1 Introduction

Customer comfort during service interactions is essential for creating enjoyable customer experiences. In fact, customer comfort, which is understood as “a sense of ease that facilitates calm and worry-free feelings within an environment” (Ainsworth and Foster 2017, p. 27), is a driver of positive customer service outcomes, such as customer satisfaction and word-of-mouth (Gaur et al. 2009; Lloyd and Luk 2011; Paswan and Ganesh 2005; Spake et al. 2003). For traditional human-to-human service settings, service providers know that they can make customers feel more comfortable by optimizing the atmosphere of the servicescape or by better training their employees (Ainsworth and Foster 2017; Lloyd and Luk 2011). However, increasingly many service encounters are “untact” and involve service robots instead of human employees (Lee and Lee 2020). As interactions with service robots can be perceived as eerie and threatening (Mende et al. 2019), it is essential to determine how customers can be made more comfortable during these novel types of service encounters. Here, especially service robots’ physical appearance and human-likeness could be important drivers. In fact, research has demonstrated that a service robot’s degree of human-likeness can impact customer service outcomes, such as customer satisfaction and loyalty, as well as the degree to which the service robot is perceived as eerie and threatening (Belanche et al. 2021b; Mende et al. 2019; Yoganathan et al. 2021).

Yet, while it seems very likely that human-likeness in service robots does impact customer comfort levels, it is currently completely unclear whether human-likeness is an enabler or an inhibitor of customer comfort. On the one hand, human-likeness could facilitate customer-robot rapport (Qiu et al. 2020). Rapport captures a personal connection and enjoyment during the interaction with the service employee (Gremler and Gwinner 2000) and higher levels of rapport could make customers feel more comfortable. On the other hand, human-likeness could also increase a service robot’s social presence, that is the degree to which the service robot is perceived as a social being (Heerink et al. 2010). Because social presence gives customers the impression that the service robot has its own intentions and objectives (Biocca 1997), we argue that higher levels of social presence make customers feel less comfortable. Therefore, based on existing literature, it remains unclear whether customers feel more or less comfortable when interacting with human-like service robots.

Importantly, whether human-likeness is an enabler or and an inhibitor of customer comfort might also be greatly influenced by service failures. Wirtz et al. (2018) point out that customers have exaggerated expectations towards human-like service robots. If a human-like service robot then makes a mistake, customers are argued to perceive this mistake as being worse than if the same error had been committed by a machine-like service robot (Choi et al. 2021; Wirtz et al. 2018). In this way, service failures might diminish the effect of human-likeness on customer-robot rapport and social presence. This could then also diminish the effect of human-likeness on customer comfort.

With our empirical study, we aim to reconcile these opposing predictions. In particular, we investigate whether the comfort-enabling effects (i.e., customer-robot rapport) or the comfort inhibiting effects (i.e., social presence) of human-likeness dominate. Moreover, we investigate how these effects are impacted by service failures.

In this way, we make three major contributions to the service robot and service technology literature. First, we introduce the concept of customer comfort to the service robot field and identify a service robot’s degree of human-likeness as one of its major drivers. In addition, we show that more customer comfort translates into higher customer satisfaction, engagement, word-of-mouth intention, and willingness-to-pay. Second, we contribute to the service robot literature by showing that human-likeness has a positive rather than a negative impact on customer comfort. Here, it is demonstrated that customers feel more comfortable interacting with human-like service robots because it is easier to build rapport with them. Increased levels of social presence do not appear to explain the discomforting effects of human-likeness observed by previous research (e.g., Mende et al. 2019). Third, we contribute to the service robot literature by demonstrating that the positive indirect effect of human-likeness diminishes in the presence of service failures. Lastly, the present work contributes to practice by demonstrating that human-like (vs machine-like) service robots are the superior choice. Importantly, even for contexts in which service robots commit many errors, we show that human-likeness does not appear to become a liability.

In the remainder of this study, we first lay out the underlying theories and hypothesized relationships. Afterward, we explain the experimental procedures and analysis techniques. Lastly, based on the reported results, we discuss implications.

2 Theory and conceptual model

2.1 Service robots in hospitality

The hospitality industry is seen as one of the most relevant contexts for service robots (Garcia-Haro et al. 2021). Service robots, which Wirtz et al. (2018, p. 909) define as “system-based autonomous and adaptable interfaces that interact, communicate and deliver a service to an organization’s customers”, could help the industry overcome its pressing labor shortages. Especially in the wake of the Covid pandemic, in the US about 50% and in Germany about 16% of service employees left the industry (Dmitrieva 2021; Sullivan 2021). By taking over tasks such as greeting guests, serving dishes, or taking customer orders (Decker et al. 2017; Pieskä et al. 2013), service robots could increasingly help alleviate these shortages.

Also service research increasingly investigates service robots in hospitality settings. For example, investigating self-service machines in restaurants, Jeon et al. (2020) show that customer acceptance is largely driven by performance expectancies. Furthermore, Chiang and Trimi (2020) show that guests prioritize assurance and reliability in their service quality evaluations of hotel service robots. Using a restaurant-based vignette study, Belanche et al. (2021a) demonstrate that customer attributions (i.e., service enhancement and cost reduction) mediate the relationship between customers’ affinity towards a service robot and intentions to use and recommend it. Lu et al. (2021) show that human-likeness in voice and language style affects several consumption outcomes, such as service encounter evaluations. Similarly, Qiu et al. (2020) show that service robots’ human-likeness contributes to the overall hospitality experience. They also show that this effect is mediated by customer-employee rapport. In one of the first field studies on service robots in a restaurant setting, Odekerken-Schröder et al. (2022) demonstrate the need for employee-service robot teams as helpful and friendly staff can compensate for service robot errors. Also collecting real-life data from customers who had been served by a robot barista, Kim et al. (2021) find that atmosphere, novelty, and consumer return on investment are important drivers of customer satisfaction and behavioral intentions.

