Feeling threatened by robots? The role of service failure on consumers’ service evaluations

Understanding how individuals react to artificial intelligence is crucial and increasingly important. In particular, “anthropomorphism” (Epley et al., 2007), that is the extent to which machines are imbued with human-like characteristics, is a key determinant of consumers’ response to AI. This is the case of service robots, which as a result are perceived has having more “agency” (Yam et al., 2021), but also as threatening “human distinctiveness” (Mende et al., 2019).

However, while this effect has been shown when robots are working as expected (i.e., service success) (e.g., Mende et al., 2019), prior research has neglected to account for the threat perceived by consumers when robots make mistakes (i.e., service failure). We fill this gap and show that in the case of service failure (vs. service success) the negative effect of the perceived threat on evaluation is mitigated. We argue that this occurs because making mistakes and being wrong is an innate human trait (Schulz, 2011) that dampens consumers’ fears of robots threatening human relations, human identity, and humanity in general. We test our prediction in four experimental studies.

Through an Implicit Association Test (IAT; Greenwald et al., 1998), Study 1 (N = 72, 81% female, Mage = 22.09, SDage = 0.94, European sample) shows that participants had the tendency to attribute agency to the robots rather than to the automatic machine (MD-score = 0.160, SDD-Score = 0.440; t(71) = 3.085, p < 0.0020, 95% CI [0.05667, 0.2638]). In this sense, we show that consumers subconsciously believe that robots are more intelligent and have more agency than traditional machines and that this belief is activated regardless of whether individuals can observe the physical appearance of such robots.

In Study 2a (N = 400, 85% female, 4% non-binary; Mage = 24.17, SDage = 6.82, US sample from Prolific), and Study 2b (N = 400, 76% female, 3% non-binary; Mage = 26.14, SDage = 7.87, US sample from Prolific), we employed a 2 (type of agent: robot vs. self-check-in machine) × 2 (service outcome: success vs. failure) between-subjects design to test the effect of agent’s type and service’s outcome on consumers’ evaluation of the service. We tested our conceptualization respectively in two different settings: hotel check-in and restaurant order. In both studies, we run a conditional moderated mediation on PROCESS (model 14) with 10,000 bootstrap analyses. Results of Study 2a show that participants attributed more agency to the robot than to the automatic machine (b = .2954, SE = .1134, 95% CI [.0664, .5124]). Moreover, an increase in perceived agency leads to an increase in hotel evaluation (b = .2346, SE = .0581, 95% CI [.1203, .3488]). Most importantly, the index of moderated mediation was significant (Index = - .0720, BootSE = .0490, 95% CI [- .1894, - .0018]). When the service is a failure, participants tend to give higher hotel evaluation when they perform the check-in with the robot instead of the automatic machine (.1055, BootSE = .0546, CI 95% [.0218, .2320]). When the service is a success, there is no significant effect of the type of agent on hotel evaluation (.0356, BootSE = .0254, [- .0044, .0940]). The direct effect of type of agent on hotel evaluation was not significant (b = - .1092, SE = .1223, 95% CI [- .3497, .1312]). Results were replicated in Study 2b.

In Study 3 (N = 400, 77% female, 3% non-binary; Mage = 35.21, SDage = 14.47, US sample from Prolific), we used the same design as Study 2a and 2b but also included a measure of human identity threat (HIT; Mende et al., 2019).

We run model 87 in PROCESS (Hayes, 2018) with perceived agency as first mediator and HIT as second mediator. The results show that participants attributed higher agency to the robot than they did to the automatic machine (b = .3130, SE = .1209, 95% CI [.0753, .5507]). Moreover, an increase in perceived agency leads to an increase in human identity threat (b = .2563, SE = .0603, 95% CI [.1379, .3747]). Most importantly, the index of moderated mediation was significant (Index = - .0265, BootSE = .0162, 95% CI [- .0669, - .0041]). When the service is a success, participants tend to evaluate the service less positively when they interact with a robot rather than an automatic machine (- .0306, BootSE = .0170, CI 95% [- .0719, - .0057]). When the service is a failure, there is no significant effect of type of agent on evaluation (- .0041, BootSE = .0065, [- .0196, .0072]). The direct effect of type of agent on evaluation was not significant (b = - .0387, SE =.0568, 95% CI [- .3026, .2253]).

Overall, our results contribute to a better understanding of consumers’ reactions to automatic machines in service contexts. In particular, we show that consumers are more likely to judge a service failure less negatively when the failure involves a robot rather than an automatic machine. We provide evidence that while robots are perceived as having agency and as threatening human identity, the negative effect of threat on consumers’ evaluations is mitigated in case of service failure.