Spinning the Wheel: The Effectiveness of Gamification in Service Recovery

The Effect of Gamification on Recovery Satisfaction

Research from the SFR literature demonstrates that a service failure can lead to negative emotional, attitudinal, and behavioral reactions and that monetary compensation can play a pivotal role in the process of reducing these negative customer reactions (Gelbrich, Gäthke, and Grégoire 2015; Nazifi et al. 2022). In particular, compensation is a key driver of customer satisfaction following service failures (Roschk and Gelbrich 2014). Drawing on the peak-end rule (Loewenstein and Prelec 1993), a service failure is considered a negative peak moment in the encounter that can be improved with a later positive episode through compensation. Prior research shows that gamifying a task can lead to positive customer outcomes, including enhanced engagement and improved satisfaction (Hammedi, Leclercq, and Steils 2024). Accordingly, we expect gamified compensation to generate a higher level of satisfaction than nongamified compensation. We therefore propose the following hypothesis:

The Mediating Effect of Enjoyment

Gamification has been identified as a powerful strategy for generating enjoyment through mechanisms that are commonly associated with games such as point or badge systems and competition (Charry et al. 2024). In contrast to incentive or compensation systems, the effectiveness of gamification relies on the uncertainty associated with obtaining a reward as a result of lotteries, contests, or challenges (Leclercq, Poncin, and Hammedi 2020a). This uncertainty triggers intrinsic motivations that generate enjoyment (Shen, Hsee, and Talloen 2019). Drawing on the peak-end rule (Ashtar et al. 2024), after a service failure (as the peak moment), offering a gamified recovery to customers may position enjoyment as the final (or most recent) moment of the experience. Thus, the overall positive trend and the “recency” effect of gamified recovery ultimately create a sense of a happy ending and enhanced enjoyment, which consequently reduces the detrimental impact of service failure on customer satisfaction. We propose the following hypothesis:

The Moderating Effect of Customer Choice

Several studies have suggested that gamification may backfire in certain contexts, thereby demotivating customers and harming their experience (Hammedi, Leclerq, and Van Riel 2017; Leclercq, Hammedi, and Poncin 2018). One factor that limits the effect of gamification is the violation of the leading property of the game, namely, the freedom to participate (Caillois and Barash 2001). If firms do not seek explicit consent from customers for their participation, they risk situations in which customers perceive gamification as external regulations of their performance, which may lead to stress and reactance (Deci and Ryan 2002). Based on the peak-end rule, the effectiveness of gamified recovery (i.e., the final or more recent moment) with regard to creating an enjoyable experience after a service failure (i.e., the negative peak moment) depends on customers’ sense of voluntary participation. Namely, the imposition of this approach on individuals who are unreceptive to gamification may amplify the disutility of the negative episode as the peak moment, thus exacerbating the negative emotions and dissatisfaction associated with the recovery (Joireman et al. 2013). However, for individuals who are receptive to gamification, this approach can reduce the disutility of the negative episode (i.e., the failure) and thus increase the enjoyment associated with the encounter and enhance the individual’s satisfaction with the recovery. In support of the important role of choice, prior research indicates that choice has a positive effect on recovery satisfaction, especially when customers may not be open to recovery (e.g., compensation in the form of cryptocurrencies; Nazifi et al. 2021). In the context of gamified recovery, which can be viewed as an innovative solution that is unfamiliar to customers, we propose the following hypothesis:

The Moderating Effect of Failure Severity

One key contextual factor that has been extensively studied in the SFR literature is failure severity. Various studies have shown that failure severity weakens the effectiveness of service recovery (Grewal, Roggeveen, and Tsiros 2008; Hess, Ganesan, and Klein 2003) and can increase the negative motives inferred by customers as well as their anger and retaliatory behavior (Joireman et al. 2013). Similarly, the literature on gamification highlights that using gamification may be counterproductive depending on the context. For instance, Hammedi, Leclerq, and Van Riel (2017) show that gamifying healthcare among patients suffering from severe disease may be viewed as a poor approach by the service provider. Drawing on the peak-end rule, the application of gamified recovery after a severe failure may lead to lower benefits than a mild failure because the disutility of a severe failure may be perceived as indicating a more pronounced negative episode (the peak moment) than that of a mild failure; as such, the positive influence of a gamified recovery (the final or more recent moment) would be weakened after a severe failure. Furthermore, severe failures may reduce the perceived enjoyment associated with gamification since this approach may be viewed as inappropriate when the failure is severe because its hedonic element (Ciuchita et al. 2023) may not be suitable for the gravity of the situation, and the corresponding uncertainty of the outcome (Hammedi, Leclercq, and Steils 2024) may reduce the perceived enjoyment and subsequent satisfaction associated with the encounter or even further aggravate customers. We propose the following hypothesis:

The Moderating Effect of Compensation Level

Prior research highlights that the amount of compensation should match the magnitude of the failure (Nazifi, El-Manstrly, and Gelbrich 2020), a notion rooted in the well-established resource exchange principles governing failure-recovery situations (Roschk and Gelbrich 2017). Following service failures, customers develop expectations based on the type and severity of the failure (Nazifi et al. 2022) and use this expectation as a baseline to compare the amount of compensation to determine whether this has been met with full compensation, exceeded through overcompensation (thus creating a sense of winning), or fell short with partial compensation (thus creating a sense of losing). As a result, if the compensation exhibits a mismatch with the severity of the service failure, it will not lead to a successful recovery (Nazifi, El-Manstrly, and Gelbrich 2020). Similarly, the offered reward has also been identified as a pivotal lever in gamified services (Hammedi, Leclercq, and Steils 2024), in which context gamification creates uncertainty regarding obtaining rewards and thus instills a sense of winning or losing.

