Service problems and recovery strategies: An experiment

Service problems and recovery strategies: An experiment

Terrence J Levesque


This experiment examines the effectiveness of recovery strategies after a service failure on customer loyalty and complaint intentions. Respondents encountered different core failures in terms of problem severity (denial or delay) and criticality levels (high or low). The results suggest the effectiveness of service recovery strategiesassistance (fixing the problem) and/or compensation (defraying the costs incurred)varied depending on the type of service, problem severity, and criticality levels. The implication is that recovery strategies need to be matched to the specific incident. Service firms should focus on avoiding or reducing core failures. Getting it right the first time is the best strategy.


La presente recherche examine au moyen dune experience l’efficacite de differentes strategies de recuperation sur la fidelite et les intentions de porter plainte de la clientele a la suite dune defaillance de service. Les participants ont ete confrontes d differences defaillances de service en termes de gravite (interruption du service ou delai) et de niveau critique (eleve ou faible). Les resultats indiquent que l’efficacite des strategies de recuperation-aide technique (resolution du probleme) et/ou compensation financiere (defraiement des coats encourus)-varie en fonction du genre de service, de la gravite du probleme et des niveaux critiques. Les resultats de l’etude laissent supposer que les strategies de recuperation doivent etre associees a un incident specifique. Les entreprises de services doivent mettre l’accent sur 1’evitement ou la reduction des defaillances. La meilleure strategie consiste encore a donner le service correctement.

Increased customer loyalty is a critical driver of a firm’s long-term financial performance (Jones & Sasser, 1995). Customer dissatisfaction reduces loyalty and erodes the firm’s reputation. A major cause of dissatisfaction is unsatisfactory problem resolution (Hart, Heskett, & Sasser, 1990) and up to 50% of customers who experience problems are not satisfied with the recovery strategy (Best & Andreasen, 1976; Zeithaml, Berry, & Parasuraman, 1990).

When customers experience a problem with a service provider, they may respond by switching to a new supplier (exit), attempting to remedy the problem by complaining (voice), or staying with the supplier anticipating that “things will get better” (loyalty) (Hirschman, 1970). They may also talk to other consumers about the experience, and this negative word of mouth can also influence a firm’s profits and reputation (Richins, 1987; Singh, 1990; Singh & Wilkes, 1996). It is thus important for the firm to develop a recovery strategy to respond to service problems. Interestingly, in spite of the importance of service recovery, according to Kelley and Davis, “a dearth of empirical research confines any theoretical discussion to anecdotal reports” (1994, p. 52).

The objective of this paper is to provide empirical evidence on the effects on customers’ future intentions toward the provider of different recovery strategies after a service failure (Figure 1). Using hospitality industry scenarios, subjects were presented with a new purchase situation where the provider’s quality was known only by reputation. Purchase decisions varied according to the importance of the purchase to the consumer. Purchase outcomes included service failures where service was either denied or delayed. The service provider offered the following recovery strategies: (a) apology only, (b) apology and assistance, (c) apology and compensation, or (d) apology, assistance, and compensation. Subjects’ intentions toward the provider were measured, including their propensity to voice their dissatisfaction and the likelihood that they would continue to do business with the provider. The study provides insight into the threats to the loyalty-building process and an opportunity to determine the value of a contingency approach in designing a service recovery strategy.


Service firms spend substantial resources to deliver a consistent offering that meets customers’ expectations. Two major challenges to maintaining consistency are variability, due to the human element involved in delivery, and inseparability, because the customer often has to be present to receive the service. Service firms recognize that failures will occur and develop recovery strategies to meet the challenges (Hart et al., 1990). Of interest here are core service failures that are manifest problems caused by the service provider’s failure to deliver a service that had been contracted for (e.g., delay or denial).

In general, service recovery strategies can consist of three distinct actions, either alone or in combination: (a) apologize (acknowledging the problem), (b) assistance (fixing the problem), and (c) compensation (paying for the “trouble” costs of the problem). The recovery strategy plays a role in the customer’s future intentions toward the provider. For both practitioners and researchers, it is important to understand which service recovery strategy is most effective in a given situation.

Customers’ response to service failures include: loyalty, exit, and voice (complaining to the service provider, friends and relatives, or third parties) (Day & Landon, 1977; Hirschman, 1970; Singh, 1988; Singh & Wilkes, 1996). Considerable research has been conducted into complaining and switching behaviour (see Richins, 1987; Singh & Howell, 1985; and Singh & Wilkes, 1996, for reviews). Few research studies have investigated the extent to which service recovery strategies moderate these customer responses.

Situation variables may also influence customers’ responses to service problems. Customers are more likely to engage in switching and complaining to the service provider and others as problem severity increases (Keaveney, 1995; Kelley & Davis, 1994; Richins, 1983, 1987; Singh & Wilkes, 1996). Different criticality levels, when a purchase situation is more important or critical (Ostrom & Iacobucci, 1995), may also influence their responses; they are more likely to complain when problems are encountered in high criticality situations (Richins, 1983, 1987; Webster & Sundaram, 1998). The following sections elaborate on service recovery strategies, problem severity, and criticality.

