_{i, j} = \sum_{n = 1}^{j - 1} ρ^{T_{i, n \to j}},

where $T_{i, n \to j}$ is the number of days between the prior transaction n and the current transaction j for customer i and $0 < ρ \leq 1$ is a decay parameter. When there is no recency weighting, that is, $ρ = 1$ , then Prior Purchases is simply the number of previous transactions.

Ratio of Prior Credits, Ratio of Prior Redeliveries, and Ratio of Prior Refunds are the ratios of recency weighted prior credits, redeliveries, and refunds (with $ρ$ as the decay parameter) to the recency weighted number of purchases. It is expected that the higher these ratios, the more likely it is that the customer will churn due to dissatisfaction. We use a 2-month warm-up period for existing customers at the beginning of the observation period to initialize variables related to prior activities. Hence, there is no problem of left truncation of transactional variables for existing customers.

Business is a dummy variable indicating whether a customer has been registered as a business entity. Although, on average, business customers spend a bit more per order than regular customers ($203 vs. $191), their purchase frequency is 67% higher than that of regular customers (45 vs. 27 for the observation period). Business customers usually have an established routine of purchasing from the same supplier, and changing suppliers involves additional costs (money and/or time wise). Therefore, we expect business customers to be less likely to churn.

Member is a dummy variable used to capture whether a given customer is a member of the company’s loyalty program. The membership itself does not considerably change the frequency of purchases (31 vs. 29 for members and nonmembers during the observation period, respectively) or the amount spent per transaction ($196 vs. $190). However, there may be a customer patronage effect and membership may reduce the likelihood of churn.

Quarter 1, Quarter 2, and Quarter 3 are control dummy variables indicating whether the transaction occurred in quarters 1, 2, or 3 of the calendar year. The fourth quarter is considered the reference category.

Tuesday, Wednesday, Thursday, Friday, Saturday, and Sunday are control dummy variables for the day of the week. The reference day is Monday.

N of SKUs is also a control variable reflecting the number of unique stock keeping units (SKU) in the basket.

Customer Experience is a continuous variable that captures customer experience and is computed as the time in years since the customer’s registration with the company.

Perishable is a ratio variable that captures the fraction of the monetary value of perishable goods in the total value of the basket. In our data, these categories are dairy, deli, and meat.

Substitution is a dummy variable indicating whether the company provided substitute products for out of stock items. Product substitution does not necessarily mean that a customer will raise a complaint, but inappropriate substitutions may lead to a complaint.

The descriptive statistics of all variables used in the model are shown in Table 4, and the correlation matrix of variables used in Equations 1 and 3 is presented in Table 5.

Table 4.

Descriptive Statistics for Explanatory Variables.

Variable	Mean	SD	Minimum	Maximum
Complaint	0.051	0.221	0	1
Goods (if complaint = 1)	0.967	0.179	0	1
Service (if complaint = 1)	0.033	0.179	0	1
Credit (if complaint = 1)	0.968	0.176	0	1
Redelivery (if complaint = 1)	0.029	0.167	0	1
Refund (if complaint = 1)	0.003	0.055	0	1
Business	0.171	0.377	0	1
Member	0.566	0.496	0	1
Prior purchases	3.746	2.860	0	43.111^a
Ratio of prior credits	0.049	0.114	0	1
Ratio of prior redeliveries	0.001	0.021	0	1
Ratio of prior refunds	0.000	0.007	0	1
N of SKUs	34.594	18.138	1	230^b
Customer experience (years)	1.535	1.173	0	4.260
Substitution (dummy)	0.410	0.492	0	1
Perishable (ratio)	0.303	0.164	0	1
Quarter 1	0.303	0.164	0	1
Quarter 2	0.246	0.430	0	1
Quarter 3	0.265	0.441	0	1
Quarter 4	0.290	0.454	0	1
Monday	0.180	0.384	0	1
Tuesday	0.166	0.372	0	1
Wednesday	0.149	0.356	0	1
Thursday	0.168	0.374	0	1
Friday	0.176	0.381	0	1
Saturday	0.123	0.329	0	1
Sunday	0.038	0.190	0	1

^a95% and 99% do not exceed 7.737 and 13.304, respectively.

^b95% and 99% do not exceed 67 and 87, respectively.

Note. SD = standard deviation.

Table 5.

Correlation Matrix.

	Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1	Member	1
2	Business	−.07	1
3	Goods with credit	.02	−.02	1
4	Goods with redelivery	.00	.01	−.01	1
5	Goods with refund	.00	.00	.00	.00	1
6	Service with credit	.00	.00	−.01	.00	.00	1
7	Service with redelivery	.00	.00	.00	.00	.00	.00	1
8	Service with refund	.00	.00	0.00	.00	.00	.00	.00	1
9	Prior purchases	.04	.17	.00	−.01	.00	.00	.00	.00	1
10	Ratio of prior credits	.05	−.04	.07	.01	.00	.02	.00	.00	.03	1
11	Ratio of prior redeliveries	.00	.01	.00	.00	.00	.00	.00	.00	.01	−.03	1
12	Ratio of prior refunds	.00	.00	.00	.00	.00	.00	.00	.00	−.01	−.01	.18	1
13	Quarter 1	−.01	.00	.00	.00	.00	.00	.00	.00	−.01	.00	.00	.00	1
14	Quarter 2	.01	.00	.00	.00	.00	.00	.00	.00	.05	.00	.00	.00	−.34	1
15	Quarter 3	.01	.00	.00	.00	.00	.00	.00	.00	.07	.00	.00	−.01	−.36	−.38	1

Results

Parameter Estimation

We estimate the model using maximum likelihood (Appendix A). To address the potential endogeneity of recovery types, we use the 2SRI method described in Appendix B. The recovery residuals included in Equation 1 were all highly insignificant, indicating there is no evidence of the suspected endogeneity. We follow the advice of Papies, Ebbes, and Van Heerde (2016) that in the absence of endogeneity, the original model (as in Equation 1) should be estimated without additional residuals from the first stage (recovery Equation B1). Therefore, all further results presented are from the model with no residuals included from the recovery equation. Table 3 contains all parameter estimates for the churn and purchase parts of the model, while estimates of complaint types are presented in Appendix C.

To understand the effects of certain model features and show that the proposed model outperforms alternative specifications, we conduct model comparisons based on log likelihood (LL) and Bayesian information criterion (BIC) (Table 6). Of all the specifications, the proposed full model has the lowest BIC and is favored over the alternatives. As Table 6 suggests, the highest fit penalties were received by models with no covariates for $λ$ , no time-related variables (quarter and day of the week), no demographic variables (member and business), and exponential hazard ( $κ = 1$ ), respectively.

Table 6.

Model Comparisons.

Model	Restrictions	LL	BIC	BIC Difference
0	No restrictions (full model)	−4,670,116	9,341,213
1	No covariates for $π_{c h u r n}$	−4,674,596	9,349,948	8,735
2	No covariates for $λ$	−4,713,452	9,427,661	86,448
3	No covariates for $π_{c o m}$	−4,672,507	9,345,505	4,292
4	No complaint variables	−4,672,309	9,345,304	4,091
5	No demographic variables	−4,682,893	9,366,584	25,371
6	No time variables	−4,704,087	9,408,805	67,592
7	Exponential hazard ( $κ = 1$ )	−4,678,951	9,358,855	17,642

Note. LL = log likelihood; BIC = Bayesian information criterion.

Effects of Complaint Recoveries on Churn Likelihood

As the effects of the covariates in the logistic model depend on the base values of the dependent variables, we assess their effect from the average level of probability to churn, which is 2.9%. The most interesting effects come from the types of strategies used to handle complaints. If the complaint was a goods type, recovery by credit and redelivery reduces churn by 0.7% and 2.6%, respectively (computed from coefficients $β_{c h u r n, 4}$ = −0.279 and $β_{c h u r n, 5}$ = −2.287 in Table 3). However, if such a complaint is recovered with a refund (coefficient $β_{c h u r n, 6}$ ), the effect on churn is positive but only marginally significant. The fact that redelivery is the most effective recovery strategy for goods-related complaints addresses the matching scenario of our research question from the churn/retention perspective and is in line with findings of Roschk and Gelbrich (2014). For service-related complaints, which represent the nonmatching scenario, only credit was found to have a marginally significant mitigating effect on churn (coefficient $β_{c h u r n, 7}$ ). Hence, results provide marginal support for addressing the nonmatching scenario of our research question from the churn/retention perspective by suggesting that, in the absence of directly matching resources, addressing complaints using tangible compensation with a low particularism level/high flexibility (i.e., money) would be the most effective approach to prevent customer churn.

Effects of Complaint Recoveries on Customer Long-Term Profitability

Customer long-term profitability is the second outcome variable of interest investigated here. As formulated in the research question, our aim is to determine the recovery strategy that can produce the highest long-term profitability depending on the type of resources involved in the complaint and available for recovery, on an individual level. This is a question that, to the best of the authors’ knowledge, has yet to be answered in the services marketing literature.

We now shift our attention to the long-term financial consequences of recovery strategies by examining the impact of different recovery strategies on RLV from a resource exchange perspective. We investigate this effect (1) when matching the recovery offer with the lost resource (i.e., mainly goods-type complaints) and (2) when the offered recovery does not match the lost resource (i.e., mainly service-type complaints). It is important to note that we did not compute CLV, which is computed either for as-yet-to-be-acquired customers or for just-acquired customers; instead, we computed RLV, which is for existing customers who have already engaged in prior transactions. This distinction is rather important and more details are provided by Fader and Hardie (2015).

