Service innovation in the restaurant sector during COVID-19: digital technologies to reduce customers ’ risk perception

Purpose – This paper investigates the extent to which the COVID-19 pandemic has accelerated service innovation in the restaurant sector. It explores the use of digital technologies (DT) as a safety-empowerment andresilientstrategyinthefood-serviceindustryduringthepandemic.ItalsoinvestigatestheimpactofDToncustomers ’ risk perception (CRP) and customers ’ intention to go to restaurants (CIR) in Italy. Design/methodology/approach – Based on the theory of planned behaviour and perceived risk theory, this study investigates a sample of customers residing in Italy. Multiple regression and mediation analyses are conducted to test the research hypotheses, adapting the logic model developed. Using the bootstrapping technique, this study also explores whether the pandemic has moderated the relationship among several variables adapted from the literature. Robustness tests are also performed to corroborate the analysis. Findings – The pandemic has accelerated the food-service industry ’ s digital transformation, forcing restauranteurstoimplementDTtosurvive.FindingsshowthatDTsupportrestauranteursinimplementinginnovativeservicesthatreduceinteractionsandempowercleanlinessamongworkersandcustomers,reducingCRPandpreservingCIR.Thus,managingriskperceptionishelpingtherestaurantsectortorecover. Practical implications – Practical implications are presented for policymakers to catalyse the digital transformation in small- and medium-sized restaurants. The results may also be beneficial for entrepreneurs who can implement innovative service practices in order to reduce interaction and empower cleanliness levels. Moreover, academics can use these results to conduct similar research in other geographical contexts. Originality/value – The present research represents the first study investigating the relationship between theuseofdigitaltechnologiesandtheintentionsofcustomerstogooutfordinnerduringtheongoingpandemicinItaly.


Introduction
The recent Covid-19 pandemic outbreak has damaged the economy worldwide. In particular, the restaurant sector (RS) -considered a high-risk industry (Song et al., 2021) has been sharply affected by Covid-19, which has raised revolutionary trials to tackle. The rapid spread of the virus led to restaurant restrictions with a consequent reduction in turnover, job loss and -in some casespermanent closure of commercial activities (Min et al., 2021). In addition, a significant change in customers' behavioural intentions to stay home has emerged (Rizou et al., 2020). Customers' attitudes to visit restaurants have significantly reduced due to the fear of being exposed to the potential Covid-19 risks (Galanakis, 2020;Esposito et al., 2021). Similarly, restaurant operators have been demonstrated to be susceptible to contracting the virus, and their willingness to go to work has been considerably reduced.
The restaurants have been called to reorganize their activities to continue in a context characterized by the need to ensure health safety, reduce the risk perception and comply with governments measures to survive. Although the World Health Organization has declared that food is not a means of Covid-19 transmission, the operations undertaken during the restaurant supply chain (SC) have been considered hazardous due to the numerous interactions among the actors involved in the (SC) (Min et al., 2021). As a result, a lot of restrictive measuressuch as the constraining on opening hours and the cutting of seating capacitieshave been implemented worldwide. However, in the more recent month following the advent of the vaccination campaign, the intensity of such measures has been reduced because society needs to help the economic and social recovery of businesses. In this perspective, the Service Innovation (SI) to enhance traditional restaurant services and create new service opportunities has emerged as the only way of coping with these critical issues. Although SI has always been considered as a strategic factor for competitiveness and at the discretion of the managers (Feng et al., 2020;Helkkula et al., 2018;Kowalkowski and Witell, 2020), the pandemic has posed an imperative to innovate in the RS in order to ensure the organization's survival and resilience (Edvardsson et al., 2018;Heinonen and Strandvik, 2021). This "imposed service innovation" has been expected to reduce negative implications by sparking a change of thinking and exciting business opportunities never considered before (e.g. Nenonen and Storbacka, 2020;Batat, 2020). In this light, the adoption of digital technologies (DT)such as contactless digital payment, advanced cleaning systems, digital menus accessible via QR codes, service robots, touchless elevators, food delivery apps, etc.has been used as service innovation tools to implement risk-reduction strategies. Through these tools, DT is expected to reduce guest and workers interactions and empower restaurants cleanliness (Shin and Kang, 2020). In this context, scholars have shown thatduring the post-pandemic eracustomers' risk perception should be considered one of the main barriers to the re-start of food service activities since it influences the purchase decision-making process (e.g. Dedeo glua and Bo gan, 2021; Leung and Cai, 2021).
Consequently, reducing customers' retention to go to the restaurant has become the new priority for the RS. Indeed, "due to the risks associated with the possibility of contracting the virus, the mitigation of customers' perceived risk also requires attention" (Min et al., 2021, p. 2). Therefore, in the wake of its relevance, scholars have started to investigate risk perception in the RSboth from a customers and workers' perspectiveto assess the levels of risk perception and the actions implemented to manage it during the post-pandemic (e.g. Roberts et al., 2020;Min et al., 2021;Yan et al., 2021). In particular, academics have shown that high levels of perceived risk determine the reduction of customers' intention to visit restaurants (e.g. Song et al., 2021;Foroudi et al., 2021). At the same time, other scholars have started to investigate the use of DT for RS recovery during the pandemic (e.g. Shin and Kang, 2020;Demaitre, 2020;Hoque et al., 2020). More specifically, Shin and Kang (2020) have explored how the use of DT could reduce the risk perception of the virus among customers and workers in the hospitality sector (i.e. hotels and restaurants within the hotels). They have highlighted that these technologies could impact food safety and hygiene practices. However, despite scholars have investigated separately the impact of DT con risk perception and the impact of risk perception on the intention to go to the restaurant, the relationship between the COVID-19 and innovation in restaurants use of DT and the customers' risk perception and its consequential effect on customers' intention to go to the restaurant remains still unexplored. Starting from this background, adopting a theoretical framework based on the integration of the theory of planned behaviour (TPB) and the perceived risk theory (PRT), this paper aims to answer the following research question:

