When humanizing brands goes wrong: the detrimental effects of personalization amid negative eWOM handlings

(1)

When humanizing brands goes wrong: the detrimental effects

of personalization amid negative eWOM handlings.

FACULTEIT ECONOMIE EN BEDRIJFSKUNDE UNIVERSITEIT VAN AMSTERDAM

Bachelor’s Thesis:

Student: Rick Boersen 10191178

Supervisor: H.H. Lee Date: 4-07-2014

(2)

Abstract

In the current business environment, companies have to deal with different factors than before. The phenomenon of negative eWOM is one of the new challenges companies face. The purpose of this study is to find an appropriate response strategy to negative eWOM. The main theoretical frameworks we use to formulate an answer to our research question are the implicit theory of personality, and brand humanization. We conducted an empirical study (N=121) to test our hypotheses regarding personalization of messages. The main finding is that companies using personalized messages to address negative eWOM, are evaluated worse, than companies who use non-personalized messages. Even though there are some drawbacks to this study, some theoretical and managerial implications can be drawn. The main theoretical contributions of this study consist out of contradictions of previously accepted standards, and implications for the implicit theory. For managers, this study implies that caution is needed when choosing an online response strategy.

(3)

1. Introduction

The rise of the internet has provided the customer with alternative ways of expressing and sharing their negative experiences with a brand (van Noort & Willemsen, 2011). Previously customers would only tell some relatives and acquaintances about bad experiences they may have had. The internet however has empowered them to share their dissatisfaction with the world (Ward & Ostrom, 2006). For example, customers nowadays engage in negative electronic word of mouth (hereafter negative eWOM) on social media platforms (Ward & Ostrom, 2006). A good example of this phenomenon can be found on the Facebook page of T-Mobile: “What a couple of scammers work on your sales division. I’m a customer for 10 years now and want to renew my contract. They sent me a completely different offer per e-mail, than we have agreed upon on the phone!” (T-Mobile’s Facebook page). T-Mobile replied as follows: “Hi *anonymous*, this can’t be our intention. I would like to investigate this further for you. Could you send us your mobile number, postal code and date of birth in a private message? (T-Mobile’s Facebook page)”

To handle this kind of negative eWOM it’s relevant for managers to know what an appropriate response strategy is. T-Mobile chose a very personalized way to respond to this particular customer. They did this by using an informal tone, and addressed the customer by his first name. But is this the best way to react to negative eWOM? In contrast, T-Mobile could have used a non-personalized way to approach this complaining customer. Their reply in this case could have looked like this: Dear Mr. *anonymous*, we apologize for the inconvenience. Would you please send us your mobile number, postal code and date of birth in a private message? This way we can find a solution to the problem. This study will delve deeper in this matter by investigating the impact of personalized- and non-personalized replies on brand perceptions of the customer after they engaged in negative eWOM.

The means by which companies try to achieve a restoration of the customer’s brand perceptions today, is mostly through the use of webcare. Webcare is defined as: “The act of engaging in online interactions with (complaining) consumers, by actively searching the web to address consumer feedback” (van Noort & Willemsen. 2011). Webcare often takes place in the form of personalized messages. These personalized messages are used to give a more human face to the brand. In current literature, brand humanization is often linked with more positive brand evaluations of the customer (Delbaere, McQuarry & Phillips, 2011). Therefore, a personalized webcare approach is widely used by most companies.

(5)

Webcare using personalized messages causes previously inanimate brands to become humanized through the online conversations with customers (Gensler, Völkner, Thompkins & Wiertz 2013). This helps companies to signal their brand humanization practices though personalized messages. Most companies engage in webcare in a personalized fashion, since literature seems to stress the positive effects of brand humanization.

However, Puzakova, Kwak and Rocereto (2013) found that in the case of brand wrongdoing, there is a more detrimental effect on the customer’s brand perceptions when the brand is humanized than when it’s not. Customers will see the humanized brand as a thoughtful entity. This leads to the tendency to hold the company responsible for its actions. For this reason, customers devaluate a humanized brand more than their non-humanized counterparts. Since personalized messages signal brand humanization to the customer, it might be that customers devaluate the company using personalized messages as well.