Service Recovery Strategies

Service recovery strategies describe the actions that service providers take to respond to defects or failures (Gronroos, 1988). The most common and frequently used actions are apology, assistance, and/or compensation (Bitner, Booms, & Tetreault 1990; Hart et al., 1990; Hoffman, Kelley, & Rotasky, 1995; Kelley, Hoffman, & Davis, 1993). Their effectiveness depends on the situation and is influenced by such factors as problem severity, criticality, and the type of service. Effectiveness also depends on how the contact employee handles the problem: responsiveness, empathy, and understanding improve the effectiveness of the strategy (Bitner et al., 1990; Hart et al., 1990; Smith, Bolton, & Wagner, 1999). Thus, both what is done (e.g., compensation) and how it is done (e.g., empathetically, quick response) contribute to the effectiveness of the recovery strategy. For this investigation, the primary focus is on what was done in terms of the relative effectiveness of assistance and compensation. Apology, a how strategy, is also included because it is the minimum action that can be taken when a problem occurs (Bitner et al., 1990) and is recommended as a requisite for service recovery (Hart et al., 1990). In this investigation, apology is regarded as the baseline and will be offered in all recovery strategies.

Underlying the effectiveness of recovery strategies is the concept of exchange. The service failure and recovery can be viewed as an exchange, in which the customer experiences a loss due to the service failure, and the firm attempts to provide a gain, in the form of a service recovery, to make up for the customer’s loss (Smith et al., 1999). Customer evaluations of the service failure and recovery depend on the type and amount of resources lost and gained during the exchange.

While an apology is better than no apology (Smith et al., 1999), an apology alone is relatively ineffective when a customer encounters a service failure (Goodwin & Ross, 1992; Hoffman et al., 1995; Webster & Sundaram, 1998). Typically, customers expect some gain (e.g., assistance, compensation) for their loss (service failure) (Smith et al., 1999; Tax, Brown, & Chandrashekaran, 1998). An apology offers little gain but may be effective when minor service problems are encountered. As this study examined core service failures (large loss), it would be expected that offering an apology only (small gain) would not be an effective recovery strategy.

Assistance involves taking actions to rectify the problem. Assistance is possibly the most effective single recovery strategy, because it can bring the customer back to the original purpose of buying the service. In the case of certain manifest core failures (e.g., denial or unavailability of service), it is argued that the service firm has little leeway; it must fix the problem quickly (Parsuraman, Berry, & Zeithaml, 1991). Here, the gain is fulfilling the basic promise, which may equal the loss from the failure.

Compensation involves monetary payment for the inconvenience the customer has experienced and may be required if the failure cannot be fixed (e.g., no room available). In terms of gain and loss, increasing compensation should lead to greater satisfaction with the service recovery. However, in particular cases, higher compensation may lead consumers to feel less satisfied because they were over-rewarded (Smith et al., 1999). While consumers want a gain in this loss situation, and increasing the gain through compensation and assistance should improve satisfaction, there may be an upper limit to the gain.

Of interest here was the relative effect of assistance versus compensation. Empirical results are mixed on the relative effectiveness of assistance and compensation. In a study of critical incidents in a retail setting Kelley et al. (1993) found high levels of satisfaction and retention with assistance in the form of replacement or correction. Compensation, in the form of discounts, reported nominally higher satisfaction ratings. In another critical incident study of restaurant service, compensation received higher satisfaction and retention ratings than assistance (Hoffman et al., 1995).

Two studies experimentally manipulated service failures and recoveries to investigate the relative effects of apology, compensation, and assistance. Darida, Levesque, and McDougall ( 1996) found that apology only was the least effective strategy, and significant improvements in respondents’ loyalty were found when assistance and compensation were added to the recovery effort. Both problem severity and criticality moderated the recovery efforts. The second study examined the relative effectiveness of apology, compensation (a 25% or 50% discount), and a form of assistance (re-perform the service) after a service failure involving delay (Webster & Sundaram, 1998). Both recovery efforts and criticality had a significant effect as well as an interaction effect on loyalty intentions. In low criticality situations, compensation was most effective, followed by assistance. In high criticality situations, assistance was most effective, followed by compensation. Clearly, the exchange, loss versus gain, was influenced by criticality with respect to assistance versus compensation. Comparing the two levels of compensation (high or low) revealed that the higher level was always more effective, supporting the exchange concept. Apology was the least effective strategy, regardless of criticality.

A third study, using two service settings, examined four service recovery strategies (compensation, response speed, apology, and recovery initiation) after a service failure (Smith et al., 1999). The failure was either a core failure (e.g., denial) or process failure (e.g., inattentive service) and either high or low problem severity (denial or wrong room type). Of interest here was the finding that distributive justice, a major contributor to satisfaction, was strongly related to higher compensation levels. As well, when customers experienced core failures, they assigned higher fairness values (which were linked to satisfaction) with compensation and quick action. Furthermore, problem severity provided mixed results; in some cases, high severity (versus low severity) reduced the effectiveness of recovery strategies; in others it had no effect. Overall, while the effectiveness of the recovery strategies varied depending on the service setting and problem severity, compensation was a strong contributor to satisfaction.

A major implication of these studies is that relative effectiveness of the recovery strategies was situationspecific: factors such as problem severity and criticality had an impact on the effectiveness of specific recovery strategies. Furthermore, the relative effectiveness of assistance versus compensation has been mixed. Finally, offering higher versus lower levels of compensation was more effective, and offering both assistance and compensation versus either assistance or compensation was more effective.

Based on the above it is hypothesized that (see Figure 1 ):

Hla: When a core service failure occurs, offering an apology only is less effective than assistance or compensation in terms of improving customers’ future intentions toward the service provider.

