What makes or breaks the success of the chatbot function in an age of online

(1)

What makes or breaks the success of the chatbot function in an age of online

customer service delivery

Name: Victor Rieff

Student number: 10670955

Contact: victor.rieff@student.uva.nl

Faculty: Economics and Business, University Of Amsterdam Education: MSc Business Administration, Marketing track Supervisor: Dhr. Dr. H. Güngör

(2)

Statement of Originality

This document is written by Student Victor Rieff (10670955) who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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)

Chatbots are an emerging phenomenon in online customer service delivery processes, but academic knowledge about the implementation of these technologies is little or non-existent. To provide more knowledge about this subject this research focuses on the effect of the use of chatbots on the online customer service experience. It is hypothesized that chatbots have a positive impact on the online customer service experience, and this relationship is strengthened by technical savviness, trust in technology, perceived potential benefit and the ease of use. While it is expected that these effects are weaker for older consumers. The hypotheses are tested with a dataset consisting of 210 consumers. It was found that using chatbots positively influences the online customer service experience, but this is negatively moderated by the perceived ease of use. It is suggested to search for a deeper understanding about the influence of chatbots in customer service delivery processes.

(5)

Chapter One: Introduction

A revolutionary change in customer service delivery is taking place due to the introduction of chatbots in service delivery processes. Chatbot technologies are increasingly being implemented in different types of customer service delivery systems (Devlin, 2017; Hill, Ford, & Farreras, 2015; Sablich, 2017). Nowadays it is possible to communicate with a chatbot to plan a holiday, to get health advice, or just for some small talk (Devlin, 2017; Sablich, 2017). These communication technologies are rapidly improving in quality. In a chatbot test in 2015 people were simplifying their own messages to make them easier to interpret for the chatbot system (Hill et al., 2015). While two years later chatbot technologies are self-learning and able to recognize and empathize with human emotions (Devlin, 2017). Due to these influences of artificial intelligence, communicating with chatbots feels more like human to human interactions (Devlin, 2017).

Since these technologies are fairly new and developing in high speed, there is a need for more academic literature about chatbots (Sablich, 2017). This research will contribute to the academic literature in this field by focussing on the use of chatbots in online customer service delivery. This will also lead to managerial implications which can be used in practice.

Making chatbots can have functional benefits for consumers, for example increasing the responsiveness of the firm and speed of the service delivery (Klaus, 2013; Sablich, 2017). On the one hand 50% of consumers tend to have some sort of anxiety for the use of computer technologies (Wilfong, 2006). While the use of new technologies can also be influenced by how experienced and capable an individual is with the use of technologies (Soares, Zhang, Proença, & Kandampully, 2017). The level of trust in technology can also play a role. This could be driven by the fact that customers can experience chatbots as less personal since there

(6)

is no human interaction (Hill et al., 2015). But on the other hand the people might be more open since they are not interacting with another human being and therefore feel more anonymous (Hill et al., 2015). Another explaining factor could be that some individuals perceive these technologies as more useful, while others think chatbots do not add any value to their service experiences (Johnson & Payne, 1985). Or chatbots can be perceived as hard to interact with and therefore consumers might avoid using them (Mahar, Henderson, & Deane, 1997). This can make it hard to successfully introduce new computer driven technologies to a broad audience. Therefore it is needed to gather more information about what factors can stimulate the success of chatbots in customer service delivery processes.

It could also be the fact that there are differences among consumer groups. These artificial intelligence techniques might be more effective to younger consumers. A study by ORC international shows that 47% of the millennials have heard about the term chatbot, where this percentage is 22% for older generations (Tomasco, 2016). So it could be the fact that because younger generations know more about these techniques, they also have a more positive attitude about using them. Where the older generations might prefer the human interaction way of customer service delivery (Leask, Fyall, & Barron, 2014). Therefore this research will also take a look on the influence of age. These subjects will be analyzed driven by the main research question: _{What is the effect of the use of chatbots on the customer} experience in online customer service situations?

This question will be answered by doing research on seven sub-questions which will be described in more detail later in this research. The used data is collected through an online customer survey and is analyzed using the statistical program SPSS and PROCESS.

(7)

Structure

The second chapter of this research will contain the literature review and theoretical framework. This chapter will provide an overview of the theory in this research field. This information will be used as a foundation for the theoretical framework that will guide this research. This chapter will provide explanations about different possible factors that influence the success of chatbots in customer service delivery. In the third chapter the used research and data collection methods are described. Here the research design, used measurements, research procedure, and the analysis methods will be elaborated. In the fourth chapter the results of this research can be found. This section includes the reliabilities of the measurement scales, the correlations between the used variables, and the results of the hypothesis tests. The fifth chapter contains the discussion and conclusion of this research. This chapter includes a short summary of the results section followed by the unpredicted results and alternative explanations for this research. Subsequently the limitations and future research suggestions can be found whereupon the theoretical and managerial contributions are described. This study is wrapped up with a general conclusion at the end of chapter five.

Chapter Two: Literature review and theoretical framework

2.1 Literature Review

Customer service

This research will focus on the customer service industry because the marketing field is shifting from a good-based economy to a more service-based economy (Rust & Huang, 2014). Which means that customer service is getting more important. The proportion of the added value determined by customer service is growing (Buera, Kaboski, Joseph, 2012).

(8)

Besides, customer service is also changing due to technological influences. The technologies used in customer service delivery are getting smarter and are used more often in practice (Rust & Huang, 2014). These new technologies provide new ways of communication, for example through social media or online chats (Hill, Ford, & Farreras, 2015). This means that the customer service delivery process is shifting from a face to face service, to an online process. According to Klaus (2013) the functional and psychological factors are determining the success of this shift. Where the functional aspects can be described as the usability and the quality of the technical attributes. And the psychological factors are attitudinal and based on the customers perception of trust and value for money (Klaus, 2013).

