The service triad: an empirical study of service robots, customers and frontline employees

Service robots in hospitality services

While still being a nascent field, various scholars have recently studied the role of (service) robots in hospitality and tourism services. Table 1 provides an illustrative overview of empirical studies using primary data sources. Almost all studies rely on laboratory experiments, while field data are rare, with some notable exceptions (e.g. Tuomi et al., 2020). Anthropomorphism is an often included construct, while only very few studies consider the service robot's social presence. Finally, the studies by Qiu et al. (2020) and Tuomi et al. (2020) take a service triad perspective by also including FLEs in their study. Extending prior research, our current field study includes both, anthropomorphism and social presence and investigates the interaction within the service triad of service robot – customer – FLE to further develop our understanding of service robots in FLE encounters. Table 1 also presents a few illustrative empirical studies in other industries that address anthropomorphism, social presence and/or the service triad. The studies by Barrett et al. (2012) and Mende et al. (2019) acknowledge the service triad and study the effects of service robots on human employees in healthcare and other settings, whereas Heerink et al. (2008) focus on robot acceptance in healthcare. None of these studies include the related but distinct concepts of anthropomorphism and social presence, which can be seen as first and second degree social responses (Lee et al., 2006). In order to enhance our understanding of the interplay between these concepts on utilitarian and hedonic value, ultimately resulting in customer repatronage, this paper introduces an exploratory observation study, a field study and a scenario-based experimental design.

To introduce a conceptual model contributing to the nascent field depicted in Table 1, we summarize the ongoing debate on the core concepts of the conceptual model below.

Anthropomorphism

The first concept is anthropomorphism. Anthropomorphism describes a main feature of humanoid robots and has its roots in the Greek words “anthropos” (human) and “morphe” (shape or form). It originally refers to the phenomenon by which nonhuman entities are given human shape or form (Wan and Aggarwal, 2015). Social psychology expands the view on anthropomorphism to the “tendency to imbue the real or imagined behavior of non-human agents with human-like characteristics, motivations, intentions, or emotions” (Epley et al., 2007, p. 864), offering a foundation for research on service robots (Xiao and Kumar, 2021).

While marketing has found anthropomorphism to increase product and brand liking (Aggarwal and McGill, 2012), it is unclear whether anthropomorphism in a frontline service triad including service robots enhances customers' repatronage. Contemporary service research acknowledges the importance of the human tendency to anthropomorphize robots (Mende et al., 2019; Van Doorn et al., 2017), but the question remains whether customers' anthropomorphism of robots facilitates or constrains use intention (Blut et al., 2021).

One stream of research argues that anthropomorphizing a nonintelligence product (e.g. service robot) is a useful strategy to increase consumer preferences because the human intentions and emotions are associated with intelligence and competence in task performance (Wan and Aggarwal, 2015). Taking this perspective would favor the use of anthropomorphized robots in the service triad of technology-customer-FLE (Duffy, 2003; Reed et al., 2012). A recent meta-analysis conducted by Blut et al. (2021) demonstrates that anthropomorphism is in the eye of the beholder rather than referring to the extent to which firms design robots as humanlike.

A second stream of research emphasizes the paradoxical effect that increased anthropomorphism can result in consumers experiencing discomfort such as feelings of eeriness or a threat to their human identity and feelings of human inadequacies (Lu et al., 2019; Mende et al., 2019; Reed et al., 2012). This view is in line with the uncanny valley theory postulating that the customer's affinity for a robot does not continuously increase with its human likeness as customers may find a highly humanlike robot creepy and uncanny (Mori, 1970; Mori et al., 2012). Strong anthropomorphic qualities may also lead to overly optimistic expectations about a robot's abilities, which can be disappointing (Wirtz et al., 2018). Fostering scholarly understanding on the service triad technology-customer- FLE and the role of anthropomorphism is an important research direction (Van Doorn et al., 2017).

Social presence

A related, but distinct concept is social presence. In virtual reality studies, Heeter (1992) indicates that presence consists of the three dimensions personal presence (extent to which you feel you are in a virtual world), environmental presence (the extent to which the environment seems to know you are there) and social presence (the extent to which someone or something, like computer generated beings, believes you are there). Social presence has the most implications for human-robot-interactions (HRI) because it is the ultimate aim of designing socially, interactive robots (Lee et al., 2006).

Origins of social presence of robots can be found in symbolic interactionism and social psychological theories of interpersonal communication (Biocca et al., 2003). The emphasis of social presence is on the agent's capacity for social interaction and verbal or nonverbal cues in communication. Therefore, physically present (e.g. sculptures) would not suffice to be perceived as socially present (e.g. beings) as social presence is mainly based on the sense that one has “access to another intelligence” (Biocca et al., 2003).

Media equation theory argues that customers equate social robots with real social actors as they rely on their natural tendency of accepting things at their face validity and react to robots as if they were human (Lee et al., 2006). The computers are social actors (CASA) research paradigm is derived from media equation theory and is frequently used to understand HRI. CASA is based on the idea that when confronted with an anthropomorphic robot, (a) humans respond to the robot socially, (b) humans are persuaded by the imitation of human characteristics of the robot and (c) humans do not process the fact that the robot is not a human (Lee et al., 2006).