However, the effect found by Puzakova and colleagues is moderated by the implicit theory of the customer. This theory states there are differences in beliefs about malleability of traits and attributes related to the self and the environment (Puzakova et al, 2013). The theory proposes that two groups of personalities can be observed. On the one hand, we have entity theorists, who believe that traits and personal attributes are subject to change. On the other hand, incremental theorists believe that personal traits are fixed, and are not malleable. Puzakova observed the effect only for entity theorists. This suggests that in practice, the relationship would be applicable for some, but not for all customers. The study of Puzakova raises questions concerning webcare. Why are companies trying to pursue a humanized brand by personalizing their responses to complaints, when the results of this study suggest more detrimental effects when they succeed to humanize?

This theory however is not tested in the situation of a response to negative eWOM in a social media context. This gap in the extant literature leads to the following research question: Do personalized replies by a humanized company to negative eWOM have a detrimental effect on the brand evaluations of the customer? Furthermore, the moderator effect of the implicit theory of personality on the relationship between the personalization of replies, and brand perceptions will also be tested.

To provide an answer to this research question, an empirical study will be conducted. We will investigate our hypotheses using an experimental survey design. This enables us to test our propositions in a suitable manor. This paper will start off with assessing the current literature on this topic. Secondly we will conceptualize our hypotheses in a model. Hereafter, we will present our methodology. Fourthly, we will present our empirical results. We will 5

(6)

conclude by discussing how our results relate to the extant literature. Also we will provide some ideas for future research on this topic.

(7)

2. Literature Review

This section will delve deeper in the concepts used in this study. In this literature review, the variables in this study will be explained. First, the concepts of eWOM, and personalization of replies will be discussed. Finally the moderator variable implicit theory of personality will be explained.

2.1 Personalization of responses to negative eWOM

The independent variable used in this study is the personalization of a company’s response to negative eWOM from the customer. This paragraph will first illuminate the concept of eWOM further. Secondly, the personalization of responses will be discussed.

2.1.1 Electronic word of mouth

Word of Mouth (hereafter WOM) is defined as face-to-face sharing of experiences about a product or a service with close others (Ritchins, 1984). The influence of WOM on companies can be enormous. This can be illustrated by noting some big companies that have been built though strong WOM. For example, Red Bull, Starbucks and The Body Shop relied on WOM as a marketing strategy (Kotler & Keller, 2012). These stories of success were based on positive experiences of customers. On the other hand, when a customer has a bad experience, he or she might engage in negative WOM as a result. Negative WOM can be very harmful to companies. The effect of negative WOM on evaluations is even bigger than the impact of positive WOM. This is known as the negativity effect (Park & Lee, 2009). Ritchins (1983) stresses the importance of negative WOM as well: “if the number of consumers experiencing dissatisfaction is high enough, such responses may have lasting effects in terms of a negative image, and reduced sales for the company. In this way, negative WOM can have substantial financial implications for the firm”.

Nowadays, the rise of the internet and social media provided customers with a new way of sharing their experiences online via electronic WOM. EWOM is defined as: any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet (Hennig-Thurau, Gwinner, Walsh & Gremler, 2004). According to this definition, eWOM also can be positive or negative. For this study, the focus will be on the negative aspect of eWOM.

(8)

In comparison to traditional WOM, eWOM has the potential to be even more harmful to companies. Negative eWOM is found to have detrimental effects on all phases of the decision-making process of the customer. This includes brand evaluations, brand loyalty, and brand purchase decisions (van Noort & Willemsen. 2011). Furthermore, the mere definition of eWOM states that eWOM is made available to a multitude of people. This wide availability of negative publicity is a phenomenon that poses new challenges for companies, because customers now share their negative experiences with the world (Hennig-Thurau et al., 2004) (Ward & Ostrom, 2006).

Not only is the wide availability of negative eWOM a problem to companies, but also the speed in which the negative message spreads. Negative eWOM messages can spread at enormous speeds due to the internet, and sharing on social media platforms (Cheung & Lee, 2008). Because the negative messages spread so fast they do not only damage existing customer’s brand evaluations, but also those of potential future customers.