These theories about online customer service delivery is especially important for e-commerce websites which are not only focussing on selling products online, but also trying to become more interactive in terms of service delivery (Hill et al., 2015; Klaus, 2013). But to make this shift successful they have to be aware about the differences between traditional and online service delivery. Online customer service delivery can to some extent be compared to traditional customer service delivery (Parasuraman, Zeithaml, Malhotra, 2005). Most importantly for the the level of customer service is the quality and amount of the facilities provided to the consumer (Parasuraman et al., 2005). This implies that when a website only offers the opportunity to email for customer service, it will generally have a lower perceived service quality (Gefen, 2002). Generally e-mails do not generate a quick response, while the responsiveness is also an important aspect of online customer service quality (Klaus, 2013). Besides, the design and appearance of the online environment are also important factors driving customer service quality (Gefen, 2002). The design and appearance can influence the perceived trustworthiness in terms of reliability and quality assurance (Parasuraman et al., 2005). Furthermore, the customer wants to feel important and wants to be taken seriously and

(9)

get personal attention (Gefen, 2002). Firms should take these things into account when they try to improve their online customer service quality. This can result in higher customer satisfaction and loyalty (Dixon, Freeman, & Toman, 2010). But there are more benefits that firms can achieve by innovating their online customer service delivery processes.

Another advantage of these innovations lies at the ease of collecting big amounts of customer data, also referred to as big data (Rust & Huang, 2014). Information Technology (IT) is needed to process this customer data. In the service industry this is mainly done through IT-enabled Customer Relationship Management (CRM) systems, which systematically collects customer data and links the front and back office of a company (Chen & Popovich, 2003; Piccoli, Lui, Grün, 2016). This data can be used in the back office to find meaningful customer insights which can be used in practice at the front office (Chen & Popovich, 2003). These insights can be used to personalize the customer service, and in this way ultimately provide higher service quality (Rust & Huang, 2014). CRM is closely related to the customer experience management (Chen & Popovich, 2003; Homburg, Jozic, & Kuehnl, 2017).

Customer Experience Management

Customer Experience Management (CEM) tries to influence the customers experience on every touchpoint from pre-purchase, to purchase, to post-purchase (Homburg et al., 2017). Customer experience is the most important factor driving the customer service quality evaluations (Klaus, 2013). Therefore it is important for firms to manage the experiences of the customers.

Within the changing marketing landscape CEM is becoming more influential in managing (Homburg et al., 2017). Due to innovating technologies there are more touchpoints

(10)

available for customers, which makes it harder for firms to consistently control the customer experiences across all touchpoints (Piccoli et al., 2017). Especially since not all touchpoints are controlled by the brand itself, for example social media. Customer experience management is growing in importance since 89% of the firms expect that firms are primarily going to compete on CEM and CRM, where in 2010 this was believed by 36% of the firms (Chen & Popovich, 2003; Homburg et al., 2017). So it is crucial for scholars and managers to be aware of the changing environment in these fields and the marketing field in general.

Therefore the main contribution of this research lies in expanding the knowledge in the CEM and CRM fields. It should first be defined what customer experiences are and how to measure them in order to manage the customer experiences.

Customer Experiences

Customer experience (CE) consists of the sensorial, affective, cognitive, relational and behavioral responses a person has to a brand and/or firm (Homburg et al., 2017). According to Brakus, Schmitt, & Zarantonello (2009) evaluating the customer experience on only 1 of these determinants is insufficient since customer experiences are not as simple as just liking or disliking a brand. They state that the sensorial response can be seen as the first experience, for example seeing the logo or name of the brand for the first time. This is followed by a general attitude towards it, which is more stable than just an emotional reaction (Verhoef,

Lemon, Parasuraman, Roggeveen, Tsiros, & Schlesinger, 2009). While the relational and behavioral aspects refer to the feeling of connection to a firm/brand and the actual purchasing actions of the consumer (Brakus et al., 2009; Verhoef et al., 2009). So for measuring customer experience all of these determinants should be taken into account (Homburg et al., 2017). Gathering customer data is crucial for measuring and improving the customer

(11)

experience (Piccoli et al., 2016). A widely used method to collect data about customer experiences and customer satisfaction is the Customer Effort Score (CES). This metric is proven to be more effective than the classic Net Promoter Score (Dixon et al., 2010). The CES is based on one question: “How much effort did you have to put forth to handle your request” (Dixon et al., 2010). This implies that if the amount of effort is low, people will generally be more satisfied about their service experience and therefore be more loyal to the firm (Dixon et al., 2010). For this research it was chosen to not focus solely on the CES since all of the factors determining the customer service experience should be taken into account (Homburg et al., 2017). By integrating these data a broader overview of the customer experience can be provided.

The process of collecting data is getting easier due to the continuously developing technologies used in customer service delivery (Rust & Huang, 2014). The data can be used to get customer insights about the experiences they had. Other purposes of these data can lie in the application in artificial intelligence.

Artificial intelligence

Artificial intelligence (AI) is a broad concept, defined by the Cambridge Dictionary as “The study of how to produce machines that have some of the qualities that the human mind has, such as the ability to understand language, recognize pictures, solve problems, and learn” (Artificial Intelligence, 2017). Derived from this definition it can be stated that it is about technologies with informational processing, learning and emotional abilities. _Technology innovations are rapidly improving and are already replacing human jobs (Miller, 2017). Not only mechanic jobs are on stake, also managerial jobs can eventually be performed by

(12)

machines instead of humans (Chui, Manyika, & Miremadi, 2015). The focus of this research will not be on the technologies behind AI, but on the application of AI in customer services.

The use of artificial intelligence can have time and speed benefits, because robots generally are quicker than humans (Hill et al., 2015). Another benefit lies in the bundling of customer data through the use of IT. This can help to personalize the customer service, in a way that firms can offer products and services which are personalized to the customers wants and needs (Piccoli et al., 2017). Since the AI concept is too broad to cover in one research, this research will only focus on the use of chatbots in customer service.

Chatbots

Online retailers and service websites are increasingly implementing different types of chatbots in online customer service delivery (Chattaraman, Kwon, & Gilbert, 2012). As Chattaraman et al. (2012) describe in their article there are 3 types of chatbots. First of all they mention the chatbot with a search support function, which is most used. Another type of chatbot has a decision support function, which means they help the consumer to compare different alternatives. The third type of chatbot is the navigational chatbot, which guides the consumer through the entire online purchasing process. The more developed chatbots can cover all of these functions.