RLV

We followed Fader, Hardie, and Lee (2005), who derive the RLV for an existing customer, conditional on observed prior behavior, by combining the model for the interpurchase time and churn processes with a separate model for profit per transaction. Assuming the distribution of average customer transaction values is independent of the transaction process (which we formally test later), the expected RLV for a customer can be expressed as:

RLV = margin × revenue per transaction \times DERT,

where the discounted expected residual transactions (DERT) is computed based on customers’ future behavior. To compute DERT, we simulate the future behavior of all customers according to the developed model, including interpurchase times, chances of service failure, failure type and response type, and, finally, churn probability. We simulate future transactions until the customer churns and then discount them to the beginning of the simulation period at a rate of 7%. We performed 1,000 replications of these simulations and averaged the results. Regarding revenue per transaction, Fader, Hardie, and Lee (2005) provide an elegant and robust solution for predicting a customer’s future transaction values from the previous transaction history using a gamma-gamma model (details are presented in Appendix D). Finally, the average margin for the focal company is 30%. Having all three components in Equation 6, we can compute the RLV for every customer.

Recovery policies and their effectiveness

To understand the effectiveness of different recovery policies on customers’ long-term profitability, we first calculate the average RLV of customers in our data set, which equals $1,908. This figure constitutes a benchmark level in further policy simulations. After establishing a benchmark for the company’s complaint recovery actions, based on observed data, we proceed to answer the main question of how the choice of different responses (credits, redeliveries, or refunds) to certain types of complaints (goods- and service-related) can influence customers’ RLV. To achieve this, we conduct nine new policy simulations to assess all possible combinations of recovery of goods-related complaints (G-Crd, G-Red, G-Ref) as well as service-related complaints (S-Crd, S-Red, S-Ref)² for a given customer. For instance, one of the policies is G-Crd/S-Ref, which stands for “recover all goods-related complaints with credits and all service-related complaints with refunds.” The full set of scenarios is presented in Table 7. It is important to note that it may not be feasible for the company to address all complaints with a certain recovery option; hence, some flexibility is required. We acknowledge this and treat these extreme scenarios as boundary conditions. Our main goal is to determine the extent to which the RLV can deviate from the benchmark. This may serve as an input for optimization to determine whether changing the current policy will pay off, and if so, how much.

Table 7.

Change in Residual Lifetime Value in % for Different Recovery Policies (All Customers).

		Service
		Credit	Redelivery	Refund
Goods	Credit	−0.08	−0.09	−0.14
	Redelivery	+3.41	+3.39	+3.35
	Refund	−4.81	−4.82	−4.85

For each of the recovery policies, we compute the RLV per customer for any given combination of complaint and recovery. The RLV figure obtained is then compared against its benchmark value of $1,908. When we simulate the G-Red/S-Crd (recover all goods-related complaints with redelivery and all service-related complaints with credit) scenario, the average RLV is $1,973, which is a 3.41% increase from the benchmark. Doing this for all nine combinations of G-{Crd, Red, Ref}/S-{Crd, Red, Ref}, we obtain the percentage changes shown in Table 7.

The greatest positive change in RLV can be achieved when we address a goods-related complaint with redelivery and a service-related issue with credit. As the results suggest, we observe a 3.41% increase in customers’ average RLV. This finding suggests that in the case of goods-related complaints, where resource matching is normally feasible, following the theory-driven norm (redelivery for goods) is advised to maximize the increase in customers’ RLV. Nonetheless, in Table 7, comparing the figures in the credit column with those in the redelivery and refund columns suggests that when deviating from the normative model of matching resources for service-related complaints, a credit, as a type of monetary compensation and a resource with a low level of particularism, would deliver the highest increase in customers’ RLV (money for service). Also, as the first and third rows of Table 7 show, in the case of goods-related complaints, if the company intends to deviate from the normative model and provide recovery with monetary compensation (instead of redeliveries), compared to refunds, offering store credits leads to a lower decrease in the RLV for affected customers.

Other Insights

Historical effects of complaint and recoveries on churn likelihood

We investigate the long-term historical effects of complaint and recovery. As results in Table 3 suggest, the coefficients for the ratios of recency weighted prior credits and redeliveries to recency weighted total number of transactions are both positive ( $β_{c h u r n, 13}$ and $β_{c h u r n, 14}$ ), indicating the deleterious effects of too many complaints (failures) in the past (Michel, Bowen, and Johnston 2009). However, their effects are quite different in magnitude. As these ratios are bounded at 0 and 1, we compute their effects for a 0.1 change in magnitude (which translates to 1 complaint in 10 transaction). On this basis, a 0.1 increase in the ratio of prior credits increases the probability of churn by 0.2% (computed from $β_{c h u r n, 13}$ = 0.618), while a 0.1 increase in the ratio of prior redeliveries increases churn by 0.9% (computed from $β_{c h u r n, 14}$ = 2.916). The ratio of prior refunds was insignificant.