Hlb: When a core service failure occurs, there will be no difference between assistance versus compensation in terms of improving customers’ future intentions toward the service provider.

Hlc: When a core service failure occurs, a service recovery strategy that offers assistance or compensation will be less effective than one that combines assistance and compensation in terms of improving customers’ future intentions toward the service provider.


Criticality is the importance of the service to the consumer (Ostrom & lacobucci, 1995). By extension, criticality is the perceived importance of successful service delivery in a given service encounter (Webster & Sundaram, 1998). When a service purchase was critical, consumers were likely to see service failure as more serious than when the purchase was less critical (Ostrom & Iacobucci, 1995). When consumers made choices in highly critical situations, they were more likely to regard price as less important than in situations that were not critical. As well, quality was more important in highly critical than in less critical purchases (Ostrom & Iacobucci, 1995). Thus, consumers are more likely to purchase high-criticality services from a high-end service provider (e.g., higher prices and quality) than a lower-end provider to ensure, in part, that the service will not fail.

Both the importance of the service and the purchase from a high-end user were employed to create the highcriticality factor in this study. For example, a subject would purchase a service for a special occasion from a provider that promised first-class service (high criticality).

The available evidence suggests that, based on the exchange concept, the amount of the customer’s loss depends on the criticality or consequences of failure. High-criticality situations (higher loss) are likely to lead to greater dissatisfaction than low-criticality situations (lower loss), given similar recovery strategies (gain). When customers encountered service failures in highcriticality situations, they were more likely to have lower satisfaction and loyalty, regardless of the recovery strategy (Darida et al., 1996; Webster & Sundaram, 1998). It is anticipated that criticality levels will affect future intentions toward the provider as proposed in the following hypothesis (see Figure 1 ):

H2: When a core service failure occurs in a low-criticality situation versus a high-criticality situation, for a given service recovery strategy customers’ future intentions toward the service provider will be more positive.

Service Failure

When asked to identify service encounters that resulted in quite dissatisfying experiences, customers frequently mentioned core service failures (Bitner et al., 1990; Hoffman et al., 1995; Kelley et al., 1993). Core failures are situations where the customer does not receive the basic service promised by the provider. Core service failures, which include “lost” hotel reservations and reserved tables that were occupied, were a major reason for customers switching service providers (Keaveney, 1995). Thus, core service failures are relatively serious, and consumers expect the service provider to rectify the problem (Hart et al., 1990). Furthermore, as problem severity increases, the customer will anticipate a recovery strategy that reflects the severity of the problem (Kelley & Davis, 1994). Core service failures are large loss situations where the recovery strategy required (gain) would be higher than situations involving minor problems.

Of interest here are two types of core failures, unavailability or denial of service (e.g., no hotel room even though customer has a reservation), and delay (e.g., table not ready even though customer has a confirmed reservation for 6:00 p.m.). Both are core failures because the provider has established manifest contracts with the customers that have been broken. However, the severity of the failures are different; denial is considerably more serious than delay. Denial is a total breach of the basic contract (e.g., a lost hotel reservation) and, relative to delay, it is a higher loss situation.

Delay can also be a core failure, depending on the circumstances. If the customer, who has set an appointment time with the service firm, arrives at the scheduled time and then waits, a failure has occurred. This investigation examines this type of failure, referred to as a pre-process wait (Taylor, 1994). Pre-process waits tend to be viewed as more unpleasant than in-process waits (Dube-Rioux, Schmitt, & Leclerc, 1988). In a pre-process wait, longer delays resulted in lower evaluation of service (Taylor, 1994).

As noted earlier, it was found that the relative effectiveness of assistance versus compensation was moderated by criticality and problem severity (Darida et al., 1996). Assistance was more effective in high-criticality situations, and compensation was more effective in a low-criticality service delay situation (Webster & Sundaram, 1998). As well, problem severity produced mixed results; in some cases high severity (versus low severity) reduced the effectiveness of recovery strategies, in others it had no effect (Smith et al., 1999).

Based on the preceding discussion, the effect of problem severity on future intentions toward the provider is hypothesized as follows (see Figure 1 ):

H3: When a high-severity core service failure versus a low-severity core service failure occurs, for a given recovery strategy, customers’ future intentions toward the service provider will be more negative.

It is also anticipated that problem severity will moderate the effects of the service recovery strategies as proposed in the following hypotheses:

H4a: When a high-severity core service failure occurs, the incremental improvement in customers’ future intentions toward the service provider will be higher for assistance than for compensation.

H4b: When a low-severity core service failure occurs, the incremental improvement in customers’ future intentions toward the service provider will be higher for compensation than for assistance.