Customers have various reasons to engage in negative eWOM. Hennig-Thurau et al. (2004) investigated a large sample of customers to find out what motivates them to engage in eWOM. The main factors they found that induce the customer to engage in eWOM are their desire for social interaction, their concern for other customers, the desire for economic incentives, and the potential to enhance their self worth.

In addition to the study by Hennig-Thurau and colleagues, Shin, Sorg and Biswas (2013) identified two underlying motives for customers to engage in negative eWOM. The first reason they found is the intrinsic regulatory focus of the customer. The regulatory focus theory states that people try to achieve their goals through two different modes: promotion, or prevention. Prevention-oriented customers are found to engage in negative eWOM more than promotion-oriented customers would. Secondly, Shin and colleagues found a contextual factor that influences the generation of eWOM; collective dissonance. Their findings suggest that customers are more motivated to engage in eWOM when their experiences are inconsistent with those of others. For example, if there is an abundance of positive reviews, one with a negative experience would be more motivated to post eWOM.

A third study on the motivations for posting eWOM suggests that there are three predictors: altruism, venting and empowerment (Willemsen, Neijens & Bronner, 2013). Venting is concerned with someone’s desire to express their feelings. Altruism is the concept of the willingness to warn someone else for the bad practices of a company. Altruism is trying to help someone else, without receiving something in return. Customers draw on the empowerment motive when they want to express their power. Negative eWOM comes with

(9)

many negative effects for the company. The customer might know this, and uses this knowledge to his advantage. The empowerment motive can be seen as a power tool to enforce a proper handling of the negative eWOM by the company.

2.1.2 Personalization of responses

As the literature suggest that negative eWOM can have detrimental effects on a customer’s behavior, and thereby the company’s performance, it’s important to know what an appropriate response strategy would be. Companies react to negative eWOM through webcare interventions. By this mean, they try to restore the customer’s evaluations of the brand. Webcare is; “The act of engaging in online interactions with (complaining) consumers, by actively searching the web to address consumer feedback” (van Noort & Willemsen. 2011). The effectiveness of webcare is dependent on the motivation of the poster of the negative eWOM. For example the study by Willemsen, Neijens and Bronner (2013) suggests that the motivations of customers to engage in negative eWOM are crucial to their receptiveness of the webcare by the company. Customers with the empowerment motive for example, are very receptive to webcare, where the altruism and venting motives are not.

Another way of looking at effectiveness of responses to negative eWOM is to look at the message itself. To restore the customer’s evaluations of the brand, companies can choose different ways to address and approach the customer. This study will focus on personalized- and non-personalized messages by the company.

Literature suggests that conversational human voice is a very important aspect in messages to restore brand evaluations (Kelleher & Miller, 2006). Conversational human voice is defined as; an engaging and natural style of organizational communication as perceived by an organization’s publics based on interactions between individuals in the organization and individuals in public (Kelleher, 2009). Furthermore, Conversational human voice is linked to increased customer satisfaction and trust in the company (van Noort & Willemsen, 2011).

In the case of a response to negative eWOM, a company could use the concept of conversational human voice in their personalized messages. Some indicators of a message written with a conversational human voice can be found in the article by Kelleher (2006). Kelleher identified familiar qualities such as being open to dialog, and being welcoming. However, it also includes factors that previously wouldn’t be associated with corporate communication before. For example, a sense of humor, admitting mistakes and treating others as human. This tone can be used to create a very personalized message. The company will address the customer in an informal and personal way. On the other hand, a company could choose a more non-personalized way to address the customer. Here, the company would

(10)

address the customer in a formal way. For example, they would initiate the message by addressing the customer by his last name. Furthermore, the company would use a more traditional form of corporate communication.

Most of the times, companies choose for the more personalized way to approach the customer. They just assume that it’s the best way. However, this may not be the case. The following section (the conceptual model) will shed more light on this.