2.2 Customer comfort during service encounters

We take a customer comfort perspective to better understand to what extent service robots should be human-like. Predominantly, (customer) comfort has been considered in healthcare and nursing contexts. For these contexts, Kolcaba et al. (2006) formulated the so-called Comfort Theory, conceptualizing four kinds of comfort: physical, psychospiritual, social, and environmental. In their conceptual framework, antecedents (e.g., comfort interventions) and consequences of comfort (e.g., health-seeking behaviors) are specified (Krinsky et al. 2014).

Also in retailing and hospitality contexts, service scholars have investigated antecedents and consequences of customer comfort. Here, service scholars traditionally distinguish between two kinds of comfort: physical comfort and psychological comfort. Physical comfort depends on the physical aspects of the environment, such as temperature, humidity, or lighting (Chua et al. 2016; Kinnane et al. 2013). In contrast, psychological comfort “represents a sense of ease that facilitates calm and worry-free feelings within an environment” (Ainsworth and Foster 2017, p. 27). In a service context, psychological comfort can further be described “as a feeling of anxiety or relaxation arising from the social interaction with the service employee.” (LLoyd and Luk 2011, p. 178). We solely focus on psychological comfort and generally refer to this kind of comfort as customer comfort.

Research has identified multiple drivers of customer comfort. For example, Ainsworth and Foster (2017) show that customer comfort is impacted by a shop’s layout and a consumer’s familiarity with the retailer. Furthermore, Lloyd and Luk (2011) find that employees’ service manner and need identification behaviors are important drivers. Lastly, Rosenbaum et al. (2018) demonstrate that social incompatibilities stemming from war, nationalism, or religious differences greatly influence customer comfort.

Also downstream consequences of customer comfort have been investigated in retail and hospitality settings. For brick and mortar shops, Ainsworth and Foster (2017) relate customer comfort to utilitarian and hedonic value. In a beauty salon and physician context, Spake et al. (2003) find that customer comfort is associated with customer satisfaction and commitment. Also Paswan and Ganesh (2005) make similar observations in a university context. In a bank context, Gaur et al. (2009) establish a positive association between comfort, satisfaction, and loyalty. Lastly, for retailing and casual dining, Lloyd and Luk (2011) find that comfort increases customer satisfaction and ultimately word-of-mouth intention.

Throughout our study, we build on these findings. In particular, we argue that also during robot-enabled service encounters, customer comfort is not an end in itself but that it translates into managerially relevant customer service outcomes. Therefore, we investigate customer comfort’s downstream effect on customer satisfaction, word-of-mouth, engagement, and willingness-to-pay. We understand engagement as a customer’s level of interaction and connection with a brand or its offerings that goes beyond purchase (Vivek et al. 2014).

2.3 The relationship between human-likeness and customer comfort

The service robot literature indicates that a service robot’s degree of human-likeness impacts customer comfort. Human-likeness captures a service robot’s physical similarity with humans (Belanche et al. 2020a). It is distinct from anthropomorphism, which is not related to a robot’s design but to a user’s attribution of human characteristics and traits to it (Blut et al. 2021; Epley et al. 2007). As we seek to understand to what extent the design of human-like service robot appearances impacts customer comfort, we investigate the concept of human-likeness.

Although the literature strongly indicates that human-likeness impacts customer comfort, it is not clear whether this impact is positive or negative. Based on existing literature, both could be hypothesized. On the one hand, human-likeness has been found to make customers feel threatened and eerie (Mende et al. 2019). Exploring explanations for the observed effects, Mende et al. (2019) suggest that people feel threatened by human-like service robots due to evolutionary mechanisms (Gray and Wegner 2012), that human-like service robots elicit mortality salience (MacDorman 2005), and that people are afraid of losing control, losing their jobs, or the world being taken over by robots (Ray et al. 2008). Also Akdim et al. (2021) show that attitudes towards service robots become increasingly negative as service robots’ human-likeness increases. It stands to reason that such negative perceptions and attitudes make customers feel less comfortable.

On the other hand, human-likeness has been theoretically and empirically associated with more positive customer service outcomes which are highly correlated with customer comfort (e.g., Lloyd and Luk 2011). Zhang et al. (2008) propose that anthropomorphism leads to higher service quality and satisfaction. This view is shared by van Doorn et al. (2017) who suggest that human-likeness can have positive effects on customer service outcomes. Murphy et al. (2019) propose three marketing outcomes that are dependent on a service robot’s human-likeness (i.e., affective reaction, acceptance, and anthropomorphic loyalty). Besides these conceptual studies, Qiu et al. (2020) empirically demonstrate that the hospitality experience is positively affected by a service robot’s human-likeness. With multiple experiments, Yoganathan et al. (2021) show that human-likeness positively impacts customers’ willingness-to-pay as well as their visit intention. Also, Belanche et al. (2021b) find a positive relationship between the human-likeness in service robots and customer loyalty intentions. Therefore, human-likeness appears to generally have a positive impact on customer service outcomes. Given the high correlation between customer comfort and customer service outcomes (e.g., Lloyd and Luk 2011), it stands to reason that customers feel more comfortable when interacting with a human-like service robot than when interacting with a machine-like service robot.

Therefore, human-likeness could be an inhibitor of customer comfort as it can provoke eerie customer reactions (Mende et al. 2019), but at the same time, it could also be an enabler of customer comfort as it can drive more positive customer service outcomes (Belanche et al. 2021b). Even though these findings point into opposite directions, they nevertheless hint at a general association between human-likeness and customer comfort. We formalize this expectation in our first hypothesis, H1. In the following subsections, we further build on this relatively broad hypothesis to investigate the direction of the relationship.