Another interesting finding is that the immediate and historical effects of complaints have opposite effects. For instance, when credit is offered as compensation, it forces customers to place another order to redeem the credit voucher, thereby reducing the probability of immediate churn. However, the more frequent the occurrence of such complaints, the greater the likelihood that the customer will leave the company.

Timing distribution and hazard analysis

In our churn-purchase model, we observe significant heterogeneity in the scale parameter $λ$ (see Table 3). Its average value is 18.6, but it also substantially differs for business versus nonbusiness customers (14.2 vs. 19.6) and for members versus nonmembers (18.2 vs. 19.2), indicating business and member customers have shorter duration between purchases.

The value of the shape parameter $κ$ in the Weibull timing distribution is 1.094 (Table 3), which is significantly different from 1 (restricted to an exponential distribution). This is rather important as it affects the behavior of the hazard function:

h (t) = \frac{f (t)}{1 - F (t)} = \frac{f (t)}{S (t)},

where $f (t)$ is a probability density function, $F (t)$ is a cumulative density function, and $S (t) = 1 - F (t)$ is a survival function.

The exponential distribution with restricted $κ = 1$ has a constant hazard rate ( $h (t)$ that does not depend on t), but the Weilbull distribution with $κ > 1$ has a hazard function that monotonically increases with t. This fact is very critical for the company, as it indicates that with the passage of time, if a customer has not yet made a purchase, the probability of an order being placed increases. In other words, customers are increasingly more likely to purchase as time passes and the company should invest more resources to prevent customer defection.

Decay rates

The decay parameter $ρ = 0.983$ is less than 1 (Table 3), indicating that customers “forget” previous events. To quantify the effect of forgetfulness, we compute the time it takes for the effect of prior events to lose half their value, in other words, the t for which $ρ^{t} = 0.5$ . We find prior purchases are “half-forgotten” in 40 days.

Probability of complaint

The part of the churn-purchase model that captures the complaint probability is not the main focus of analysis; rather, it was required for model completeness. Therefore, we discuss only the main significant effects, while the complete list of estimates can be found in Appendix C.

On average, business customers are more likely to encounter service-related failures and complain ( $β_{c o m, 2}$ for service = 0.193), while, on average, participating in the company’s loyalty program proved to increase customers’ tendency for both types of complaints ( $β_{c o m, 1}$ = 0.190 and 0.218 for goods and service, respectively). Table C1 also suggests product substitution can lead to customers’ goods-related complaints ( $β_{c o m, 6}$ for goods = 0.115). Furthermore, baskets with a higher monetary value ratio of perishable products also proved to be more prone to goods-related complaints ( $β_{c o m, 7}$ for goods = 0.308).

Discussion and Conclusions

Summary of Findings

The high level of human involvement as well as the real-time nature of most service encounters make the latter extremely dependent on the characteristics of those involved, leaving almost no opportunity for “quality control” (Patterson, Cowley, and Prasongsukarn 2006). Hence, in the event of a failure, service customers evaluate companies based on the way they handle (recover) the situation and not necessarily on the failure incident per se. Nevertheless, while complaint recovery studies are generally favored in the literature, their contradictory findings in terms of the effectiveness of recovery practices (Cambra-Fierro, Melero, and Sese 2015) have affected the usefulness of their findings, especially for managers who wish to answer the burning issue of “what needs to be offered in which situation.”

Recent studies have tried to address this question from a resource exchange theory perspective (Foa and Foa 1974) and have prescribed matching the resource type of the complaint and recovery sides as the most effective complaint recovery strategy (Roschk and Gelbrich 2014, 2017). Nevertheless, in practice, we still observe companies’ tendency to deviate from this theory-driven model and adopt a mix of matching and nonmatching compensation for reasons stemming from operational constraints or strategic decisions. Building on resource exchange theory and focusing on tangible compensation (i.e., credits, refunds, and redeliveries), we develop a joint probabilistic model of customer purchases, occurrence of complaints, and customer churn to empirically investigate the types of recovery actions that should be offered for certain types of complaints (under both matching and nonmatching scenarios) and how particular strategies affect the churn/retention of customers as well as their long-term profitability. In particular, we aimed to contribute to the complaint recovery literature by introducing a framework for offering tangible compensation (i.e., Mix&Match) to address the significant question of which tangible compensation can best prevent customer churn and maximize customer lifetime value when we match or do not match the resources involved in complaint and recovery.