2.2 Implicit theory of personality

The moderating variable of this study is the implicit theory of personality of the customer. Implicit theory of personality can be defined as beliefs about the malleability of traits and attributes related to the self, and the environment (Puzakova et al, 2013). Implicit theories influence the way how different persons understand and react to human actions and outcomes (Dweck, Chiu & Hong, 2006). Literature suggests there are two different implicit theories of personality. These are the entity- and the incremental theory of personality (Levi, Dweck & Schroessner, 1998).

The entity theory of personality suggests that a person’s traits are fixed. In other words, entity theorists believe that one’s attributes are stable, and therefore not malleable (Levi, Dweck & Schroessner, 1998). This leads them to interpret outcomes as results from these fixed traits (Dweck, Chiu & Hong, 2006). An example of a statement from an entity theorist could be as follows: “I failed the test because I’m dumb”. This example illustrates the interpretation of an event by an entity theorist. The reason of the failure at the test, is an underlying attribute or trait of the person himself.

The incremental theory of personality forms a contrast to the entity theory of personality. Incremental theorists view one’s attributes and traits as malleable (Levi, Dweck & Schroessner, 1998). This means incremental theorists try to understand outcomes and actions in terms of other factors that could have influenced the results (Dweck, Chiu & Hong, 2006). A statement made by an incremental theorist would therefore look more like this: “I failed the exam because of my effort or learning strategy”. An incremental theorist would not directly blame the underlying trait (intelligence) but instead address the malleability of the trait (effort).

(11)

3. Conceptual Model

Now the main concepts related to this study are explained, this section will present the conceptual model. This section forms the hypotheses that this study will test. First, our hypotheses related to the effects of personalization on brand evaluations are presented. Second, the hypotheses related to the predicted moderation are addressed. This section will conclude with a graphic of the conceptual model.

3.1 The effect of personalized replies on brand evaluations

Personalization often is considered to have positive influences on a customer’s brand evaluations. For example, Delbaere and colleagues (2011) found that personalization made a positive emotional response to the brand more probable relative to what a non-personalized metaphor could achieve. Knowing that personalization is found to have positive influences on brand perceptions, companies use this in their online response strategies. As previously stated, companies engage in webcare as a response to negative eWOM from the customer. In this way, they try to restore the brand evaluations of customers with bad brand experiences. In the application of webcare, conversational human voice is often used in the online conversations with customers. This application of conversational human voice is found to lead to a more humanized perception of the brand (van Noort & Willemsen. 2011).

In contrast to the current widely accepted opinion that the humanization of brands leads to positive perceptions, literature also addresses situations in which humanization practices backfire. For example, a study by Puzakova, Kwak and Rocereto (2013), found that a customer’s brand evaluations can decrease under a certain condition. They found that in the case of brand wrongdoing, negative publicity is found to have a more detrimental effect on brand evaluations when the brand is humanized, in comparison to when it’s not. The reason for this is that customers see the humanized brands as thoughtful entities. This leads to the perception of the customer that these brands should be held responsible for their actions. As a result, customers tend to punish the humanized brand more than they would punish a non-humanized brand for their wrongdoings (Puzakova et al., 2013).

The findings of Puzakova suggest that online humanization of the brand is not positive in all situations after all. Since negative eWOM is often the result of wrongdoing or malfunctioning of the company or its products, it could be that the humanization practices of the company will backfire. The results imply that companies who have established themselves

(12)

as humanized brands by using personalized messages, and signal this by communicating in a personalized manor, will be damaged more by the negative eWOM. This in turn would result in a bigger devaluation of the humanized brand in comparison to their non-humanized counterparts in the case of brand wrongdoing. This leads to the first hypothesis of this study:

Hypothesis 1: In a company’s replies to negative eWOM of the customer, companies who use personalized messages will be evaluated less favorably than their counterparts using non-personalized messages.

3.2 The moderating effect of implicit theory of personality

However, the study by Puzakova and colleagues (2013) identified a moderator for the observed relationship. They found that in the case of brand wrongdoing, persons react differently depending on their implicit personality. The brand evaluations of entity theorists differed from those of incremental theorists after they were exposed to brand wrongdoing. Entity theorists were found to devaluate the brand, where for incremental theorists no differences in brand evaluations were observed. The reason the researchers propose is that entity theorists believe in fixed personality traits. Therefore entity theorists view a single event of brand wrongdoing as a sign of underlying bad personality traits of the humanized brand. In contrast, incremental theorists believe in the malleability of personality traits. Therefore, incremental theorists will not punish the humanized brand for a single event of bad practice.