Chatbots can also be classified on basis of how they are programmed. Some are pre-programmed and can perform certain tasks for which it is programmed to perform. The more developed chatbots make use of AI and are self-learning (Burgess, 2017). Which makes them capable to learn and perform more activities over time since they can learn from their past activities and from the input of more data (Burgess, 2017). An example of a pre-programmed chatbot is Apple’s iOS Siri which can perform certain pre-programmed

(13)

tasks such as setting an alarm, checking the weather and sending a text message (Apple inc., 2018). An example of a self-learning AI driven chatbot is IPsoft’s Amelia (IPsoft, 2018). According to IPsoft (2018) the Amelia chatbot can be hired by businesses like it is an employee. Therefore Amelia also has to be trained in how the company processes work, just like a human employee. This chatbot is able to work with human colleagues and will learn faster than humans when the business grows. IPsoft (2018) states that Amelia can do the job of thousand normal employees, combined with a low risk rate. These AI driven chatbots are more costly to integrate in practice, but will eventually result in a decrease in the total costs and will improve the customer experience (Burgess, 2017; IPsoft, 2018).

This also applies to the online customer service experience, since offering more service providing facilities to the customer improves the online customer service experience (Parasuraman et al., 2005). Providing a chatbot function can be seen as an additional facility to interact with the firm. This implies that when a firm provides this facility the perceived customer service experience would increase, as long as the chatbot functions properly (Gefen, 2002). Also in terms of responsiveness chatbots are quicker than other types of online customer service such as email (Hill et al., 2015; Klaus, 2013).

But on the other hand chatbots are not human. This could make people feel like they are not getting the personalised service they desire (Gefen, 2002). But chatbots are developing in a way that consumers do not even notice they are talking to a robot (Devlin, 2017). The attitude towards communicating with a chatbot could also vary across different consumer groups.

Younger generations are more affiliated with AI and chatbots (Tomasco, 2016). This could also have an impact on how they experience the interaction with a chatbot in comparison with older generations.

(14)

Technical savviness

The amount of use of a technology like a chatbot can be influenced by the technical savviness of the individual (Soares et al., 2017). Technical savviness can be described as the level of which an individual is able to control technological machines and to what extent they are experienced using technologies (Soares et al., 2017; Venkatesh, Thong, & Xu, 2012; Wilfong, 2006). Technical savviness is closely related to self-efficacy (Wilfong, 2006). Because when people get more experienced with the use of a certain technology they will more likely be able to complete particular tasks using those technologies (Venkatesh et al., 2012). So it can be stated that technical savviness is a broader concept than self-efficacy. Self-efficacy is solely referring to the capability of performing the actions which are necessary to complete a task (Wilfong, 2006). While technical savviness includes this capabilities and the experience with technologies (Venkatesh et al., 2012). To reduce confusion there will only be referred to technical savviness in the continuation of this research report.

Individuals who have a higher level of technical savviness generally adopt new technologies quicker than individuals with a lower level of technical savviness (Soares et al., 2017). This can be explained by the fact that they probably have more experience with the use of technologies in the past (Venkatesh et al., 2012). But people high on technical savviness also have a higher intention to try out new products and are more willing to learn about new technologies (Soares et al., 2017). The technical savviness also relates to computer anxiety.

On the one hand, people who are more experienced with the use of technologies in their past will be less anxious to try out new technologies (Rosen & Maguire, 1990). But on the other hand, people who are more anxious for new technologies will avoid using new

(15)

technologies in the future and therefore will not become more experienced (Mahar et al., 1997; Wilfong, 2006). This indicates that prior experiences with technologies generally result in a higher technical savviness and reduce anxiety for technologies (Chua, Chen, & Wong, 1999). Besides, this computer-anxiety can also be caused by the level of trust an individual has in the technologies (Rosen & Maguire, 1990).

Trust in technology

Consumer trust can be a critical stimulator for consumer behavior (Jarvenpaa, Tractinsky, & Vitale, 2000). The perceived risk is one of the main determinants for the level of ones trust (Rousseau, Sitkin, Burt, & Camerer, 1998). When the perceived risk is low, people do not need a high level of trust to overcome this risk barrier (Jarvenpaa et al., 2000). This indicates that when the perceived risk is higher, the individual will need a higher level of trust to be willing to take the risk (Rousseau et al., 1998).

Trust is a dynamic concept, which means it can change over time (Bozic, 2017; Rousseau et al., 1998). People can have a low trust level on forehand, but this can later be influenced by the experiences they had with a certain product or person (Rosen & Maguire, 1990). Therefore Bozic (2017) defines trust as: “the willingness to rely on an exchange partner in whom one has confidence”. This exchange partner can be human, business or a machine (Jarvenpaa et al., 2000; Rosen & Maguire, 1990; Venkatesh et al., 2012).

The level of trust an individual has in the technology can be measured by the performance expectancy and the effort expectancy (Venkatesh et al., 2012). According to Venkatesh et al. (2012) the performance expectancy relates to the potential benefit someone expects to achieve by using a particular technology. And the effort expectancy is related to the perceived ease of use of this technology (Johnson & Payne, 1985). Which translates to

(16)

how much effort should be taken to make the technology work in a favorable way (Dixon et al., 2010). According to this theory, people who perceive the technology as beneficial and easy to use will have a higher trust level in the technology. People who have a higher level of trust are more willing to accept new technologies (Bozic, 2017). This translates to the fact that people with a high performance expectancy and a low expectancy for the amount of effort adopt a new technology quicker (Dixon et al., 2010; Venkatesh et al., 2012).

Perceived potential benefit

The perceived potential benefit can be described as the extent to which something is relevant for everyday life (Turner, Turner, & Van de Walle, 2007). Or how the use of a product can increase the performance or efficiency for a specific task (Bruner & Kumar, 2005). The perceived potential benefit is also influenced by the expected performance of a product or technology (Venkatesh et al., 2012). When a product is expected to provide good performances then it will generally be perceived as more useful (Turner et al., 2007; Venkatesh et al., 2012). Therefore it can be stated that the usefulness of a product is the perceived potential benefit a product can offer.

The perceived potential benefit of a product significantly correlates with the usage level of a product according to the ‘Technology Acceptance Model’ (Davis, 1989). Which means that when someone perceives a product as beneficial, the amount of use will generally be higher. This model also states that products perceived as useful will also be used more in the future. This can be explained by the cost-benefit paradigm described by Johnson & Payne (1985). According to this paradigm people make choices by outweighing the amount of effort they have to put in and the amount of potential benefit they can generate from this (Maslowsky, Buvinger, Keating, Steinberg, & Cauffman, 2011). This indicates that the

(17)

perceived potential benefit should be higher than the perceived amount of effort the individual has to put in to generate this benefit. When the amount of effort needed is low people are more willing to adopt a new product or technology, especially if the risk of use is low (Johnson & Payne, 1985). So the perceived ease of use also plays a role in the adoption of a new product or technology (Dixon et al., 2010)).