Churn/retention as the outcome variable

When churn is the outcome variable, the findings of our analyses suggest that matching complaint and recovery resources (e.g., redelivery for goods-related complaints) delivers the best outcome. This finding is in line with those of Roschk and Gelbrich (2014) and Smith, Bolton, and Wagner (1999). Furthermore, when deviating from the theory’s advice and using monetary compensation in the case of goods-related complaints, despite being less fungible, we find credits are more effective in preventing customer churn than refunds. This can be attributed to the nature of this type of monetary compensation, which ties the redemption of the offer to a future purchase (Roschk and Gelbrich 2014). Nonetheless, in the case of service-related complaints, when resources are not matched on the two sides of the exchange, we find only marginal support for credit’s effect on reducing churn probability.

Long-term profitability as the outcome variable

In terms of the long-term profitability of recovery practices, we also find that matching resources as a means of recovery delivers the highest increase in customers’ RLV. For goods-related complaints, the redelivery response creates the highest RLV. However, when deviating from the resource matching practice in recovering goods- and service-related complaints, offering the resource with the lowest particularism level (money in the form of credit) proved to be the most promising strategy for increasing customers’ long-term profitability. As Table 7 suggests, the focal company has an opportunity to increase the average RLV by 3.41% by simply following the suggested strategy of offering redelivery for goods-related complaints and credit for service-related complaints. Although it may look like a small figure, it translates to millions of dollars for companies as large as the focal one.

Managerial Implications

The unavoidability of service failure is now an accepted fact for both service companies and customers. This “reality” has motivated companies to invest more in effective ways of recovering from failures and compensating for complaints. In this endeavor, the main challenge for managers is to understand that not all complaints are the same; hence, one standard offer cannot recover all situations. This implies that having access to a “roadmap” of “when to offer what” is of great significance to service firm managers and can positively affect customer retention and long-term profitability (de Matos, Henrique, and Alberto Vargas Rossi 2007).

The current study contributes to the decision-making process concerning complaint recovery by shedding light on the effectiveness of various types of tangible compensation, in terms of both churn prevention and long-term profit generation (RLV). We consider two important situations where (1) the type of recovery offer matches the resource lost because of the failure and (2) the company decides not to provide matching compensation.

Our findings suggest that when using tangible compensation to deal with goods- and service-based complaints, the optimum approach to increase retention and long-term profitability is to offer, where possible, matching resources on the two sides of the exchange (e.g., redelivery for goods complaints) and to choose the recovery resource with the highest flexibility (universalism) where offering nonmatching compensation (e.g., money for service complaints). For companies that use a mix of matching and nonmatching compensation as their complaint recovery strategy, we strongly advise against deviating from the normative model of matching resources, where possible, as it harms customers’ lifetime value. Moreover, for companies that adopt the blanket approach of offering monetary compensation for the majority of complaints, our findings vouch for offering store credit (as opposed to refunds) to optimize customers’ RLV.

While these findings are based on limited types of resources for complaints and recoveries (i.e., goods and service for complaints, and money and goods for recoveries), the theoretical grounds on which our results are based suggest they will hold up in other cases of complaint recovery where other resources are involved. In this regard, our findings suggest that, to maximize customer retention and long-term profitability (RLV) in complaint recovery, companies should try to match the lost resources with similar ones when providing compensation. However, as this matching is not always possible, especially in the case of service-related failures, companies should attempt to recover the situation by offering monetary compensation in the form of credit. This offer gives affected customers a high level of freedom of choice and at the same time encourages them to return.

Our findings are particularly valuable for companies with complaint recovery strategies similar to this study’s focal company: following a blanket approach of addressing the majority of complaints with monetary compensation regardless of their nature (Table 2). As our analysis suggests, such an approach will not deliver the optimum customer retention and profitability.

The results of our financial analysis provide managers with the boundary effects of potential recovery strategies. Knowing these boundaries, managers can better plan and budget for activities associated with handling customer complaints and their resolution. For instance, the results show that for goods-related complaints, redelivery is the best recovery action. However, this type of resolution may incur the extra transport cost of another delivery. Managers now have more information, which enables them to decide whether redelivery is a financially better option than offering a credit voucher.

Finally, companies are advised to ensure prevention of repeated service failures, particularly for those customers who have chosen to remain loyal despite their previous negative experience. Our findings suggest one way to achieve this is to lower product substitution incidents by better predicting demand. Also, when the product ordered is not available, companies can give customers the opportunity to choose alternatives as opposed to deciding on the replacement product without their consent. In addition, companies could establish a procedure to better monitor “outgoing baskets,” especially when the order is placed by a customer who is a member of the company’s loyalty program, as such customers have higher chances of raising both goods- and service-related complaints.