In our research, we expect the same effect. We expect that entity theorists who receive a personalized message will see this as a sign of brand humanization practices. As we know from previous research, entity theorists see the wrongdoing by a humanized brand as a sign of underlying badness. Therefore, entity theorists tend to punish the humanized brand for their wrongdoings. We expect the same to happen in our research. Therefore, we propose that entity theorists will punish the humanized brand with a lower brand evaluation. Incremental theorists on the other hand, will not hold the humanized brand responsible for their wrongdoings. Therefore, incremental theorists will not devaluate the brand.

This leads to the following hypotheses regarding the effects of implicit theory of personality on brand evaluations.

Hypothesis 2a: When the company’s reply to a customer’s negative eWOM is personalized, entity theorists will devaluate the brand more than when the reply is not personalized.

(13)

Hypothesis 2b: For incremental theorists, personalized, or non-personalized replies to the customer’s negative eWOM will not lead to differences in brand evaluations.

Beneath, in figure 1, a graph of the conceptual model of this study is presented. It displays the different concepts and the relationships between each of the variables.

Figure 1: Conceptual model

(14)

4. Methodology

The previous sections discussed the extant literature related to the research question. Also the conceptual model that will be tested was provided. This section will address the methods, and the research design through which the data will be obtained. To start this section, the research design will be discussed first. After this, the sample used in this study is addressed. Third, the procedure by which the data was obtained is explained. Finally, the scales through which the variables were measured are discussed.

4.1 Research Design

This paragraph will delve deeper in the research design. First the general design of this study will be explained. Secondly, some pros and cons of this particular design are addressed.

The general design of this study is an experimental setting combined with a survey. The survey will use a questionnaire as the strategy for obtaining data. This strategy fits the purposes of this study the best, because it allows receiving large amounts of comparable data from a sample (Saunders, Lewis & Thornhill, 2012). This is exactly what is needed to test the hypotheses. In order to obtain comparable data from the questionnaire, a standardized questionnaire was designed. This implies that every respondent will answer the same questions, and in our case, by using a standardized answer set. By standardizing the questionnaire, respondents will give answers that are comparable using statistic techniques. By obtaining this quantitative data, relationships between variables can be tested (Saunders et al., 2012). Surveys are used to obtain large amounts of data in a relatively low-cost way. This makes it attractive for our purposes as well. The survey contained questions about the respondent’s perceptions of the brand, his implicit theory, and a manipulation and reality check.

The survey design using a questionnaire will be combined with an experimental setting. The experiment is used to expose the respondents to different conditions. For this study, this means that respondents will be randomly assigned to two different experimental groups. This way we can manipulate the personalization of the message. One half of the sample will be exposed to the condition of a humanized brand using personalized replies. In contrast, the other half of the sample will be assigned to a non-humanized version of the same brand using non-personalized replies. This way, one half of the sample will get the impression of a humanized brand while the other half will be confronted with a non-humanized brand

(15)

impression before filling in the questionnaire. This experimental setting within the questionnaire allows us to test our hypotheses adequately.

The two manipulations were adapted from the Facebook page of the company. The actual negative eWOM comments that are shown to respondents are authentic. However, we adapted the respective reactions of the company. We succeeded in making the manipulations look as realistic as possible, so that respondents would feel that they watch actual authentic company replies.

The way this all translates into our actual research is as follows. Respondents are first presented with an introduction to this study. Hereafter, the manipulations of the brands are shown. Respondents are randomly assigned to one of the two manipulation groups. After this, the questionnaire started by asking questions about the reality of the situations, brand evaluations of the customer, and implicit theory items. The study concluded by asking the respondents about their relationship with the brand, eWOM, and some demographic information.