The Study Design

The study design was guided by five decisions. First, because the purpose of the research was to investigate the relative effectiveness of recovery strategies in different situations, an experiment in the hospitality industry was conducted. Consumers are generally familiar with hotels and restaurants and this service category is typically characterized by variability in the quality of service provided (Boulding, Kalra, Staelin, & Zeithmal, 1993). Second, the decision was made to examine core service failures because they are manifest, they are common, they cause customers to switch firms, and they present a challenge for managers. Third, interviews were conducted with service managers in the hospitality sector to identify realistic service failures and service recovery strategies that had been used to respond to these types of service failures. This advice increased the external validity of the experiment. Fourth, the cooperation of a hotel was obtained, and the survey was administered to hotel guests. This provided a sample that helped ensure realism and again increased external validity (McCullough, 1995). Finally, manipulation checks were conducted before finalizing the design to ensure that both problem severity (two levels) and criticality levels (two levels) were appropriate,

To test the hypotheses, the study had two experiments that used manipulations of hotel and restaurant scenarios. A subject was presented with two separate service failure scenarios, one involving hotel reservations and the second involving restaurant reservations. The research design for each scenario (hotel and restaurant) was a 4 (levels of recovery) by 2 (levels of criticality) by 2 (levels of problem severity) between subjects design. Table 1 provides the design for the hotel scenario and Appendix 1 provides an example of the scenarios used in the study. The restaurant scenario was identical in terms of design but varied in terms of operational definitions.

The service recovery strategy was operationalized as a four-level factor with the following levels: (a) apology only, (b) compensation, (c) assistance, and (d) compensation and assistance. Assistance and compensation were matched to problem severity. For example, with “delay” in the restaurant, compensation was a voucher for complimentary refreshments; with “denial,” compensation was a voucher for a complimentary meal at another restaurant. “Apology only” was operationalized as the service provider saying, “There is nothing I can do to help you.” As noted earlier, all recovery strategies included the apology, “I am very sorry for the inconvenience.” Again, the guidance of service managers in the hospitality industry helped shape the recovery strategies to match the problem severity levels.

Two manifest core service failures-denial (no reservation in a fully booked hotel/restaurant) and delay (room/table not ready at the reserved time)-were used to represent problem severity. Discussions with service managers in the hospitality industry helped to ensure that the descriptions were representative of actual situations.

The two criticality levels for the hotel scenarios were operationalized as a destination hotel for a 4-day stay (high) and a no-frills motel chain on the way to a vacation (low). For the restaurant scenarios, the levels were operationalized as a family celebration at a restaurant known for excellent food and service in a formal setting (high) and a casual family meal in a restaurant known for good food and service in a relaxed setting (low).

Manipulation Check of Factors

Before finalizing the experimental design, separate manipulation checks were performed for (a) criticality and (b) problem severity. Briefly, the criticality results showed that respondents viewed the descriptions of the destination hotel (special-occasion restaurant) and the no-frills motel (casual restaurant) to be different and in the predicted direction. Hence, the factors represented valid manipulations. For problem severity, the results showed that service denial is more annoying than service delay. While the restaurant situation was less pronounced, the manipulation check was considered acceptable for the experiment. Appendix B provides more details on the manipulation checks.

The Questionnaire

In the study, each subject was assigned to one of the 16 treatment combinations. The questionnaire instructions asked subjects to assume that the situation (scenario) had just happened to them and they were asked how they would react to it. One of the scenarios and the eight items used to measure subjects’ reactions are reproduced in Appendix A.

Data Collection

The sampling frame for the study was the guest list for a 1,600-room hotel located in downtown Toronto, Ontario. The questionnaires were distributed in the rooms over a 2-week period to capture a broad range of hotel guests. An incentive, in the form of a draw for cash prizes, was offered to encourage participation. A total of 1,811 questionnaires were distributed and 636 were completed, for an overall response rate of 35%. Prior to the analysis, the questionnaires were screened for missing data, response patterns that indicated low discrimination ability, and logic inconsistencies.

Sample Characteristics

Briefly, the demographic characteristics of the sample were the following: the gender split was 52% male, 48% female; the majority of respondents (59%) were between 30 and 49 years of age; annual family income was bimodal with approximately 25% of the sample between $50,000 and $70,000 and 25 % over $ 100,000. Over 60% of the respondents had an undergraduate or a graduate degree. Relative to larger populations, the sample was more affluent and had a higher education level. This was expected, given that the hotel is part of a well-known chain that targets business travellers and tourists seeking amenities.

For the hotel scenario, the percentage of respondents who had experienced each of the two service failures was the following: 40% of respondents experienced “no room even with a reservation” and 70% of respondents experienced “room not ready.” The relatively high proportion of respondents having experienced these two problems indicates that the problems are relatively common. Thus, the study has presented problems that respondents might expect to occur.


Analysis of Future Intentions

The study operationalized future intentions by using eight items (7-point scales) that measured subjects’ likelihood of engaging in exit, voice, and loyalty behaviours ranging from 1 (not likely at all) to 7 (extremely likely).

The items were drawn from a previous study on behavioural intentions in service industries (Zeithaml, Berry, & Parasuraman, 1996). The items selected were most appropriate to the hospitality situations; they also span the exit, voice, and loyalty behaviour typology identified by Hirschman (1970). Table 2 summarizes the distribution of the scale scores for each item. It was apparent from the means that the experimental conditions provoked definite responses tending toward the unfavourable ends of the scales from the provider’s viewpoint.

Generally, the score distributions were quite skewed. For the favourable items, the distributions were rightskewed, with most of the scores tending toward the riot at all likely response. The distributions were left-skewed for unfavourable items, with most scores tending toward extremely likely. The normal range of the Pearsonian skewness coefficient is [-1, 1]. Muthen and Kaplan ( 1985) suggested that maximum likelihood or GLS parameter estimation for data falling in the normal range is robust. However, for data outside the range, there is a risk estimates will be biased. In this case, half the items exhibited skewness outside the normal range, and most were close to the limits. There was a risk that analysis of the original scales would obscure important structures in the responses because of the influence of observations in the tails of the distribution. Analysis of the transformed data revealed important interactions among the experimental factors that were not detected using the original scales.