Although in service research Van Doorn et al. (2017, p. 43) refer to automated social presence (ASP) as “the extent to which technology makes customers feel the presence of another social entity”, the original construct of social presence can be either a human or artificial intelligence evoking reactions to social cues (Biocca et al., 2003). Therefore, in the current study, we focus on social presence. For engineers and designers of social robots, increasing the experience of social presence is typically a design goal (Biocca et al., 2003). Recently, Gambino et al. (2020) summarize that engineers and designers aim for natural forms of social interaction between service robots and users to minimize the cognitive effort it takes human actors to use service robots.

Utilitarian and hedonic value

Anthropomorphism and social presence are expected to result in utilitarian and/or hedonic value. Motivation theory suggests that customers behave to satisfy their needs. Rooted in motivation theory, the more recent self-determination theory (SDT) provides a substantive basis for human behavior, distinguishing between extrinsic (utilitarian/instrumental) and intrinsic (hedonic) motivations (Deci, 1975; Deci and Ryan, 1985; Ryan and Deci, 2001). In marketing, Hirschman and Holbrook (1982) introduced a more experiential view of consumption, including hedonic reasons to the more traditional utilitarian reasons for a purchase. Likewise, contemporary studies investigate the effect of utilitarian and hedonic value on repeat patronage (Hepola et al., 2020) or as dimensions of experiential value of robots in the service encounter (Wu et al., 2021).

This study focuses on the value of service robots in hospitality which is inherent to the service perspective implying that “value is created collaboratively in interactive configurations of mutual exchange” (Vargo and Lusch, 2008, p. 145). The concept of value has its origins in other disciplines. Sociology, psychology and economics, for example, have a long tradition of investigating instrumental and hedonic dimensions of attitude (Voss et al., 2003).

In restaurants, it is also commonly known that the value customers perceive is not merely based on utility (utilitarian) but to a large extent also on gratification (hedonic) (Noone et al., 2009). The distinction between utilitarian and hedonic value also found its way into recent research on service robots. In their study on service robots' value co-creation and value co-destruction potential, Čaić et al. (2018) argue that service robots offer new value propositions, where value is created when engaging in the service leaves actors better off relative to their initial conditions. Demonstrated in the context of elderly care, socially assistive robots positively impact both utilitarian (e.g. effectiveness) as well as hedonic value (e.g. fun) (Čaić et al., 2019).

FLE interaction quality

Taking a service triad perspective consisting of service robots, customers and FLEs, implies that the interaction quality of FLEs plays a role during the service encounter. The nature of interactions is widely seen as the nucleus for value creation during the service encounter exercising a strong impact upon customer responses. Early, service researchers positioned service encounters as role performances in which the so-called service script would contain information about the role set related to one's own expected behavior as well as to the expected complementary behavior of others reflecting the prototypical service experience (Hui and Bateson, 1991; Solomon et al., 1985).

While service research evolved, scholars in marketing and organizational behavior were giving increasing attention to the personal interaction between the customer and the FLE of service businesses. The service encounter became a focal point in consumer evaluations of the entire service organization and implied a great opportunity for a service firm to customize the delivery of its service to help the individual consumer. This customization opportunity is a potential source of competitive advantage for the service firm, which can lead to favorable service quality evaluations by consumers (Bettencourt and Gwinner, 1996; Bitner et al., 1990; Bock et al., 2016).

FLE performance quality is concerned with how the service is delivered, especially emphasizing the demand for emotional labor. For example, a service employee is expected to express positive emotions when interacting with a customer and act in a way to build trust, demonstrate promptness and reliability, and give a sense of personal attention (Singh, 2000). Therefore, we define quality of FLE interaction as consumers' perception of the interpersonal interactions with human employees that take place during service delivery (cfr. Brady and Cronin, 2001) and the FLEs willingness to help (cfr. Singh, 2000). In turn, a high-quality performance is thought to enhance customer intentions (Singh, 2000).

In the 2017 special issue on organizational frontlines, service scholars acknowledge the emerging role of technology resulting in a triadic (technology-customer-FLE) rather than a dyadic service encounter. Research recognizes that the human connection between staff and consumers can be challenging in technology-infused service interactions, and there will be a greater desire for employees who can connect with customers (Rafaeli et al., 2017).

Along the continuum ranging from technologies that replace FLEs to those that augment FLEs to provide service, smart technologies (e.g. service robots) provide value and an important question is how such technologies can be leveraged and integrated in the triadic service encounter technology-customer-FLE to create value. Marinova et al. (2017) define frontline interactions to also include interactions between a customer and an artificial intelligence-powered machine, which connects the customer with the organization by replacing or augmenting FLEs to coproduce value. In a similar line, Singh et al. (2017) describe organizational frontline as interactions and interfaces at the point of contact between an organization and its consumers that promote, facilitate or enable value creation and exchange. They explicitly argue that interfaces refer to the characteristics of modes, agents (or robots), artifacts and servicescapes that serve as the medium for the contact between the customer and the organization, acknowledging the role of service robots.