Perceived ease of use

The concepts of the perceived ease of use and the perceived potential benefit are closely related (Davis, 1989). When a person perceives something as easy to use it will most generally also be perceived as being more useful (Bruner & Kumar, 2005). The perceived ease of use combined with the ease of learning how to use the product can be defined as the usability of a product (Venkatesh & Davis, 1996). This means that a product or technology should be user friendly to be perceived as easy to use, and therefore has a higher usability level (Ellis & Kurniawan, 2000). Especially technical products are generally perceived as more difficult to use (Turner et al., 2007; Wilfong, 2006). The level to which a technology is self-explanatory, or can be used intuitively, are important indicators of the perceived ease of use (Ellis & Kurniawan, 2000).

According to the ‘Technology Acceptance Model’, the perceived ease of use is also significantly correlated to the acceptance of new technological products (Davis, 1989). Which indicates that products which are perceived as easy to use will generally be used more often. On the other hand this is also influenced by the technical savviness (Soares et al., 2017). Because individuals who are more technical savvy will generally perceive technologies as easier to use (Venkatesh et al., 2012). This is also influenced by the age of the individual (Homburg et al., 2017).

(18)

Age and technology

Generational theory states that people can be profiled according to the time period they were born in (Gurău, 2012). People from the same generation tend to have the same type of values, preferences and behaviors (Bilgihan, 2016). According to this theory the generations can be separated in different ways (Brosdahl & Carpenter, 2014; Gurău, 2012; Leask et al., 2014; Soares et al., 2017). For this research four generations will be discussed in more detail: the silent generation, the baby boomers, generation X, and generation Y.

The silent generation is born between 1925 and 1945 (Brosdahl & Carpenter, 2011). A large part of this generation has consciously experienced the second world war. According to Brosdahl & Carpenter (2011) this could have made an impact in a way that makes this generation more conventional thinkers. This generation is generally not highly involved with modern technologies (Bolton et al., 2013). They will be almost unrepresented in this research and therefore will not be described in more detail.

The baby boomers generation can be described as: a generation, born between 1946 and 1960, which consists of people who tend to be idealistic thinkers and are self-focused (Brosdahl & Carpenter, 2011). This generation is aged between 57 and 72 while taking part in this research in 2018. Which means that people from this generation are retiring or already are retired. This has influence on their purchasing behavior, because they will become more family focused (Brosdahl & Carpenter, 2011). On the other hand this also means they generally do not use computers on a day-to-day basis anymore since they stop working. So they get less experienced using technologies and lack behind in developing their technical savviness (Venkatesh et al., 2012). This generation did not grow up surrounded by technology and therefore have not been dealing with technology since a young age (Bilgihan,

(19)

2016). Which means they had to adapt to the emerging technologies and therefore generally have a lower technical savviness than younger generations, such as Generation X and Y, who have more experience using technologies (Venkatesh et al., 2012).

Generation X is sometimes referred to as the thirteenth generation (Brosdahl & Carpenter, 2011). This generation, born between 1961 and 1981, is described as being individualistic (Brosdahl & Carpenter, 2011). These consumers also tend to have a more negative view of the world and are generally distrustful (Brosdahl & Carpenter, 2011; Losyk, 1997). Since the internet was open for use since 1992 this is the first generation which was using the internet, but they did not grow up with it (Brosdahl & Carpenter, 2011). Due to the internet the amount of choice in terms of products and jobs got bigger (Bilgihan, 2016). This results in the fact that people from generation X are less loyal and committed and like to keep their eyes open for new opportunities (Losyk, 1997). This decreasing loyalty trend is also visible in the generation born after generation X (Bilgihan, 2016).

These humans of generation Y are born between 1982 and 2000 (Gurău, 2012). This generation is focused on self-development, and believe education is the key to success (Brosdahl & Carpenter, 2011). Nowadays they are growing in spending power and are almost outperforming the large baby boomer generation on this aspect (Bilgihan, 2016). They are generally more technical savvy than the older generations since they grew up interacting with technologies (Gurău, 2012; Soares et al., 2017). Which makes them more involved in the use of technologies like social media (Leask et al., 2014). This can also be due to the age of people from this generation. Because younger people generally take more risk than older people, and might therefore be more open to try out new things (Arnett, 2000). But it is most likely that even if other people from other generations would be of the same age, generation Y’s behavior would still be different than the other generations (Soares et al., 2017). This

(20)

especially applies to the use of new technology-driven products (Bolton et al., 2013). So it can be concluded that generation Y is a unique generation which is different than the older generations (Soares et al., 2017). On the other hand there is still a discussion about whether it is useful to profile people in these age categories (Brosdahl & Carpenter, 2011). And if generations are separated, then still there are different ways of describing these generations (Leask et al., 2014). Because a single generation can be influenced by different environmental factors, for example due to different influences in different parts of the world (Brosdahl & Carpenter, 2011). But since it is expected that most of the participants in this research will be dutch (or at least european), these environmental factors will not be described in further detail. So the political, environmental, cultural and legal factors will be assumed to be equal (Hofstede, 1984; Soares et al., 2017).

2.2 Theoretical Framework

The main research question in this research is: _{What is the effect of the use of chatbots on the} customer experience in online customer service situations? This question will be answered by doing research on the following sub-questions:

Sub-question 1: What is the effect of the use of chatbots on the online customer service experience?

Sub-question 2: How does technical savviness influence the relationship between the use of chatbots and online customer service experience?

Sub-question 3: How does trust in technology influence the relationship between the use of chatbots and online customer service experience?

Sub-question 4: How does the perceived potential benefit of the technology influence the relationship between the use of chatbots and online customer service experience?

(21)

Sub-question 5: How does the perceived ease of use of the technology influence the relationship between the use of chatbots and online customer service experience?

Sub-question 6: Does age moderate the moderation effects of technical savviness, trust in technology, perceived potential benefit and ease of use?

Sub-question 7: Are there any differences between generation Y and older generations in terms of technical savviness, trust in technology, perceived potential benefit and ease of use?

Sub-question 1: What is the effect of the use of chatbots on the online customer service experience?