Transformations are more effective at inducing normality when the coefficient of variation is large, in excess of 0.25 (Afifi & Clark, 1990). The coefficients of variation ranged from 0.33 to 0.45 for the left-skewed unfavourable items and from 0.66 to 0.74 for the rightskewed favourable items. To avoid the consequences of this departure from normality, the data was transformed.

The original items were transformed in a two-step procedure. Because it is desirable that all scales be transformed by the same function, the left-skewed unfavourable items were reversed, producing rightskewed scales. All scales were then transformed by the monotonic function y = (x -1)Ix, where x is the original scale. The transformation had the desired effect of lowering the skewness coefficients and the consequent effect of increasing the variation of the scores in relation to their means. The transformed scales were used in the remainder of the analysis.

With the transformed scales, all of the correlations were significantly different from zero at the 1 % level. The correlations among the favourable items were generally higher compared to the unfavourable items. It was also apparent that items 2 and 4, “switch to a competitor” and “do less business,” were more closely associated with favourable items. There were significant low to moderate correlations between favourable and unfavourable items. This suggests that the responses were not manifestations of a simple underlying dichotomy, and that at least two latent constructs were present in the data.

Principal components were extracted from the transformed scales. Two potential factors were identified from the initial solution. These were subjected to an oblique rotation to account for the likely correlation between the underlying factors evident in the correlation analysis. The two-factor solution explained 73% of the variation in the transformed scales for both the hotel and the restaurant experiments.

Table 3 shows the factor pattern and structure of the final factor solution. The interpretation of the two factors was consistent with the literature on customer response to dissatisfactory service encounters (Hirschman, 1970; Singly 1988). Items describing the subjects’ intentions to return to the provider in the future and to describe it in favourable terms measured Factor 1. This factor related to the loyalty intentions of the subjects. Factor 2 was measured by items that had to do with the subjects’ reactions to a dissatisfactory experience: complaining, negative word-of-mouth, and, to a lesser extent, switching. These actions have been classified as customer complaint behaviour, or CCB. The tendency of the item “switch to a competitor” to load on both factors was noted. Subsequently, all analyses were performed with and without the item included. No differences were found in the test results or the magnitudes of the effects. Consequently, the switching item was retained for the results reported below. The remainder of the analysis treats the factor scores associated with the loyalty and CCB factors as dependent variables.

Experinmental Results

Table 4 shows the cell means of the loyalty and CCB factors for the hotel and restaurant experiments. In general, when core service failure (denial or delay) occurred, no service recovery strategy, including assistance plus compensation, led to positive future intentions toward the service provider. Intentions did not attain the mean values associated with the midpoints of the original 7-point intention scales.

As the dependent variables were moderately correlated (r = .545 for the hotel experiment, r = .516 for the restaurant experiment) and the design was unbalanced, the dependent variables, loyalty and CCB, were analysed by a multivariate general linear model (GLM). The analysis considered all main effects and two-way interactions.

Table 5 shows the results of the GLM analysis. The multivariate statistics indicated that the joint distribution of loyalty and CCB intentions was affected by recovery, severity, and criticality for both the hotel and restaurant experiments. Except for the interactions of severity with criticality, all effects were significant at the .10 level. The results of the univariate tests suggests the nature of the main and interaction effects for each intention. Recovery and criticality affected the marginal distributions of the loyalty and CCB intentions for both experiments, while severity affected all intentions except CCB for the restaurant experiment. Severity interacted with recovery to affect loyalty in both experiments and CCB intentions for the hotel experiment. However, on its own, severity did not affect intentions in the restaurant experiment. Criticality influenced the impact of recovery in the restaurant experiment and the levels of loyalty and CCB intentions directly in both experiments.

Effects of Recovery on Intentions

H1a implied the means of apology only were less than the means of any of the other three treatment levels for recovery. The hypothesis was examined by a joint test of the following effects contrasts: (a) apology versus assistance, (b) apology versus compensation, and (c) apology versus both assistance and compensation. Contrasts were designed for all hypothesis tests so that differences in positive effects were consistent with the hypothesis. Table 6 reports the results in terms of estimated differences between the means of the effects being compared. As well, the upper and lower limits of the 95% confidence interval for the population mean difference of effects is reported.

With respect to Hla, all estimates of the effects differences were positive, consistent with the proposition that the effects of the apology only treatment would be smallest. Furthermore, the confidence intervals for both experiments uniformly supported Hla. The effect of each level of recovery beyond apology could be said to exceed the effect of apology with at least 95% probability.

H1b implied the assistance treatment effect would always equal the compensation effect. Table 6 shows the differences were positive in three of four cases. However, the confidence intervals indicated the evidence was not sufficient to reject the hypothesis of no difference between the effects of compensation and assistance as all included zero.

H1c suggested the combination of assistance and compensation would be more ameliorative than either offered alone. All of the effects differences support the proposition. Additionally, all confidence intervals indicated the evidence rejected the hypothesis of no effects differences at least at the 95% level.

