The rise of artificial intelligence : the influence of virtual assistants on brand attachment

(1)

The rise of Artificial Intelligence

The influence of Virtual Assistants on brand attachment

Student: Miruna-Ilinca Vlad

Student number: 11720743

Thesis coordinator: Dr. H. Güngör

MSc BA, Marketing

(2)

Statement of Originality

This document is written by Student Miruna-Ilinca Vlad, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

Virtual Assistants are Artificially Intelligent entities which are programmed to perform certain tasks such as making a reservation or retrieve information in a short amount of time. Virtual Assistants possess human-like traits such as speech capabilities, and are often given a name and a personality, only to make them more likable.

The purpose of this study is to see exactly how much of a disruptive force emerging technologies such as Virtual Assistants can be. More specifically, if the frequency of using Virtual Assistants can impact the feelings of brand attachment consumers perceive (brand attachment is defined as an emotional bond between a consumer and a brand). Furthermore, the study aims at analysing if the performance risk a consumer perceives when using a Virtual Assistant can influence this interaction, or if age has a significant moderating effect. The proposed research question is: what is the effect of the frequency of use of Virtual Assistants

on consumer brand attachment? How is this effect mediated by the perceived performance risk generated by Virtual Assistants? Does age have a moderating role?

Data was collected through an online survey. The main sample of the study consists of 172 respondents who have used Virtual Assistants at least once. The results of the study show a significant positive interaction between the frequency of use of Virtual Assistants and brand attachment. This means that the more often a consumer uses a Virtual Assistant, the more brand attachment will be perceived by that consumer. Furthermore, it is discovered that perceived performance risk also has a significant negative mediating role. An increase in frequency of use determines a decrease in perceived performance risk. Moreover, an increase in perceived performance risk determines a decrease in brand attachment. Finally, the moderating roles of age generations and age in general were both analysed, however, no significant interactions were found.

(4)

1. Introduction

In December 2017, Amazon announced that their more affordable Amazon Echo version: Amazon Echo Dot, became the bestselling product on their website (Wong, 2017). During that same time, another popular press article reports that consumers were asking Amazon‘s Artificially Intelligent Virtual Assistant named Alexa to, apparently, order more Echo Dot Alexa devices (Balakrishnan, 2017).

This recent consumer frenzy for Artificial Intelligence (AI) driven technology such as Virtual Assistants (VAs) is becoming more apparent as consumers are starting to realize that, soon, Artificial Intelligence will be integrated in almost all of our devices, not just personal computers or smartphones (Wakabayashi & Wingfield, 2018). Firms are also shifting their focus toward smart technology, such as Chatbots and Virtual Assistants (Eydman, 2018). For example, automobile companies such as Toyota, Nissan, and Ford are already working on integrating Alexa into their new car models (Mutisi, 2018), while companies such as Google or Amazon aim at integrating their VAs into consumer homes, thus contributing to the growth of the smart homes market (Weinreich, 2017).

Taking into account the relevance of this topic, literature regarding Virtual Assistants is starting to emerge. Some studies look into the implications of VAs in healthcare or the education sector (Vollmer, et al., 2017; Shiban et al., 2015; Lovato & Piper, 2015). Other studies focus on the way consumers evaluate VAs based on their performance, in contrast to the expectations of consumers (Luger & Sellen, 2016). However, literature regarding the marketing implications of consumer-Virtual Assistants interaction is still scarce, with very few studies on this particular topic. This research gap represents the main focus of this thesis.

(7)

2

To be able to measure these marketing implications, a specific dependent variable is taken into account. This variable is represented by brand attachment, which can be defined as a powerful emotional relationship established between a consumer and a brand (Belaid & Temessek Behi, 2011). Brand attachment fits well within the current research question, given the fact that there still is an existing research gap regarding newer ways of enhancing brand attachment (Park, et al., 2010; Wu et al., 2017).

Consumer frequency of use of Virtual Assistants is taken into account as an independent variable. Despite the rapid growth in popularity of voice-operated VAs, studies are emerging regarding the infrequency of use of this new technology (Reisinger, 2017; Cowan, et al., 2017). Based on Cowan et al.‘s study, the frequency of use of VAs has very well-established implications. Their study shows that infrequent users have a harder time trusting their Virtual Assistant, which can arguably influence consumer brand attachment.

Existing Virtual Assistants literature could potentially be enriched by studying the mediating role of perceived performance risk. Perceived performance risk is defined by Bauer (1960) as a certain amount of risk felt by an individual in regard to the functionality of a product. By linking this construct to existing literature, it is concluded that performance risk can be damaging to the attitudes a consumer has toward a product or a brand (Batra, et al., 2012). Since brand attachment is an attitudinal construct itself, it can be assumed that high amounts of perceived performance risk have the ability to lower brand attachment.

Furthermore, the moderating role of age is also studied. Extant literature shows that consumers from contrasting age generations respond very differently to different types of technologies (Wang et al, 2009; Chung et al, 2010).

All information above is compiled into the following research question: what is the

(8)

3

this effect mediated by the perceived performance risk generated by Virtual Assistants? Does age have a moderating role?

This thesis aims to provide both theoretical and managerial contributions. From a theoretical perspective, it can potentially enrich existing literature regarding VAs. For example, Wu et al. (2017) investigate the interaction style effect of „personalized, autonomous, and optimized services provided by smart and connected objects‖ on brand warmth and brand attachment (Wu et al., 2017, p. 61). Their current study is lacking a real-life research setting and real-real-life brand examples. To contrast that, this thesis aims to investigate if a unique variable combination has an impact on brand attachment, while also using real brand examples (Apple, Google, Samsung, Microsoft, and Amazon). Furthermore, most literature on the topic of brand attachment always mentions the need of identifying innovative ways to enhance brand attachment (Wu et al, 2017; Park et al, 2010; Esch, et al., 2006).

From a managerial perspective, creating emotional bonds between brands and consumers is an important step for companies (Malär, et al., 2011). Brand attachment is an important element which helps create brand equity and attitudinal consumer loyalty (Batra, et al., 2012; Vogel et al., 2008). By investigating newer ways of enhancing brand attachment, a large number of companies, especially companies which are looking forward to AI driven technology and markets, can potentially benefit from the insights this thesis aims to uncover.

The thesis is structured according to standard academic research papers. The first part of the thesis is represented by an in-depth literature review. The literature review provides an overview of all the relevant concepts used within this paper. Following this, the methodology and data collection techniques used are described. This section also includes all measurements of the variables used in the analysis. Afterward, all assumptions are analysed and the results

(9)

4

are presented. Conclusions are drawn from the results and their implications are discussed in the ―Conclusions and implications‖ section. The final part of this thesis presents the limitations of the study and how future research could approach the study in order to bring further improvements.