RQ1.
To what extent do digital technologies support restaurants in managing risk perception and enticing customers to go to restaurants during the Covid-19 pandemic?
RQ2. How the Covid-19 pandemic has affected customers' perception of digital technologies?
To answer the research questions, a survey on a customers' resident in Italy has been carried out with the intent to explore: (1) The potential mediating role of customers' risk perception in the relationship between DT and customers' intention to go to the restaurant; (2) The potential mediating role of expected interactions and expected cleanliness in the relationship between DT and customers' risk perceptions and (3) The potential moderating role of the COVID-19 pandemic in the relationship between DT and (a) expected interactions; (b) expected cleanliness and (c) customers' risk perceptions.
This paper is divided into seven sections. After the introduction, the second section presents the literature review focused on service innovation in the restaurant sector. Moreover, prior research on customers' risk perception during the COVID-19 is provided. The third section develops the theoretical framework and the research hypotheses. The fourth section describes the research methodology. In the fifth section, the results are presented. The sixth section presents the managerial discussion. Lastly, the main conclusions, limitations and managerial implications of the study are discussed.
2. Literature review 2.1 Service innovation and digital technologies in the restaurant sector Over the years, the idea that firms can achieve sustainable and competitive advantages through the innovation of products and services has been widely recognised (Xin et al., 2006;Lee et al., 2010;Maglio, 2017). In turn, companies have started to commit to SI, and scholars have increased their attention towards this issue (Ciasullo, 2018;Helkkula et al., 2018;Mele and Russo Spena, 2018). Accordingly, the body of literature on SI has significantly grown (Lush and Nambisan, 2015;Edvardsson et al., 2018). Scholars have adopted many assumptions about SI over time. Coombs and Miles (2000) have classified SI research into three distinct approaches: assimilation, demarcation and synthesis. The assimilation approach focuses on the impact of DT as the main driver of service innovation (Gallouj, 2002;Toivonen and Tuominen, 2009). Therefore, according to the assimilation perspective, existing theories on product innovation should also analyse SI research without adaptation (e.g. Evangelista, 2000;Miozzo and Soete, 2001). On the contrary, the demarcation approach argues that SI essentially differs from product innovation (Coombs and Miles, 2000), requiring the development of new specific-service theories for its investigation (Drejer, 2004;Tether, 2005). The synthesis approachovercoming both the assimilation and the demarcation perspectivessuggests that theories on SI should provide an integrated perspective that combines innovation in manufacturing and services, including both TQM 34,7 technological and non-technological innovations (Coombs and Miles, 2000;Gallouj and Savona, 2009;Wittel et al., 2016). Although these perspectives contributed to the development of the literature on SI, a more holistic approach to investigating this topic was required. From this perspective, servicedominant (S-D) logic emerged as the most useful answer to this theoretical need (Vargo and Lusch, 2004, Lusch and Vargo, 2014. S-D logic is a theoretical view applied to management research, together with economics and society Lush, 2004, 2006). According to S-D logic, the exchange of goods and money lie down on a service-for-service exchange (Pels et al., 2014). In this light, scholars have provided broader definitions of service innovation from an S-D logic perspective (Edvardsson and Tronvoll, 2013). One of the most recognised concept definitions describes SI as "the re-bundling of diverse resources that create novel resources that are beneficial (i.e. value experiencing) to some actors in a given context; this involves a network of actors, including the beneficiary (e.g. the customer)" (Lusch and Nambisan, 2015, p. 161). Therefore, following this approach, SI comprehends three elements of S-D logic: (1) the service ecosystem; (2) service platforms and (3) value cocreation (Barile and Saviano, 2010;Gr€ oonroos and Ravald, 2011;Polese et al., 2015;Frey et al., 2019). The service ecosystem includes all the actors involved in the service's activities that proactively participate in the value co-creation process (Polese et al., 2011;Vargo and Lusch, 2017). The service platform comprises tangible and intangible resources that facilitate service exchange, providing an organisational structure for resources (Lush and Nambsian, 2015). The last element of SI is the value co-creation linked to the ability of social and economic actors to integrate resources of service ecosystems and service platforms in order to collectively create value (Barile and Polese, 2010;Lush and Nambsian, 2015;Rajala et al., 2016;Polese et al., 2019).
A recent stream of research has begun to recognise the impact of DT on SI (Lush and Nambisan, 2015;Kaartemo and Helkkula, 2018). Therefore, DT have become a pivotal component of value creation in service systems (Maglio and Spohrer, 2008;Nambisan et al., 2017). A lot of enabling technologiessuch as the fifth-generation mobile network (5G), radio frequency identification (RFID), artificial intelligence (AI), mobile devices and smartphones, wearables, applications or apps (along with application programming interfaces), cryptocurrency and blockchainsupport firms in enhancing services, customer engagement and expectation . From an S-D logic perspective, DT acquire a different role, according to the ability of the technology to lead service innovation (Gummesson, 2012;Kindstrom et al., 2013;Frey et al., 2019). Thus, DT should be analysed in the double role of "operand" and "operant" resources (Nambisan, 2013;Lusch and Nambisan, 2015). On the one hand, actors use operand resources to gain support. Accordingly, operand resourceswhich are generally tangible and static (e.g. natural resources)should be considered enablers or facilitators of SI (Nambisan, 2013). On the other hand, operant resources are used to act with other resources in order to produce beneficial effects. Therefore, operant resources are seen as initiators or actors of SI (Akaka and Vargo, 2014;Lush and Nambisan, 2015). Different to operand resources, operant resources are intangible and dynamic (e.g. knowledge, skills). These characteristics make operant resources more difficult to transfer and, thereby, a source of competitive advantage (Capon and Glazer, 1987).
The pandemic has fostered SI dynamics, using DT as an instrument for organisational survival and resilience (Heinonen and Strandvik, 2021;Mele et al., 2020). In this "imposed service innovation" (Heinonen and Strandvik, 2021), DT have supported the restaurant sector in creating new business models and services to survive. In addition, restaurants and hotels have started to use DT to implement risk-reduction strategies (Shin and Kang, 2020). Mitchell (1999) defines a risk-reduction strategy as a process that attempts to decrease the perceived risk of purchasing a product or service. Therefore, a level of perceived risk higher than the acceptable level leads consumers to adopt a risk-reduction strategy in the decision-making COVID-19 and innovation in restaurants process (Gem€ unden, 1985;Pappas, 2016). Accordingly, DT can reduce interactions between customers and employees, transforming the restaurant sector (RS) from traditional "hightouch and low-tech" service experiences into "low-touch and high-tech" service experiences (Bitner et al., 2000;Shin et al., 2019;Shin and Kang, 2020). Moreover, DT can implement advanced cleaning systems to enhance the cleanliness and safety of restaurant surfaces (Xenex, 2020). Therefore, in the pandemic scenario, DT for enhancing cleanliness and reducing interactions can influence perceived risk and the resulting behavioural intention to go to restaurants (Shin and Kang, 2020). From this background, understanding the impact of risk perception on customers' behaviour becomes crucial for answering the research questions.
In adverse events, people make decisions instinctively according to their perceived risk (Slovic, 1987). Therefore, the level of risk perception should produce two opposite behaviours. On the one hand, people with low-risk perceptions may exhibit a negative attitude towards compliance with safety guidelines and preventive measures imposed by governments (Hakim et al., 2021). On the other hand, people with a high-risk perception may slow down the global economy's recovery by avoiding behaviours considered usual before the pandemic. Due to the relevance of risk perception in shaping behaviours, its investigation has become essential in understanding customers' decision-making.
In keeping with these arguments, scholars have begun to investigate risk perception during COVID-19 (e.g. Cori et al., 2020;Dryhurst et al., 2020;Neuburger and Egger, 2021). In this scenario, research on risk perception in the RSwhich is proving to be one of the hardest hit by the pandemic due to the high risk of contaminationis progressively increasing, with both customer and worker risk perspectives being investigated (Yang et al., 2020). The majority of studies retrieved in the literature focus on customers' risk perception (e.g. Foroudi et al., 2021;Min et al., 2021;Song et al., 2021;Villace-Molinero et al., 2021;Yost and Cheng, 2021). In particular, Villace-Molinero et al. (2021) have demonstrated that customers' risk perception has changed since the COVID-19 pandemic. Accordingly, Sung and King (2021) have shown that media reporting has positively influenced risk perceptions during the outbreak. Other scholars have demonstrated that perceived risk affects customers' purchase intention (e.g. Leung and Cai, 2021;Pham Minh and Mai, 2021). Meanwhile, another stream of literature has turned attention towards the endeavour that the RS has to make to achieve and maintain customers' trust in authorities and restaurateurs to recover quickly from the shock (e.g.. Dedeo glua and Bo gan, 2021; Foroudi et al., 2021;Hakim et al., 2021;Yost and Cheng, 2021). At the same time, other researchers have explored workers' risk perception TQM 34,7 (e.g. Roberts et al., 2020;Asefa et al., 2020;Yan et al., 2021). Roberts et al. (2020) have shown that workers perceived a moderately higher risk of contracting the virus than customers. However, Asefa et al. (2020) demonstrated that workers' risk perception was positively associated with age, knowledge about the virus and preventive behaviours. Moreover, the pandemic has also strongly impacted workers' mental health. In this line, Yan et al. (2021) have shown that job satisfaction reduces the link between workers' risk perception and their potential depressive symptoms. Lastly, Wut et al. (2021) have investigated the risk perception of both customers and workers in order to understand how crisis-management practices have been implemented in the RS.
Despite the fact that the literature on this issue in the RS during and post the COVID-19 pandemic (when there were no restrictions/lockdowns) is growing, research on the role of DT in managing risk perception is still scant. Table 1 provides the main results of the literature review.