Most recently, Yoganathan et al. (2021) argue that robots and human staff can deliver services in collaboration. The interaction quality between robots and FLE in a service setting is expected to influence consumer service outcomes differently. Knowledge on the conditions under which service robots-FLE collaboration generate positive or negative outcomes is still scarce (Larivière et al., 2017). For that reason, the current study investigates the role of FLE interaction quality in the triadic service encounter including service robots, customers and FLE.

Customer repatronage

The knowledge gap on how repatronage intentions in the service triad evolve is central to our study. In the contemporary service industry, facing numerous alternative offerings, service providers first encourage consumers to make an initial purchase, and in a second stage, they encourage existing customers to revisit or repurchase, based on their previous experiences (Ho and Chung, 2020). In highly competitive hospitality services such as restaurants, repatronage is an important loyalty indicator (Wirtz and Mattila, 2004). Customer repatronage reflects the likelihood that a customer will visit the restaurant again (Atulkar and Kesari, 2017). The service robot literature recently studied the effect of human likeness of the robot service (Lu et al., 2021) and customer satisfaction with service robots (Jia et al., 2021) on repatronage intentions (e.g. revisit intentions and purchase intentions) resulting in mixed findings.

Moderating hypotheses

Contemporary service research views augmentation from the perspective that technology enhances human actors (De Keyser et al., 2019; Larivière et al., 2017; Marinova et al., 2017). Taking a service triad perspective, we reason that actors can augment each other's tasks in the service encounter, implying that a service robot can augment FLEs or that FLEs can augment service robots. Augmentation involves assisting and complementing other actors in the service triad to perform their tasks better and achieve their goals in the service encounter.

Service triads, consisting of service robots, customers and FLEs by definition imply that the physical encounter between customer and FLE is augmented by technology (De Keyser et al., 2019; Hilken et al., 2017). In order to guide service managers in setting-up these service triads, an increased understanding is needed as to how human and nonhuman actors work in tandem.

Previous studies show that customer needs for a human touch can be especially relevant when handling failures (De Keyser et al., 2015). In case the service robot's utilitarian or hedonic value is low, which can be thought of as some kind of failure, we expect the FLE to compensate for this failure and augment the service robot's value resulting in repatronage intentions.

More specifically, for our context of hospitality service, triads with a strong emphasis on the provision of the core service such as serving food and drinks (utilitarian elements) we expect that high-quality interactions with FLEs can augment lower levels of service robot utilitarian value. In other words, customers in the service triad who lack the robot's accuracy and consistency feel supported by employees' efforts in the interaction (Stein and Rameseshan, 2019) and decide to revisit the venue. In a similar vein, for our people-oriented service encounter (Li et al., 2021) in a restaurant that people typically visit for enjoyment (hedonic elements), we expect that high-quality interactions with FLEs can also augment lower levels of service robot hedonic value (Qiu et al., 2020).

Summarizing, a positive customer perception of interpersonal interactions with FLEs (Brady and Cronin, 2001) and their demonstrated willingness to help (Singh and Sirdeshmukh, 2000) can increase customer repatronage intentions in cases when the service robot functional performance and entertainment are relatively low. Therefore, we hypothesize the following:

The conceptual model is visualized in Figure 1.

Empirical context

Our empirical context reflects a triadic encounter consisting of service robots – FLEs – customers. It entails a fast casual dining restaurant in Europe that offers Asian-style dining. The restaurant promises worldly food for small town prices and strives for revenue management, described by Noone et al. (2009) as reducing service encounter duration to welcome more customers and generate more revenues during high demand periods. The restaurant can typically be recognized by a long waiting line outside that customers gladly accept in return for an affordable and fast casual dining experience.

In the COVID-19 pandemic, this restaurant implemented two frontline service robots, resulting in a service triad of service robots, FLEs and customers. First, these service robots minimized human-to-human contact and, thereby, the risk of spreading the virus (Davis, 2020). Second, substituting human FLEs with service robots increased the maximum number of customers that could be seated as the particular government only allowed a maximum number of people in a restaurant at the same time, including staff. Third, this limited amount of customers allowed, (normal maximum capacity of the restaurant is approximately 300 customers) created a smaller setting which was an excellent environment to experiment with the service robots. Both service robots – Amy and Akatar, displayed in Plate 1 – can be considered humanoid, which refers to a robot with humanlike features (Mende et al., 2019). Namely, they both have a face and a name (van Pinxteren et al., 2019). Moreover, they can communicate unilaterally with the customers with a humanlike voice (they can speak to the customers, but they do not respond) (Złotowski et al., 2015). Each service robot has its own shape that supports a distinctive set of tasks: Amy serves drinks and picks up the empty glasses, and Akatar delivers dishes from the kitchen to the customer's table.