H1

A service robot’s degree of human-likeness impacts customer comfort.

2.3.1 The mediating roles of rapport and social presence

The present work seeks to resolve current tension in the service robot literature and determine whether human-likeness makes customers feel more or less comfortable. To do so, it is examined how human-likeness impacts customer comfort. Two underlying mechanisms are investigated. Because customer comfort is driven mainly by the social interaction with the service employee (Lloyd and Luk 2011), these mechanisms are of a social and relational nature. The first investigated mechanism, rapport, is argued to be a driver of increased customer comfort. The second investigated mechanism, social presence, is argued to be a driver of decreased customer comfort. The two mechanisms are discussed in the following subsections.

2.3.2 Rapport

The first mechanism argued to underlie the relationship between human-likeness and customer comfort is customer-robot rapport. According to Gremler and Gwinner (2000), rapport is “a customer’s perception of having an enjoyable interaction with a service provider employee, characterized by a personal connection between the two interactants” (p. 92). Multiple studies demonstrate rapport’s high explanatory power for understanding customer satisfaction, word-of-mouth, and loyalty (Gremler and Gwinner 2000; Macintosh 2009; Delcourt et al. 2013). Consequently, next to functional and outcome-related elements, customer-employee rapport appears to play a central role during service encounters (Gremler and Gwinner 2000). As a result, also service robot research has increasingly shown interest in customer-robot rapport. For example, Wirtz et al. (2018) see rapport as one of two major relational drivers of customer acceptance of service robots. Also Qiu et al. (2020) investigate rapport to better understand hospitality experiences involving service robots.

The literature provides many indications that rapport can be built between a service robot and the customer. Wirtz et al. (2018) propose human–robot rapport as a key driver of service robot acceptance. This link finds empirical support by Fernandes and Oliveira (2021). Furthermore, Nomura and Kanda (2016), Seo et al. (2018), and Lubold et al. (2019) all show that users can build rapport with robots. Seo et al. (2018) also find that human–robot rapport-building behavior is similar to what is predicted by the human–human rapport literature. Lastly, Qiu et al. (2020) show that human-likeness positively impacts human–robot rapport. Therefore, following the results by Qiu et al. (2020), the present work argues that customers more easily build rapport with a human-like than with a machine-like service robot.

Based on the understanding of rapport as defined by Gremler and Gwinner (2000) as well as of customer comfort as defined by Lloyd and Luk (2011), there is reason to believe that customer-robot rapport makes customers feel more comfortable. Gremler and Gwinner (2000) define rapport as a “personal connection” (p. 92) between the customer and the service employee that is accompanied by an “enjoyable interaction” (p. 92). Customer comfort is defined as being largely driven by such enjoyable interactions (LLoyd and Luk 2011). Therefore, because human-like service robots can build more rapport (i.e., create more enjoyable interactions), customers are hypothesized to feel more comfortable when interacting with a human-like service robot.

H2

Customer-robot rapport mediates the positive relationship between a service robot’s human-likeness and customer comfort.

2.3.3 Social presence

The second investigated mechanism is social presence. Social presence captures the extent to which a machine or robot is perceived as another social entity (Heerink et al. 2010). In this way, social presence goes beyond physical presence which merely captures the extent to which something is in the same physical or virtual space (Biocca et al. 2003). For example, a sculpture can be physically present in a room. However, due to its lack of social interactivity and non-apparent intelligence, it would not be perceived as being socially present (Biocca et al. 2003; Odekerken-Schröder et al. 2022). In contrast, a robot could provoke perceptions of social presence by interacting with its environment (Biocca et al. 2003). As a result, people would have the impression to not be interacting with a machine, but with another (real) person that has own feelings, thoughts, and intentions (Biocca et al. 2003; Heerink et al. 2010).

It has been argued that a service robot’s level of human-likeness is a key driver of perceived social presence in service robots (e.g., Yoganathan et al. 2021). This view also finds empirical support in the literature. Kontogiorgos et al. (2020) find that a human-like embodiment of an AI assistant is perceived as having a higher social presence than a smart speaker embodiment of the same AI assistant. In addition, Barco et al. (2020) find that children perceive an anthropomorphic robot as more socially present than a zoomorphic robot. Lastly, with a field study in a restaurant context, Odekerken-Schröder et al. (2022) find a positive relationship between anthropomorphism and social presence. Therefore, the present study expects human-likeness to increase a service robot’s social presence.

Contrasting the above discussion about customer-robot rapport, the present work argues that social presence makes customers feel less comfortable. Mende et al. (2019) show that customers feel eerie and threatened when interacting with human-like service robots. In this way, they provide empirical support for multiple studies suggesting such a threatening effect of human-likeness (Gray and Wegner 2012; MacDorman 2005; Ray et al. 2008). Here, we argue that these threatening and eerie effects do not directly stem from human-likeness itself but rather from the social presence of the service robot. In particular, Biocca (1997) argues that social presence provokes the attribution of intentions. Similar to Gray and Wegner (2012) who show that people find it unnerving when a robot appears to have its own thoughts, the present work argues that people feel threatened and uncomfortable if a service robot appears to have its own goals and objectives. Therefore, it is hypothesized that social presence mediates the negative effects of human-likeness on customer comfort observed by Mende et al. (2019).

H3

A service robot’s social presence mediates the negative relationship between the service robot’s human-likeness and customer comfort.