Theoretical framework and hypotheses development
The present research was developed following an approach that integrates two behavioural theories, that is the theory of planned behaviour and perceived risk theory. Rahmafitria et al. (2021) suggest that these theories might be considered complementary in underpinning customers' behaviour. Accordingly, several studies have employed a theoretical framework that combines the TPB and PRT (e.g. Liao et al., 2010;Rahmafitria et al., 2021).
The TPB is a popular and extensively validated framework employed in a variety of instances of human behaviour. According to Ajzen (1985Ajzen ( , 1991, behavioural intentions are established by attitude, subjective norms and perceived behavioural control. Attitude "refers to the degree to which a person has a favourable or unfavourable evaluation or appraisal of the behaviour in question" (Ajzen, 1991, p. 188). It is described as the appearance of beliefs due to certain actions (Liao et al., 2010). As a result, if a person is certain of an action's positive or negative effects, they will take a specific attitude. Subjective norms are linked to social factors that enclose the perceived social pressure to undertake, or not, a specific behaviour (Vesci and Botti, 2019). In particular, the subjective norm is a subjective belief according to which behaviours depend on others' opinions or judgements (Ajzen, 1991;Yuzhanin and Fisher, 2016). Lastly, perceived behavioural control refers to a customer's perception of their ability to fulfil a specific action. Perceived behavioural control reflects past experience with the conduct's performance and expected impediments that could prevent the behaviour (Kiriakidis, 2017).
The literature suggests that the TPB may be useful in explaining customers' behaviour in a pandemic scenario (e.g. Prati et al., 2011;Rahmafitria et al., 2021). During the COVID-19 outbreak, the TPB may be adopted to explain the effects of attitude, perceived behavioural control and subjective norms to comply with protocols and measures affecting customer intentions to go to restaurants (CIR) (Rahmafitria et al., 2021). The pandemic has trigged perceived risks for restaurant customers, impacting their purchase behaviour (Foroudi et al., 2021). Indeed, when someone perceives a high risk of contracting the virus, they will avoid going to the restaurant. Therefore, when investigating customers' decision-making, it appears to be essential to take into account their perceived risks. Perceived risk is associated with perceived uncertainty (Bauer, 1960;Rehman et al., 2020). A higher level of uncertainty makes customers feel a higher level of risk. Scholars have widely adopted PRT to investigate consumer decision-making behaviours (Buratti and Allowood, 2019). Recently, a growing number of studies have started to explore the impact of risk perception in the hospitality industry (e.g. Lepp and Gibson, 2003;Adam, 2015). Perceived risk has been measured as a multi-dimensional construct (Liao et al., 2010). Researchers have identified different risk dimensions to explain customers' purchase behavioural intentions in the hospitality industry. For example, Adam (2015) has categorised perceived risk into six dimensions; that is political