Of course this research design comes with some limitations. The first limitation of using a questionnaire for data collection would be that only a limited number of questions can be asked before the respondent gets bored, and starts giving unreliable answers. So if the questionnaire gets too long, respondents can lose their motivation to complete it in a usable way (Saunders et al., 2012). However, in order to test our hypotheses we only need a small amount of questions. This will lead to a short questionnaire. For this reason, this drawback of the design will not create too much problems.

Another drawback of the survey design is that the questionnaire has to be well designed from the start. This, because changing the questionnaire in the middle of the study makes the data incomparable, and unreliable for our purposes. In order to solve this problem, a pre-test will make sure the questions are clear to respondents, and every measurement needed for analysis is included.

The manipulations and questionnaire were in the Dutch language. This, because it increases the amount of possible respondents that could be reached.

The brand we chose to use in this study is de Nederlandse Spoorwegen (NS). The NS is responsible for public transport in the Netherlands. The core operation is moving people via train. The main reason for using NS for this study is their behavior on social media. The NS uses personalized and non-personalized messages to react to negative eWOM. This made the brand attractive to manipulate, because possible respondents most likely don’t have an image about the humanization practices of the brand. If we chose a company that is known for its 15

(16)

personalized messaging, this could have influenced the results because respondents don’t trust the manipulations. In order to test if the respondents interpreted the manipulations correctly, a pre-test was conducted. This provided us with information about the manipulation, and the perceived reality of the manipulations. In other words, the pre-test informed us if respondents recognized the personalized message- and the non-personalized message as such. Also the perceived reality of the manipulations was tested.

4.2 Sample

The sample will mostly consist out of contacts in the personal network of the researcher. This has the advantage that potential respondents are familiar with the researcher and will therefore be more motivated to participate in the study. Since the researcher has contacts within different groups of the Dutch population, we expect a wide spread of participant characteristics. The actual demographic details will be presented in the next section.

4.3 Procedure

This paragraph explains how respondents were reached. Furthermore, this section will address how respondents provided their answers to the questionnaire.

Respondents were contacted via a different set of channels. The first method used was a direct, face-to-face approach. We did this both prior to the study, and during the actual study. Some respondents were contacted before the actual research started. This was mainly to raise awareness. This way, respondents knew that they could be contacted in the future, and that their cooperation would be appreciated. During the collection of data, we also used a face-to-face approach. In this case, we provided the respondents with an iPad so they would be able to fill in the questionnaire on the spot.

Another way we reached respondents was through the use of Facebook. This channel enabled us to use direct, as well as mass communication. We sent private messages to potential respondents to ask them for their participation in the study. Also, a public message was sent out to reach much larger amounts of people.

The third way respondents were reached was via e-mail. This mainly included relatives, acquaintances and colleagues. Finally, Whatsapp was used to reach respondents.

The timing of the usage of channels was a thoughtful process. We used the different channels in a consecutive fashion. This way, we could reach respondents multiple times via different channels. This way, a message had two functions. The message would be a first encounter with the study for some, but a reminder to cooperate for others.

(17)

Respondents filled in the questionnaire on different platforms. Each of the channels provided the respondents with an URL link. This enabled respondents to participate with the study on a platform of their choice. This includes mobile phones, iPads and computers. No paper and pen questionnaires were distributed.

4.4 Measures

To conclude this section, this paragraph will explain which scales are used to measure our variables. The measures will be discussed in the order of appearance in the questionnaire. The examples of items provided in this section are in English. However, in the study all items were translated to Dutch.

4.4.1 Manipulation and reality checks

In order to test if the manipulations were successful, respondents were asked to answer a couple of questions. An example of an used item is: “The NS uses a corporate voice” The possible answers were provided in a seven-point Likert scale reaching from (1) “strongly disagree, to (7) “strongly agree”. A low score (for example 1) indicates that the respondent perceives the company to be humanized.

Data about the reality of the manipulations was obtained using a couple of questions. An example of an item is: “How do you estimate the probability of this situation occurring in real life”? Answer possibilities were again provided by a seven-point Likert scale. This time however, answer possibilities ranged from (1) “very small” to (7) “very large”. A high score (for example 7) on this item indicates a high reality perception of the manipulation by the respondent.