2. Literature review

In order to identify and examine a relevant literature gap, multiple concepts are analysed within this literature review. This chapter explains concepts such as: Artificial Intelligence (AI), Virtual Assistants (VAs), voice-operated VAs, anthropomorphism, anthropomorphism in voice-operated VAs, consumer frequency of use of VAs, brand attachment, and perceived performance risk generated by VAs. Furthermore, the age moderator is explained and linked to the variables of this study, based on extant existing theory. All concepts are presented individually, while also establishing a link between them, in order to provide a coherent literature gap which the research question can potentially fill in.

2.1 Artificial Intelligence

Artificial Intelligence (AI) is defined as a program that has the ability to display cognitive functions similar to humans (Rajani, 2011). In other words, Artificial Intelligence is commonly viewed as a manifestation of human intelligence within machines (Huang & Rust, 2018). In some cases, Artificially Intelligent entities are also referred to as intelligent agents (Russell, et al., 1995).

The marketing implication of Artificial Intelligence is sparking the interest of many scholars and researchers. An example of the current and future marketing implications of AI is presented by Allen (2017). The author takes into account the marketing implications of AI and presents them as tactics applied across customer lifecycles. Allen (2017) believes that AI

(10)

5

has great potential to improve customer satisfaction within a firm. Other examples look into the implementation of AI as a support system for marketing-related decision making (Li & Li, 2010), or research the effect of AI on value proposition delivery for companies (Syam & Sharma, 2018).

A potential explanation for the increasing interest for AI-related literature could be related to the recent popularity of AI within the structures of organizations (Agrawal, et al., 2017). Furthermore, AI is considered to have strong commercial potential (Brooks, 1991). According to Hyken (2017), AI is becoming more common in areas such as customer service support. This is due to the fact that AI assistants are programmed with self-learning capabilities and connectivity to the Internet, making them highly adaptable and desirable to consumers. (Chung, et al., 2009; in Chung, et al., 2016; in Huang & Rust, 2018).

Taking into account the potential of AI, there is no surprise that, instead of interacting with other humans, consumers nowadays are establishing relationships with digital devices more than ever before (Gummerus, et al., 2017). The human-AI interaction is advancing quickly, almost becoming as natural as human-human interaction (Nowak, 2004). A potential explanation for this phenomenon could be that certain AI driven devices are designed to incorporate a number of anthropomorphic features (human-like features) (Fan, et al., 2016). These features positively enhance consumer perceptions and stimulate social interaction (Aggarwal & McGill, 2007). The implications of anthropomorphism are further analysed in an upcoming sub-chapter.

The topic of this thesis focuses on the marketing potential of human and AI interaction. In order to make the project feasible, a certain product which incorporates Artificial Intelligence is considered. This product will be presented in the following sub-chapter: Virtual Assistants.

(11)

6

2.2 Virtual Assistants

A Virtual Assistant (VA) is an Artificially Intelligent entity that can be accessed by a user on a personal computer or a similar device, such as a mobile phone (Sarda, et al., 2017). The same authors mention that VAs can be activated using speech (voice), a keyboard or other peripheral devices. Virtual Assistants evolved from Chatbots, a similar type of AI that is able to carry on a conversation with a user, and adapt its responses based on the information it receives (Chung, et al., 2017).

A Virtual Assistant is able to provide information that a user seeks and perform certain tasks, while still maintaining a socially interactive interface. (Claessen, et al., 2017). In some cases, Virtual Assistants are being associated with personal secretaries (Gruber, 2013). For example, Siri is a Virtual Assistant implemented in iPhones and other Apple devices. Siri is designed to perform certain tasks such as scheduling a meeting or making a reservation (Apple Inc, 2018).

Studies on the topic of Virtual Assistants are starting to emerge. However, these studies mostly focus on the implementation of VAs in healthcare (Vollmer, et al., 2017) or in the education sector (Shiban, et al, 2015). Other VAs research focuses on consumer privacy issues, and how these issues affect consumer evaluations toward VAs (Matsui & Yamada, 2017). Furthermore, VAs studies focus on consumer evaluations toward VAs based on frequency of use (Cowan, et al, 2017) or the age of users (Luger & Sellen, 2016; Lovato & Piper, 2015). Even though the topic of Virtual Assistants is receiving quite some attention in recent literature, the marketing potential of VAs still remains an untapped research domain.

In order to provide a specific area of interest regarding VAs, a certain type of Virtual Assistants has been selected for this thesis: voice-operated VAs. This selection was made because voice-operated VAs are the most common VAs owned and used by consumers

(12)

7

(Reisinger, 2017). Thus, from now on, only VAs which can be operated using the user‘s voice will be referred to. This concept is further discussed in a following sub-chapter.

2.2.1 Voice-operated Virtual Assistants

A voice-operated VA is a certain type of VA that is controlled by a user through voice commands (Sarda, et al, 2017). Most well-known voice-operated VAs, also known as Intelligent Personal Assistants (IPAs), Intelligent Virtual Assistants (IVAs) or Virtual Personal Assistants (VPAs) (Cowan, et al, 2017), are: Apple Siri, Amazon Alexa, Microsoft Cortana, Google Assistant, and Samsung Bixby (Digital Trends, 2017).

Voice-operated Virtual Assistants such as Siri are used by millions of people. In 2017 it is estimated that over 700 million people owned an iPhone with Siri incorporate into (Reisinger, 2017). Their popularity only seems to be growing, with sources claiming that VAs devices such as Alexa or Google Assistant are being purchased in large quantities (Hao, 2018). Predictions for the Virtual Assistants market seem to confirm this. An online report by Grand View Research claims that the VAs market is expected to exceed 12 billion USD by the year 2024, and voice-operated Virtual Assistants appear to be the ones dominating the industry (Grand View Research Inc, 2016). Voice-operated Virtual Assistants are also expected to be soon implemented into new car models (Mutisi, 2018). Mercedes has recently joined forces with IBM to develop the ideal Virtual Assistant for a car (Scheunert & Hildesheim, 2017). Officially, at the beginning of 2018, Mercedes Benz announced that their upcoming A-class car models have their own Virtual Assistants incorporated into them (Davies, 2018).

What can be concluded about voice-operated VAs, and VAs in general, is they are gaining popularity very fast and companies are responding quickly to this trend. This is illustrated by specific examples such as Mercedes Benz, where being able to give commands

(13)

8

to your car by simply saying the activating phrase ―Hey, Mercedes‖ is becoming an important point-of-difference (Stover, 2018).