In the case of partial compensation through a gamified recovery, a combination of the initial uncertainty associated with the game-like activity and the sense of loss that represents the final moment of the experience may damage the customer’s overall enjoyment of the process and subsequent satisfaction. In that case, gamifying the recovery process becomes counterproductive because the final or more recent moment (gamified recovery) does not compensate for the negative impact of the peak moment (service failure) due to the emergence of a sense of loss. Based on the peak-end rule, partial compensation, as a subpar outcome, does not constitute a happy ending, thus resulting in reduced enjoyment and subsequent satisfaction. Conversely, in cases of full compensation or overcompensation, the combination of the uncertainty introduced by gamified recovery and the creation of a sense of winning is in line with a positive trend or a happy ending according to the peak-end rule, thus enhancing the customer’s perceived enjoyment and subsequent satisfaction (i.e., the final moment; Shen et al. 2019). We propose the following hypothesis:

The Moderating Effect of Time Pressure

The extant marketing literature has revealed that time pressure can play a key role in influencing human judgments and decision-making (Dhar and Nowlis 1999). According to the type of information that customers consider in urgent contexts, time pressure may have adverse effects (Thomas, Esper, and Stank 2010). Research suggests that time pressure causes customers to overestimate negative events more strongly (a phenomenon known as negativity bias) because it facilitates biases that cause negative events to be outweighed (Yang and Unnava 2016). Accordingly, whereas under normal circumstances, the disutility of a service failure (i.e., the peak moment) can be reduced through a gamified recovery (i.e., the final or more recent moment), thus leading to enhanced enjoyment and increased satisfaction with regard to the encounter, under conditions of time pressure, customers tend to overestimate the adversity of the service failure, which may disrupt the enjoyment generated by the gamified recovery. Thus, in individuals’ minds, the negative peak moment becomes more pronounced, and the positive final moment may become less pronounced. We propose the following hypothesis:

Design

This study examined the effectiveness of gamified compensation in a real-life setting in which customers experience a service failure. We also considered whether the gamified compensation was optional or imposed. Therefore, we employed a 2 (compensation type: nongamified vs. gamified) by 2 (choice: present vs. absent) between-subjects design. The field setting focused on a French fast-moving consumer goods (FMCG) retailer. Retailing represents a relatively underresearched consumption setting in the failure-recovery context and thus demands further research (Khamitov, Grégoire, and Suri 2020).

The customers in a queue at the after-sale service desk were approached by research assistants. We presented this interaction as an attempt to reduce the time spent in the queue by treating their service failure alternatively. Before inviting respondents to participate in our study, we ensured that they were facing a failure by asking them to indicate why they were in the queue. The reported failures ranged from the pricing of items to the delivery of expired products. In the absence of a choice, respondents were assigned to one compensation type (nongamified vs. gamified). In the presence of a choice, they could choose between these two compensation types, and customers were allocated to the compensation type based on their choice, which exhibited an approximately even split (nongamified: 52.0%, gamified: 48.0%). In the gamified condition, customers were invited to spin a wheel on a tablet. They were told that they could win a 20 Euro coupon, a 5 Euro coupon, or no coupon to spend on their next visit. The coupons were randomly distributed, and the wheel was programmed to yield an average compensation of 5 Euros (see Figure 2 for a sample of the wheels).

OPEN IN VIEWER

Figure 2.

Sample wheels.

In the nongamified condition, participants received a fixed coupon for 5 Euros to spend on their next visit. This amount was defined to ensure equivalent compensation across the conditions. Neither the customers were informed of the compensation they would receive before completing the questionnaire nor what other customers had received.

Data Collection

Customers at the FMCG retailer were recruited in the winter of 2022. An attention check was included by asking the respondents to select the retailer brand that they were currently visiting. Six respondents who failed the attention check were excluded from the analysis. The sample size was determined through a power analysis conducted through G*Power (α = .05, power = .80). We expected an effect size of f = 0.20 for the interaction between gamification and choice on playfulness, which lies between the small (f = 0.10) and medium (f = 0.25) conventions (Cohen 1988). The power analysis indicated a minimum required sample of 199. Supplemental Appendix Table A6 provides the full details regarding the sample size calculations for this study and subsequent studies. Our final sample consisted of 227 valid questionnaires (M_Age = 41.15, SD = 14.34; Female = 61%). Customers accessed the survey before receiving the rewards to avoid any resulting biases.