The online customer service experience builds on the functional and psychological aspects of the service delivery (Klaus, 2013). Chatbots especially focus on the functional part of the customer service. They can improve the service quality because they can provide a quick and easy way of communication with the firm (Hill et al., 2015). Chatbots generally decrease the amount of effort the consumer has to put in to get the desired level of service (Klaus, 2013). According to the CES, this will improve the experience and satisfaction of a customer (Dixon et al., 2010). Besides, since chatbots are an additional communication facility they can improve the responsiveness of the firm (Gefen, 2002). Therefore it is expected that the use of chatbots should be beneficial for the online customer service experience.

On the other hand, communicating with a chatbot can be perceived as impersonal (Hill et al., 2015). While customers want to be taken serious and get personalized service (Gefen, 2002). Consumers could be unsatisfied when they do not perceive the service as personalized to their wants and needs. So the use of chatbots could have a negative impact on the psychological aspects of the online customer service experience (Klaus, 2013). But

(22)

chatbots are getting more personal and are developing empathic feelings (Devlin, 2017). This could take away the feeling of talking to a robot, wherefore the psychological aspects are not perceived as negative anymore. Since chatbots are fairly new, it could be the fact that they were not functioning optimally from the start (Chattaraman et al., 2012). Prior users who had negative experiences with chatbots in the past might avoid using them in the future (Wilfong, 2006). These past experiences with an underdeveloped and not properly functioning chatbot might have influenced the perceived online customer service experience in a negative way (Gefen, 2002). But this research will focus on the current level of chatbot usage. Therefore it is expected that the chatbots have been developing and are working properly now. So these negative aspects of using chatbots are not relevant anymore. Therefore the first hypothesis (H1) for this research is: _{More frequent users of chatbots} generally are more satisfied about their online customer service experiences.

Sub-question 2: How does technical savviness influence the relationship between the use of chatbots and online customer service experience?

People differ on the level of how experienced and capable they are with the use of technologies (Soares et al., 2017). This means that to what extent someone can use technologies in a beneficial way is influenced by the technical savviness of that person (Wilfong, 2006). A technical savvy person will be more capable to benefit from the use of technologies than someone who has a lower level of technical savviness (Venkatesh et al., 2012). Therefore it is expected that this same relationship will hold for the use of chatbot technologies. This means that people with a high level of technical savviness will be more capable to use the chatbot function in a beneficial way than people with a low level of technical savviness. And people who are more technical savvy are also more open to the use

(23)

of new technologies than people who are less technical savvy (Soares et al., 2017). Therefore it is also expected that a higher level of technical savviness relates to a higher amount of use of chatbots.

So individuals with a low level of technical savviness will generally have a lower usage level for the chatbot function (Soares et al., 2017). This means they will not benefit from this online customer service facility, which results in a less positive perceived service experience (Gefen, 2002). On the other hand, people who do use these functions can benefit from it in terms of the functional aspects the chatbot function has to offer (Klaus, 2013). Such as the increased level of responsiveness through the use of chatbots, compared to using email (Hill et al, 2015).

To summarize this, it is expected that a high level of technical savviness results in a higher amount of use of chatbots. And these people are also generally more capable with the use of chatbots and can therefore benefit from the functions it has to offer. Which results in a more positive online customer service experience. Therefore the second hypothesis (H2) for this research is: _{Technical savviness will positively moderate the relationship between the use}

of chatbots and customer service experience.

Sub-question 3: How does trust in technology influence the relationship between the use of chatbots and online customer service experience?

The third expectation is that amount of trust in technology will influence the relationship between the use of chatbots and online customer service experience. This is expected because people who have a higher level of trust in technologies are more open to accept and use new technologies (Bozic, 2017). However, people who have a higher level of trust, tend to have higher expectations (Venkatesh et al., 2012). So if the chatbot function

(24)

does not provide the service quality matching the expectations, this could result in dissatisfaction (Parasuraman et al., 2005). Based on this it can be stated that the level of trust could have a negative and positive effect on the customer service experience. But since chatbots do not require a lot of effort, it is not needed to have a high level of trust in technology to use the chatbot function (Venkatesh et al., 2012). People who do have a high level of trust tend to have a more open attitude towards the adoption of new technologies (Klaus, 2013). This means that these people score better on the psychological aspects for the use of chatbots (Klaus, 2013). They will feel more comfortable using the chatbot function and perceive it as more easy to use (Venkatesh et al., 2012).

Using a chatbot function generally is free of charge for the consumer and does not require much time and effort. So people do not have to invest much to make use of this technology. Therefore, the risk barriers for using chatbots are low (Rousseau et al., 1998). Since the risk barrier is low, there is no need for a high level of trust to overcome this barrier (Jarvenpaa et al., 2000). But people who are more trusting will be even more open to take this risk (Bozic, 2017). So it can be concluded that people with a high level of trust are expected to adopt chatbots quicker, and therefore can benefit from its functions. This implies that people with a high amount of trust in technology are expected to be more satisfied with the online customer service they get. So the third hypothesis (H3) for this research will be:_Trust

in technology will positively moderate the relationship between the use of chatbots and customer service experience.

Sub-question 4: How does the perceived potential benefit of the technology influence the relationship between the use of chatbots and online customer service experience?

(25)

The perceived potential benefit of a chatbot refers to the extent that people think using a chatbot could be beneficial (Turner et al., 2007). This is the case if people believe the use of chatbots can improve the quality of customer service (Bruner & Kumar, 2005). This is also influenced by the performance expectations of the chatbots (Turner et al., 2007; Venkatesh et al., 2012). If it is expected that chatbots will perform well, the perceived potential benefit is higher (Davis, 1989).

Another way of determining the perceived potential benefit of a chatbot is by making a cost-benefit trade-off (Johnson & Payne, 1985). The amount of effort for using chatbots is low, which can be seen as the costs. So the benefits do not have to be high to exceed these low costs. The potential benefits are the performance expectations the consumer has (Turner et al., 2007). When the perceived benefits are higher than the costs, the consumer will be more likely to adopt the chatbot technology.

So a higher level of perceived potential benefit is expected to result in a more frequent amount of use of chatbots. Since it is expected that chatbots improve the customer service experience, it is expected that this relationship will be even stronger when someone perceives the chatbot function as useful. So therefore the fourth hypothesis (H4) for this research is:

The perceived potential benefit of a chatbot will positively moderate the relationship between the use of chatbots and customer service experience.