2.3 Anthropomorphism

A relevant theory that could explain the tendency of humans to view computers or other types of AI systems as entities capable of social interaction is represented by the theory of social response. This theory suggests that humans behave in a social manner only when interacting with technologies that specifically possess human characteristics (Moon, 2000).

The human trait of seeing or attributing human-like features to inanimate objects is called anthropomorphism (Aggarwal & McGill, 2007). The concept of anthropomorphism is developed by Guthrie (1993) in „Faces in the could: A new theory of religion‖, where the author explains how humans attribute human-like features to objects in order to make them easier to understand and connect to on a social level.

From a marketing perspective, it is believed that anthropomorphism has a direct influence on consumers and the way they evaluate products. This idea is presented in the 2007 study by Aggarwal & McGill, and expanded in the 2011 study by the same authors. What can be concluded from these studies is that product or brand anthropomorphism has positive attitudinal implications on consumers. This information is useful for the current research question since it provides further support to the main assumption that using VAs can influence brand attachment.

2.3.1 Anthropomorphism in voice-operated Virtual Assistants

According to Wu et al., (2017), smart products such as VAs can be easily perceived as human-like because of their highly interactive design. Firms that commercialize VAs make use of the positive implications of anthropomorphism (Aggarwal & McGill, 2007) by

(14)

9

designing Virtual Assistants with certain human features such as natural language (Porcheron, et al., 2016) and an interactions style that is friendly-like (Wu, et al., 2017). Furthermore, other human-like features such as names, for example: Alexa or Siri, are implemented in order to activate a human schema and improve consumer product evaluations, similar to Aggarwal and McGill (2007)‘ study.

It can be concluded that anthropomorphism is a prevalent feature of voice-operated Virtual Assistants, a feature that has deeply rooted marketing implications. Furthermore, there is an overall interest regarding the behaviour of consumers toward anthropomorphic products. However, there is an overall lack of literature specifically about Virtual Assistants. Thus, this thesis aims to enrich literature regarding anthropomorphic products, literature about VAs, and brand attachment, by investigating what are some of the marketing implications of Virtual Assistants.

2.4 Consumer brand attachment

Extant literature focuses on the conceptualization of brand equity and brand equity drivers (including brand attachment) (Vogel, et al., 2008). The model for consumer brand equity is presented in the work of Keller (1993), where the author defines brand equity as the response of consumers to a brand, based on the brand knowledge they possess. Brand equity is deemed an important concept for marketers and companies alike, mainly due to its contributions to the competitive advantage of firms and to the improvement of the customer lifetime value (CLV) (Berry, 2000; Rust, et al., 2004).

As mentioned before, brand attachment is an important element of brand equity. Brand attachment can be defined as a bond that is formed between consumers and brands (Park, et al, 2010). Many studies are conducted on the topic of brand attachment, and overall results determine that, even though it is mainly an attitudinal construct, it has the power to influence

(15)

10

the behaviour of consumers in a favourable manner (Japutra, et al., 2014). Extant literature establishes the fact that feelings of attachment toward a brand are very strong and „less vulnerable to disruption‖ (Grisaffe & Nguyen, 2011, p. 1052). In some cases, the feelings of consumers toward brands are so strong that some authors even compare them to feelings of love (Batra, et al., 2012; Loureiro, et al., 2012).

Brand attachment is selected for this thesis as a dependent variable because there is an existing literature gap regarding newer ways of enhancing brand attachment (Park, et al, 2010; Wu et al., 2017). This idea is further supported by a recent Keller article, an article in which the author mentions that ―with the deep penetration and extensive daily usage of smart phones and with the Internet of things looming, understanding how to factor all things digital into marketing and branding is unquestionably a top research priority‖ (Yadav & Pavlou, 2014; in Keller, 2016, p. 11).

2.5 Consumer frequency of use of voice-operated Virtual Assistants

Despite their rapid growth in popularity of Virtual Assistants, only a small number of studies regarding the infrequent consumer use of this new technology are emerging. However, based on what it is available right now, it can be argued that the frequency of use of Virtual Assistants is an important topic that needs further investigation (Shankar, et al., 2003; in Aldas-Manzano, et al., 2011).

Batra et al. (2012) link emotional attachment toward a brand, among an array of possible antecedents, to high usage frequencies. The same authors mention how a long-term relationship with a brand can be correlated with attitudinal and behavioural brand loyalty. This further strengthens the idea that using a VA more frequently can likely influence the level of brand attachment (part of attitudinal brand loyalty) of a consumer. Given the growing

(16)

11

popularity of Virtual Assistants discussed in previous sections, it is expected for many people to frequently use these types of products.

Despite this belief, it seems that there is actually a smaller sample of consumers who actually utilize this technology on a daily basis than expected. This is determined by looking at statistics regarding how often consumers use Virtual Assistants (Creative Strategies Inc, 2016). It is apparent that Virtual Assistant are owned by a large amount of consumers, however, only a small amount of these consumers frequently use VAs. For example, only 30% of iPhone users utilize Siri regularly, despite the fact that 98% of iPhone owners have tried using Siri at least once (Creative Strategies Inc, 2016). Another popular press report also brings up the fact that most people seem to use voice-operated VAs in their car, mainly for convenience reasons (Dunn, 2017). Cowan, et al. (2017) mention that consumers refrain themselves from using voice-operated VAs in public. This idea is confirmed by a recent US-based study, where only 6% of voice-operated VAs owners use VAs in public, while only 1% of owners use them in work environments (Dunn, 2017). Further investigation reveals infrequent VAs users have a hard time accepting this technology, due to them not trusting that the assistant will perform well the tasks it is given (Cowan, et al., 2017). This means that a low frequency of use can lead to negative perceptions of the product, which could potentially transfer to the brand.

As mentioned before, the amount of studies conducted on this particular topic is quite limited, with most of the information stemming from popular press articles and not academic research. Therefore, in order to formulate a proper starting hypothesis, other similar academic studies that utilize frequency of use of different technologies are also a considered a good reference. An example of such a study looks at frequency of Internet use and separates it into three main categories: low, moderate and high Internet usage patterns (Emmanouilides & Hammond, 2000). The authors conclude that, depending on the frequency of Internet use,

(17)

12

consumers not only utilize different types of Internet applications more or less frequent, but have a tendency to utilize different types of applications. This can arguably support that idea that, depending on the frequency with which consumers use VAs, their behaviour is also influenced accordingly. Another example of academic research on this topic is represented by a study about frequency of technology use based on the age of the participants (Olson, et al., 2011). The results of the study show that, in most cases, frequent technology users have an easier time utilizing a diverse array of technologies and have a better understanding of them.