Exploratory field observations

To gain a better understanding of this triadic service encounter, data collection started with a field observation during the first three days of the implementation of the service robots (June 3 until June 5, 2020). Field observations typically clarify and focus initial ideas and give concrete insights into the context and the people involved (Goodman et al., 2012). A semi-structured observation protocol was followed that allowed for deviation and comments, allowing a rich description of the hospitality context at hand (Denzin, 2001). In total, data were collected during 9 h of field observation, spread across three researchers. Field observations in the restaurant were covert, with permission of the restaurant owner, to ensure that interactions with the service robot were not influenced by the observer, avoiding the Hawthorn effect (Jones, 1992). The field observation enabled the research team to get a rich understanding of the service triad and resulted in two main insights. First, the field notes uncovered dyadic and triadic interactions in the triad “service robot-customer-FLE”. Second, the field notes revealed two potentially different benefits of the service robot: (1) utilitarian value: service robot serves food and drinks to the customers and by doing so also offers functional support to the FLE and (2) hedonic value: service robot offers entertainment and enjoyment to customers, which can for example be observed by customers taking selfies with the service robot. These insights were used as an input for the survey development of our field study and subsequent scenario-based experimental design.

Sample and measures

Based on extensive discussions with the restaurant owner and store manager, it became clear that the typical segment of the restaurant consists of relatively young customers such as students, young couples and families with young kids. Therefore, we decided for a QR-code that quickly and efficiently converse the survey URL to customers. The main reasons underlying this decision are: (1) the free Internet access in the restaurant, (2) the high likelihood of customers bringing their smart phone, (3) aim for minimal human-human interaction in the COVID-19 pandemic and (4) environmental friendliness.

Before the first day of our data collection, we prepared a podcast with instructions for the team of human FLEs. The store manager shared this podcast with his team via the team's Whats App group to emphasize the importance of timing of showing the flyer with QR-code (i.e. after customers completed their main course to make sure they experienced FLEs and service robot interactions). In addition to the podcast, we also provided instruction flyers for the team including the steps they had to recall in the data collection stage. These flyers were located at various backstage locations in the restaurant, reminding the human staff of the research taking place.

The FLEs showed a plasticized flyer (Appendix 1) to customers after they finished their main course. This ensured that customers did experience the triadic service encounter.

As an incentive, the customers were offered a free homemade iced tea in return for completing the online survey on their mobile device. Data collection took place over the course of one month, from September 14 to October 14, 2020. In total, 124 customers who interacted with the service robot completed the survey, resulting in a final dataset of 108 responses after elimination of incomplete answers. Of the respondents, 70.8% were female, and in terms of age, 81.5% fell within the range of 18 and 34 years. In addition, 69.4% of the sample consisted of repeat customers (i.e. had visited this fast casual dining restaurant before). The respondents mainly visited the restaurant with friends (62%), their partner (21.3%) or family (13.9%).

All items in the survey were adapted from existing measurement scales, which were partially reduced to fit our context of fast casual dining. The items were assessed on a seven-point Likert scale (1 = “strongly disagree”, 7 = “strongly agree”). Our dependent variable, customer repatronage intention, was captured by the respondent's intention to revisit the restaurant within the next six months and was measured with a two-item scale adapted from Maxham and Netemeyer (2002). The moderating variable, FLE interaction quality, was adapted from three items of Brady and Cronin's (2001) interaction quality construct. The service robot's utilitarian and hedonic value were both assessed based on four items adapted from the recently developed service robot adoption willingness scale (Lu et al., 2019). Specifically, the utilitarian value construct was composed of items focusing on the service robot's accuracy and consistency in performance, whereas hedonic value was assessed through customer's fun and entertainment experienced while served by the robot (Lu et al., 2019). The service robot's social presence comprised of five items adapted from Lee et al. (2006). Lastly, anthropomorphism was captured by five items developed by Lu et al. (2019). To answer the questions related to the service robot, we asked the respondents to answer these questions while keeping in mind the robot they interacted with the most. We included this baseline service robot as a control variable in our PLS model. A complete list of the items and their factor loadings can be found in Table 2, whereas their scale reliabilities are displayed in Table 3.

Data analysis

We turn to partial least squares structural equation modeling (PLS-SEM) to test our hypotheses. PLS-SEM is an estimation technique based on OLS regressions. It focuses on the prediction of a specific set of hypothesized relationships that maximizes the explained variance in the dependent variables, similar to OLS regressions (Hair et al., 2016). This makes PLS-SEM particularly useful for success driver studies (Hair et al., 2011). The decision to apply this method of analysis was driven by two main reasons. First, PLS-SEM can handle small sample sizes of less than 200 respondents (Bacile, 2020; Chin, 1998; Hair et al., 2012). Hair et al. (2016) provide minimal sample size requirements to detect various R² values at a 5% significance level while taking the complexity of the PLS path model into account. The maximum number of arrows pointing at a construct in this study is three, so we need at least 37 respondents to pinpoint R² values of at least 0.25 at a 5% significance level. Thus, we can conclude that our sample size of 108 is sufficiently large. Second, the method is nonparametric in nature and can therefore deal with nonnormal data (Chin, 1998; Hair et al., 2016). Hair et al. (2012) recommend performing Shapiro–Wilk or Kolmogorov–Smirnov tests to evaluate whether data are normally distributed. Both tests in SPSS indicate that our anthropomorphism, service robot's hedonic value, FLE interaction quality and customer repatronage variables are nonnormally distributed. Additional checks for skewness and kurtosis (Hair et al., 2016) confirm that our data are nonnormally distributed. For these reasons, we use PLS-SEM. More specifically, SmartPLS 3.3.2 software (Ringle et al., 2015) was applied to conduct the analyses. We used the standard, recommended algorithm and settings, and administered case-wise deletion for missing variables.