COVID-19 and innovation in restaurants
risk, financial risk, socio-psychological risk, environmental risk, expectation risk and physical risk. Moreover, Bentley and Page (2008) have highlighted the importance of health risks in hospitality research. After the SARS pandemic, the Asian tsunami and the twin towers terrorist attack, many researchers started to investigate health and safety risk perception (e.g. Yang and Nair, 2014). The COVID-19 outbreak has generated an increasing degree of risk from a multi-dimensional perspective, especially in the RS. Thus, understanding and managing customer risk perception is a key point for RS survival. Therefore, it has emerged that the integration of the TPB and PRT permits a deeper insight into the research objectives on which the conceptual model described in Figure 1 is built.
In the pandemic context, the TPB could be used as an alternative to interpreting the effect of attitude, perceived behavioural control and external influences in compliance with Covid-19 protocols on the customers' intention to go to the restaurant. Other studies in the outbreak scenario have attempted to enlarge the TPB (e.g. Agarwal, 2014;Myers and Goodwin, 2012;Ho et al., 2013). However, to the best of our knowledge, no research has expanded the TPB involving risk perception factors (i.e. socio-psychological, security, privacy, hygienic, environmental and health risk), integrating the PRT. Thus, TPB has never been used to explain the relationship between the impact of DT on risk perception and the intention of customers to visit restaurants. Accordingly, this background leaves a gap on which our study is built, aiming to provide a theoretical novelty.

Expected interactions and expected cleanliness
DT are expected to underpin the RS in three different domains: physical, cognitive and social (Gomes de Freitas and Stedefeldt, 2020). More specifically, DT may empower the services' infrastructures (i.e. physical domain), support the sector in understanding customers' and workers' perceptions and beliefs (i.e. cognitive domain) and, lastly, enhance social support to co-create resilient strategies (i.e. social domain; Galanakis, 2020).
Scholars have demonstrated that DT may help the RS manage health and safety issues (Esposito et al., 2021). Accordingly, Albert and Hayes (2003) and Hoque et al. (2020) have shown that robots and other DT have allowed restaurants to facilitate physical distancing practices. Furthermore, scholars have contended that DT have also been implemented to boost delivery practices and food preparation hygiene (e.g. Demaitre, 2020;Meisenzahl, 2020). In this light, DT may support restaurants in achieving logistical and disinfection tasks and giving COVID-19 information to people, minimising the fear of the virus (Meisenzahl, 2020).
Previous research on risk perception during the pandemic in the RS has mainly explored cultural, socio-psychological and demographic factors (e.g. Choe et al., 2021;Foroudi et al., 2021;Hakim et al., 2021). External sources of perceived health risks during the outbreak have been investigated only in the more extensive hospitality industry. In particular, Zeng et al.(2020) have demonstrated that DT may reduce customers' perceived health risks. Furthermore, Shin and Kang (2020) have explored how DT have been used in the hospitality sector as a risk-reduction strategy to manage booking intentions during the pandemic. They have shown that risk perception is interconnected with the level of expected interactions and expected cleanliness. Moreover, before the pandemic, Nicholls (2010) highlighted that the implementation of technology-mediated systems may enhance physical distancing and reduce the opportunity for interactions. Therefore, the following hypotheses aim to investigate how the use of DT may affect expected interactions and the consequent relationship between risk perception and customers' intention to go to restaurants.
Accordingly, the following hypotheses are developed: H1a. There is a positive relationship between DT use in restaurants and the reduction of expected interactions among customers and workers during the COVID-19 pandemic.
H1b. There is a positive relationship between the reduced level of expected interactions and the reduction of customers' risk perception in restaurants during the COVID-19 pandemic.

COVID-19 and innovation in restaurants
H1c. The expected interactions mediate the relationship between DT and customers' risk perception.
During the COVID-19 pandemic, the perception of cleanliness has played a crucial role in restaurants. The RS needs to guarantee a higher level of cleanliness to attract customers (Shin and Kang, 2020). Guests are greatly interested in restaurant cleaning quality (Wang and Hung, 2015). Indeed, the literature has shown that cleanliness impacts service quality (Barber et al., 2011), customer satisfaction (Liu and Jang, 2009), security (Amblee, 2015) and perceived risk (Shin and Kang, 2020). In this light, the implementation of DT to cleaning systems can enhance the cleanliness perception, resulting in a lower level of customers' risk perception (Shin and Kang, 2020;Esposito et al., 2021). Thus, the following hypotheses are proposed: H2a. There is a positive relationship between DT and the level of expected cleanliness in restaurants during the COVID-19 pandemic.
H2b. There is a positive relationship between expected cleanliness and the reduction of customers' risk perception in restaurants during the COVID-19 pandemic.
H2c. The level of expected cleanliness mediates the relationship between DT and customers' risk perception.
Through the theoretical lens of PRT and the TPB, perceived risk represents a pivotal factor in interpreting how consumers make decisions. The theoretical framework adopted in the present paper has hinted at the role of attitude, subjective norms and perceived behavioural control -concerning the use of DTin determining customers' decision-making under uncertainty (Rahmafitria et al., 2021). In the context of this research, consumers are likely to evaluate risks according to the cleanliness and interactions level perception during the pandemic, which impacts their intention to go to restaurants. Accordingly, the following hypotheses are developed: H3a. There is a positive relationship between the use of DT in restaurants and the reduction of customers' risk perception during the COVID-19 pandemic.
H3b. There is a positive relationship between reducing customers' risk perception and their intention to go to restaurants during the COVID-19 pandemic.
H3c. There is a positive relationship between the use of DT and customers' intention to go to restaurants during the COVID-19 pandemic.
H3d. There is a positive relationship between the level of expected cleanliness and customers' intention to go to restaurants during the COVID-19 pandemic.
H3e. There is a positive relationship between the reduced level of expected interactions and customers' intention to go to restaurants during the COVID-19 pandemic.
The conceptual model depicted in Figure 1 shows the existent interconnections among the constructs already discussed. In keeping with these arguments, we assume the existence of mediation relationships between the variables. Accordingly, the following hypotheses are developed: H4a. The reduction of customers' risk perception mediates the relationship between DT and customers' intention to go to restaurants.
H4b. Customers' risk perception mediates the relationship between expected cleanliness and customers' intention to go to restaurants.