In conclusion, past literature argues that the frequency of use usually leads to less technology aversion. This can arguably shape the attitude of consumers toward a better view on both the product, in this case the Virtual Assistant, and the brand. In most cases, the frequency of use is usually included in conceptual models as a dependent variable, rather than an influencing factor (Aldas-Manzano, et al., 2011). To further contrast past research, the current study looks at frequency of use as an independent variable and its potential impact. Based on above-mentioned information, this variable is believed to have a significant influence on the behaviour and attitude of a consumer. In other words, it is expected that the more frequently a consumer utilizes a VA, the more brand attachment that consumer perceives. Thus, the initial hypothesis of this study is:

H1: The frequency of use of Virtual Assistants has a direct positive relationship with

consumer brand attachment.

2.6 Perceived performance risk

The concept of perceived risk is used in many studies as a tool for predicting and explaining the behaviour of consumers in purchase and use situations (Bauer, 1960). McCorkle (1990) recognizes six dimensions of perceived risk. These six dimensions include: perceived financial risk, perceived performance risk, perceived social risk, perceived time risk,

(18)

13

perceived psychological risk, and perceived physical risk. While the perceived risk model is complex and could yield interesting results, only one dimension is included in the current study: perceived performance risk. The reasoning behind this choice is based on existing literature which states that a positive attitudinal valence toward a brand can have an impact on the feelings of brand attachment of consumers (Batra, et al., 2012). In other words, if a product performs well, it will likely reinforce a positive attitudinal valence toward the brand, thus presumably influencing brand attachment.

Bauer (1960) defines perceived performance risk as the risk a consumer perceives in relation to the functionality of a product. Studies show that certain elements such as price, or risk relievers such as private testing can actually minimize perceived performance risk (Peterson & Wilson, 1985; Mitra, Reiss, & Capella, 1999). However, it is hard to predict the effectiveness of such risk relievers, since their effect depends on the amount of risk involved with a specific product (Cases, 2002). Cowan‘s (2017) qualitative study on frequency of use of VAs supports the idea that infrequent Virtual Assistants users have a harder time trusting VAs because they believe that they will not perform well. Thus, it is hypothesised that the more frequently a consumers uses a Virtual Assistants, the less likely it is for that consumer to perceive high amounts of performance risk. Using a specific Virtual Assistant frequently can help the consumer understand the capabilities and limits of that Virtual Assistant and judge its utility more accurately. All above-mentioned information is compiled into the following hypotheses:

H2: The frequency of use of Virtual Assistants has an indirect positive relationship with

consumer brand attachment. This relationship is mediated by perceived performance risk of

(19)

14

H2a: There is a negative relationship between the frequency of use of Virtual

Assistants and perceived performance risk.

H2b: There is a negative relationship between perceived performance risk and

2.7 The influence of age

Regardless of the capabilities of new Artificially Intelligent entities such as Virtual Assistants, the age of consumers plays an important role in the diffusion and acceptance of these technologies. There are numerous studies that provide evidence regarding this phenomenon (Wang, et al., 2009; Chung, et al., 2010). What can be concluded from these studies is that younger consumers are, in most cases, more comfortable when it comes to using different types of technologies, especially new technologies such as Virtual Assistants.

Each generation of consumers has its own unique and defining features, features that are very well highlighted by Gibson and Sodeman (2014). The authors explain how older generations of consumers such as Baby Boomers (1943-1960) and Generation X (1961-1980) have not been exposed to as much technology as younger generations of consumers, also known as Millennials (1981-1999) and Generation Z (2000-onward). Millennials and Generation Z have experienced technology in almost all aspects of their lives (Gibson & Sodeman, 2014). Being often around technology has its own effects. Extant studies show a significant difference in the way each generation adopts and responds to technology (Chung et al, 2010; Blackburn, 2011). For instance, existing literature provides evidence regarding the fact that older consumers have a harder time adapting to change (Gilly & Zeithaml, 1985), while younger generations are constantly looking forward to it (Emeagwali, 2011). Younger generations embrace advancing technology, and are used to coexisting with technology so much that they even developed new abilities such as multitasking (utilizing multiple

(20)

15

technologies at once) (Emeagwali, 2011; Fromm & Garton, 2013). Regarding the use of Virtual Assistants, there are a few popular press articles and studies which provide a glimpse into the VAs use patterns of different generations of consumers. In a survey-based study by Morning Consult (2017), consumers were asked whether they use different types of Virtual Assistants. The most popular VA from this survey is Apple‘s Siri, with 17% of the consumers (ages between 18-29) admitting to using Siri multiple times a day. This is contrasted with older consumer generations, where only 8% in total (ages 45-65, and 65>) utilize this VA multiple times a day.

It is important to note that, while there are a multiple studies conducted on the topic of age generations and technology adoption, studies that particularly focus on a direct positive relationship between frequency of use of Virtual Assistants and consumer brand attachment moderated by age are not yet present in the academic field. However, based on the current information highlighted in this sub-chapter, it can be deductively hypothesized that younger generations of consumers are more likely to adopt and frequently utilize technologies such as Virtual Assistants, hence the probability of them developing more brand attachment toward a certain brand. All-above mentioned information is compiled into the following hypotheses.

H3: The direct positive relationship between frequency of use of Virtual Assistants and

consumer brand attachment is moderated by age, so that the relationship is stronger for

younger consumer generations (Generation Z and Millennials) than for older generations

(Generation X and Baby Boomers).

H4: The frequency of use of Virtual Assistants has an indirect positive relationship with consumer brand attachment. This relationship is mediated by perceived performance risk

(21)

16

stronger for younger consumer generations (Generation Z and Millennials) than for older

generations (Generation X and Baby Boomers).

2.8 Research question and conceptual model

Even though the topic of Virtual Assistants is receiving attention from scholars and researchers, the marketing potential of Virtual Assistants is still an untapped research area. The idea that VAs have the potential to enhance a customer-based brand equity driver (brand attachment) (Aaker, 1992), is a starting point for this thesis. Brand attachment is believed to have strong implications for consumers and companies, being a good indicator of the overall performance of a firm (Park, et al, 2010; Lassar, et al., 1995; Berry, 2000; Rust et al., 2004). Thus, future research regarding newer ways of enhancing brand attachment is required (Park, et al, 2010; Wu, Chen, & Dou, 2017).