The Role of Criticality and Severity on Intentions

H2 implied that the effect of low criticality was less than high criticality on both loyalty and CCB intentions. Table 6 shows consistent differences for CCB intentions in both experiments and loyalty intentions in the restaurant experiment. Inspection of the confidence intervals indicated the hypothesis that low criticality and high criticality had the same effect on intentions could be rejected at least at the 95% level. The result for loyalty intentions in the hotel experiment was unexpected. It points to service problems having a more negative impact on loyalty in low-criticality situations than in high-criticality situations. The result was significant at the 95% level.

H3 implied that the effect of high severity was larger than that of low severity for both loyalty and CCB intentions. Table 6 show that all four differences were positive. The confidence intervals were consistent with the proposition except for loyalty intentions in the restaurant experiment. When a 90% confidence interval was calculated for the difference of high- and low- severity effects on restaurant loyalty intentions, the results rejected the hypothesis of no difference.

H4a postulated compensation to be less effective than assistance in high-severity incidents. Table 6 shows consistent signs in three of the four cases. However, the hypothesis of no difference cannot be rejected at the 5% level for either experiment. H4b reversed the effectiveness of assistance and compensation in low-severity incidents. The results in Table 6 indicated no support for the proposition. The only significant result favoured the greater effectiveness of assistance.

Figures 2 through 4 show the role of severity and criticality as moderators of the effects of recovery for loyalty intentions. The interactions between recovery and severity were significant for the hotel experiment (F3.586 = 6.051, p = .000) and the restaurant experiment (F3.586 = 4.049, p = .007). The univariate statistics indicated the effect concerned only loyalty intentions. An important feature of the interaction of severity and recovery was the apparent diminishing returns to recovery effort for low-severity problems. Coupled with the finding that the effects of assistance and compensation were not statistically different (H4a), the evidence in Figures 2 and 3 suggests there is no gain in loyalty intentions from offering both assistance and compensation after service delay. In contrast, offering both was productive in each experiment after service denial.

While the study proposed no a priori effect of the interaction between criticality and recovery, the analysis investigated the possibility that criticality moderated the effect of recovery. Criticality determined the effect of recovery only on loyalty intentions in the restaurant experiment (F3,586 = 3.898, p = .009). Figure 4 suggests diminishing returns to recovery effort in low criticality situations.


This study examined core service failures, incidents where the service firm had broken a promise to the custourer. These failures are common, manifest, serious, and cause customers to switch service providers (Keaveney, 1995). In many industries, like the hospitality industry, customer retention is critical to growth and profitability. Customer loyalty depends on the provider’s ability to establish a positive track record in the early stages of the relationship. Customer dissatisfaction with a transaction at this stage reduces the opportunity to build loyalty and erodes the firm’s reputation. When core failures occur, customer retention is at risk. The provider’s recovery from a problem is crucial to rebuilding the relationship. Because the situations presented to subjects modelled new purchases based on reputation alone, the results bear upon the process of building customer loyalty.

With some exceptions, the results support the concept that customer reaction to service failure and recovery strategies depend on the type and amount of resources lost and gained during the exchange (Smith et al., 1999). For any given service failure (loss), offering the subject more (gain) increased positive future intentions toward the provider. Offering assistance or compensation was more effective than an apology only, and offering assistance plus compensation was most effective. High loss situations (high criticality and/or high problem severity) resulted in less positive future intentions than did lower loss situations. As with previous studies (Darida et al., 1996; Smith et al., 1999; Webster & Sundaram, 1998), situational factors influenced the relative effectiveness of the recovery strategies.

As hypothesized, the outcome of a service failure was conditional on the incident. The likelihood of a breakdown in the relationship with a new customer was higher after service failure for a high-criticality purchase. The findings concur with those of prior research (Kelley & Davis, 1994; Parasuraman et al., 1991; Webster & Sundaram, 1998). Similarly, the negative impact of service failure was higher for a high-severity failure compared to a low-severity failure, consistent with earlier findings (Keaveney, 1995; Ostrom & Iacobucci, 1995; Richins, 1987). While the firm can control severity, criticality is the customer’s prerogative. The results point to the value of identifying the customer’s view of the importance of the purchase, especially if the appropriate recovery strategy depends on criticality.

Interestingly, the effectiveness of the recovery effort depended on the type of failure. For low-severity failures, the results indicated diminishing returns to effort. Offering either compensation or assistance was as effective as offering both. In contrast, offering either was less effective than both for high-severity failures. There was also evidence of diminishing returns to effort in low-criticality failures for the restaurant experiment. The firm’s choice of a recovery strategy depends on the characteristics of the purchase and failure incident.

There are several managerial implications to the results. Generally, to improve the probability of retaining new customers, the firm needs to reduce core failure rates. “Getting it right the first time” optimizes value for both the customer and the firm. Chance service failures are always possible, and recovery strategies need to be designed to address specific situations. Systematic failures are another matter. Core service failures, such as delay and denial, are caused by overbooking, reservation errors, insufficient staff to prepare for guests’ arrivals, and poor management controls. Overbooking is a common strategy in the hospitality industry to maximize capacity utilization and reduce the impact of “no shows.” Service firms might want to consider the short-term gains versus the long-term costs of the strategy. Customer defection and negative word-of-mouth are the potential consequences of core service failures where service is delayed or denied.

Firms need to recognize problem severity and criticality as factors in developing recovery strategies. In broader terms, recovery strategies need to be matched to the specific incident (Smith et al., 1999; Webster & Sundaram, 1998). The implication is that generic recovery strategies or by-the-book approaches may not be appropriate, particularly when the service employee and customer are in a face-to-face encounter. Here, the opportunity exists for the employee to assess the customer’s situation and judge what specific approach to take to resolve the problem to the customer’s satisfaction.