Since the data from both our dependent and independent variables come from the same source, common method bias could be a potential threat (Podsakoff et al., 2003). To evaluate the extent to which our data suffers from common method bias, we employ the procedure suggested specifically for PLS-SEM research by Kock (2015). As our estimations indicate that our highest VIF is 1.73, we can confirm our VIF values do not exceed the 3.3 threshold, suggesting that common method bias is not a concern for this study.

In the next section, we first evaluate our measurement model, which attaches manifest variables to their latent variables. After that, we test the relationships between the latent variables by assessing the structural model (Fornell and Larker, 1981; Hulland, 1999).

Measurement model – validity and reliability

To ensure construct reliability, we check the item loadings, composite reliability and Cronbach's alpha values. First, for individual item reliability, we investigate the loadings. A generally accepted heuristic is that item loadings should be 0.7 or higher (Hair et al., 2016). All our items exceed this threshold. For construct reliability, Hair et al. (2016) detail that the composite reliability and Cronbach's alpha values should exceed 0.7. As Table 3 shows, construct reliability was established with strong composite reliability values ranging from 0.89 to 0.95 and Cronbach's alpha ranging from 0.83 to 0.91.

The AVE values for all constructs highly exceed 0.50 (see Table 3), indicating sufficient levels of convergent validity (Bagozzi and Yi, 1988; Hair et al., 2016). To ensure discriminant validity, we follow both the Fornell-Larcker criterion and the Heterotrait-Monotrait (HTMT) ratio criterion. For the Fornell-Larcker criterion, each construct must share more variance with its own measures than with any of the other constructs. This is reflected by a higher square root of the AVE for each construct compared with its correlations with other constructs (Fornell and Larker, 1981; Hair et al., 2016). In addition, the square root of the AVE should not be lower than 0.7 (Chin, 1998). As Table 3 shows, all constructs meet these criteria. Following the HTMT ratio criterion, the HTMT values for all pairs of constructs should be below 0.85 (Voorhees et al., 2016). The HTMT values for our constructs range from 0.07 to 0.81 and are below the accepted threshold. Lastly, we can confirm that multicollinearity was not a threat to the measures as none of the variance inflation factor values exceeded the threshold level of 5 (Hair et al., 2016).

To evaluate the predictive relevance of the model, we examine the effect size and explained variance of the endogenous constructs. Table 4 indicates the R² values of the endogenous constructs range from 0.31 to 0.57, all exceeding the commonly accepted thresholds set by Falk and Miller (1992), Chin (1998) and Hair et al. (2011). In addition to the R², it is increasingly encouraged to report the f² effect sizes (Hair et al., 2016). The f² effect sizes for the supported hypotheses range from 0.04 to 0.73 and, thereby, vary from small to large effects (Hair et al., 2016). As such, the model's predictive relevance is supported.

Structural model – hypotheses testing

To evaluate the structural model and test the significance of the path coefficients, we ran a bootstrapping procedure with 5,000 samples (Hair et al., 2011). The effect of the control variable related to the baseline service robot was insignificant on all endogenous variables (p > 0.1). The estimation results supported a positive effect (β = 0.65; p < 0.001; f² = 0.73) of anthropomorphism on social presence, in support of H1. To test H2 and H3, we employed Rodríguez-Entrena et al.'s (2018) approach to test the statistical differences between path coefficients. Using 95% basic bootstrap confidence intervals, we find support for H2, in that anthropomorphism has a statistically significant (CI [0.1190, 0.4748]) stronger effect on utilitarian value (β = 0.49; p < 0.001; f² = 0.32) than on hedonic value (β = 0.21; p < 0.05; f² = 0.04). Despite the larger positive effect size of social presence on the service robot's hedonic value (β = 0.45; p < 0.001; f² = 0.18) than its utilitarian value (β = 0.33; p < 0.001; f² = 0.15), this difference is not statistically significant as its confidence interval includes zero (CI [–0.3165, 0.0637]). Therefore, we cannot find support for H3. We did find support for H4, with a positive effect of a robot's utilitarian value on customer's repatronage intention (β = 0.23; p < 0.05; f² = 0.05). Surprisingly, the path between the robot's hedonic value and customer's repatronage intention was not significant and failed to provide support for H5 (β = 0.02; p > 0.05; f² = 0.00). Further, we find a negative moderation effect of human employees' interaction quality on the relationship between the service robot's utilitarian value and customer's repatronage intention (β = −0.26; p < 0.05; f² = 0.04), supporting H6. Finally, we report an insignificant moderation effect of human employees' interaction quality on the relationship between the service robot's hedonic value and customer's repatronage intention (β = 0.22; p < 0.05; f² = 0.05), thereby rejecting H7. Figure 2 and Table 4 summarize the results of the hypothesis testing.