The research question looks into the idea that consumers establish deep relationships with products and brands (Fournier, 1998; Fournier & Alvarez, 2012), but adapts it to more modern products: Virtual Assistants. The main hypothesized interaction is further supported by studies regarding anthropomorphism in products or brands, and the overall positive effect anthropomorphism has on consumers (Aggarwal & McGill, 2007; Aggarwal & McGill, 2011).

Taking into account the strong implications of human-computer interaction and existing literature proof in that domain, it can be concluded that consumers often establish relationships with anthropomorphic products such as Virtual Assistants. However, literature that investigates the marketing implications of this phenomenon is lacking, especially an empirical investigation regarding the possible correlation between the frequency of use of VAs and consumer brand attachment. The formulated research question is: what is the effect of the frequency of use of Virtual Assistants on consumer brand attachment? How is this

(22)

17

effect mediated by the perceived performance risk generated by Virtual Assistants? Does age have a moderating role?

In order to paint a clear picture regarding the scope of this empirical study, a conceptual model figure is required. Figure 1 contains all expected variable interactions as outlined in the hypotheses.

Figure 1: Conceptual Model

H1: The frequency of use of Virtual Assistants has a direct positive relationship with

H2: The frequency of use of Virtual Assistants has an indirect positive relationship with

consumer brand attachment. This relationship is mediated by perceived performance risk of

Virtual Assistants.

H2a: There is a negative relationship between the frequency of use of Virtual

(23)

18

H2b: There is a negative relationship between perceived performance risk and

H3: The direct positive relationship between the frequency of use of Virtual Assistants and

consumer brand attachment is moderated by age, so that the relationship is stronger for

younger consumer generations (Generation Z and Millennials) than for older generations

H4: The frequency of use of Virtual Assistants has an indirect positive relationship with consumer brand attachment. This relationship is mediated by perceived performance risk of

Virtual Assistants. This relationship is moderated by age, so that the relationship is stronger

for younger consumer generations (Generation Z and Millennials) than for older generations

3. Methodology

3.1 Survey Design

The study is designed as a mono-method, correlational research. More about the limitations of the design of the study is discussed in the chapter ―Limitations and future research‖. To deductively answer the research question, it is required to test the causal relationship between the frequency of use of Virtual Assistants (independent variable) and consumer brand attachment (dependent variable). This is done using primary data collected from a survey in a form of an online questionnaire. The study is of cross-sectional nature, meaning that it only addresses a particular relationship between selected variables in a certain point-in-time. This particular research design has a lower internal validity than its counterpart (longitudinal research). More about this limitation is discussed in the ―Limitations and future research‖ section.

(24)

19

Qualtrics.com is used as a platform for creating the questionnaire. A pre-test is administered to see if the questionnaire has any mistakes and is executed properly. This pre-test is conducted by sending out the questionnaire to a few individuals willing to provide feedback. Once finalised, the questionnaire is sent out to respondents using an anonymous link in order to maintain any personal information as private as possible, and to ensure respondents that their identity is protected. Once collected, the data is analysed using the IBM statistical software SPSS.

3.2 Sample

Due to the fact that a non-probability sample provides an easily available sample of respondents, it is implemented as a sampling technique for this study. The sample type selected for this study is a convenience sample. Despite the high availability of this sample, it must be noted that a convenience sample has deeply rooted limitations such as low external validity. These limitations are further addressed in the ―Limitations and future research‖ section. Overall, the sample consists of 223 respondents. However, after deleting missing data entries, the number of respondents dropped to 203. From the 203 respondents, 109 (53.6%) are male respondents, 94 (46.4%) respondents are females. No respondents selected any other gender. For this particular study, having a high age variance is required in order to accurately test the moderating role of age. Thus, respondents of multiple age groups were encouraged to participate. Even so, finding an almost equal amount of respondents from each age generation is not an easy task given the sampling method. Age-wise, the final sample consists of: 7 Generation Z participants (3.5%), 138 Millennials (68%), 49 Generation X participants (24.1%), and 9 Baby Boomers (4.4%). In terms of education level, 42 (20.8%) of respondents indicate High school or less as their education level, 98 (48.4%) have a University Bachelor Degree, 54 have a University Master‘s Degree (26.3%), and 9 (4.5%) have a PhD/Doctorate

(25)

20

Degree. In terms of nationality, 21 (10.4%) respondents are Dutch, 145 (71.4%) are Romanian, and 37 (18.2%) have other nationalities. The questionnaire contains a filter question which separates respondents who have used a Virtual Assistant at least once from those who have never used a Virtual Assistant. Only respondents that have used VAs at least once are used in the main analysis. The final sample size of respondents who have used VAs is 172 (84.7%), and the sample size of respondents who have never used a VA is 31 (15.3%).

3.3 Procedure

Given that the sample used is a convenience sample, participants are approached either verbally or on social media platforms in order to participate in the study. An anonymous link is provided to all participants. When the link is accessed, participants are able to see a message which states the purpose of the study, assures participants that their information will remain completely private, provides contact information for any questions or concerns, and provides the approximate time for completing the questionnaire (5-7 minutes).

After the general introduction of the study, the level of awareness of Virtual Assistant is measured using the following item: ―Have you ever heard of Virtual Assistants?‖. From the overall sample of 203 respondents, 15 (7.3%) respondents indicated that are not at all aware of VAs, 23 (11.3%) are slightly aware, 29 (14.3%) are moderately aware, 48 (23.8%) are very aware, and 88 (43.3%) are extremely aware of VAs. Follow-up questions measure the level of awareness of the 5 most popular Virtual Assistants used in this study: Siri, Bixby, Google Assistant, Alexa, and Cortana. Respondents are also allowed to add any other Virtual Assistant suggestion in a blank space on the sixth item ―Other‖. However, no other Virtual Assistant suggestion was added by the respondents. Figures 2.1 and 2.2 provide an overview of the level of awareness of Virtual Assistants perceived by the 203 respondents. As expected, most respondents are very aware or extremely aware of popular Virtual Assistants such as

(26)

21

Siri, Google Assistant, and Alexa. Furthermore, not many respondents are aware of other Virtual Assistants, other than the ones used in the current study.