Furthermore, the finding that recovery exhibits diminfishing returns, coupled with the fact that recovery is not completely restorative, suggests that in some cases more is not better. Instead, the gap between full restoration and the outcome of the exchange may hinge on process.

The process aspect of recovery emphasizes the how of service recovery-the interaction between the contact employee and the customer during the incident. These strategies have been effective in turning service problems into satisfying experiences for customers (Bitner et al., 1990). It is possible that a combination of what (compensation and assistance) and how (responsiveness, empathy, and understanding) may be more effective than either strategy in isolation. The knowledge that the effectiveness of what strategies depends on identified situation factors, combined with the how strategies’ ability to tailor-make the recovery offer, should considerably improve their effectiveness. As an example, knowing that compensation was more effective than assistance in delay situations where expectations are low, the contact employee is more likely to match the best option with the appropriate delivery of the recovery.

The ineffectiveness of the standard assistance and compensation strategies suggests that a significant gap remains to be filled by more costly tangible actions (the what) and/or by managing the tone of the relationship (the how). This represents an interesting challenge for the firm. On the one hand, costlier recovery efforts affect the firm’s profitability, but can be standardized. On the other hand, skilled service employees can mean lower outlays for specific recovery incidents, but may be difficult to standardize as the number of employees increases. This issue is particularly relevant for firms like hotels and restaurants, which operate several locations. As well, the short- term costs must be balanced against the longterm benefits of customer retention.

The limitations of the study should be recognized. While the results were strong, the study dealt only with core service failures. These results may not be generalized to other types of service problems including those that deal with minor problems or process problems (i.e., rude or unresponsive employees). The experiment measured stated future intentions and, while researchers have found these measures to be reasonable correlates for actual behaviour (Rust, Zahorik, & Keningham, 1995), it is recognized that there may be discrepancies between stated intentions and actual behaviours. Furthermore, the levels of compensation were bounded by industry practices. Larger compensation levels are likely to lead to more positive intentions toward the provider.

Another limitation was that, because it was a paper and pencil study, it did not capture the tone of the relationship between the service provider and the customer in the situation. On the other hand, anecdotal evidence can capture the service provider’s willingness to go beyond the call of duty, to take ownership of the problem and to tailor the response to the customer’s needs. In these instances, the customer may bond to the organization because of the special way in which the service provider handled the situation. The critical incident technique allows the researcher to capture some of the relationship, or the how side, as witnessed by findings that show the importance of the service provider in problem situations (Bitner et al., 1990). In this investigation, the results reflected a scenario that contained a brief conversation that could be described as friendly and helpful (depending on the scenario), but did not offer a strong emotional content.

Three suggestions for future research are offered. First, the effectiveness of how recovery strategies should be explored, either alone or, preferably, with what strategies. the tone of the relationship can be captured using videos as part of the experimental design (Sparks & Bradley, 1997). Initially, it might be useful to contrast different employee responses (e.g., neutral versus responsive, empathetic) against compensation and/or assistance. The objective is to understand the relative effectiveness of how and what strategies, particularly the combination of strategies that is most effective in a given situation.

A second suggestion is to broaden the types of service problems studied. While core service failures are serious and cause customers to switch service firms, other common service problems, particularly how the service was delivered, can also lead to customer exit. The effectiveness of how and what recovery strategies in tackling problems with rude or uncaring contact employees versus core service failures would offer interesting insights into customer retention.

Finally, the contrasting results between the two service scenarios suggest that extending the research to additional scenarios such as financial or professional services could help clarify the impact of what and/or how strategies when core service failures occur. Financial and professional services are typically higher risk for consumers with higher switching costs. The insights to be gained here include future intentions toward service firms when customers are faced with high switching costs.

In summary, the research examined an important issue for service providers: what happens when a core service failure occurs and various recovery strategies that reflect current industry practices are offered? Customers’ future intentions toward the provider depended on the service recovery strategy and the context, problem severity, and expectations. From the service provider’s perspective, the results are troubling. Customers held negative future intentions toward the service provider regardless of the recovery strategy, and this includes assistance plus compensation. Considering core service failures, getting it right the first time is the best strategy for a service firm.


Afifi, A.A., & Clark., V. (1990). Computer-aided multivariate analysis. New York: Van Nostrand Reinhold.

Best, A., & Andreasen, A.R. (1976). Consumer response to unsatisfactory purchases: A study of perceiving defects, voicing complaints, and obtaining redress. Law and Society II, 701- 742.

Bitner, M.J., Booms, B.H., & Tetreault, M.S. (1990). The service encounter: Diagnosing favourable and unfavourable incidents. Journal of Marketing, 54, 71-84.

Boulding, W., Kalra, A., Staelin, R., & Zeithaml, V.A. (1993). A dynamic process model of service quality: From expectations to behavioral intentions. Journal of Marketing Research, 30, 7-27.

Darida, T, Levesque, T., & McDougall, G. (1996). Service problems and recovery strategies: An exploratory investigation in the hospitality sector. In C. Berneman (Ed.), Proceedings of the Administrative Sciences Association Conference, 1997 (pp. 101-110). Montreal: Administrative Sciences Association.