Limitations and Further Research

The findings of this study are derived from one dataset for a specific context (online grocery retail) with limited types of compensation (i.e., redeliveries, refunds, and credits), where the focal company shows a tendency to offer store credit in the majority of complaint incidents (see Table 2). Despite the strength and robustness of the effects we found, one should be cautious about applying our findings in other contexts. A possible direction for future research is to conduct similar analyses of a new data set with a broader variety of complaint and recovery options, especially when high involvement products are concerned. The model can also be tested in cases where there is a bias in favor of other types of compensation (e.g., refunds).

Furthermore, because in our data the company recovered all previous complaints, it was impossible to compare the effects of the proposed recovery strategies with those of nonrecovery cases. In addition, due to the lack of information on the severity of failures (Weun, Beatty, and Jones 2004) in this study’s data set, this determining factor could not be incorporated into the empirical model. Future research can shed light on how severity factors can potentially influence the way(s) customers react to different recovery offers as discussed here, under different situations.

Supplemental Material

Supplemental Material, JSR-17-299.R4_-_Executive_Summary - Mix&Match: A Resource-Based Complaint Recovery Framework for Tangible Compensation

Supplemental Material, JSR-17-299.R4_-_Executive_Summary for Mix&Match: A Resource-Based Complaint Recovery Framework for Tangible Compensation by Stanislav Stakhovych and Ali Tamaddoni in Journal of Service Research

Footnotes

Appendix A

Appendix B

Appendix C

Table C1.

Model Parameter Estimates for $π_{c o m}$ .

Parameter	Description	Estimate	SE	Estimate	SE
Parameter	Description	Goods		Service
$π_{c o m}$	Probability of complaint^a	0.050		0.002
$β_{c o m, 0}$	Intercept	−3.456***	.019	−6.653***	.100
$β_{c o m, 1}$	Member	0.190***	.009	0.218***	.051
$β_{c o m, 2}$	Business	−0.032*	.018	0.193***	.090
$β_{c o m, 3}$	Member × Business	−0.075***	.024	−0.134	.118
$β_{c o m, 4}$	N of SKUs	0.012***	.000	0.010***	.001
$β_{c o m, 5}$	Customer experience	−0.023***	.004	−0.070***	.020
$β_{c o m, 6}$	Substitution	0.115***	.009	0.025	.046
$β_{c o m, 7}$	Perishable	0.308***	.028	0.168	.142
$β_{c o m, 8}$	Tuesday	−0.074***	.014	−0.117	.075
$β_{c o m, 9}$	Wednesday	−0.116***	.015	−0.075	.076
$β_{c o m, 10}$	Thursday	−0.088***	.014	−0.034	.073
$β_{c o m, 11}$	Friday	−0.138***	.014	−0.122*	.074
$β_{c o m, 12}$	Saturday	−0.222***	.016	−0.243***	.085
$β_{c o m, 13}$	Sunday	−0.351***	.026	−0.084	.124
$β_{c o m, 14}$	Quarter 1	−0.003	.013	−0.133**	.066
$β_{c o m, 15}$	Quarter 2	−0.035***	.013	−0.146**	.065
$β_{c o m, 16}$	Quarter 3	−0.049***	.012	−0.096	.064
$β_{c o m, 17}$	Prior purchases	−0.001	.002	0.015*	.009

^a Average across i and j.

*p < .10. **p < .05. ***p < .01.

Note. SE = standard error.

Appendix D

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Stanislav Stakhovych

Ali Tamaddoni

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Ansari

Asim

Mela

Carl F.

Neslin

Scott A.

(2008), “Customer Channel Migration,” Journal of Marketing Research, 45 (1), 60–76.

Bitner

Mary Jo

Booms

Bernard H.

Tetreault

Mary Stanfield

(1990), “The Service Encounter: Diagnosing Favorable and Unfavorable Incidents,” Journal of Marketing, 54 (1), 71–84.

Cambra-Fierro

Jesus

Melero

Iguacel

Javier Sese

(2015), “Managing Complaints to Improve Customer Profitability,” Journal of Retailing, 91 (1), 109–124.

Chandon

Pierre

Wansink

Brian

Laurent

Gilles

(2000), “A Benefit Congruency Framework of Sales Promotion Effectiveness,” Journal of Marketing, 64 (4), 65–81.

Che

Hai

Chen

Xinlei

Chen

Yuxin

(2012), “Investigating Effects of Out-of-Stock on Consumer Stockkeeping Unit Choice,” Journal of Marketing Research, 49 (4), 502–513.

Consumer Federation of America and North American Consumer Protection Investigators. (2015), 2014 Consumer Complaint Survey Report. Washington, DC: Consumer Federation of America and North American Consumer Protection Investigators.