Figure 2.1: Level of awareness of Virtual Assistants (Siri, Bixby, Google Assistant)

Figure 2.2: Level of awareness of Virtual Assistants (Alexa, Cortana, Other)

In order to be able to access all the questions from the questionnaire, respondents are required to indicate that they have used Virtual Assistants in the past. If not, they are

27 14 16 31 115 116 21 21 13 32 60 30 ₂₅ 34 54

Not at all aware Slightly aware Somewhat

aware Moderatelyaware Extremelyaware

Siri Bixby Google Assistant Level of Awareness 79 18 ₁₆ 26 64 104 17 21 20 41 191 6 4 ₀ 2

Not at all aware Slightly aware Somewhat aware Moderately aware Extremely aware Alexa Cortana Other Level of Awareness

(27)

22

redirected toward the end of the survey, where they are able to indicate their age, gender, country of origin, and level of education.

The respondents who have used VAs in the past are then asked which Virtual Assistant they prefer the most (―From the following list, please indicate the Virtual Assistant

you prefer the most‖). Figure 3 provides an overview of the answers of 172 respondents. As

expected, Siri (Apple) is the most preferred Virtual Assistants due to the overall popularity of Apple devices such as iPhones and iPads. Surprisingly, the second most popular Virtual Assistants is Google Assistant (Google), not Alexa (Amazon). The reason for this could be that the sample of the study is mainly European-based. Alexa is not yet completely functional in most European countries due to it being region locked.

Figure 3: Level of preference of Virtual Assistants

Furthermore, respondents are asked to indicate how frequently they use their favourite VA. Based on the answers of 172 respondents, the following frequencies of use can be observed: very low frequency of use (11 respondents- 6.4%), low frequency of use (54 respondents- 31.4%), medium frequency of use (61 respondents- 35.5%), high frequency of use (27 respondents- 15.7%), and very high frequency of use (19 respondents- 11%). As expected from the literature review section, most respondents tend to have low or medium

90 14 37 20 9 2 Siri Bixby Google

Assistant

Alexa Cortana Other

Level of Preference of Virtual Assistants

(28)

23

frequencies of use of VAs. However, the number of respondents with high or very high frequencies of use is also quite large.

The final questions from the questionnaire respondents are asked to answer measure constructs such as perceived performance risk and brand attachment. For these constructs, previously validated scale items are used. The measurements of all constructs and control variables are discussed in detail in the upcoming sub-chapter ―Measures‖. Finally, after answering all questions, participants are thanked for their participation. For an overview of the questionnaire, please see Appendix 1.

3.4 Measures

3.4.1 Frequency of use

The frequency of use of VAs is measured using a 5-point Likert scale („How often do you use

the Virtual Assistant you prefer the most?‖) with answers ranging from „Never‖ to „A great

deal‖. This way, five different frequency levels can be measured: very low, low, medium, high, and very high.

3.4.2 Consumer brand attachment

The brand attachment of consumers is measured using the brand attachment scale from Park et al.‘s article (Park, et al, 2010). In their study, the authors mention a reliability coeficient (Cronbach‘s α) of 0.95. Answers are anchored by a 5-point Likert scale, ranging from „Not at all‖ to „Completely‖. The construct developed by Park et al. (2010) is considered to be a reliable and in-depth scale, often used or referenced by other authors in the field (Batra et al., 2012; Malär, et al., 2011; Stokburger-Sauer, et al., 2012). A self-conducted reliability check for this scale reveals a reliability coefficient of 0.941, a value very close to the one from the original paper. At a closer investigation, all 10 individual scale items also appear to have high

(29)

24

reliability, with all Cronbach‘s α values over 0.30. It can be safe to assume that all 10 items measure brand attachment sufficiently. Thus, all 10 items are kept for further analyses.

3.4.3 Perceived performance risk

This construct is measured using a 3-item scale adapted from Grewal, Gotlieb, and Marmorstein (1994). The proposed scale by Grewal et al. (1994) is also an adaptation of Shimp and Bearden‘s scale (1982). It is a 5-point Likert scale anchored by answers which range from ―Strongly Disagree‖ to ―Strongly Agree‖. In the original paper, Shimp and Bearden (1982) mention a reliability coefficient (Cronbach‘s α) for this construct of 0.90. After self-conducting the reliability check for this scale, the analysis reveals a reliability coefficient of 0.849. This value is close to the original Cronbach‘s α from the validation paper. It has a value higher than 0.70 which means that the scale measures the construct in a reliable and consistent manner. A reliability check for each individual scale item reveals that each item of the scale is individually reliable, with values exceeding 0.30. Thus, all three items are kept and used in the final analysis.

This scale has counter-indicative items (items have reversed values). This means that, if a respondents answers with ―Strongly agree‖, this actually indicates a low perceived performance risk. Thus, before conducting the analysis, the values of this scale are recoded to match their reversed state.

3.4.4 Age

Age is measured by instructing respondents to fill a blank space with their age in number. This way any pre-recorded answers (age intervals), which could potentially be confusing for respondents or contain overlapping answers, are avoided. Age is measured using the following item: ―Please indicate your age (in numbers)‖. Age generations are determined as

(30)

25

such: respondents born 2000 onward are representatives of Generation Z (18 years old); born between 1981-1999 are Millennials (ages 19-37), born between 1961-1980 are Generation X respondents (ages 38-57); born between 1960-1943 are representatives of the Baby Boomers age generation (ages 58-75).

3.4.5 Control variables

Multiple control variables are included in the study. In an extant study by Park et al. (2010) on consumer brand attachment, control variables such as gender and relationship length are utilized. Their research also includes a third control variable represented by past purchase behaviour, however, given the cross-sectional nature of the current study, this control variable is excluded. Aside from gender and relationship length, other control variables are introduced. These include education level and nationality. Education level is measured by asking respondents to indicate their current level of education (High school or less =1; Bachelor degree =2; Master‘s degree =3; Doctorate/PhD =4). Nationality is measured by asking respondents to indicate their country of origin. Given the fact that the sample used is a convenience sample, it is expected that most participants are either from The Netherlands or Romania. Thus, the answer choices are labelled as such: The Netherlands (=1), Romania (=2), Other (=3).

3.4.5.1 Gender

According to a previous study on the topic of consumer brand attachment, it is important to control for the ―positive effect of gender on brand choice‖ (Meyers-Levy & Sternthal 1991, in Park et al, 2010, p. 12). Items are labelled using values of 1 for male respondents, 2 for female respondents, and 3 for ―other‖.

(31)

26

3.4.5.2 Relationship length

Park et al. (2010) mentions relationship length as a valuable control variable when measuring consumer brand attachment. It is used to control for consumers feeling brand attachment toward the VA brand, without actually being influenced by frequency of use of their VA or the perceived performance risk generated by VAs. The Virtual Assistants and brand relationship lengths are measured using the following items administered in the questionnaire: ―Please indicate the approximate number of years since you have first used your most

preferred Virtual Assistant‖, and ―Please indicate the approximate number of years you have

known the brand of your most preferred Virtual Assistant‖. The above mentioned control

variables measures are adapted from a past paper which measures the relationship length impact on perceived trustworthiness (Levin, et al., 2006).