TQM 34,7
H4c. Customers' risk perception mediates the relationship between expected interactions and customers' intention to go to restaurants.

The potential moderating role of COVID-19
Starting from the previous hypotheseswhich assume a relationship between the use of DT and (1) the level of expected interactions; (2) the level of expected cleanliness and (3) the level of customers' risk perceptionthe present paper also aims to investigate if the COVID-19 outbreak has impacted on these relationships. Moreover, it attempts to explore which relationship is most affected by the pandemic. Accordingly, the following hypotheses are proposed: H5a. The COVID-19 pandemic has moderated the relationship between DT and the level of expected cleanliness.
H5b. The COVID-19 pandemic has moderated the relationship between DT and customers' risk perception levels.
H5c. The COVID-19 pandemic has moderated the relationship between DT and the level of expected interactions.

Research methodology 4.1 Data collection and coding process
This paper aims to analyse whether and to what extent the use of DT in the RS during the COVID-19 pandemic has affected customers' risk perception, considering the level of expected interactions and expected cleanliness. The research attempts to investigate how these factors should influence customers' intention to go to restaurants. Moreover, the potential moderating role of COVID-19 on the relationship between the use of DT and the indirect effects of customer risk perception (CRP), expected cleanliness (EC) and expected interactions (EI) are explored. To this end, an exploratory study of a quantitative nature was carried out by underpinning multiple regression analyses to test the research hypotheses. Data were collected from a sample of customers resident in Italy. According to the literature analysis, a questionnaire was built to analyse which of these dimensions have more impact on CIR. Furthermore, we attempt to explore which indirect effect is more influenced by COVID-19. The questionnaire comprised 33 questions, including personal data and closedended questions using a five-point Likert scale (Matell and Jacoby, 1971;Ciasullo et al., 2021). The questionnaire's reliability and usability were checked with a pre-test by three researchers. A random sampling from the mailing list of three restaurantslocated in the north, centre and south of Italywas performed. Restaurants were chosen from the Gambero Rosso gastronomic guidebook 2021. The restaurants were selected according to the following criteria. The researchers have chosen one restaurant for the north of Italy, one restaurant for the south and one restaurant for the centre of Italy to explore customers' perceptions all over the Italian country and avoid biases related to the location. Moreover, the researchers have preferred restaurants located in big cities where tourism is also active (i.e. Milan, Rome and Naples). In this way, researchers have ensured that the customers potentially come from all Italian regions despite the restaurants being located in specific cities.
Furthermore, the researchers have selected the size of the restaurant as another criterion to skim the number of restaurants. It has been checked if the restaurants that have fitted the selection criteria have a website with a newsletter section. This choice is justified because authors are sure that the restauranteurs have a mailing list and the permission to send an e-mail to their customers.

COVID-19 and innovation in restaurants
Three restaurants were contactedone for each geographical locationto ask if they wanted to collaborate in the research. Restaurants that showed availability for the study were selected. The total number of e-mails collected from the three restaurants was 1980. Accordingly, the population size (N) was composed of 1980 customers. The sample size (n), calculated with the formula for finite population, was considered sufficiently representative (i.e. confidence interval: 2.09; confidence level: 95%). Table 2 provides a detailed description of the sampling process.
The questionnaire was disseminated using a computer-assisted web interview (CAWI) technique (Bartosik-Purgat, 2018). Customers were invited by e-mail to take part in the research in May 2021. To avoid cross-cultural methodological questionsas all the interviewees were Italian speaking (Smaldone et al., 2020)the survey was developed in the Italian language and then translated into English. The questionnaire comprised six sections. Section 1 concisely described the purpose of the study. Section 2 comprised 12 questions on customers' attitudes towards DT, their perceived behavioural control and the external influences that lead customers to use DT during the pandemic. Section 3 contained six questions on customers' risk perception when DT are implemented in restaurants. Section 4 contained six questions regarding the level of cleanliness and interactions expected by restaurant customers. Section 5 comprised five questions relating to customers' willingness to return to restaurants and their perception of DT during the COVID-19 pandemic. Lastly, four questions' personal data section (Section 6) was administrated. Respondents stated their perceived risk or the level of agreement for each item or sentence (i.e. 1 5 least/strongly disagree; 5 5 maximum/strongly agree). Each construct was built using multiple indicators adapted from the literature (e.g. Li et al., 2020;Nardi et al., 2020;Shin and Kang, 2020;Amin et al., 2021;Foroudi et al., 2021;Zhong et al., 2021). The items developed in each section have been adapted from previous studies developed in different contexts and rephrased according to our research scenario. Only items linked to the "Digital technologies" section and "Customers' risk perception" section have been taken verbatim from previous studies since they have developed according to the TPB and PRT. Moreover, some new items were developed to fulfil the information needs to answer the research questions. Table 3 shows the coding framework designed.