2.3.4 The moderating role of service failures

Whether human-likeness is an enabler or an inhibitor of customer comfort might also be greatly influenced by the presence or absence of service failures. Service failures occur “when customers’ expectations are not met, or service performance falls below a customer’s expectation” (Geum et al. 2011, p. 3127). As Belanche et al. (2020b) point out, customers might perceive service failures by human-like and machine-like service robots very differently. Also Wirtz et al. (2018) expect service failures to weigh more heavily for human-like than for machine-like service robots. In particular, they argue that customers have relatively low expectations towards a machine-like service robot, while they have exaggerated expectations towards a human-like service robot. Consequently, they argue that compared to a machine-like service robot, customers would be considerably more disappointed when a human-like service robot makes a mistake. Based on this argumentation, Wirtz et al. (2018) suggest that human-likeness in service robots should be limited. Finding empirical evidence for this interplay between human-likeness and service failures, Choi et al. (2021) show that service failures do indeed have a more negative impact for human-like service robots than for machine-like service robots. Consequently, the present work hypothesizes that service failures diminish the effects of human-likeness on customer-robot rapport and social presence.

2.3.5 Rapport

Following the argumentation by Wirtz et al. (2018), we argue that customers have exaggerated expectations toward human-like service robots while they have relatively low expectations towards machine-like service robots. This means that customers would be greatly disappointed if a human-like service robot makes a mistake, while they would be much less disappointed if the same mistake was committed by a machine-like service robot (Wirtz et al. 2018). We argue that this increased disappointment offsets the positive effects of human-likeness on customer-robot rapport. This expectation is also in line with Choi et al. (2021) who find that service failures weigh more heavily for human-like service robots than for machine-like service robots.

Therefore, while human-like service robots would generally establish more rapport with a customer than machine-like service robots, a service failure would diminish the additional rapport that human-likeness helped to build. As a consequence, following a service failure, we no longer expect a human-like service robot to have more rapport with a customer than a machine-like service robot. In this way, service failures are also expected to diminish the positive indirect effect of human-likeness on customer comfort.

H4a

Service failures diminish the indirect effect of human-likeness on customer comfort mediated by customer-robot rapport.

2.3.6 Social presence

Consistent with Wirtz et al. (2018), we argue that also the effect of a service robot’s human-likeness on its social presence diminishes in case the service robot makes a mistake. When people perceive a service robot as socially present, they feel as if they are with another social and intelligent entity (Biocca et al. 2003; Heerink et al. 2010). If a service robot makes a mistake, this could disillusion customers by reminding them that the service robot is merely a machine but no intelligent or social being. Hence, it is expected that service failures diminish the social presence that the service robot’s human-likeness helped to build (Odekerken-Schröder et al. 2022). Consequently, following a service failure, human-like service robots would no longer be perceived as more socially present than machine-like service robots.

We had previously argued that customers would feel less comfortable when interacting with a human-like service robot because they perceive it as having its own thoughts and intentions (Biocca 1997). However, as customers would no longer perceive a human-like service robot as being socially present following a service failure, this realization would also take their fears. As a result, we hypothesize that service failures diminish the negative indirect effect of human-likeness on customer comfort. The complete conceptual model including all hypotheses is shown in Fig. 1.

Fig. 1
figure 1

Conceptual model

Full size image

H4b

Service failures diminish the indirect effect of human-likeness on customer comfort mediated by social presence.

3 Methodology

To validate our hypotheses, we conducted an online experiment. This experiment was set in a restaurant context, which is seen as one of the most relevant contexts for service robots (Garcia-Haro et al. 2021). Our experiment and especially our human-likeness manipulation are greatly inspired by one of Europe’s first robot-restaurant chains. This restaurant chain employs two different kinds of service robots, of which one is very machine-like and the other is much more human-like (see Fig. 2). This difference in appearance did not only inspire our study, but we also used images of the restaurant’s service robots as stimuli for our experiment.

Fig. 2
figure 2

Machine-like and human-like service robot stimuli

Full size image

3.1 Procedure

We employed a 2 (human-like service robot vs machine-like service robot) × 2 (no service failure vs service failure) between-subjects design. 438 participants were recruited through the online platform MTurk in May 2021. To minimize cultural differences between respondents and make sure that they would understand our English questionnaire, respondents had to be residents of the USA. After informing participants about the purpose of the study, they saw an image of either a human-like or a machine-like service robot (see Fig. 2). Respondents were then asked to read a short scenario in which the shown service robot served them (see Online Appendix A). This narrative was heavily based on Smith et al. (1999) and either described a situation in which the customer received a correct dish or an incorrect dish (i.e., failure vs no failure condition). Afterward, respondents were asked to complete a questionnaire.

3.2 Measurement

Table 1 shows all measures used in our study, their original authors, and the original studies’ Cronbach’s alphas. We coded human-likeness and customer comfort on 7-point bipolar scales. All other constructs were coded on 7-point Likert scales. To match the service robot context, we slightly adapted the wording of some scales. In addition, we omitted items that could not be answered by respondents based on a text-based scenario.

Table 1 Scales for construct measurement
Full size table

3.3 Data cleaning

Before the analysis, we removed all invalid responses. This included responses containing incorrect values (e.g., NAs). In addition, the survey contained six attention checks to make sure respondents paid attention from beginning to end. The first three attention checks made sure respondents read the scenario carefully. The last three attention checks asked respondents to set the answer to a specific question to either “strongly agree” or to “strongly disagree”. In this way, respondents who sped through the questions without reading them were identified.

To ensure high data quality, respondents who did not answer all six attention checks correctly were excluded. This resulted in 161 valid responses. While the number of excluded respondents seems initially high, it matches very well the expectations by Aguinis et al. (2020) who review and advise on the use of MTurk. In particular, they find that on average every attention check is missed by 15% of MTurk respondents. As we included six attention checks, our exclusion rate is almost exactly as predicted by Aguinis et al. (2020).