To further expand on the moderation effects found in H6, Figure 3 illustrates the relationships between the constructs. It displays the relationship between the service robot's utilitarian value, customer repatronage intentions and FLE interaction quality. The figure shows that in situations where the service robot's utilitarian value is at the mean and – especially – at lower levels, the FLE interaction quality does have a pronounced effect on customer repatronage intentions. In other words, FLE interaction quality can compensate for suboptimal levels of service robot utilitarian value, and FLEs can augment the service robots. However, in situations where service robot utilitarian value is high, there is not a pronounced relationship between the FLE interaction quality and customer's repatronage intentions.

Scenario-based experimental design

To test the robustness of the findings related to hypotheses 4–7 from our field study, we conducted a scenario-based online experimental design. This setup allowed us to ensure more variation in FLE interaction quality and recruit a sufficiently large sample size during the 2020/2021 COVID-19 lockdowns.

Design, procedure and stimuli

We adopted a 2 (service robot utilitarian value: high, low) × 2 (service robot hedonic value: high, low) × 3 (FLE interaction quality: high, low, no interaction) between-subject design. For the high (low) service robot utilitarian value condition, the service robot took orders and served food and drinks in a highly (in)consistent and very (in)accurate manner. With respect to the service robots high (low) hedonic value, the service robot brought (did not bring) fun while serving drinks and foods by, for example, making jokes and was (not) entertaining by, for example, posing for pictures, making the interaction with the robot very (un)enjoyable. For high (low) FLE interaction quality, human employees were very (un)helpful, and how they interacted with the customers was excellent (horrible). For the FLE control condition of NO FLE, customers did not interact with any of the human employees and were only served by the robot.

At the start of the survey, participants were asked the following: “Imagine you visit a fast casual dining restaurant. The restaurant promises wordly food for small town prices. Customers typically come here for healthy dishes and fast service at an affordable price hence fast casual dining. In addition to the human employees that work at the restaurant, they recently also employed a new service robot, Akatar. Together with human employees, Akatar is serving the customers of the restaurant. A picture of the service robot Akatar is shown below (see Plate 1). Thereafter, participants were randomly assigned to one of the experimental conditions. The exact information provided to the participants is shown in Appendix 2 for each experimental scenario.

Sample and measures

Participants were recruited via Amazon Mechanical Turk (MTurk). We took several measures to ensure the quality of our data. First, we included an attention check (open ended question asking what the scenario was about) next to the standard manipulation checks. Second, we determined a priori that we only considered MTurkers from the US, as a native English-speaking country (Aguinis et al., 2021). Third, we designed a short questionnaire (Hamby and Taylor, 2016). Fourth, we avoided using scales that only have the “end” points labeled (Goodman et al., 2013). Fifth, only participants who passed the attention check and did not take less than 230 s or more than 10 min were retained as part of the final sample. Taking response times into consideration is a method to screen MTurk data for careless responding (Aguinis et al., 2021). This resulted in a final sample of 361 useable responses (all from the US) (M_age = 43.9, 51% male).

After exposure to the scenarios, our dependent variable customer repatronage was identical to our field study. In addition, we included prior experience with service robots, prior experience with fast casual dining restaurants and participant's gender and age as control variables. We used items from our field study constructs, which were based on existing measurement scales as manipulation checks. The manipulation check for utilitarian value was “To what extent would you rate the service robot Akatar as effective?”. We used the statement “I have fun interacting with the robot” as a manipulation check for service robot hedonic value. As a manipulation check for FLE interaction quality, we included the item “Overall, I'd say the quality of my interaction with this restaurant's employees is excellent.”

Additional moderation analyses including control condition

The service triad of technology-customer-FLE is central to study 1 and study 2. So far, the setup of our studies allowed us to investigate possible augmentation between FLE and service robot. To potentially isolate a substitution role in the scenario-based experimental design as well, we included a control condition in which customers were only served by the robot and not by human FLEs. We employed the same procedure as in hypotheses 6 and 7 (PROCESS custom model 1) but coded the three categories of our moderator FLE interaction quality (no interaction, low interaction quality and high interaction quality) using the indicator method (Hayes and Preacher, 2014).