Measures

The questionnaire included measures of satisfaction, playfulness as a dimension of perceived enjoyment, blame attributions, service failure severity, and procedural justice. Unless otherwise specified, all constructs were captured using established scales that were scored on 7-point Likert-type scales ranging from 1 (“strongly disagree”) to 7 (“strongly agree”). The participants responded to three items pertaining to satisfaction (e.g., I was satisfied with the handling of my complaint; α = .97) adapted from Roschk and Gelbrich (2017). The respondents further reported the playfulness of the resolution through four items (e.g., The way the company provided compensation was playful; α = .96) amended from Hwang and Choi (2020). Playfulness has been identified as an important dimension of the enjoyment generated by gamified settings (Leclercq, Poncin, and Hammedi 2020a) and thus is used as a proxy for perceived enjoyment in the field study. Blame attribution and failure severity were measured as control variables because they have been shown to influence recovery satisfaction (Gelbrich, Gäthke, and Grégoire 2016). Blame attribution was reported based on one item (i.e., the service failure that I encountered was entirely the service provider’s fault) from Gelbrich, Gäthke, and Grégoire (2015). Failure severity was assessed using a single bipolar item, which was scored on a seven-point scale ranging from 1 (minor problem) to 7 (major problem) from Maxham III and Netemeyer (2002). A measure of procedural justice was included because of the experimental setting. Indeed, participants recruited in the waiting queue were aware that they would experience alternative treatment for their service failure compared to other clients. Procedural justice was measured via a two-item scale (e.g., I believe the store has fair guidelines to handle problems) adapted from Roschk and Gelbrich (2017). Finally, we collected the respondents’ sociodemographic information. Supplemental Appendix provides a list of all measurement items (Supplemental Appendix Table A1) and correlational matrices for the tested variables (Supplemental Appendix Table A2) across the studies.

Propensity Score Matching

We used the propensity score matching procedure (Fröberg, Kolesova, and Rosengren 2024) to alleviate the potential self-selection bias induced by respondents’ choice of the gamified or nongamified recovery. This procedure aims to create groups revealing similar characteristics regarding the covariates across conditions. Therefore, we performed a 1:1 nearest neighbor propensity score matching without replacement with a propensity score estimated using logistic regression of the gamification treatment on the covariates—perceived service severity, blame, and procedural justice. The function caliper was activated at the level of 0.20 of the standard deviation of the propensity score. After matching, all standardized mean differences for the covariates were less than 0.10, thus indicating adequate balance. A total of 166 responses showed a match across conditions and were used for hypothesis testing. Although the sample size was below our target, it should provide reasonable power at p < .10 and was thus used for hypothesis testing.

Main and Interaction Effects

Two analyses of covariance (ANCOVAs) featuring compensation type (gamified vs. nongamified recovery) and choice (choice vs. no choice) as the independent variables and satisfaction and playfulness as the dependent variables were conducted; these analyses included failure severity, blame attribution, and procedural justice as control variables. Regarding satisfaction, the results indicated a marginally significant main effect of choice (M_{Choice/NoChoice} = 3.94/3.71, SE = 0.08/0.09, F(1, 159) = 3.50, p = .063, d = 0.15) and a nonsignificant effect of compensation type (M_{Gamified/Nongamified} = 3.82/3.83, SE = 0.08/0.08, F(1, 159) = 0.25, p = .874, d = −0.01). Importantly, we observed a marginally significant choice × compensation type interaction (F[1, 159] = 3.38, p = .068, d = 0.15), although, pairwise comparisons revealed no significant difference in satisfaction between compensation types in the condition in which it was imposed (M_{Gamified/Nongamified} = 3.59/3.82, SE = 0.17/0.17, p = .14, d = −0.16) and the condition in which it was not (M_{Gamified/Nongamified} = 4.04/3.85, SE = 0.15/0.15, p = .22, d = 0.13). The interaction is illustrated in Figure 3 (Panel A).

OPEN IN VIEWER

Figure 3.

Study results.

With regard to playfulness, the type of compensation had a significant main effect (M_{Gamified/Nongamified} = 4.82/3.31, SE = 0.15/0.15, F(1, 159) = 46.13, p < .001, d = 0.54), while the choice did not (M_{Choice/NoChoice} = 3.93/4.21, SE = 0.17/0.15, F(1, 159) = 1.30, p = .26, d = 0.11). We found a significant interaction between these factors (F[1, 159] = 7.49, p = .007, d = 0.22). ¹ Pairwise comparisons showed that in the choice condition, greater playfulness was felt with gamified compensation than with nongamified compensation (M_{Gamified/Nongamified} = 4.97/2.88, SE = 0.21/0.20, p < .001, d = 1.31) and to a lesser degree in the no-choice condition (M_{Gamified/Nongamified} = 4.67/3.74, SE = 0.25/0.22, p = .047, d = 0.66). ²

Mediated Effect

We first tested the indirect effect of gamification on satisfaction. The mediation analysis was conducted using Hayes and Preacher (2014) Model 4. Compensation type (gamified vs. nongamified recovery) was integrated as the independent variable, playfulness as the mediator, satisfaction as the dependent variable and blame, failure severity, procedural justice, and choice as covariates. The results revealed a significant positive indirect effect (b = .16, SE = 0.06, 95% CI [0.04, 0.29]) and no significant direct effect, suggesting that gamified recovery leads to greater satisfaction by increasing enjoyment.