Sub-question 5: How does the perceived ease of use of the technology influence the relationship between the use of chatbots and online customer service experience?

The perceived ease of use for chatbots does not only indicate the direct perceived ease of use, but also refers to the learnability of using chatbots (Venkatesh & Davis, 1996). The expectations concerning the ease of use of chatbots are closely related to the expectations

(26)

relating to the benefits of chatbots (Davis, 1989). The perceived ease of use can also be seen as part of the cost-benefit trade-off (Johnson & Payne, 1985). The costs for using chatbots can be seen as the amount of effort the consumer has to put in. According to the CES a low amount of effort should have a positive effect on the service experience (Dixon et al., 2010). This implies that when the consumer perceives chatbots as easy to use, the effort expectations are low (Maslowsky et al., 2011). If this is the case it can be stated that the perceived ease of use score is high (Ellis & Kurniawan, 2000). So a high score on ease of use means that the chatbot function is perceived as easy to use, or as easy to learn. This translates to the expectation that consumers who perceive chatbots as easy to use, will generally adopt chatbots quicker than consumers who perceive chatbots as difficult to use.

According to H1 it is expected that the use of chatbots generally result in a better online customer service experience. Therefore it is expected that the use of chatbots, combined with a high perceived ease of use, will result in a more positive service experience. This is expected because a high level of ease of use translates to a better understanding of how to work with chatbots. Therefore the fifth hypothesis (H5) for this research is: _The

perceived ease of use of a chatbot will positively moderate the relationship between the use of chatbots and customer service experience.

Sub-question 6: Does age moderate the moderation effects of technical savviness, trust in technology, perceived potential benefit and ease of use?

The influence of age will be tested in two different ways. This sub-question will focus on the influence of age as a moderator on the expected moderating effects discussed in sub-question 2, 3, 4 and 5.

(27)

According to theory, the more experienced an individual is with using technologies, the more capable this individual will be to work with technologies (Venkatesh et al., 2012). This could imply that older people have more time to gain experience and develop their capabilities with technologies. But the most important factor driving this technical savviness is experience with technologies while growing up (Brosdahl & Carpenter, 2011; Gurău, 2012). Therefore it is expected that a higher age will have a negative effect on the technical savviness of an individual. So the first hypothesis for this sub-question (H6a) is: _{There will be} a negative moderated moderating effect of age on the on the effect of technical savviness.

The amount of trust in technology is also related to how accustomed the individual is to technologies (Rosen & Maguire, 1990). Younger consumers are raised surrounded by technologies, this will probably have a positive impact on their level of trust (Soares et al., 2017). So it is expected that

(H6b): _{There will be a negative moderated moderating effect of}

age on the effect of trust in technology.

In terms of the perceived potential benefit and ease of use it can be expected that younger consumers will more likely perceive chatbots as beneficial than older consumers. Because younger consumers are more involved with technical communication methods like mobile phones, social media and mobile apps because these are quick and easy to use (Leask et al., 2014). They will probably also perceive communication with chatbots as beneficial since these can also facilitate time and effort savings (Klaus, 2013). From this statement it can be concluded that the potential benefit and ease of use are closely related to each other, which is supported by the cost-benefit trade-off (Johnson & Payne, 1985). Furthermore, younger people are expected to be more capable to work with technologies (Soares et al., 2017). This implies that chatbots will be perceived as more easy to use by younger people than older people. Therefore it is expected that (H6c): _{There will be a negative moderated}

(28)

moderating effect of age on the effect of the perceived potential benefit.

And (H6d):_There

will be a negative moderated moderating effect of age on the effect of the perceived ease of use.

Based on these expectations it can be stated that older consumers will generally have a less positive customer service experience while using chatbots than younger consumers. Therefore it is expected that (H6e): _{Age will negatively moderate the relationship between the}

use of chatbots and online customer service experience.

Sub-question 7: Are there any differences between generation Y and older generations in terms of technical savviness, trust in technology, perceived potential benefit and ease of use?

For the seventh sub-question the consumers are separated into four different generations based on generational theory (Brosdahl & Carpenter, 2011; Gurău, 2012). This sub-question aims to identify differences in technical savviness, trust in technology, and the perceived potential benefit and ease of use among the different generations. Since it is expected that generation Y is significantly different than the older generations, the hypotheses will be formulated in a way that compares generation Y with the older generations (Soares et al., 2017).

The baby boomers generation did not grow up surrounded by technologies (Bilgihan, 2016). Therefore it is expected that these consumers are less experienced using technologies than generation Y consumers (Gurău, 2012; Soares et al., 2017). This same reasoning holds for generation X consumers, but these consumers did grow up during the rise of the internet (Brosdahl & Carpenter, 2011). So the difference in technical savviness between generation Y and baby boomers is expected to be bigger than the difference between generation Y and X. Since consumers from generations Y are more experienced and capable of using technologies

(29)

than the older generations (Soares et al., 2017). Therefore it is expected that (H7a) _:

Generation Y consumers have a higher technical savviness than consumers from older generations.

The level of trust in technology can also be influenced the experience an individual has (Rosen & Maguire, 1990). Since generation Y consumers are generally more experienced with technologies, it is also expected that they have a higher level of trust in technology (Soares et al., 2017). A higher level of trust generally results in quicker adoption of new technologies (Rousseau et al., 1998). This is supported by the fact that younger consumers are generally more open to adopt new technologies than older consumers (Bolton et al., 2013). Therefore it is expected that (H7b): Generation Y consumers have a higher level of trust in technology.

The potential benefit and ease of use expectations are closely related to each other since they can both be based on the cost-benefit trade-off (Johnson & Payne, 1985). The costs for using chatbots is referred to as the perceived ease of use and effort the consumer has to make (Venkatesh & Davis, 1996). Generation Y consumers are more involved in using technical communication methods than older generations (Leask et al., 2014). Therefore it can be stated that the cost-benefit trade-off is positive, so the perceived benefits are higher than the costs (Johnson & Payne, 1985). Since a chatbot is a type of communication technique it can be expected that this reasoning will hold for the use of chatbots (Chattaraman et al., 2012). This can be explained by the fact that generation Y consumers tend to be individualistic and prefer working with technologies for efficiency reasons (Brosdahl & Carpenter, 2011). Since they generally have a higher technical savviness it is also expected that they will be more capable using chatbots. So it can be concluded that they think chatbots are beneficial because of the improved responsiveness, which is time efficient (Hill et al.,

(30)

2015). And generation Y consumers are generally more capable of using technologies and therefore will perceive chatbots as easier to use, compared to older generations (Ellis & Kurniawan, 2000). So the last hypotheses for this research are: (H7c): consumers from

generation Y perceive chatbots as more useful than consumers from older generations.