3.4 Analysis

We analyzed the data using the PROCESS macro (Hayes 2017). To create bias-corrected estimates for the indirect effects, 5000 bootstrap samples were run. As discussed in Online Appendix D, we calculated heteroskedasticity-robust standard errors. We included respondents’ age, gender, and previous interactions with service robots as controls for all tested models. Previous robot interaction was measured on an ordinal scale (0 = no interactions, 1 = 1–3 interactions, 2 = 4–10 interactions, 3 = More than 10 interactions).

4 Results

4.1 Descriptive statistics

The final analysis includes 161 responses. With regards to gender, the sample is balanced with 82 female and 79 male participants. As discussed in more depth in Online Appendix D, the sample is relatively young (M = 41.48, σ = 11.92, Min = 19, Max = 75). 105 respondents reported no previous interactions with service robots, 46 reported 1–3 interactions and ten respondents reported 4–10 interactions. 80 participants were randomly assigned to the human-like service robot condition and 81 participants to the machine-like service robot condition. In addition, 89 participants were randomly exposed to a service failure, 72 were not. Thus, the four resulting subgroups consist of 31 to 50 respondents. All measures show high internal consistency as measured by Cronbach’s alpha: αhuman-likeness = 0.94, αrapport = 0.93, αsocial presence = 0.78, αcustomer comfort = 0.97, αsatisfaction = 0.94, αword-of-mouth = 0.96, αengagement = 0.96, αwillingness-to-pay = 0.92.

4.2 Pretest and manipulation check

To verify the human-likeness manipulation, a pretest with 105 respondents was conducted via MTurk. The results show that the human-like service robot was perceived as significantly more human-like (M = 3.60) than the machine-like service robot (M = 2.02, p = 0.00). Also in the actual experiment, a manipulation check confirms that the human-like service robot is perceived as significantly more human-like (M = 3.54) than the machine-like service robot (M = 2.74, p = 0.00).

4.3 Analyses

4.3.1 Model 0: association between human-likeness and customer service outcomes

It is still debated whether human-likeness in service robots is advantageous or disadvantageous (Akdim et al. 2021; Mende et al. 2019; Wirtz et al. 2018; Yoganathan et al. 2021). To offer additional insights regarding the general effect of human-likeness, Model 0 investigates the association between human-likeness and customer service outcomes. Except for customer satisfaction (b = 0.35; HC3 SE = 0.25; p = 0.16), the human-like service robot is associated with significantly higher customer engagement (b = 0.61; HC3 SE = 0.24; p = 0.01), word-of-mouth (b = 0.60; HC3 SE = 0.27; p = 0.03), and willingness-to-pay (b = 0.67; HC3 SE = 0.25; p = 0.01). Figure 3 visually presents the results of all tested models.

Fig. 3
figure 3

Overview of models with results

Full size image

4.3.2 Model 1: mediation through customer comfort

Model 1 investigates how a service robot’s human-likeness impacts customer comfort and assesses to what extent customer comfort mediates the relationship between human-likeness and customer service outcomes. In support of hypothesis H1, human-likeness is positively associated with customer comfort (b = 0.46; HC3 SE = 0.24; p = 0.06). In turn, customer comfort is positively associated with customer satisfaction (b = 0.73; HC3 SE = 0.07; p < 0.001), word-of-mouth (b = 0.70; HC3 SE = 0.08; p < 0.001), engagement (b = 0.60; HC3 SE = 0.08; p < 0.001), and willingness-to-pay (b = 0.61; HC3 SE = 0.07; p < 0.001).

To investigate whether customer comfort mediates the impact of human-likeness on customer service outcomes, a mediation analysis is conducted using Model 4 of the PROCESS macro (Hayes 2017). The 90% confidence intervals around the indirect effects exclude 0 for customer satisfaction (LLCI = 0.05; ULCI = 0.64), engagement (LLCI = 0.04; ULCI = 0.54), word-of-mouth (LLCI = 0.05; ULCI = 0.62), and willingness-to-pay (LLCI = 0.04; ULCI = 0.54). While we fail to observe a mediation at the 95% confidence level, these results nevertheless indicate that customer comfort mediates the impact of human-likeness on customer service outcomes.

4.3.3 Model 2: mediation through rapport and social presence

Model 2 investigates whether rapport and social presence mediate the impact of human-likeness on customer comfort. Human-likeness significantly impacts rapport (b = 0.58; HC3 SE = 0.23; p = 0.01), however, it does not significantly impact social presence (b = 0.21; HC3 SE = 0.22; p = 0.35). In turn, rapport (b = 0.74; HC3 SE = 0.12; p < 0.001) and social presence (b = -0.21; HC3 SE = 0.11; p = 0.06) both significantly influence customer comfort.

Interestingly, while human-likeness is not a significant driver of a service robot’s social presence, previous robot interactions drive these perceptions (b = 1.00; HC3 SE = 0.20; p < 0.001). Therefore, customers who had previously interacted with service robots perceive both service robots as significantly more socially present.

A mediation analysis is conducted using Model 4 of the PROCESS macro (Hayes 2017). The 95% confidence interval around the indirect effect through rapport excludes 0 (see Table 2). Therefore, in support of hypothesis H2, a mediation through rapport can be observed.

In contrast, the 95% confidence interval around the indirect effect through social presence does not exclude 0 (see Table 2). Therefore, social presence does not mediate the effect of human-likeness on customer comfort. Hence, hypothesis H3 does not find support.