Data analysis and empirical model
This research applied multiple regression analysis implementing the ordinary least square (OLS) model to test the research hypotheses. OLS is a multivariate statistical method used to describe the linear dependence of the outcome variable from one or more predictor variables (Tripepi et al., 2008). According to the conceptual model described in Figure 1, SPSS 23.0 software was used to test the hypotheses. A normality test (i.e. skewness and kurtosis) was performed for each variable. Moreover, to evaluate potential multicollinearity problems, the variance inflation factor (VIF) was tested. In order to avoid possible heteroscedasticity problems, a White test was performed. Furthermore, a Durbin-Watson testto detect probable autocorrelation in the residualswas carried out.
Furthermore, in order to address the common method bias, an estimator bias was analysed through the standard error methodology     . Lastly, before testing the conceptual model, the maximum shared squared variance (MSV) and the average variance extracted (AVE) were estimated to define discriminant validity and validate the constructs. The results show an AVE of 0.36 is higher than the value of MSV of 0.20. Accordingly, the discriminant validity is set.
To test the mediating role of the variables expected cleanliness, expected interactions and customers' risk perception in the relationship between DT and customers' intention to go to restaurants, the following empirical model was developed: where C_int_to_go is the variable customers' intention to go to restaurants, γ 0 is the constant, γ 1, γ 2, γ 3 and γ 4 are the coefficients of the explanatory variables (DT; CRP; EI and EC) and ε i is the error.

TQM 34,7
Moreover, an additional analysis was performed to analyse if COVID-19 has influenced customers' perceptions. A moderation analysis for each effect between DT and the dependent variables expected cleanliness, expected interactions and customers' risk perception was carried out, based on the following equations: where DT * Covid_EC, DT * Covid_EI and DT * Covid_CRP represent the interactions between the respective independent variables. These relationships were tested according to the bootstrapping technique (Awang et al., 2015) since this method allows the researchers to test the model holistically, providing better performance compared to the other possible analytical approach (e.g. casual steps) (Nevitt and Hancock, 2001).

Results and discussion
This section presents the main findings. The descriptive results and the results of the dependency model are shown. Moreover, the results of the moderation analysis are displayed.

Descriptive results
Descriptive statistics are presented in Table 4. Of the respondents, 48.52% were female, and 47.57% were male, while 3.80% did not provide this information. The average age of the interviewees was 48 years. Moreover, it is shown that the largest number of respondents had a high school diploma (i.e. 35.24%) and a bachelor's degree or master's degree (i.e. 53.34%).
The concentration of interviewees for the survey is shown in Figure 2. Demographic statistics for the sample are described in the legend. The largest portion of respondents came from Campania (i.e. 190), Lazio (i.e. 200) and Lombardy (i.e. 3,030). The regions with less aggregation were Veneto (i.e. 3), Molise (i.e. 1), Liguria (i.e. 5) and Marche (i.e. 5). Sardinia and Sicily did not present answers. Table 5 presents the standardised factor loading. All the variables are considered acceptable since the skewness and kurtosis values are, in all cases, lower than 2.

Correlation matrix and fit indices
In order to evaluate the goodness-of-fit of the model, the correlation matrixes for the latent construct and fit indices were computed.

COVID-19 and innovation in restaurants
variables. No multicollinearity problems were retrieved since the correlation values are lower than the threshold (i.e. 0.8). Table 7 provides the fit indices. Significant correlations were found between (1) DT and CRP; (2) EI and CRP and (3) EI and CRP. The VIF (in all cases lower than 2.625) suggests no multicollinearity problems (Miceli et al., 2014). To avoid heteroscedasticity problems, a White test was carried out. The resulting p-value (19.038) was not significant. As a consequence, no robust standard errors were retrieved. Moreover, a strong causal relationship between the dimensions analysed is granted by an R coefficient equal to 0.782 (Emerson, 2015). As shown in Table 7, the R-squared coefficientscalculated to assess the variability level among datashow a good variability explanation (i.e. R square: 0.776 and adjusted R square: 0.719). Furthermore, the standard error was 0.870. The analysis suggests that our model has a good explanation of all the variables. Moreover, Table 7 presents the White test results and the Durbin-Watson test. The White test retrieved a non-significant p-value (i.e. 19.038), suggesting no heteroskedasticity problems. In addition, the value obtained from the Durbin-Watson test (i.e. 1.841) reveals that there is no autocorrelation among the residuals.

Results of the dependency models
The results of the dependency models are presented in Table 8. Model 1 shows the results that test the relationship between DT and expected cleanliness (H2a). Model 2 tests the relationship between DT and expected interactions (H1a). Model 3 represents the results that test the impact of DT on the level of customers' risk perception (H1b, H2b, H3a). Lastly, Model 4 provides the results regarding the impact of DT on customers' intention to go to restaurants (H3b, H3c, H3d, H3e). All the models are statistically significant, with a p-value lower than 0.01. Moreover, the coefficient of determination for each model is between 0 and 1 (Model 1-R 2 5 0.4321; Model 2-R 2 5 0.5202; Model 3-R 2 5 0.7104; Model 4-R 2 5 0.2783). These values suggest that the regression models provide adequate explanatory power.
The impact of DT on the level of expected cleanliness (Model 1) and expected interactions (Model 2) is, in both cases, positive and significant (i.e. Model 1 5 0.7819, Model 2 5 0.8675). In line with previous studies (i.e. Shin and Kang, 2020;Zeng et al., 2020), the relationship between expected cleanliness and customers' risk perception is positive and significant (i.e. 0.3203). Similarly, the relationship between the variable expected interactions and customers' risk perception shows a positive and significant coefficient (i.e. 0.5731). Results from the OLS confirm that customers perceive a lower level of expected interactions and a higher level of expected cleanliness when DT are implemented in restaurants (Esposito et al., 2021).