Danaher

Peter J.

Smith

Michael S.

Ranasinghe

Kulan

Danaher

Tracey S.

(2015), “Where, When, and How Long: Factors that Influence the Redemption of Mobile Phone Coupons,” Journal of Marketing Research, 52 (5), 710–725.

del Río-Lanza

Ana Belén

Vázquez-Casielles

Rodolfo

Díaz-Martín

Ana M.

(2009), “Satisfaction with Service Recovery: Perceived Justice and Emotional Responses,” Journal of Business Research, 62 (8), 775–781.

de Matos

Celso Augusto

Henrique

Jorge Luiz

Rossi

Carlos Alberto Vargas

(2007), “Service Recovery Paradox: A Meta-Analysis,” Journal of Service Research, 10 (1), 60–77.

10.

Dorsch

Michael J.

Törnblom

Kjell Y.

Kazemi

Ali

(2017), “A Review of Resource Theories and Their Implications for Understanding Consumer Behavior,” Journal of the Association for Consumer Research, 2 (1), 5–25.

11.

Fader

Peter S.

Hardie

Bruce G.S.

(2015), “Simple Probability Models for Computing CLV and CE,” in Handbook of Research on Customer Equity in Marketing, Kumar

Shah

Denish

, eds. Cheltenham, UK: Edward Elgar, 77–100.

12.

Fader

Peter S.

Hardie

Bruce G. S.

(2009), “Probability Models for Customer-Base Analysis,” Journal of Interactive Marketing, 23 (1), 61–69.

13.

Fader

Peter S.

Hardie

Bruce G. S.

Lok Lee

(2005), “RFM and CLV: Using Iso-Value Curves for Customer Base Analysis,” Journal of Marketing Research, 42 (4), 415–430.

14.

Foa

Edna B.

Foa

Uriel G.

(2012), “Resource Theory of Social Exchange,” in Handbook of Social Resource Theory, Törnblom

Kjell

Kazemi

Ali

, eds. New York: Springer-Verlag, 15–32.

15.

Foa

Uriel G.

Foa

Edna B.

(1974), Societal Structures of the Mind. Oxford, UK: Charles C Thomas.

16.

Foa

Uriel G.

(1971), “Interpersonal and Economic Resources,” Science, 171 (3969), 345–351.

17.

Gelbrich

Katja

Gäthke

Jana

Grégoire

Yany

(2015), “How Much Compensation Should a Firm Offer for a Flawed Service? An Examination of the Nonlinear Effects of Compensation on Satisfaction,” Journal of Service Research, 18 (1), 107–123.

18.

Gelbrich

Katja

Roschk

Holger

(2011), “A Meta-Analysis of Organizational Complaint Handling and Customer Responses,” Journal of Service Research, 14 (1), 24–43.

19.

Goodwin

Cathy

Ross

Ivan

(1990), “Consumer Evaluations of Responses to Complaints: What’s Fair and Why,” Journal of Consumer Marketing, 7 (2), 39–47.

20.

Gupta

Sunil

Zeithaml

Valarie A.

(2006), “Customer Metrics and Their Impact on Financial Performance,” Marketing Science, 25 (6), 718–739.

21.

Holloway

Betsy B.

Beatty

Sharon E.

(2003), “Service Failure in Online Retailing: A Recovery Opportunity,” Journal of Service Research, 6 (1), 92–105.

22.

Kelley

Scott W.

Douglas Hoffman

Davis

Mark A.

(1993), “A Typology of Retail Failures and Recoveries,” Journal of Retailing, 69 (4), 429–452.

23.

Khamitov

Mansur

Grégoire

Yany

Suri

Anshu

(2020), “A Systematic Review of Brand Transgression, Service Failure Recovery and Product-Harm Crisis: Integration and Guiding Insights,” Journal of the Academy of Marketing Science, 48, 519–542. https://doi.org/10.1007/s11747-019-00679-1.

24.

Knox

George

Oest

Rutger Van

(2014), “Customer Complaints and Recovery Effectiveness: A Customer Base Approach,” Journal of Marketing, 78 (5), 42–57.

25.

Lopes

Evandro Luiz

Silva

Marcos Antonio da

(2015), “The Effect of Justice in the History of Loyalty: A Study in Failure Recovery in the Retail Context,” Journal of Retailing and Consumer Services, 24 (3), 110–120. https://doi.org/10.1016/j.jretconser.2015.03.003

26.

Mahajan

Jayashree

Churchill

Gilbert A.

(1990), “Alternative Approaches for Investigating Contingency-Based Organizational Predictions in Personal Selling,” International Journal of Research in Marketing, 7 (2-3), 149–169.

27.