4.4.2 Brand evaluations

Brand evaluations were tested using the scales used in the study by van Noort & Willemsen (2011). An example of an item used in the measurement is: “The NS offers a good quality”. Respondents could react to this statement using a seven-point Likert scale reaching from (1) “strongly disagree, to (7) “strongly agree”. A high score on this item (for example 7) would indicate the respondent has positive brand evaluations regarding NS.

4.4.3 Implicit theory of personality

The implicit theory of personality of the respondents was tested using the measurements used in the article by Puzakova and colleagues (2013). An example of one of the measurements used would be the statement: “People can change even their most basic qualities”. The scales were a seven-point Likert scale reaching from (1) “strongly disagree, to (7) “strongly agree”. A low score on this item (for example 1) would indicate the respondent has entity theorist’s 17

(18)

ideas. On the other hand a high score (for example 7) indicates the respondent has the ideas of an incremental theorist.

4.4.4 Control variables

We asked for factors that could influence our findings. This included questions about familiarity with the brand, eWOM awareness and eWOM engagement of the respondent. In this section, all items were measured using a seven-point Likert scale. The definition of the Likert scale points was dependent on the formulation of the question.

4.4.5 Demographics

Finally we asked for some demographic information. We only asked for age and gender

4.5 Analysis Methods

This paragraph will describe the methods for analysis we will use in order to test our hypotheses. First, we have to test if our manipulations were successful. In order to do this, we test the reliability of our measures using the crohnbach’s alpha. If our variables are reliable, we can compare our two manipulations based on our manipulation check. This will be done using an independent samples t-test.

After we confirmed the reliability of our measures and manipulations, we can test our hypotheses. For hypothesis 1, we have to compare two groups. This will be done using the independent samples t-test as well.

For our second hypotheses, we have to compare a couple of groups. We will do this with a factorial ANOVA. This will show us if groups differ. However, if we want to analyse which groups are different, we need some extra information. Therefore we will run a simple effect analysis. This analysis will provide us with comparisons of our experimental groups.

(19)

5. Analysis & Results

First, this section will address the descriptive statistics of the final sample. Hereafter, the reliabilities of the scales will be discussed. Finally, the results of this study are presented.

5.1 Descriptive statistics

The final sample consisted of 121 respondents. These respondents were targeted in the Netherlands. The targeted group included several layers of the Dutch population, including for example students, and average aged persons. Education levels were not considered when targeting respondents. The sample included 55 percent female, and 45 percentage male respondents. The average age in the sample was 25 years. The standard deviation of the mean age was 10.8 years. To ensure the anonymity of the respondents, only this limited amount of demographic information was collected.

Finally, 63 respondents completed the questionnaire with the personalized message condition. The questionnaire with the non-personalized message condition was completed by 58 respondents.

5.2 Reliability analysis

Before the analysis of the obtained data could be started, the raw data had to be edited. This included counter-balancing of a couple of items in different measurements. When the raw dataset was adapted for analysis, the manipulations were tested. The data to test this was obtained using four manipulation check items in the questionnaire. These items included questions about the level of humanization and personalization of the brand. This measurement had a crohnbach’s alpha of .72. This can be considered as a reasonable value. Furthermore, the crohnbach’s alpha could not be increased by deleting items.

The analysis on how the manipulations were interpreted was done using an independent samples t-test. Since Levene’s test was not significant, equal variances were assumed. The data showed that the manipulations had different levels of personalization. For the personalized version of the brand, respondents perceived the brand level of brand humanization higher (M=4.6, SE=1.1) than for the non-personalized version of the brand (M=3.1, SE=.85) (p=.000<.05).

(20)

Furthermore, an analysis on the reality of the manipulations was adopted. The crohnbach’s alpha of this measure was .646. With a mean score of 5.8 (SD=1.1), both manipulations were found realistic.

Now that the reliability and reality of the manipulations is confirmed, the reliability of the main variables will be discussed. The brand evaluations of the respondent were measured using a four item measurement, adapted from van Noort & Willemsen (2011). This measurement had a crohnbach’s alpha of .902. This good score could not be improved by deleting items from the scale.