Table 2 PROCESS results for indirect effects of Models 2, 3 & 4
Full size table

4.3.4 Model 3: moderating effect of service failures

Model 3 extends Model 2 by introducing service failures as a moderator for the effect of human-likeness on rapport and social presence. For rapport, the overall effect of human-likeness remains highly significant (b = 1.04; HC3 SE = 0.35; p < 0.01) and also the interaction effect between human-likeness and the service failure dummy is significant (b =  − 0.94; HC3 SE = 0.46; p = 0.04). For social presence, neither the main effect of human-likeness (b = 0.55; HC3 SE = 0.35; p = 0.11) nor the interaction effect between human-likeness and the service failure dummy are significant (b = -0.70; HC3 SE = 0.46; p = 0.13). The effects of rapport (b = 0.74; HC3 SE = 0.12; p < 0.001) and social presence (b = −0.21; HC3 SE = 0.11; p = 0.06) on customer comfort remain statistically significant.

To test for a moderated mediation, Model 7 of the PROCESS macro is used (Hayes 2017). As the indices of moderated mediation in Table 3 show, service failures moderate the indirect effect that goes through rapport. More specifically, in the absence of service failures, human-likeness positively affects customer comfort through rapport. In the presence of service failures, this indirect effect disappears. These findings provide support for hypothesis H4a.

With regards to social presence, service failures do not moderate the hypothesized mediation through social presence (see Table 3). Therefore, social presence does not mediate the effect of human-likeness in the absence nor in the presence of service failures. Hence, hypothesis H4b does not find support.

Table 3 Indices of moderated mediation for Models 3 & 4
Full size table

4.3.5 Model 4: complete model

Model 4 investigates whether the effect of human-likeness through rapport/social presence and customer comfort is sufficiently strong to impact downstream customer service outcomes. In addition, Model 4 examines whether service failures moderate these indirect effects. For this analysis, Model 80 of the PROCESS macro is used with a customized w-matrix (Hayes 2017).

Model 4 extends Model 3 by showing that customer comfort significantly impacts customer satisfaction (b = 0.42; HC3 SE = 0.08; p < 0.001), engagement (b = 0.23; HC3 SE = 0.08; p = 0.01), word-of-mouth (b = 0.33; HC3 SE = 0.07; p < 0.001), and willingness-to-pay (b = 0.21; HC3 SE = 0.06; p < 0.001).

As the indices of moderated mediation in Table 3 show, service failures moderate the indirect effect that goes through rapport and customer comfort for all tested customer service outcomes. Therefore, only in the absence of service failures, human-likeness positively affects customer service outcomes through rapport and customer comfort.

With regards to social presence, service failures do not moderate the hypothesized indirect effect that goes through social presence and customer comfort (see Table 3). Therefore, social presence and customer comfort do not mediate the effect of human-likeness on customer service outcomes in the absence nor in the presence of service failures. An overview of the tested hypotheses is provided in Table 4.

Table 4 Hypothesis overview
Full size table

5 Discussion

Customer comfort is a major driver of customer satisfaction, commitment, loyalty, and word-of-mouth (Gaur et al. 2009; Lloyd and Luk 2011; Paswan and Ganesh 2005; Spake et al. 2003). Thus, it is of great importance to make customers feel comfortable during the service delivery. This raises the question of how customer comfort can be ensured during novel types of service encounters that involve interactions with service robots. In the present study, we investigate human-likeness as a key driver of customer comfort during robot-enabled service encounters. Our goal is to assess whether human-likeness is an enabler or an inhibitor of customer comfort and whether the answer to this question depends on the presence of service failures.

First, it is still a matter of debate whether human-likeness in service robots is generally desirable (Akdim et al. 2021; Belanche et al. 2021b; Wirtz et al. 2018; Yoganathan et al. 2021). Model 0 shows that increased levels of human-likeness are associated with higher customer engagement, word-of-mouth, and willingness-to-pay. Also Model 1 shows that human-like service robots make customers feel more comfortable. Therefore, the present study adds to the debate by demonstrating positive rather than negative (Akdim et al. 2021; Mende et al. 2019) effects of human-likeness. In addition, Model 1 shows that customer comfort mediates the positive effect of human-likeness on customer satisfaction, engagement, word-of-mouth, and willingness-to-pay. These results highlight customer comfort’s central role during robot-enabled service encounters.

To better understand the relationship between human-likeness and customer comfort, we investigate two potential mediators. In this way, we seek to better understand why some studies suggest a negative (Mende et al. 2019) and others a positive (Yoganathan et al. 2021) relationship between human-likeness and customer comfort. The first investigated mediator is customer-robot rapport. As Model 2 shows, rapport does indeed mediate the relationship between human-likeness and customer comfort. Therefore, customers appear to feel more comfortable when interacting with a human-like robot because they build more rapport with it. Importantly, in line with our reasoning that service failures weigh more heavily for human-like service robots (Wirtz et al. 2018), Model 3 shows that the positive effect of human-likeness disappears if a service failure occurs.

The second investigated mediator is social presence. As Model 2 shows, social presence is not significantly impacted by a service robot’s human-likeness. Also when accounting for service failures in Model 3, human-likeness does not appear to drive social presence perceptions. This is surprising as previous studies found a positive relationship between the two concepts (Odekerken-Schröder et al. 2022). One possible explanation for the lack of significant results could be that the human-likeness manipulation was not salient enough. Yoganathan et al. (2021) point out that human-likeness is a relative concept that depends on customers’ reference points. Therefore, participants might have still perceived the human-like service robot as relatively machine-like. If they did, social presence evaluations might have been equally low. As the manipulation checks show, the human-like service robot was rated significantly higher on human-likeness, but still only received a rating of 3.5 out of 7. Therefore, more extreme or salient differences in human-likeness might lead to detecting a relationship between human-likeness and social presence as we would expect based on previous studies (e.g., Odekerken-Schröder et al. 2022). Nevertheless, the human-likeness manipulation was sufficient to provoke differences in multiple customer service outcomes (Model 0), customer comfort (Model 1), and rapport (Models 2, 3, & 4). Hence, human-likeness does simply not appear to be a main driver of social presence.