And

(H7d): Consumers from generation Y perceive chatbots as more usable than consumers from older generations.

2.3 Conceptual Model

Figure. 1.

(31)

Chapter Three: Methods

3.1 Research design

The data for this research was gathered through an online survey. The aim was that it would take the respondents about 10 minutes to cover all the questions. These choices were made to find enough respondents in a 2 week period. To enhance the reliability of this research a minimum of 200 participants were needed to fulfill the survey. This resulted in a total of 210 participants, 46,7% male. Data is collected through social media, e-mail and personal contacts. The only constraint for taking part in this research is that the respondent had an basic idea about what chatbots are and how they function.

The year of birth of the respondents varies from 1951 to 2001, wherefrom 50% is born in 1994 or later. The largest proportion (90,5%) of respondents have a dutch nationality, and 95,4% was born in a European country. In terms of educational level it is noticeable that 41% of respondents has a university bachelor degree or higher.

The survey consists of seven parts. The first part includes questions about the demographics. The second part of the survey consists of questions about the past online customer service experiences the respondent had. Where in the third part questions about the amount of use of chatbots are stated. The fourth part of the survey measures the technical savviness of the respondents. While the fifth part measures the level of trust the respondents have in technologies. The sixth and seventh part of the survey consists of questions about the perceived potential benefit and ease of use of chatbots.

All parts of the survey, except the demographics, are measured on a 5-point Likert scale. This choice was made to reduce bias and because this type data is easier to analyze

(32)

than results from open questions. By doing analysis, using SPSS, it was tried to answer the sub-questions. By combining the sub-question results, an answer can be given on the main research question: _{What is the effect of the use of chatbots on the customer experience in} online customer service situations?

3.2 Measurements

Dependent variable online customer service experience

The online customer service experience was measured on a 6-item scale with a reliability of α = .800, which makes this a reliable scale. Using these questions it was tried to measure the customers’ satisfaction level about their past online customer service experiences. The questions were addressing both the speed and the quality of the customer service. The used measurements for this variables were based on the scales used by Grönroos (1984) and

Parasuraman et al. (2005). These scales also have a reliability higher than α = .7. An example item is: _{“I feel taken seriously when I try to solve a problem through an online customer} service point”

. Each item is measured on a 5-point Likert scale, ranging from (1) totally

disagree to (5) totally agree. A high score on this scale means a high satisfaction level about online customer service experiences. Two out of the six items were counterbalanced and had to be recoded for analysis.

Independent variable usage of chatbots

This part of the survey measures the current frequency of using chatbots, and the intention to use chatbots in the future. The questions are based on the measurement scales used by Parasuraman (2000). This scale consists of four items which are also measured on a 5-point Likert scale, ranging from (1) totally disagree to (5) totally agree. This reliability for this

(33)

scale is α = .771. This could be increased by .047 by excluding one item of this scale. But since the effect is considered as small, and this would imply that 25% of the scale would be excluded, it was chosen to keep the original 4-item scale. An example question for this scale is: _“

I always use the chatbot function when I have a question”. A high score on this scale

indicates a high level of usage for chatbots. One item was counter-indicative, so this has to be recoded before doing analysis.

Moderating variable technical savviness

The technical savviness is measured on a 8-item scale. This scale was found to be reliable since the Cronbach’s alpha is α = .804. By deleting one item the reliability could increase to α = .809, but it is chosen to keep the original scale since this increase was perceived as too small. This scale also uses the 5-point Likert scale, ranging from (1) totally disagree to (5) totally agree. The questions in this part of the survey are about how confident the respondents are about using technological products and services. The questions are based on the research methods used by Soares et al. (2017) and Venkatesh et al (2012). An example question for this scale is: _{“I can usually figure out new high-tech products and services, without help from}

others”

. A high score on this scale means a high level of technical savviness. One item had to

be recoded for analysis because this was reversely phrased.

Moderating variable trust

The trust in technology is measured on a 6-item scale. All of the questions were phrased negatively so they had to be recoded before doing analysis. One item was excluded, this resulted in a 5-item scale with a reliability of α = .617. This means that the reliability of this scale is questionable. Since the scale is not perceived as unacceptable it will still be used in

(34)

the rest of this research. But it should be taken into account that the reliability for this scale is lower than the other used scales when interpreting the results. The scale uses a 5-point Likert scale, ranging from (1) totally disagree to (5) totally agree. The questions in this part of the survey were addressing the amount of trust the consumer has in the technological developments. The questions are based on methods used by Parasuraman (2000) and Venkatesh et al. (2012). An example question is: “People can solve problems more

effectively than computers”.

A high score on this scale indicates a high level of trust in

technology.

Moderating variable perceived potential benefit

The perceived potential benefit of chatbots is measured at a 5-item scale, with a reliability of α = .842. The questions were answered on a 5-point Likert scale, ranging from (1) totally disagree to (5) totally agree. This part of the survey tried to measure how useful the consumers thinks that chatbots are. The scales developed by Davis (1989) and Soares et al. (2017) were used for developing this scale. An example question is: _{“communicating with a} chatbot saves me time”

. A high scores on this scale means that the perceived potential benefit

of chatbots is high. One item had to be recoded before doing analysis.

Moderating variable perceived ease of use

For measuring the ease of use of chatbots a 7-item scale is used, with a reliability of α = .844. This scale also used the 5-point Likert scale, ranging from (1) totally disagree to (5) totally agree. This part of the survey tried to measure the ease of use and the expectations about chatbots. The same ideas as for the potential benefit scale were used to develop this scale (Davis, 1989; Soares et al., 2017). An example question is: _{“Generally I know how to use a}

(35)

chatbot”.

A high score on this scale indicates a high perceived ease of use level. Five out of

the seven items had to be recoded before doing analysis, because these were negatively phrased.

Moderating variable age

The age of the respondents is measured by their birth date. This data was recoded to come up with the age in years. The dummy variable representing the different generations was also based on this data. This dummy is created to check for differences between the generations. The Cronbach’s alpha could not be determined for this scale since it consists of one item.