Overall, we find that the relationship between the service robot's utilitarian value and customer repatronage intentions is moderated by the multicategorical moderator FLE interaction (p < 0.05). The effect of the service robot's utilitarian value on customer repatronage intentions is positive when there is no FLE interaction (β = 1.3167; p < 0.001; CI [0.6110, 2.0224]), similar to when FLE interaction quality is low (β = 1.5298; p < 0.001; CI [0.9554, 2.1042]). In contrast, the effect is not statistically significant if FLE interaction quality is high (β = 0.4571; p > 0.1; CI [–0.1053, 1.0195]). This indicates that service robots can potentially substitute customers' interaction with human FLEs if their utilitarian value is optimized. We find that the effect of the service robot's hedonic value on customer repatronage is not moderated by the multicategorical moderator FLE interaction (p > 0.1). The results of these additional analyses are depicted in Figures 6 and 7.

Theoretical contributions

Our empirical findings from the field study of the triadic interactions between customers, service robots and FLEs in a fast casual dining restaurant provide three important theoretical insights. First, we provide empirical evidence for the interplay between different actors in the “customer-FLE-technology” triad (De Keyser et al., 2019), resulting in favorable customer outcomes. In the modern-day Service Encounter 2.0, customer-company interactions that take place in service systems are comprised of interrelated technologies, human actors, physical/digital environments and company/customer processes (Lariviere et al., 2017). In these settings, technology can both augment and substitute human FLEs (Marinova et al., 2017; Li et al., 2021). Companies that are able to find the right balance and roles for the different actors in the customer-FLE-technology triad will be able to attain a competitive advantage (Lariviere et al., 2017). However, so far little is known in the service literature about how companies must strike a balance between the different actors and their roles. To the best of the authors' knowledge, this is the first empirical study to provide insight into how perceived characteristics of different actors within the service triad (i.e. service robots and human employees) work in tandem to affect customer repatronage intentions. This has important implications for the current debate on the augmenting versus substituting role of frontline service technology within the service triad (Larivière et al., 2017; Li et al., 2021; Ostrom et al., 2021). We show that high-quality human FLE interactions in the service triad can augment the low utilitarian value of a service robot. In contrast, as the technology matures and service robots exhibit more utilitarian value to customers, the need for compensation through high-quality FLE interactions decreases and service robots can potentially substitute the human FLEs.

Second, the empirical findings advance service management literature by unraveling the relationship between anthropomorphism and social presence and their effect on perceived value. The study provides evidence for the fact that anthropomorphism – the humanlike emotions and intentions of the service robots – has a positive impact on the perceived social presence of the service robot. Extant research is inconclusive with respect to the effects of anthropomorphism. It posits that humanlike emotions and intentions can either inspire trust and bonding (Lu et al., 2020; van Pinxteren et al., 2019) or following the uncanny valley theory, customers may find a highly humanlike robot creepy and uncanny (Mori, 1970; Mori et al., 2012), creating feelings of eeriness or a threat to (a customer's) human identity (Mende et al., 2019). Our research shows that increasing anthropomorphism directly leads to social presence – a higher “sense of being with another” (Biocca et al., 2003; Heeter, 1992). This is an important finding as it suggests that not only human FLEs (Wirtz et al., 2018) but also service robots could be capable of building rapport with customers through their social presence. Moreover, we provide evidence for the important role that anthropomorphism and social presence play in hospitality services as utilitarian and hedonic value drivers. In particular, we conclude that anthropomorphism as a first degree social response (Lee et al., 2006) has a stronger effect on the utilitarian value of the service robot compared to its hedonic value. In other words, anthropomorphism impacts perceived quality of the core services provided such as serving food and drinks, stronger than perceived entertainment of customers.

Third, our studies provide strong empirical evidence for utilitarian value of service robots as a driver of customer repatronage to fast casual dining restaurants. Existing research on robots in hospitality services (see Table 1) is either conceptual in nature or uses laboratory experiments with hypothetical scenarios. Lu et al. (2020) indicate that field study research is needed to actually understand the extent to which and how service robots influence customers' outcome variables. Our field study as well as our scenario-based experimental design indicates that in the context of fast casual dining restaurants, service robot's utilitarian value has a pronounced effect on customer repatronage. Understanding the important role of service robot's utilitarian value in fast casual dining restaurants adds to our theoretical knowledge of how service robots can influence customer repatronage in hospitality.

Managerial implications

This study provides service managers of triadic service encounters with valuable insights on the implementation of service robots in frontline services and in particular, in restaurants. First, we find evidence that in hospitality services which used to be a “game of people” (Bowen, 2016), FLEs no longer always need to take an active role in the service encounter as there is a potential for service robots to substitute FLEs. Namely, we find that in fast casual dining restaurants, service robots that achieve high levels of functional performance (i.e. utilitarian value) can replace the need for customers to engage in high-quality interactions with FLEs. From the restaurant owner's perspective, implementing service robots can lead to cost reductions and productivity gains (Wirtz and Zeithaml, 2018). Especially in the social distancing era of the COVID-19 pandemic, service robots could contribute to minimizing the risk of spreading the virus. Also, services robots can be a solution to ensuring sufficient capacity to deliver consistent service in times of high staff shortages.