3.4.6 Careless response index

When dealing with survey responses, it is important to control for careless responses. This ensures better quality data, hence more accurate results (Meade & Craig, 2012). Careless response is measured using the following item: ―Which Virtual Assistant did you prefer the

most?. This question is related to an early question in the questionnaire: ―Please indicate the

Virtual Assistant you prefer the most‖. After a quick scan through the data, it is concluded

that no respondent chose a different answer to the careless response question, compared to the original question. This somewhat ensures that respondents paid attention to the questions when completing the questionnaire.

(32)

27

4. Results

4.1 Reliability check

Before conducting the main analysis, a reliability test is required. Reliability is a term which expresses the capabilty of a measurement scale of providing consistent results (Sounders, et al., 2009). A reliability check is conducted for the following, more complex, constructs: brand attachment and perceived performance risk. Table 1 shows an overview of the Cronbach‘s α of each scale. All values are above 0.7, which indicates that the scales have high reliability. The reliability coefficients of each item of the two scales are also analyzed, and all values are above 0.3. Thus, all items are included in the final analysis. The high Cronbach‘s α values are somewhat expected due to the fact that both constructs are measured using adaptations of previously validated scales.

Table 1: Cronbach’s α

Variable Cronbach‘s α Brand Attachment 0.941 Perceived Performance Risk 0.849

4.2 Exploratory factor analysis

An exploratory factor analysis is performed in order to determine the total variance of the two main constructs used in the analysis. It is also used to determine if any of the items which measure the two constructs bring a sufficient contribution, or load more onto other constructs. A principal axis factoring analysis is conducted on the 2 scales: perceived performance risk and brand attachment. Based on the Barlett‘s test of sphericity, a value of 1582.424 can be observed, and a p value of 0.000 (<0.001). This indicates that there is a significantly large

(33)

28

correlation between items. Thus, the factors are rotated with a direct Oblim (or oblique) rotation. This function allows correlation between factors in order to provide more accurate results. Furthermore, the Kaiser-Meyer-Olkin (KMO) test is conducted. KMO is a technique used to verify if the sample size is sufficient for the analysis. In this case, the KMO value is 0.926, which is higher than |.60|. This indicates that the sample size (n=172) of consumers is, indeed, high enough to provide clear results.

Looking at the eigenvalues that have a value over 1 (based on Kaiser‘s criterion of 1), two components can be observed. These components combined explain around 69% of the total variance. This is confirmed by the scree plot, where a cut-off point can be observed close to factor 2.

Figure 4: Factor analysis scree plot

Furthermore, looking at the rotated factor loadings from Table 2, it seems that all items have a sufficient measuring contribution. The items of each factor do not overlap, since each rotated factor loading value is significantly different than its counterpart. Thus, all items

(34)

29

of the two factors are kept for further analyses. The means of all the items are computed using SPSS compute means function, thus creating two final constructs: perceived performance risk total and brand attachment total.

Table 2: Rotated factor loadings

Item Rotated Factor Loadings

Brand attachment

Perceived performance

risk

How confident are you that your favourite Virtual Assistant will always perform well? (R)

-.127 .845

How certain are you that your favourite Virtual Assistant will work satisfactorily? (R)

-.015 .876

Do you feel that your favourite Virtual Assistant will always perform the functions that are described in advertisements? (R)

.057 .864

To what extent is (favourite Virtual Assistant brand) part of you

and who you are? .607

-.242 To what extent do you feel personally connected to (favourite

VA brand)? .856 .012

To what extent do you feel emotionally bonded to (favourite VA

brand)? .810 -.009

To what extent is (favourite VA brand) part of you? .756 -.115

To what extent does (favourite VA brand) say something to other

people about who you are? .834

-.007 To what extent are your thoughts and feelings toward (favourite

VA brand) often automatic, coming to mind seemingly on their own?

.788 -.081

To what extent do your thoughts and feelings toward (favourite VA brand) come to your mind naturally and instantly? .762

-.001 To what extent do your thoughts and feelings toward (favourite

VA brand) come to mind so naturally and instantly that you don't have much control over them?

.880 .165

To what extent does the word (favourite VA brand name) automatically evoke good thoughts about the past, present, and future?

.803 .017

To what extent do you have many thoughts about (favourite VA

brand)? .893

.026

Eigenvalues 7.352 1.618

(35)

30

4.3 Control Variables

When testing relationships between variables, it is important to control for other factors that could have a significant effect on the final results. The control variables used in this analysis include: gender, education level, country of origin (nationality), Virtual Assistant relationship length and Virtual Assistant brand relationship length. After a quick scan through the results, it can be concluded that the only control variable that seems to have a significant effect is gender. The impact of gender is further discussed in sub-chapter 4.6 „Control variable: gender‖.

4.4 Means, standard deviations and correlations

Looking at the correlation coefficients represented in Table 3, the following correlations can be observed. It seems that frequency of use of Virtual Assistants and brand attachment are positively correlated (Pearson coefficient= 0.539). Frequency of use of VAs is also correlated with perceived performance risk. As expected, their correlation is negative (Pearson correlation coefficient= -0.610). Perceived performance risk is also negatively correlated with brand attachment (Pearson correlation coefficient= -0.497). Finally, age appears to be positively correlated with perceived performance risk (Pearson coefficient= 0.219). However, this correlation is smaller than the previously mentioned correlations. Moreover, a correlation at a 5% level of significance can be observed between gender and perceived performance risk (Pearson coefficient= -0.162). In other words, there is reason to expect that women perceive less performance risk than men.

(36)

31

Table 3: Means, Standard Deviations, Correlations

Means, Standard Deviations, Correlations

Variables M SD 1 2 3 4 5 1. Frequency of use 2.94 1.08 1

2. Perceived performance risk 2.75 0.89 -0.610** 1

3. Brand attachment 2.15 0.89 0.539** -0.497** 1

4. Age 2.26 0.60 -0.002 0.219** 0.039 1 5. Gender 1.45 0.49 0.064 -0.162* -0.058 -0.11 1

**. Correlation is significant at the 0.01 level (2-tailed) *. Correlation is significant at 0.05 level (2-tailed)

4.5 Hypotheses testing

After conducting the reliability check and exploratory factor analysis, it can be concluded that the scales used to measure the constructs are reliable and their items do not overlap. Furthermore, the sample size is sufficiently large for the conceptual model analysed.