Even more interestingly, the present study only manipulated the service robot’s human-likeness and the exhibited service failure (i.e., failure vs no failure). However, considerable variance in social presence perceptions appears to be explained by participants’ number of previous robot interactions. One reason for this relationship might be that respondents who previously interacted with robots were able to better imagine how it feels to be in the presence of a service robot. Another explanation could be that social presence perceptions become stronger through repeated interactions with service robots. As these are merely speculations, this relationship should be further investigated by future research.

Models 2 and 3 both show that increased levels of social presence make customers feel less comfortable. The observed negative relationship is in line with our theorizing that customers find it unsettling when a service robot appears to have its own thoughts and intentions (Biocca, 1997). As social presence had previously been perceived as a desirable feature in service robots (van Doorn et al. 2017), the present study provides a new perspective on the phenomenon.

The purpose of Model 4 is to investigate whether the impact of human-likeness through rapport/social presence is sufficiently strong to impact downstream customer service outcomes. Model 4 shows that in the absence of service failures, human-likeness does indeed positively impact customer service outcomes by facilitating customer-robot rapport and customer comfort. If a service failure occurs, this positive indirect effect disappears. For social presence, we do not observe a significant indirect effect of human-likeness on customer service outcomes, independent of whether a service failure occurs or not.

5.1 Theoretical contributions

The present study makes several theoretical contributions. First, it contributes to the service robot literature by demonstrating that human-likeness is an enabler rather than an inhibitor of customer comfort. It was previously unclear whether human-likeness in service robots makes customers feel more or less comfortable. On the one hand, Mende et al. (2019) found that human-like service robots can provoke feelings of eeriness, indicating that customers would feel uncomfortable interacting with human-like service robots. Also Akdim et al. (2021) found that customer attitudes towards service robots decrease as the service robots become more human-like. On the other hand, Belanche et al. (2021b) and Yoganathan et al. (2021) both found human-likeness to drive more positive customer service outcomes which are highly correlated with customer comfort (LLoyd and Luk 2011). Investigating this tension in the literature, our results consistently show that customers are significantly more comfortable interacting with the human-like than with the machine-like service robot.

Second, the study further contributes to the service robot literature by showing that the positive impact of human-likeness on customer comfort can be explained by increased levels of customer-robot rapport. Customer-employee rapport takes on a central role during human-to-human service encounters (Gremler and Gwinner 2000). Also for service robots, studies have indicated that people do build rapport with robots (Lubold et al. 2019; Nomura and Kanda 2016; Seo et al. 2018). However, while Qiu et al. (2020) investigate downstream consequences of customer-robot rapport, they did not yet identify any such consequences. Therefore, the present study contributes to the service robot literature by empirically demonstrating that increased levels of customer-robot rapport result in increased levels of customer comfort, satisfaction, engagement, word-of-mouth, and willingness-to-pay. In addition, providing support for the findings by Qiu et al. (2020), the present study contributes to the service robot literature by showing that human-likeness facilitates human–robot rapport building.

Third, to explain the previously observed eerie and threatening perceptions of human-likeness (Mende et al. 2019), we investigate social presence as a second potential mediator. While our analyses do not support this explanation, we nevertheless make important contributions to the service robot literature. In particular, we contribute to the service robot literature by providing a starting point for future research to investigate why some scholars do (Kontogiorgos et al. 2020; Odekerken-Schröder et al. 2022; Yoganathan et al. 2021) and we do not find a relationship between the two concepts. Moreover, social presence had previously been assumed to be desirable in service robots (van Doorn et al. 2017). Here, we contribute to the service robot literature by providing a theoretical argumentation and empirical support that questions this previously held belief.

The fourth major contribution lies in the introduction of customer comfort as a central element during robot-enabled service encounters. Previous studies found that human-likeness positively impacts customer service outcomes (e.g., Belanche et al. 2021b; Yoganathan et al., 2021). Yoganathan et al. (2021) explained the observed positive relationship with higher service quality expectations and Belanche et al. (2021b) with higher functional, emotional, and monetary value. Our study contributes to the service robot literature by offering another explanation. In particular, our findings suggest that increased levels of customer comfort largely explain the positive impact of human-likeness on customer satisfaction, engagement, word-of-mouth, and willingness-to-pay.

By demonstrating such a central role of customer comfort, the present study contributes toward better understanding robot-enabled service encounters in restaurant settings. For example, increasingly many studies focusing on service robots in restaurant settings seek to understand drivers of service encounter evaluations (Lu et al. 2021), the general service experience (Qiu et al. 2020), customer loyalty (Belanche et al. 2021b), or repatronage intention (Odekerken-Schröder et al. 2022). For these outcomes, customer comfort might be an important puzzle piece to understand better what drives them and how they can be optimized.

Fifth, the present study contributes to the service robot literature by demonstrating that service failures are an important boundary condition for the effect of human-likeness. Wirtz et al. (2018) advocate that human-likeness in service robots should be limited as it might provoke exaggerated expectations, which could amplify the negative effect of service failures. As a result, Wirtz et al. (2018) perceive human-likeness as a liability that should be limited. Similarly, Belanche et al. (2020b) suggest that service failures by human-like and machine-like service robots might be perceived differently. With an empirical investigation, Choi et al. (2021) show that service failures do indeed weigh more heavily for human-like than for machine-like service robots. Our results confirm these findings. However, going beyond the findings of Choi et al. (2021), we show that if a service failure occurs, human-likeness merely loses its positive effects and does not turn into a liability. Therefore, in case of a service failure, customers do not appear to be additionally dissatisfied just because the service robot looked human-like. With these findings, the present study contributes to the service robot literature by showing that human-like service robots remain preferred even in contexts where service failures are common. This offers a novel perspective on the suggestion by Wirtz et al. (2018) to avoid human-likeness in service robots.