COVID-19 and innovation in restaurants
The results are in line with previous literature, such as the work of Demaitre (2020), demonstrating that DT represent the only way to cope with CIR and fear of being exposed to the virus. Accordingly, our findings suggest that the use of DT positively impacts reducing customers' risk perception, confirming Shin and Kang's (2020) results in the hotel industry field. Moreover, Models 1, 2 and 3 show positive and strong significant relationships among the DT variable and the independent variables customers' risk perception, expected interactions and expected cleanliness, confirming the following hypotheses: H1a, H2a, H1b, H2b and H3a. These findings are in agreement with Esposito et al. (2021), who conducted a study on the impact of information and communication technologies on EC, EI and safety perception in the restaurant industry during the pandemic. Furthermore, Table 8 highlights that risk perception significantly impacts customers' intention to go to restaurants. These findings are in line with the theoretical framework adopted, according to which the perceived risks impact customers' purchase behaviours (Rahmafitria et al., 2021). In order to test the mediation hypotheses (i.e. H1c, H2c, H4a, H4b and H4c), both the direct and indirect effects were analysed. Table 9 shows the direct and indirect influences of dependent and independent variables. Path A shows that both the relationship DT→EC (a 5 0.7819, p < 0.001) and EC→C_int_to_go (b 5 0.2361, p < 0.001) are significant and positive. The indirect effect of DT on customers' intention to go to restaurants through expected cleanliness is significant (a*b 5 0.2723, bootstrap CI [0.0584, 0.15376]). The direct effect is significant even after controlling for the mediator (c' 5 0.2024, p < 0.001). The total effect of DT is significant and positive (c 5 0.2024, p < 0.001) and is equal to (a*b) þ c' 5 0.2723 þ 0.2024 5 0.4747. In keeping with this explanation, expected cleanliness significantly mediates the effects of DT on customers' intention to go to restaurants.
The same argument should also be provided for paths B, C, D, E and F. More specifically, Table 9 shows that all the direct and indirect effects coefficients are positive and significant. As a result, expected interactions significantly mediate the effects of DT on customers' intention to go to restaurants. Moreover, customers' risk perception significantly mediates the effects of DT on customers' intention to go to restaurants. Lastly, the indirect effects of DT on customers' intention to go to restaurants through customers' risk perception and (1) expected cleanliness and (2)  6. Discussion and implications The analysis results show that risk perception largely influences Italian customers' intention to visit restaurants. This confirms that, within the pandemic and post-pandemic context, restauranteurs have been called upon to adopt strategies and implement measures to manage risk perception (Foroudi et al., 2021;Rahmafitria et al., 2021). In line with our expectations, the regression results show that CRP is influenced by the level of expected cleanliness and the level of expected interactions, confirming previous studies in the hospitality industry (e.g. Nicholls, 2010;Shin and Kang, 2020). However, in order to manage customers' perceptions concerning expected cleanliness and expected interactions, our findings show that the implementation of DT may be considered an efficient tool for this purpose (Demaitre, 2020;Hoque et al., 2020). Moreover, our findings show that the implementation of DT has a significant and positive impact on the level of CRP in the Italian context. More specifically, the mediation analysis results highlight that the use of DT determines a lower level of expected interaction and a lower level of expected cleanliness and leads to a lower level of perceived risk, resulting in higher levels of customers' intention to go to restaurants. These results extend the previous literature on CRP during the COVID-19 pandemic, which did not investigate how to manage risk perception for restaurants' recovery (Byrd et al., 2021;Choe et al., 2021;Hakim et al., 2021).
In addition, these empirical shreds of evidence confirm the PRT and TPB arguments. Accordingly, customers' subjective norms, attitude and behavioural control explain customers' compliance with the use of DT in a context characterised by uncertainty. In this scenario, the impact of DT on perceived risks unveils recovery strategies for the sector. This peculiar relationship between DT and CRP, in turn, stimulates customers in returning to restaurants, overcoming their fear of being exposed to the virus. However, the considerable effect of perceived risk on customers' intention to go to restaurants in the post-pandemic scenario may suggest that full RS recovery could take several years because more customers will be unwilling to go to restaurants due to the fear of contracting the virus. Accordingly, risk perception is likely to be one of the most relevant elements in understanding the customer decision-making process in foodservice research. From this perspective, the implementation of DT may be considered useful to manage risk perception and positively affect customers' decision-making processes.
Consistent with previous studies (Roberts et al., 2020;Villac e-Molinero et al., 2021), CRP increased during the COVID-19 outbreak. Our results show that the pandemic has empowered the relationships between the use of DT and the dependent variables customers' risk perception, expected interactions and expected cleanliness. Accordingly, during the  Esposito et al. (2021). However, our findings highlight that COVID-19 impacts customers' risk perception more than the effects of the other examined variables.
In keeping with these arguments, the use of DT emerges as a beneficial form of service innovation to manage the re-start of the RS (Mele et al., 2020). From a S-D logic perspective (Lusch and Nambisan, 2015), DT create a novel value in terms of safety perception for customers in the RS. This kind of service innovation involves not only customers but also all the actors of the restaurant supply chain. As a result, DT should be considered service infrastructures that allow the RS to have value co-creation processes for restaurant survival (Nambisan, 2013;Vargo and Lusch, 2017). From this perspective, the massive disruption generated by the pandemic has acted as an efficient catalyst for innovation (Heinonen and Strandvik, 2021). Thus, in a context characterised by uncertainties, the disruptive forces of change have to be considered through a strategic lens, with DT as drivers for SI.
This research provides several implications. From a theoretical perspective, this study contributes to developing the integrated framework of PRT and the TPB by highlighting the pivotal role of DT in perceived risks and their impact on customers' decision-making processes. Indeed, as highlighted by the literature review, previous research on customers' risk perception has predominantly focused on other industrial sectors. Therefore, there is still a need to explore how restaurant customers perceive risks and how they impact customers' intention to go to restaurants. Accordingly, this research aims to fill this gap by highlighting the critical impact of risk perception on customers' behavioural intention during the pandemic. Moreover, this study displays that service innovation based on DT can be a strategic vehicle for restaurant risk management to reduce customers' risk perception. Furthermore, this research analysed the perceived risk in both pandemic and post-pandemic scenarios to investigate if COVID-19 has affected the perceived risk mechanism. This article enlarged the context of DT innovation research. While previous research on technology innovation has focused on adopting DT from a managerial or employee perspective, this research attempts to explore the role of innovation in reducing the perceived risks from a customer perspective. In addition, our results also suggest that the restaurant sector needs several years before the full business recovery since customers, in conditions of uncertainty, will be reluctant to go to restaurants. Accordingly, this reflection suggests that safety and health risks could be considered pivotal elements in understanding the customer decisionmaking processes in hospitality studies.
This study also provides practical implications for managers and RS operators. Due to the pandemic, "imposed service innovation" has forced restaurants to adopt measures in order to provide safety for customers. Importantly, DT as tools for service innovation has a pioneering role in the changing process that the foodservice sector will go through in the years after the COVID-19 outbreak. Accordingly, restaurants need to implement DT in their working practices to enhance cleanliness levels and social distancing. Moreover, adopting advanced cleaning systems to disinfect rooms and all high-touch surfaces is fundamental for reducing perceived risks. Restaurants need to demonstrate to their customers, using social media and all their communication channels, that integrating DT can provide a higher level of safety and ensure social distancing. Managers have to reduce interactions among customers and workers using technologies and, at the same time, enhance the cleanliness level through the use of advanced cleaning technologies. Thisin turnwill reduce the perception of health risk since customers have developed a greater sensitivity to social distancing and cleaning measures that persists even after the pandemic period, impacting the choice of restaurants. However, despite the implementation of DT providing several advantages for restauranteurs, these technology-mediated services present also as many trials. Among them, the loss of employment and the increase in operating costs linked to technology investment could be considered the most relevant. TQM 34,7 Notwithstanding these disadvantages, our findings revealed the pivotal role of DT in attracting customers. Accordingly, the implementation of DT in restaurants could be seen as a long-term investment. Moreover, the loss of employment could be managed by reallocating workersfollowing training and education coursesinto new positions for DT implementation and management.