For this particular analysis, an Ordinary Least Squares (OLS) regression is used to test the hypothesized interactions between variables. OLS is a linear regression model commonly used in fields such as economics, social sciences, and psychology. It is used to test the relationship between a dependent variable and one independent variable (simple linear regression), or multiple independent variables (multiple linear regression). What OLS does is it minimizes the sums of the encountered in an existing data set. When using an OLS regression, the error term is expected to not be correlated with the dependent variables in order to avoid any bias results.

Due to the fact that the conceptual model incorporates a moderated mediation interaction, the best way to approach the analysis is to use a specific SPSS add-on developed

(37)

32

for this particular reason. The expected interactions from the conceptual model are analysed using the SPSS add-on called PROCESS. PROCESS is an SPSS extension developed by Andrew F. Hayes and The Guilford Press, typically used in moderated mediation regression analysis (Hayes, 2017). The PROCESS add-on organises each variable interaction into a path, and tests using OLS regressions if these interactions are statistically significant or not.

PROCESS Model 4 and PROCESS Model 8 are used in the analysis. The overview of the models can be seen in figure 5 (simple mediation) and figure 6 (moderated mediation).

Figure 5: PROCESS Model 4

Figure 6: PROCESS Model 8

4.5.1 Direct effect (X on Y=c1’)

Hypothesis 1 states that the frequency of use of Virtual Assistants has a direct positive relationship with consumer brand attachment. In order to test this hypothesis, PROCESS Model 4 is used. At a confidence interval of 95%, R² has a value of 0.2991. This means that the frequency of use of VAs quantifies 29% of the total variance of brand attachment

(38)

33

explained by the overall model. In other words, the frequency of use explains 29% of the variance of brand attachment. It is also statistically significant based on the p value of 0.000 (p < 0.01). The c1‘ direct path shows that two respondents who differ by one unit on frequency of use of VAs and who experience the same level of perceived performance risk, are estimated to differ by 0.45 units in brand attachment (are with 0.45 units more attached to the VA brand). Based on the above-mentioned, hypothesis 1 is supported.

4.5.2 Indirect effect (X on M=a1)

Two respondents who differ by one unit on frequency of use of VAs are estimated to differ by a= -0.49 units on perceived performance risk. The sign of a1 is negative, meaning that those relatively higher in frequency of use are estimated to perceive lower levels (-0.49 units) of performance risk. At a confidence interval of 95%, R² has a value of 0.3878. The p value is =0.000 (< 0.01), which determines a statistically significant relationship. Hypothesis 2a states that there is an expected negative relationship between consumer frequency of use of VAs and perceived performance risk. Therefore, based on the results, hypothesis 2a is supported.

4.5.3 Indirect effect (M on Y=b1)

Two respondents who experience the same level of frequency of use of VAs (in other words, who use VAs equally as much) but who differ by one unit in their level of perceived performance risk are estimated to differ by b= -0.29 units in brand attachment. The negative impact of perceived performance risk on brand attachment is lower than the impact of the frequency of use of VAs on perceived performance risk. Even so, based on the p value = 0.003, which is lower than 0.01, it is determined that there is statistical significance. The sign of b1 is negative, meaning that those relatively higher in perceived performance risk are estimated to perceive less brand attachment. Moreover, R² has a value of 0.3518. Hypothesis

(39)

34

2b states that there is an expected negative relationship between perceived performance risk and consumer brand attachment. Thus, based on the results, hypothesis 2b is supported.

4.5.4 Total direct effect (X on Y =c1)

As mentioned previously in sub-chapter 4.5.1 ―Direct effect‖, there is a strong reason to support hypothesis 1. By looking at the total direct effect, it can be interpreted that two respondents who differ by one unit in frequency of use of VAs are estimated to differ by 0.45 units in their brand attachment. This effect is in line with the effect of path c1‘. This provides further reasoning to support hypothesis 1.

4.5.5 Total indirect effect (paths a1 and b1)

The results of this path (a1b1) can be interpreted as such: two respondents who differ by one unit in their frequency of use of VAs, are estimated to differ by 0.14 units in their brand attachment, as a result of them feeling less perceived performance risk. This interpretation is in line with the hypotheses stated by the model and supporting literature. Hypothesis 2 states that the frequency of use of VAs has an indirect positive relationship with consumer brand attachment. This relationship is mediated by the perceived performance risk generated by Virtual Assistants. This hypothesis is in line with the results yielded by the PROCESS 4 OLS regression analysis. Based on this information, hypothesis 2 is supported.

For further support of hypothesis 2, bootstrap values can also be analysed. Bootstrapping is a process which runs the analysis 5000 times at a confidence interval level of 95% in order to control for high result variances. The bootstrap values BootLLCI = 0.0870 (lower interval value) and BootULCI = 0.2463 (upper interval value) are both positive, meaning that there is an overall positive effect. The fact that both intervals have the same sign (positive) provides further evidence that the total indirect effect is significant.

(40)

35

4.6 Control variables: gender

In the case of the direct effect (X on Y=c1‘), gender has a p value of 0.1519, which means that it does not have a significant effect. This means that gender does not significantly affect the relationship between frequency of use and brand attachment. However, in the case of both indirect effect paths (a1 and b1), gender appears to have a significant effect (p value= 0.0421 and 0.0419 < 0.05).

Based on the results, it seems that women are more inclined to perceive -0.22 units of performance risk than men. This effect is interesting because it is contrasted by extant literature regarding the effect of gender on perceived risk generated by products or services. Garbarino et al. (2004)‘s study suggests that women perceive substantially more performance risk than men. The authors attribute this to the fact that women tend to be more fearful of failure and, as a result, approach an uncertain situation with more precaution than men. While the implication of gender is not the main subject of this thesis, further elaboration on this topic could be an interesting subject for future research.

4.7 Moderated mediation

Using PROCESS Model 8 the moderating role of age generation is analysed. Based on extant literature on this topic, it is expected the relationship between variables is much stronger for younger age generations, than for older age generations.

4.7.1 Moderated mediation direct effect (X on Y, moderated by W)

Hypothesis 3 states that the direct positive relationship between frequency of use of Virtual Assistants and consumer brand attachment is moderated by age, so that the relationship is stronger for younger consumer generations (Generation Z and Millennials) than for older generations (Generation X and Baby Boomers). Looking at the results, there is reason to

The rise of artificial intelligence : the influence of virtual assistants on brand attachment