3.3 Procedure

Respondents for the survey were gathered according the convenience sampling method. Participants were contacted through social media, mobile phone, e-mail and in person. Some participants shared the survey with their personal contacts to reach a larger audience. This strategy was chosen for time and costs efficiency reasons. For these same reasons there was chosen to use an online questionnaire, because this can be spread widely in a short period of time. During data collection it was tried to manage that all the age categories were represented. The result was that the complete data set was gathered in a time period of 14 days. The required minimum amount of 200 respondents was achieved since 210 participants completed the questionnaire. On average participants took 720 seconds (12 minutes) to answer all questions for the questionnaire (3 outliers excluded).

Participating in this research was completely voluntary and no rewards were given to the respondents. At the first page of the questionnaire there was a clear statement containing

(36)

where the research was about. This page also contained a promise that the data will be handled with care and used anonymously for this research.

3.4 Analysis and predictions

The data will be analyzed using SPSS complemented by PROCESS (v.2.16.3). In this way the reliability of the measurement scales can be tested. PROCESS and linear regression analysis can be used to test the hypotheses.

It is expected that a positive relationship will be found between the use of chatbots and online customer service experience. Furthermore it is expected that technical savviness, trust in technology, perceived potential benefit and ease of use will positively moderate this relationship.

The influence of age will be measured in two ways. First, the moderating effect of age on the moderating effects of technical savviness, trust in technology, perceived potential benefit and ease of use will be tested (H6a, H6b, H6c, H6d). This is done using the moderated moderation model (model 3) of PROCESS. After that the consumers will be grouped into different generations. In this way the hypotheses (H7a, H7b, H7c, H7d) will be tested in terms of differences between generation Y and older generations. It is expected that generation Y will have a higher level of technical savviness and trust in technology. It is also expected that this generation will perceive chatbots as more useful and easier to use.

(37)

Chapter 4: Results

4.1 Reliabilities and correlations

Before doing analysis the reliabilities of the scales have to be tested. The internal consistency of the scales was checked by calculating the Cronbach’s alpha for the constructs: online customer service experience, use of chatbots, technical savviness, trust in technology, potential benefit, perceived ease of use. All of the scales have a score above α = .7, except for the trust in technology scale (α = .617). This makes the internal consistency of the scale questionable. For the purpose of this research this scale will still be used, but the questionable reliability should be taken into account when deriving the conclusion for this research. To do an extra reliability check, the reliability coefficient for every single item was checked. These were all above the minimum of .30, except for the already excluded items. The scale reliabilities can be found on the diagonal axis in table 1.

By observing table 1, it can be derived that online customer service experience (OCSE) is significantly correlated to all other variables in this research. OCSE is positively correlated with the use of chatbots ( _{r = .277, p = .01**), technical savviness (r = .216, p =} .01**), trust in technology ( _{r = .165, p < .05*), potential benefit (r = .251, p < .01**) and} ease of use ( _{r = .262, p < .01**). OCSE is significantly negatively correlated with age (r =} -.161, _{p < .05*). The use of chatbots is also positively correlated with technical savviness (r =} .139, _{p < .05*), trust in technology (r = .140, p < .05*), potential benefit (r = .741, p < .01**)} and perceived ease of use ( _{r = .421, p < .01**). Technical savviness is positively correlated} with trust in technology ( _{r = .265, p < .01**), potential benefit (r = .154, p < .05*) and} perceived ease of use ( _{r = .318, p < .01**). Trust in technology is positively correlated with}

(38)

the perceived ease of use ( _{r = .184, p < .01**). And as expected, the potential benefit and} perceived ease of use are also significantly positively correlated with each other ( _{r = .544, p} < .01**).

Table 1.

Descriptives and correlations between the variables (Cronbach’s Alpha on diagonal)

4.2 Exploratory Factor Analysis

As an addition to the scale reliability check, an exploratory factor analysis has been conducted on the scales for OCSE, use of chatbots, technical savviness, potential benefit, and perceived ease of use. From the Barlett’s Test of Sphericity it can be derived that the different items are significantly high in correlation (χ2_{= 3132.978,}_{p <}

.01**).

Then it was analyzed whether the sample size (_{N =}

210) is sufficient for this research.

This is done using the Keyser-Meyer-Olkin (KMO) test. The KMO score (.834) is higher than the the minimum requirement of |.60|, therefore the sample size is proven to be of sufficient size for this research.

4.3 Testing hypotheses

Before testing the hypothesis the right way of modelling had to be chosen. This was done by studying the possibilities in PROCESS (Hayes, 2017). Since a statistical model which can test the conceptual model as a whole is not provided by PROCESS, a different approach of testing the hypothesis is required. Therefore it was checked if it would make a significant difference if 2 moderators would be included in the model at the same time. This was done

(39)

for every possible combination of moderating variables using Model 2 of PROCESS. No significant change in effects were found.

Figure 2.

PROCESS model 2_.

Therefore it was chosen to do a separate analysis for every possible moderating effect, since it is expected that these variables are independent, using model 1 in PROCESS. After that model 3 was used to check for the moderated moderating effects of age (H6a, H6b, H6c, H6d).

One-way ANOVA analysis was used to check for the difference between generations in terms of technical savviness, trust in technology, perceived potential benefit and ease of use (H7a, H7b, H7c, H7d). The next step in analysis is testing the hypotheses according to the chosen methods.

(40)

Figure 3.

PROCESS Model 1_. Figure 4. PROCESS model 3_.

H1: More frequent users of chatbots generally are more satisfied about their online customer service experiences.

For testing this first hypothesis linear regression analysis is used. In this way the direct influence of the use of chatbots on OCSE can be measured. This results in significant support for the first hypothesis_{(β = .200, p < .01**)}

. This means that individuals who differ by one

unit of frequency of using chatbots will differ by 0.2 units of OCSE. At a 95% confidence interval the R² value is 0.077, which means that 7.7% of the total variance in OCSE can be derived from the use of chatbots. The model used to test this direct relationship is a simplified version of the conceptual model in figure 1. No possible moderating effects were involved in these calculations.

H2: Technical savviness will positively moderate the relationship between the use of chatbots and customer service experience.

The second hypothesis will test whether technical savviness will moderate the relationship found by testing the first hypothesis. This is done by using PROCESS model 1, where technical savviness is tested as the moderating variable (M). This results in a raise of the R²

What makes or breaks the success of the chatbot function in an age of online