Moderated Mediation Effects

We tested the effects of gamified compensation (as the independent variable) on satisfaction (as the dependent variable) contingent on choice (as the moderator), which was explained through playfulness (as the mediator). Therefore, we conducted moderated mediation analyses using PROCESS Model 8 at the 95% confidence level and 5,000 bootstraps. Blame, failure severity, and procedural justice were included as control variables. A significant index of moderated mediation was also found (b = .10, SE = 0.06, 95% CI [0.01, 0.25]). The conditional indirect effects indicated that offering customers a choice regarding service recovery increased the effect of gamified compensation on satisfaction through playfulness (b = .19, SE = 0.08, 95% CI [0.02, 0.35]) compared to when no choice was offered (b = .08, SE = 0.05, 95% CI [0.01, 0.18]). No significant direct effect of gamified recovery on satisfaction was found at any level of the moderator (b_{no_choice} = −0.31, SE = 0.18, 95% CI [−0.65, −0.04]; b_choice = −0.00, SE = 0.17, 95% CI [−0.34, 0.34]). Supplemental Appendix Table A3(a) shows the results of moderated mediation analyses.

Design

This study examined the effects of gamified compensation on satisfaction (H1) and the mediating effects of enjoyment (H2) when service failures are perceived to be more severe (H4). A 2 (compensation type: gamified vs. nongamified) by 2 (failure severity: mild vs. severe) between-subjects design was employed. We used a restaurant setting to ensure that we could observe the effects of gamification in a setting with which most customers are familiar (Roschk and Gelbrich 2017). Gamified compensation was operationalized using the spin-the-wheel game, as in Study 1.

Participants were presented with a scenario adapted from Grégoire et al. (2018), imagining dining at a French restaurant (see Supplemental Appendix Table A4 for the detailed scenarios of all studies). Under mild failure conditions, the couple was greeted by a waiter and ordered drinks. A mistake was made regarding the drinks, which were quickly rectified. There was a 30-minute wait for the food. Although it was not of high quality, they were hungry and decided to eat. In severe failure conditions, the couple waited 15 minutes to be greeted by a waiter, and the mistake with the drink order took 15 minutes to rectify. The wait for the food was nearly 60 minutes. When the food arrived, it was cold, overcooked, and overly salty, but since they were hungry, they decided to eat. Before paying, participants in the gamified condition received an apology and an opportunity to “spin the wheel” to receive some discount. Here, an animated wheel was spun, resulting in 20% being taken off their bill. The wheel was blank to avoid any anchoring effects (see Supplemental Appendix A5 for details).

In the nongamified conditions, respondents received an apology and a 20% discount. The scenario depicted a situation in which the diners ultimately accepted a service failure. Failures are commonly accepted, a phenomenon that is associated with increased uncertainty in terms of the compensation amount since the service has been (partly) consumed (Gelbrich, Gäthke, and Grégoire 2015). This circumstance is suitable for gamified compensation because no strong recovery expectations are involved.

Data Collection

A sample of 239 adults residing in the United States were recruited in the winter of 2023 via Prolific, a panel platform that is commonly used in service research (Béal and Grégoire 2022; Hammedi, Leclercq, and Steils 2024). An attention check (“How much compensation were you offered?”) was included to ensure that the respondents actually spun the wheel. Additionally, two instruction checks (i.e., “Please click on ‘disagree’ for this question” and “Please click on ‘Always’ for this question”) were included. Respondents who did not pass these checks were excluded from the survey (n = 40). Our final sample included 199 participants (M_Age = 42.26, SD = 14.98; Female = 49.7%).

Measures

All items were measured on 7-point Likert-type scales ranging from 1 (“strongly disagree”) to 7 (“strongly agree”), unless otherwise specified. Following the scenario, participants responded to the same satisfaction scale that was used in Study 1. The respondents reported their enjoyment of the resolution through five items (e.g., Playing the game was fun; α = .99) amended from Eppmann, Bekk, and Klein (2018). We further included six mediators to account for alternative explanations. We measured perceived justice (e.g., the outcome I received was fair; α = .93) based on Roschk and Gelbrich (2017) and exhilaration (e.g., I feel invigorated, α = .94) adapted from Steptoe and Bolton (1988) by three items for each construct. Reassurance was assessed using two items (e.g., I feel reassured that the company considered my concern, r = .92) from Tung, Chen, and Schuckert (2017), and reactance was assessed with three items (e.g., The company’s way of handling my complaint is intrusive, α = .94) from Gelbrich, Voigt, and Nazifi (2023). Blame was measured as per Study 1. Customer control was assessed using three items (e.g., The likelihood of receiving the compensation was in my hands, α = .81) amended from Leclercq (2022). ³ Service importance and technology aversion were included as controls because they influence customers’ recovery expectations and outcomes (Hess, Ganesan, and Klein 2003). Service importance was reported with a single item (i.e., overall, restaurants are a very important service for me) from Hess, Ganesan, and Klein (2003). Technology aversion was reported with four items (e.g., I feel apprehensive about using technology, α = .91) adapted from Lin and Hsieh (2012). The questionnaire ended with manipulation and realism checks and demographic questions. Supplemental Appendix (Table A1) provides all the measurement items.