Day, R.L., & Landon, L.E. (1977). Toward a theory of consumer complaining behavior. In A.G. Woodside, J.N. Sheth, & PD. Bennett (Eds.), Consumer and industrial buying behavior (pp. 425-37). New York: North-Holland.

Dube-Rioux, L., Schmitt, B. H., & Leclerc, R (1988). Consumer’s reactions to waiting: When delays affect the perception of service quality. In T. Srull (Ed.), Advances in consumer research (Vol. 16, pp. 59-63). Provo, UT: Association for Consumer Research.

Goodwin, C., & Ross, I. ( 1992). Consumer responses to service failures: Influence of procedural and interactional fairness perceptions. Journal of Business Research, 25, 149-163.

Gronroos, C. (1988). Source quality: The six criteria of good perceived service quality. Review of Business, 9, 10-13.

Hart, C.W.L., Heskett, J.L. & Sasser, W.E., Jr. (1990). The profitable art of service recovery. Harvard Business Review, 68, 148-156.

Hirschman, A.O. (1970). Exit, voice, and loyalty: Responses to decline in firms, organizations and states. Cambridge, MA: Harvard University Press.

Hoffman, K.D., Kelley, S.W., & Rotasky, H.M. (1995). Tracking service failures and employee recovery efforts. Journal of Services Marketing, 9 (2), 49-61.

Jones, TO., & Sasser, W.E., Jr. (1995). Why satisfied customers defect. Harvard Business Review, 73, 88-99.

Keaveney, S.M. (1995). Customer switching behaviour in service industries: An exploratory study. Journal of Marketing, 59, 71-82.

Kelley, S.W., & Davis, M.A. (1994). Antecedents to customer expectations for service recovery. Journal of Academy of Marketing Science, 22, 52-61.

Kelley, S.W., Hoffman, D.K., & Davis, M.A. (1993). A typology of retail failures and recoveries. Journal of Retailing, 9, 429-452.

McCullough, M.A. (1995). The recovery paradox: A conceptual model and empirical investigation of customer satisfaction and service quality attitudes after service failure and

recovery. Unpublished doctoral dissertation, Texas A&M University.

Muthen, B., & Kaplan, D. (1985, May). A comparison of some methodologies for the factor analysis of non-normal Likert variables. British Journal of Mathematical and Statistical Psychology, 38, 171-189.

Ostrom, A., & Iacobucci, D. (1995). Consumer trade-offs and the evaluation of services. Journal of Marketing, 59, 17-28.

Parasuraman, A., Berry> L.L., & Zeithaml, V.A. (1991). Understanding customer expectations of service. Sloan Management Review, 33, 39-48.

Richins, M.L. (1983). Negative word-of-mouth by dissatisfied consumers: A pilot study. Journal of Marketing, 47, 68-78.

Richins, M.L. (1987). A multivariate analysis of responses to dissatisfaction. Journal of the Academy of Marketing Science, 15, 24-31.

Rust, R.T., Zahorik, A.J., & Keningham, TL. ( 1995). Return on quality (ROQ): Making service quality financially accountable. Journal of Marketing, 59, 58-70.

Singh, J. (1988). Consumer complaint intentions and behavior: Definitional and taxonomical issues. Journal of Marketing, 52, 93-107.

Singly J. (1990). Voice, exit, and negative word-of-mouth behaviours: An investigation across three service categories. Journal of the Academy of Marketing Science, 18, 1-15.

Singh, J., & Howell, R.D. (1985). Consumer complaining behavior: A review and prospectus. In H.K. Hunt & R.L. Day (Eds.), Consumer satisfaction, dissatisfaction and complaining behavior (pp. 59-66). Bloomington, IN: Indiana University.

Singh, J., & Wilkes, R.E. (1996). When consumers complain: A path analysis of the key antecedents of consumer complaint response estimates. Journal of the Academy of Marketing Science, 24, 350-365.

Smith, A.K., Bolton, R.N., & Wagner, J. (1999). A model of customer satisfaction with service encounters involving failure and recovery. Journal of Marketing Research, 36, 356-372.

Sparks, B., & Bradley, G. (1997). Antecedents and consequences of perceived service provider effort in the hospitality industry. Hospitality Research Journal, 20 (3), 17-34.

Tax, S.S., Brown, S.W., & Chandrashekaran, M. (1998). Customer evaluations of service complaint experiences: Implications for relationship marketing. Journal of Marketing, 62, 60-76.

Taylor, S. ( 1994). Waiting for service: The relationship between delays and evaluations of service. Journal of Marketing, 58, 56-69.

Webster, E., & Sundaram, D.S. (1998). Service consumption criticality in failure recovery. Journal of Business Research, 41, 153-159.

Zeithaml, V.A., Berry, L.L., & Parasuraman, A. (1990). Delivering quality service: Balancing customer perceptions and expectations. New York, NY: The Free Press.

Zeithaml, VA., Berry, L.L., & Parasuraman, A. (1996). The behavioral consequences of service quality. Journal of Marketing, 60, 31-46.

Terrence J. Levesque

Gordon H.G. McDougall

Wilfrid Laurier University

Address all correspondence to Terrence J. Levesque, School of Business and Economics, Wilfrid Laurier University, Waterloo, ON, Canada, N2L 3C3.

Copyright Administrative Sciences Association of Canada Mar 2000

Provided by ProQuest Information and Learning Company. All rights Reserved