Maxham

James G., III

. (2001), “Service Recovery’s Influence on Consumer Satisfaction, Positive Word-of-Mouth, and Purchase Intentions,” Journal of Business Research, 54 (1), 11–24.

28.

Michel

Stefan

Bowen

David

Johnston

Robert

(2009), “Why Service Recovery Fails: Tensions among Customer, Employee, and Process Perspectives,” Journal of Service Management, 20 (3), 253–273.

29.

Mittal

Vikas

Kamakura

Wagner A.

(2001), “Satisfaction, Repurchase Intent, and Repurchase Behavior: Investigating the Moderating Effect of Customer Characteristics,” Journal of Marketing Research, 38 (1), 131–142.

30.

Orsingher

Chiara

Valentini

Sara

Angelis

Matteo

(2010), “A Meta-Analysis of Satisfaction with Complaint Handling in Services,” Journal of the Academy of Marketing Science, 38 (2), 169–186.

31.

Papies

Dominik

Ebbes

Peter

Heerde

Harald J. Van

(2016), “Addressing Endogeneity in Marketing Models,” in Advanced Methods in Modeling Markets (Series in Quantitative Marketing), Leeflang

Peter S. H.

Wieringa

Jaap E.

Bijmolt

Tammo H. A.

Pauwels

Koen H.

, eds. Berlin, Germany: Springer, 581–627.

32.

Patterson

Paul G.

Cowley

Elizabeth

Prasongsukarn

Kriengsin

(2006), “Service Failure Recovery: The Moderating Impact of Individual-Level Cultural Value Orientation on Perceptions of Justice,” International Journal of Research in Marketing, 23 (3), 263–277.

33.

Prasongsukarn

Kriengsin

Patterson

Paul G.

(2012), “An Extended Service Recovery Model: The Moderating Impact of Temporal Sequence of Events,” Journal of Services Marketing, 26 (7), 510–520.

34.

Pugh

Harrison B.

Brady

Michael K.

Hopkins

Lucas M.

(2018), “A Customer Scorned: Effects of Employee Reprimands in Frontline Service Encounters,” Journal of Service Research, 21 (2), 219–234.

35.

Roschk

Holger

Gelbrich

Katja

(2017), “Compensation Revisited: A Social Resource Theory Perspective on Offering a Monetary Resource after a Service Failure,” Journal of Service Research, 20 (4), 393–408.

36.

Roschk

Holger

Gelbrich

Katja

(2014), “Identifying Appropriate Compensation Types for Service Failures: A Meta-Analytic and Experimental Analysis,” Journal of Service Research, 17 (2), 195–211.

37.

Rust

Roland T.

Zahorik

Anthony J.

Keiningham

Timothy L.

(1995), “Return on Quality (ROQ): Making Service Quality Financially Accountable,” Journal of Marketing, 59 (2), 58–70.

38.

Simons

Jacob V.

Kraus

Mark E.

(2005), “An Analytical Approach for Allocating Service Recovery Efforts to Reduce Internal Failures,” Journal of Service Research, 7 (3), 277–289.

39.

Smith

Amy K.

Bolton

Ruth N.

Wagner

Janet

(1999), “A Model of Customer Satisfaction with Service Encounters Involving Failure and Recovery,” Journal of Marketing Research, 36 (3), 356–372.

40.

Smith

Amy K.

Bolton

Ruth N.

(1998), “An Experimental Investigation of Customer Reactions to Service Failure and Recovery Encounters: Paradox or Peril?” Journal of Service Research, 1 (1), 65–81.

41.

Terza

Joseph V.

Basu

Anirban

Rathouz

Paul J.

(2008), “Two-Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling,” Journal of Health Economics, 27 (3), 531–543.

42.

Van Diepen

Merel

Donkers

Bas

Franses

Philip Hans

(2009), “Dynamic and Competitive Effects of Direct Mailings: A Charitable Giving Application,” Journal of Marketing Research, 46 (1), 120–133.

43.

Van Vaerenbergh

Yves

Varga

Dorottya

Keyser

Arne De

Orsingher

Chiara

(2019), “The Service Recovery Journey: Conceptualization, Integration, and Directions for Future Research,” Journal of Service Research, 22 (2), 103–119.

44.

Weun

Seungoog

Beatty

Sharon E.

Jones

Michael A.

(2004), “The Impact of Service Failure Severity on Service Recovery Evaluations and Post-Recovery Relationships,” Journal of Services Marketing, 18 (2), 133–146.

45.

Wooldridge

J. M.

(2010), Econometric Analysis of Cross Section and Panel Data, 2nd ed. Cambridge, MA: MIT Press.

46.

Zhu

Zhen

Sivakumar

Parasuraman

Anantharanthan

(2004), “A Mathematical Model of Service Failure and Recovery Strategies,” Decision Sciences, 35 (3), 493–525.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.01 MB