Manipulation and Realism Checks

The gamified (vs. nongamified) compensation manipulation was tested using three items (e.g., The activity seemed similar to a game, α = .89) in this study and all subsequent studies. The failure severity manipulation was assessed using the same severity item as used in Study 1. Two analyses of variance (ANOVAs) with the two manipulation check scales as dependent variables supported the manipulations of the independent variables of failure severity (M_Severe/Mild = 6.01/3.84, SE = 0.13/0.14, F(1, 195) = 126.36, p < .001, d = 1.25) and compensation type (M_{Gamified/Nongamified} = 4.92/2.36, SE = 0.15/0.13, F(1, 195) = 171.17, p < .001, d = 1.38). No cross-effect of compensation type on failure severity was found (p = .12); however, a very small difference in perceived gamification was seen across severity conditions (M_Severe/Mild = 3.70/3.58, SE = 0.14/0.13, F(1, 195) = 126.36, p = .03, d = 0.07). Participants perceived the scenarios as realistic (M = 5.57), and the ratings were significantly above the scale midpoint (ps < .001) across all conditions.

Main and Interaction Effects

Two ANCOVAs featuring compensation type and failure severity as the independent variables and satisfaction and enjoyment as the dependent variables were conducted; these analyses included all the control variables listed above. Regarding satisfaction, the results indicated a significant main effect of failure severity (M_Severe/Mild = 2.55/4.62, SE = 0.15/0.15, F(1, 191) = 89.86, p < .001, d = −1.07) and a marginally significant effect of compensation type (M_{Gamified/Nongamified} = 3.37/3.79, SE = 0.16/0.13, F(1, 191) = 3.70, p = .06, d = −0.22). Importantly, we observed a significant failure severity × compensation type interaction (F[1, 191] = 15.53, p < .001, d = 0.26), as illustrated in Figure 3 (Panel B). Pairwise comparisons revealed no significant difference in satisfaction between compensation types in the mild failure condition (M_{Gamified/Nongamified} = 4.82/4.41, SE = 0.24/0.20, p = .18, d = 0.25). However, in the severe failure condition, gamified compensation led to significantly lower satisfaction than nongamified compensation (M_{Gamified/Nongamified} = 1.93/3.17, SE = 0.23/0.19, p < .001, d = −0.82).

Regarding enjoyment, while failure severity (p = .31) and compensation type (p = .52) did not have a significant main effect, we found a significant interaction between these factors (F[1, 191] = 7.96, p = .005, d = 0.19), as illustrated in Figure 3 (Panel B). Pairwise comparisons showed that in the mild failure condition, greater enjoyment was felt with gamified compensation than with nongamified compensation (M_{Gamified/Nongamified} = 4.73/3.55, SE = 0.24/0.20, p < .001, d = 0.75). However, no such significant difference was observed in the severe failure condition (M_{Gamified/Nongamified} = 2.98/3.02, SE = 0.19/0.23, p = .90, d = −0.03).

Mediated Effect

We first tested the indirect effect of gamification on satisfaction to replicate the results of Study 1 while including enjoyment as a mediator. The mediation analysis was conducted using PROCESS Model 4. Compensation type was selected as the independent variable, enjoyment as the mediator, satisfaction as the dependent variable, and the same controls as above with the addition of failure severity. The results revealed a significant positive indirect effect (b = 0.25, SE = 0.12, 95% CI [0.04, 0.51]), thus indicating that gamified compensation leads to greater satisfaction by increasing enjoyment.

Moderated Mediation Effects

Moderated mediation analyses were conducted based on Process Model 8. The model was specified in a manner similar to the mediation model discussed above, albeit with the addition of severity as the moderator. The full mediation results are provided in Supplemental Appendix Table A3(b). The data revealed a significant index of moderated mediation (b = −0.52, SE = 0.21, 95% CI [−0.98, −0.14]). Specifically, a significant indirect effect of gamified (vs. nongamified) compensation on satisfaction through enjoyment was observed when failure severity was mild (b = 0.51, SE = 0.16, 95% CI [0.22, 0.86]), whereas this indirect path was not significant when the failure was severe (b = −0.02, SE = 0.14, 95% CI [−0.29, 0.27]).

To assess alternative mediators, we used Hayes’ Model 8 to investigate justice, reassurance, exhilaration, reactance, customer control, and blame individually while including the same control variables as previously employed. Supplemental Appendix A5 lists the results. None of the mediators exhibited a significant moderated mediation index, which would suggest that they were rival explanations for enjoyment. However, when the indirect paths were examined under mild and severe conditions, isolated effects were observed. The collective results indicated that the rival mediators did not compete with enjoyment, and their role in the effects of gamification thus remained inconclusive, pending further research.