These findings also contribute to the increasing body of literature that investigates service failures by service robots in restaurant contexts. For example, taking an attribution perspective, Belanche et al. (2020b) show that restaurant guests attribute service failures by human staff to the employee, while they attribute a service failure by a service robot to the restaurant. Choi et al. (2021) demonstrate that human-like service robots can recover service failures if they apologize. Lastly, Odekerken-Schröder et al. (2022) find that friendly interactions by a restaurant’s staff can compensate for low functional value of a service robot. The present study contributes to this emergent literature by highlighting service failures as an important boundary condition for the positive effect of human-likeness.

Surprisingly, we do not observe a moderating role of service failures for social presence perceptions. In particular, service failures do not have a larger effect for human-like than for machine-like service robots. This is an important insight as it had been theorized that customers would be disillusioned by service failures, diminishing the human-like service robot’s social presence (Biocca et al. 2003; Heerink et al. 2010). The present study contributes to the service robot literature by indicating that this is not the case.

5.2 Practical contributions

With these findings, the present study greatly contributes to practice. First, we show that human-like (vs machine-like) service robots are the superior choice for service providers and service robot producers. In particular, by choosing human-like service robots, it is not only possible to make customers feel more comfortable throughout their robot-enabled service encounters, but to also increase customer satisfaction, word-of-mouth intention, engagement, and willingness-to-pay. In this way, the present study provides additional support for Belanche et al. (2021b) and Yoganathan et al. (2021) who made similar suggestions.

In addition, as customer comfort appears to take on a central role during customer-robot interactions, practitioners are advised to include comfort-enabling elements in their service robots. Besides human-likeness, one comfort-enabling element appears to be customer-robot rapport. Therefore, practitioners are advised to include rapport facilitating characteristics and features in their service robots, such as emotion mimicking (Lakin and Chartrand 2003), body language (Le May 2004), nonverbal communication (Le May 2004), or eye contact (Kim and Baker 2017). Here, future research is needed to validate the effectiveness of these mechanisms.

Based on our results, social presence appears to be an inhibitor of customer comfort. Therefore, practitioners are advised to avoid social presence in service robots. This makes it important to determine either service robot features that provoke social presence or identify customer characteristics that amplify social presence perceptions. Importantly, human-likeness does not appear to be such a driver. Hence, practitioners do not need to forgo the benefits of human-likeness in the attempt of reducing a service robot’s social presence.

Lastly, Wirtz et al. (2018) had advocated against human-likeness in service robots as it could turn into a liability in case a service failure occurs. Our results show that if a service failure occurs, the service robot’s human-likeness loses its positive effect but does not additionally harm customer comfort, satisfaction, engagement, word-of-mouth, or willingness-to-pay. Hence, even in unstructured environments where unforeseeable situations can occur that provoke a service failure, human-like service robots remain the superior choice. Nevertheless, for highly error-prone environments, human-likeness can receive lower priority as it is no longer a benefit if the service robot frequently makes mistakes.

5.3 Limitations and future research

Like most studies, also the present study faces some limitations. In the following, we highlight how these can be explored by future research. We also outline more general avenues for future research surrounding the two core constructs of our study (i.e., customer comfort and service failures). Potential research questions are summarized in Table 5.

Table 5 Research agenda for core concepts
Full size table

5.3.1 Methodology and data collection

Future research should seek to verify our findings using lab experiments with physical service robots or through field studies. In addition, future research could investigate how robot-related perceptions or customer-robot relationships develop over repeated interactions (Gutek et al. 1999). Lastly, the present study’s sample size can be considered rather small. As a result, our analyses are most reliable in detecting effect sizes of at least a moderate magnitude (Green 1991). Therefore, we encourage future research to collect larger but also more diverse (e.g., older) data samples to further generalize the present study’s findings.

5.3.2 Measurements and manipulations

Future studies could include more than two human-likeness stimuli to allow for the detection of a possibly U-shaped relationship between human-likeness and customer comfort (Akdim et al. 2021; Mori 1970). In addition, as human-likeness is a relative concept (Yoganathan et al. 2021), future studies could include reference points. Similarly, investigating different failure types could help to determine the most adequate tasks for service robots. Lastly, future research could investigate spill-over effects from service failures happening to other customers.

5.3.3 Drivers and mediators

Future studies could examine ways to increase customer comfort, such as the mimicking of emotions (Lakin and Chartrand 2003), body language (Le May 2004), and eye contact (Kim and Baker 2017). In addition, as rapport comprises the two dimensions of personal connection and enjoyable interaction (Gremler and Gwinner 2000), it would be of value to understand which of these dimensions is more important in a service robot context. Future research could also investigate why service failures moderate the effect of human-likeness. Here, our argumentation could be a good starting point (i.e., disillusionment).

5.3.4 Outcomes and consequences

Besides investigating additional customer service outcomes, it might be even more interesting for future studies to investigate customer comfort’s impact on repeat use of the technology or customers’ willingness to teach or help the service robot.

5.3.5 Context factors and moderators

Future research could investigate a potential moderating role of past experience (e.g., knowing about the flaws of a service robot from previous encounters). Other moderating factors could include customer characteristics (Epley et al. 2007) or the presence of human staff (Yoganathan et al. 2021). A possible limitation of the present study is that it did not consider customer perceptions about the service provider. For example, customers’ attributions of cost cutting and service enhancement could influence how customers perceive the service encounter and in this way determine how different robot-related aspects influence customer comfort or customer service outcomes (Belanche et al. 2021a). Lastly, future research could investigate different service failure recovery strategies for restoring customer comfort (Choi et al. 2021).