Second, our empirical findings have implications for service settings in which service robots should not substitute but rather be augmented by FLEs. We find that FLEs can compensate for lower levels of functional performance (i.e. utilitarian value) of service robots by engaging in high-quality interaction with customers. By demonstrating a high willingness to help and having excellent interactions with customers, FLEs can augment service robots that exhibit lower levels of utility to achieve customer repatronage. This advocates the joint service delivery by FLE – service robot teams in situations where service robot technology is not fully optimized. In this sense, technology and FLE can be used in tandem to provide a better service outcome (Froehle and Roth, 2004; Li et al., 2021).

Third, we provide essential insights for robot engineers and designers, gathered from a real-life setting (Mende et al., 2019) on the human likeness design parameter of service robots. The findings from our field study show that the more service robots in restaurants evoke the perception of having thoughts and emotions, the higher customers evaluate the robots' utilitarian and hedonic value. This indicates that service robots should be designed in a way to display social presence by having the ability to have thoughts and convey emotions in order to create customer value.

Fourth, our results have implications for policy makers as well. Recently, the Future of Jobs report published by the World Economic Forum (2020) articulated that the surge in digital technologies and automation largely transforms tasks, jobs and skills within the next five years. In line with these developments, Larivière et al. (2017) emphasized the importance of role readiness for employees to acclimate in the new service environment. This demands a completely new set of skills and a proactive attitude from the public sector to support reskilling and upskilling for employees (Huang and Rust, 2020; World Economic Forum, 2020). This study shows that the jobs of FLEs in hospitality will be subject to change, such that they in some cases will be substituted and in other cases augmented by service robots. Policy makers should prepare the workforce in hospitality for this change by providing FLEs with the opportunity to reskill (in case of job substitution) or upskill (in case of job augmentation). We advocate for training specific collaborative skills on how to work with a service robot in a team.

Limitations and future research

This research offers several avenues for future research. First, the empirical context of our field study entails a European, fast casual dining restaurant. Next to this, the sample is skewed since most of the respondents were female (70.8%), between 18 and 34 years old (81.5%), repeat customers (69.4%) and visiting the restaurant with friends (62%). Moreover, we carried out our research during the 2020/2021 COVID-19 pandemic. This warrants caution regarding the generalizability of our findings. Future studies should shed more light on this by conducting similar investigations across different cultural settings, types of restaurants and beyond the pandemic. In particular, it would be interesting to obtain insight into whether service robot's utilitarian and hedonic value play a more or less pronounced role in hospitality settings other than fast casual dining restaurants, and how this potentially affects the interplay between the different actors of the service triad.

Second, the service robots that were employed by the fast casual dining restaurant in our field study were endowed with limited hedonic characteristics. Namely, they communicated unilaterally and could not respond to customers, make jokes or pose for pictures. This may explain the lack of a significant relationship between the service robots' hedonic value and customer repatronage, contrary to the findings of our scenario-based experiment. Contemporary service scholars postulate that service robots will be able to deliver cognitively complex service tasks and low emotional service tasks (Lu et al., 2020; Paluch and Wirtz, 2020; Wirtz et al., 2018). Building on these insights, we encourage future service scholars to develop field studies to further disentangle the service triad and the link between service robot hedonic value, customer repatronage intentions and FLE interaction quality. Another interesting avenue for future research is the analysis of actual customer behavior demonstrating perceived hedonic value, such as taking a picture or video of the service robot or dancing with the robot, instead of mere customer perceptions.

Third, in our field study, we base our findings on a cross-sectional sample of customers in a triadic service encounter, obtained in the early stages of service robot implementation. This opens up the opportunity for further research to take a longitudinal perspective on the effects of service robot implementation in hospitality as it would be valuable to understand the extent to which our findings hold for revisiting customers over time.

Fourth, future research could further expand our knowledge on factors – beyond FLE interaction quality – that affect the relationship between service robot's utilitarian and hedonic value and customer outcomes. Interesting research questions could be: what is the impact of the utilitarian and hedonic value of the FLE, or to what extent do customers' prior experiences with the robot or the type of party (friends versus family versus business relations) play a role in the interactions with service robots and the effects it has on customer outcomes?

Fifth, it is worthwhile to study how augmentation or substitution by service robots in the service triad for certain tasks affects the employee experience. Do employees feel empowered by their robotic counterpart or rather threatened to become obsolete? While the customer experience has received major academic interest, so far research in the domain of the employee experience has been scarce (Lariviere et al., 2017).

Lastly, we encourage researchers to further expand the service triad by investigating how third parties – such as other employees or other customers – are influenced by and influence the interplay between customers and a team of service robots and frontline employees. Researchers increasingly consider the role of third parties who interact with customers and/or service providers (Abboud et al., 2020), and future research can explore how employees fulfill the third-party roles of bystander, connector, endorser, balancer or partner role in indirect interactions (Abboud et al., 2020). This research direction builds on Bowen's (2016) call for further investigation of employee roles in an evolving service context characterized by growing technologies augmenting employees. In this context, future research can investigate whether and how frontline employees can create value by adopting a third-party role when service robots are directly interacting with customers.