Design

This study examined the effectiveness of gamified compensation across different levels of compensation. A 2 (Compensation type: gamified vs. nongamified) by 3 (Compensation level: partial vs. full vs. overcompensation) between-subjects design was employed. In this study, we used a different hedonic setting, specifically a gym, since fitness settings are relatively underresearched (Khamitov, Grégoire, and Suri 2020).

As in Study 2, a vignette was used to frame the scenario. Participants imagined that they were members of a gym that costs $100 a month. After a few months, the gym informed them that it would be closing for 1 month for renovations and upgrades. During this time, the gym was not accessible, which was inconvenient for people who usually visited the gym 2–3 times per week. They complain about the issue. In the nongamified compensation condition, the complainers were offered either partial compensation ($80), full compensation ($100), or overcompensation ($100 + $20 off the following month’s membership). In the gamified compensation condition, they were offered the opportunity to spin the wheel to receive a certain amount of compensation. Participants spun the wheel (as per Study 2), randomly receiving one of the three compensation amounts listed above.

Data Collection

The sample criteria and recruitment procedure mirrored those used in Study 2. We included two attention checks (“What was the monthly cost of the gym?” and “How much compensation were you offered?”) and two instruction checks (“Please click on ‘agree’ for this question” and “Please click on ‘Maybe’ for this question”). An initial sample of 604 U.S. adults was recruited via Prolific in the winter of 2023, but respondents who did not pass these checks were excluded from the survey (n = 53). The final sample included 551 participants (M_Age = 41.95, SD = 14.47; Female = 49.0%).

Measures

The same measures for model constructs and controls, as well as manipulation and realism checks were used as per Study 2 (see Supplemental Appendix Table A1).

Manipulation and Realism Checks

We confirmed the appropriate compensation amount for the three levels based on a pretest with 82 participants (M_Age = 38.23, SD = 14.10; Female = 61.0%) who were recruited in the same way as the participants in the main study. The respondents read the study vignette and indicated the amounts of money that would (1) “put them on a par with a failure-free situation” (full compensation), (2) “make them truly delighted about the situation” (overcompensation) and (3) “make them feel at a loss regarding the situation” (partial compensation). The median amounts with 95% confidence intervals were calculated as $100 (CI: 100, 100) for full compensation, $120 (CI: 100, 125) for overcompensation, and $77.5 (CI: 70, 85) for partial compensation. Accordingly, we assigned figures of $100, $120, and $80 (round numbers) to the full compensation, overcompensation, and partial compensation conditions, respectively. The gamification of compensation was verified as in Study 2, and the results confirmed the manipulation (M_{Gamified/Nongamified} = 5.21/2.75, SE = 0.08/0.09, F(1, 549) = 439.15, p < .001, d = 1.33). Participants perceived the scenarios as realistic (M = 5.40), exhibiting scores that were significantly above the scale midpoint (ps < .001) across all conditions.

Main and Interaction Effects

Two ANCOVAs featuring compensation type and compensation level as the independent variables and satisfaction and enjoyment as the dependent variables were conducted. The control variables were the same as those used in Study 2 with the addition of failure severity. The results revealed significant main effects of both compensation type (M_{Gamified/Nongamified} = 4.86/4.53, SE = 0.10/0.10, F(1, 540) = 4.79, p = .029, d = 0.18) and compensation level (M_{Partial/Full/Over} = 3.81/4.89/5.39, SE = 0.12/0.13/0.12, F(1, 540) = 44.61, p < .001, d = 0.41) on satisfaction.

Importantly, the interaction effect between compensation type and level was significant (F[2, 540] = 11.36, p < .001, d = 0.20), as shown in Figure 3 (Panel C). Pairwise comparisons indicated that gamified compensation led to greater satisfaction than nongamified compensation in both the full compensation (M_{Gamified/Nongamified} = 5.37/4.41, SE = 0.18/0.18, p < .001, d = 0.55) and overcompensation conditions (M_{Gamified/Nongamified} = 5.70/5.07, SE = 0.17/0.17, p = .009, d = 0.43). However, at the partial compensation level, gamification led to decreased satisfaction among complainers (M_{Gamified/Nongamified} = 3.52/4.11, SE = 0.18/0.17, p = .020, d = −0.32). Interestingly, the results indicated no significant difference in satisfaction between the gamified full compensation condition and the nongamified overcompensation condition (M_{Gamified-Full/Nongamified-Over} = 5.37/5.07, SE = 0.12/0.12, p = .23, d = 0.16), thus indicating that gamification is an alternative strategy to overcompensation.