Conclusions
As stated in the previous sections, this research aimed to investigate the supportive role of DT in the foodservice field during the COVID-19 pandemic and how this affected customers' risk perception and consequential decision-making behaviours both in the pandemic and in the post-pandemic period. This study has conducted quantitative analysis on a sample of customers resident in Italy, showing that the implementation of DT leads restaurants to adopt measures that can balance the need to preserve health safety and reduce customers' risk perception to enhance their willingness to go to restaurants.
Despite the multiple benefits of implementing DT in the post-pandemic period, several challenges may be considered. In particular, the adoption of these technologies requires high costs for restaurant owners. However, considering the pivotal role played by DT in attracting customers during the pandemic, these investments are essential for restaurants' survival. Furthermore, informing stakeholders about the risk-reduction strategies implemented is essential to orient customers' decision-making behaviours. Lastly, policymakers might rely on our findings to establish guidelines, standards and regulations to support the RS during the reopening process.
This research is not without limitations, which can represent useful suggestions for further investigations. First, the study is based on a sample of customers resident in Italy. Scholars can implement the same analytical framework in other geographical contexts. Second, the analysis was carried out using an OLS model investigating the linear effect of each variable on the dependent variable. Considering the existing interconnections among the examined dimensions, an analysis based on a structural equation model may explore further relationships among latent variables. Third, the research investigates DT in general without exploring each technology's effectiveness in reducing customers' risk perception. Further research could analyse the specific impact of each innovation tool on risk perception. Forth, the present paper is focused on customers. However, from an S-D logic perspective, all the actors of the "foodservice-scape" have a pivotal role in co-creating value for the system (Polese et al., 2019). Accordingly, scholars may replicate this analysis to investigate the role of DT in reducing the risk perception of other actors in the supply chain. Fifth, this study was conducted at the beginning of the vaccination campaign in Italy. Future research may compare our findings with the results provided at the end of the vaccination journey. Lastly, considering the strategic role of UN Sustainable Development Goals in addressing the worldwide downturn, it would be interesting to explore if DT should be used as tools for the sustainable recovery of the RS (Ciasullo and Troisi, 2013;Ciasullo et al., 2017;Esposito et al., 2020).
About the authors Benedetta Esposito is a PhD Student in Big Data Management in the Department of Management and Innovation Systems at the University of Salerno. Her research is mainly focussed on sustainability and circular economy in the agri-food sector. She is also involved in corporate social responsibility research. Benedetta Esposito is the corresponding author and can be contacted at: besposito@unisa.it Maria Rosaria Sessa is a PhD in Management and Information Technology and a Research Fellow in commodity science at the University of Salerno, where she teaches Quality Management and Control of Tourist Service as a lecturer. Her research is mainly focused on the development of quality and environmental management systems, corporate social responsibility, environmental assessment tools and certification of competence.
Daniela Sica is a PhD in Commodity Science and lecturer in Quality Management and Control of Tourist Service at the University of Salerno. Her research interests are mainly focused on the sustainable production process, Quality Management Systems and corporate social responsibility.
Ornella Malandrino is Full Professor of Commodity Science at Salerno University, where she teaches and conducts research on quality, sustainability and corporate social responsibility. She is the director of the Inter-departmental Observatory for Gender Studies and Equal Opportunities at the University of Salerno.
For instructions on how to order reprints of this article, please visit our website: www.emeraldgrouppublishing.com/licensing/reprints.htm Or contact us for further details: permissions@emeraldinsight.com TQM 34,7