Regarding enjoyment, significant main effects of both compensation type (M_{Gamified/Nongamified} = 4.44/3.87, SE = 0.09/0.09, F(1, 540) = 17.18, p < .001, d = 0.34) and level (M_{Partial/Full/Over} = 3.57/4.34/4.56, SE = 0.11/0.11/0.10, F(1, 540) = 23.67, p < .001, d = 0.30) were observed, as was a significant interaction (F[2, 540] = 7.71, p < .001, d = 0.17), which is illustrated in Figure 3 (Panel C). Pairwise comparisons indicated that gamified compensation led to increased enjoyment in both the full compensation (M_{Gamified/Nongamified} = 4.86/3.83, SE = 0.16/0.16, p < .001, d = 0.70) and overcompensation conditions (M_{Gamified/Nongamified} = 4.95/4.17, SE = 0.15/0.15, p < .001, d = 0.52). Regarding partial compensation, we found no significant differences in enjoyment across compensation types (M_{Gamified/Nongamified} = 3.51/3.62, SE = 0.16/0.15, p = .62, d = −0.07). Furthermore, the gamified full compensation condition resulted in greater enjoyment than the nongamified overcompensation condition (M_{Gamified-Full/Nongamified-Over} = 4.86/4.17, SE = 0.16/0.15, p = .002, d = 0.32).

Moderated Mediation Effects

Moderated mediation analyses were performed using PROCESS Model 8. Compensation type was included as the independent variable, enjoyment as the mediator, compensation level as a multicategorical moderator (reference category: partial compensation), satisfaction as the dependent variable, and the same control variables as discussed above. Full mediation results are provided in the table within Supplemental Appendix Table A3(c). The findings revealed a significant index of moderated mediation for both full compensation versus partial compensation (b = 0.39, SE = 0.12, 95% CI [0.16, 0.65]) and overcompensation versus partial compensation (b = 0.30, SE = 0.12, 95% CI [0.09, 0.55]). Specifically, a significant indirect effect of gamified (vs. nongamified) compensation on satisfaction through enjoyment was observed for full compensation (b = 0.35, SE = 0.09, 95% CI [0.19, 0.53]) or overcompensation (b = 0.27, SE = 0.08, 95% CI [0.12, 0.43]), while no such significant indirect effect was observed for partial compensation (b = −0.04, SE = 0.08, 95% CI [−0.21, 0.12]).

Design

This study examined the effectiveness of gamified compensation when customers are pressed for time when seeking a resolution. We employed a 2 (compensation type: nongamified vs. gamified) by 2 (time pressure: low vs. high) between-subjects design in a hotel setting. This context has been extensively used in prior studies because participants are highly familiar with such a context (Gelbrich, Gäthke, and Grégoire 2016).

A vignette encouraged respondents to imagine that they had booked a hotel for an extended weekend break. On the first day, they ask the receptionist to make a reservation for a theatre performance the following evening. The receptionist informed them that this reservation would be made by the service department and that the tickets would be available at 4 p.m. the following day. The following day, they try to pick up the tickets at 4 p.m. to be ready for their dinner reservation before the show. However, it took nearly 1 hour to find the tickets. They then complained that they had waited too long. In the low-time pressure condition, participants were told they still had 3 hours until their dinner reservation, which ensured that they had plenty of time to reach their destination. In the high-time pressure condition, respondents were told they were already late for their dinner reservation. Following the complaint, those in the nongamified compensation condition received an apology and were offered complimentary access to the spa as compensation. However, in the gamified compensation condition, they received an apology and were offered a chance to spin the wheel to win a reward, which resulted in complimentary spa access.

Data Collection

The sample criteria and recruitment procedure mirrored those used in Study 2. An attention check (“Based on the scenario, what were the lost ticket reservations for?”) and two instruction checks (i.e., “Please click on ‘agree’ for this question,” and “Out of the following answers, please click ‘Never’ for this question”) were included in the questionnaire. An initial sample of 264 U.S. adults was recruited via the Prolific platform in the winter of 2023, but respondents who did not pass these checks were excluded from the survey (n = 22). The final sample included 242 participants (M_Age = 39.68, SD = 14.14; Female = 53.3%).

Measures

The same measures of model constructs, controls, gamification of compensation manipulation, attention checks, and realism checks employed in Study 2 were used in this study (see Supplemental Appendix Table A1).

Manipulation and Realism Checks

To check the time pressure manipulation, participants responded to a single item, i.e., “How much time pressure did you feel for making dinner reservations on time?”, on a seven-point scale (1 = not at all to 7 = very much). The gamification of compensation was checked similar to Study 2. Two ANOVAs including compensation type and time pressure as independent variables and their respective manipulation check scales as dependent variables were conducted. The results indicated that both the manipulation of compensation type (M_{Gamified//Nongamified} = 5.54/2.35, SE = 0.10/0.09, F(1, 238) = 551.05, p < .001, d = 1.67) and time pressure were successful (M_{HighTP/Low TP} = 6.20/2.59, SE = 0.11/0.11, F(1, 238) = 547.11, p < .001, d = 1.67). No cross-effects of the time pressure manipulation on perceived gamification (p = .93) or of compensation type on perceived time pressure (p = .96) were observed. The scenarios were perceived as realistic (M = 5.53), with scores above the scale midpoint (ps < .001) across all conditions.

Main and Interaction Effects

Two ANCOVAs featuring compensation type and time pressure as the independent variables predicting the dependent variables (satisfaction and enjoyment) were conducted. The same controls used in Study 3 were included. Regarding satisfaction, the results indicated a significant main effect of time pressure (M_{High TP/Low TP} = 3.64/4.90, SE = 0.15/0.15, F(1, 233) = 35.79, p < .001, d = −0.69) but no significant main effect of compensation type (p = .78); time pressure and compensation type interacted with one another (F[1, 233] = 18.54, p < .001, d = 0.28), as depicted in Figure 3 (Panel D). Pairwise comparisons revealed that compared with nongamified compensation, gamified compensation led to increased satisfaction when time pressure was low (M_{Gamified/Nongamified} = 5.36/4.43, SE = 0.22/0.19, p = .001, d = 0.59) and decreased satisfaction when it was high (M_{Gamified/Nongamified} = 3.23/4.05, SE = 0.22/0.20, p = .007, d = −0.43).

Regarding enjoyment, significant main effects of both compensation type (M_{Gamified/Nongamified} = 3.97/3.37, SE = 0.14/0.13, F(1, 233) = 9.18, p = 0.003, d = 0.36) and time pressure (M_{High TP/Low TP} = 3.10/4.25, SE = 0.14/0.14, F(1, 233) = 34.29, p < .001, d = −0.68) were observed, as was a significant interaction (F[1, 233] = 17.76, p < .001, d = 0.28), which is illustrated in Figure 3 (Panel D). Pairwise comparisons indicated that gamified compensation led to greater enjoyment when time pressure was low (M_{Gamified/Nongamified} = 4.95/3.55, SE = 0.20/0.18, p < .001, d = 0.85) but resulted in no difference in enjoyment when time pressure was high (M_{Gamified/Nongamified} = 3.0/3.20, SE = 0.20/0.19, p = .47, d = 0.01).

Moderated Mediation Effects

PROCESS Model 8 was specified with compensation type as the independent variable, satisfaction as the dependent variable, and enjoyment as the mediator. Time pressure was included as the moderator, and the same control variables used in Study 3 were employed. Supplemental Appendix Table A3(d) provides the full mediation results. The results indicated a significant index of moderated mediation (b = −0.56, SE = 0.18, 95% CI [−0.96, −0.24]). A significant positive indirect effect of gamified compensation (vs. nongamified compensation) on satisfaction through enjoyment was observed when time pressure was low (b = 0.49, SE = 0.15, 95% CI [0.24, 0.80]), but no significant indirect effect was observed when time pressure was high (b = −0.07, SE = 0.11, 95% CI [−0.29, 0.14]).

Gamification-Specific Factors

First, we only used a spin-the-wheel game, which is a common luck game. While this operationalization tends to capture the essence of gamification mechanisms, a broader set of game types could help enhance the generalizability of gamification in recovery situations. Gamified recovery using quizzes, or contests, might generate nuanced results because they relate to the participants’ performances or abilities, instead of pure luck. Thus, subsequent research should investigate the efficacy of quiz gamified recovery using a field study, for instance, in an airline context where passengers are waiting for a delayed flight at the airport. In addition, various designs of gamified recovery should be compared. For example, to avoid any bias, the potential rewards were not displayed on our wheels. Thus, future research should investigate whether reward visibility could influence recovery outcomes as it may trigger comparisons with other customers. This can be examined with a field experiment in a physical retail context where customers are queuing at a customer service desk. Furthermore, the level of difficulty associated with the gamified activity should be further explored to calibrate the offered rewards. As such, future research can investigate the difficulty of a quiz game to determine the compensation level following a service failure. Finally, further research may explore the reasons for consumers’ willingness to engage in a gamified recovery using a qualitative approach. Here, consumers’ inner emotional states and the prospect of a potential loss may be promising areas to pursue.

General Service Recovery Factors

Despite our efforts to examine different situational factors and service settings, a more diverse assessment of the failure context (e.g., utilitarian vs. hedonic setting) may enrich our understanding of this topic by considering the appropriateness of gamification according to the context. Additionally, in all our studies, customers accepted the failures. However, it would be interesting to examine how customers respond to gamified recovery when they reject a service (Gelbrich, Gäthke, and Grégoire 2015). Future research may examine how gamified recovery changes if it is offered proactively rather than reactively. Finally, in our field experiment, we manipulated whether customers could choose between gamified or regular recovery. This “choice/no choice” manipulation may have introduced self-selection bias. Although the propensity score matching procedure we applied limits this bias, future research should examine factors such as participants’ openness to gaming and their emotional state.

Process-Related Factors

Our findings show that perceived enjoyment explains the positive effects of gamification but does not account for its negative effects. Future research should examine the conditions under which a given process exhibits superior predictive ability (Khamitov, Grégoire, and Suri 2020). Thus, future research may clarify the roles of justice, reassurance, and perceived control, the results of which are inconclusive. In particular, gamification may trigger a sense of control through performance-based activities (e.g., contests and quizzes, Leclercq et al., 2020b). This feeling may compensate for the lack of control perceived by customers when they experience service failures. Indeed, while regular service recovery offers compensation, gamified recovery conditions compensation to win a game-like activity, inducing uncertainty, which may be negatively perceived. Future research may examine complementary explanations, such as personal connection, which constitutes another rapport component in addition to enjoyment. This approach would greatly improve our overall understanding of gamification strategies in light of the exchange-based nature of failure-recovery interactions (Roschk and Gelbrich 2017).