The moderating variable used in this study is the implicit theory of the customer. As stated before, this was measured using the same scales as Puzakova et al. (2013). We found a crohnbach’s alpha of .799 for this measurement. This can be considered as a good reliability. The reliability could not be improved by deleting some of the items.

5.3 Results

Now that we confirmed the reliability of the measurements, the main analyses could be started.

5.3.1 Brand evaluations across personalized- and non-personalized messages The first hypothesis we formulated was that in a company’s replies to negative eWOM of the customer, companies who established themselves as a humanized brand, using personalized messages will be evaluated less favorably than their non-humanized counterparts who use non-personalized messages. The hypothesis was as follows:

Hypothesis 1: In a company’s replies to negative eWOM of the customer, companies who use personalized messages will be evaluated less favorably than their counterparts using non-personalized messages.

We tested this proposition using an independent samples t-test. Again, Levene’s test was not significant. This indicates that equality of variances should be assumed.

The results show a slight difference in brand evaluations for the different conditions. The company using personalized messages shows a lower evaluation (M=3.962 SD=1.279) than the brand using non-personalized messages (M=4.266 SD=1.181). This would be in line with the hypothesis. These differences were found significant (p= 0.0895 >p=.1) on the 10% confidence level. These results imply that there is a difference between the brand evaluations across the humanized- and the non-humanized brand. Therefore, on the 10% confidence level, 20

(21)

hypothesis 1 is confirmed. Table 1 summarizes the findings of this study regarding our first hypothesis.

Table 1: Independent samples t-test results for hypothesis 1

N Mean SD Brand Evaluations Personalized Messages 63 3.962 1.279

Non-Personalized

Messages 58 4.266 1.181

Note: N=121

5.3.2 The moderating effect of implicit theory.

The expected moderating effects of implicit theory were proposed in hypotheses 2a and 2b. These two hypotheses were as follows:

Hypothesis 2a: When the company’s reply to a customer’s negative eWOM is personalized, entity theorists will devaluate the brand more than when the reply is not personalized.

Hypothesis 2b: For incremental theorists, personalized, or non-personalized replies to the customer’s negative eWOM will not lead to differences in brand evaluations.

In order to test these hypotheses, we expected the sample to have a dichotomy with respect to the implicit theories of the respondents. We expected a clear group of incremental-, and entity theorists to appear. In other words, we expected the data to have a very large standard deviation.

However, in contrast to our expectations, we found our data to be very centrally distributed. The standard deviation of the distribution we observed was .834. Since no clear groups of incremental- and entity theorists appeared, we divided the sample on the basis of this standard deviation. We assigned respondents who deviated more than the standard deviation from the mean, to the two different groups. In other words, respondents with a score on the implicit theory scales lower than 3.32 were allocated to the incremental theorists. On the other hand, respondents who scored higher than 4.99 were allocated to the entity theorists. This reduced the sample size considerably. Table 2 summarizes how much respondents have which implicit theory, and to which manipulation they were exposed.

(22)

Table 2: Frequencies of implicit theories and manipulations Manipulation Personalized Messages Non-Personalized Messages Implicit

Theory Entity Theorist 9 10

Incremental Theorist 9 8

Note: N=36

Now that we identified the two groups based on their implicit theory, we can test our hypotheses. Hypothesis 2a and b will be tested using a factorial ANOVA. In order to do this, we created a dummy variable for implicit theory. Hereafter, we conducted the main analysis. First, we obtained some descriptive statistics about the mean brand evaluations of our four experimental groups. These results are summarized in tables two and three.

The main analysis showed that message personalization predicts differences in brand evaluations of the customer (F=3.593, p=0.067). In other words, the personalization of messages has influence on the brand evaluations of the customer. This is further proof of hypothesis 1. Secondly however, there was no effect of implicit theory on brand evaluations Table 3: Mean brand evaluations of entity theorists (hypothesis 2a)

N Mean SD Brand Evaluations Personalized Messages 9 4.156 1.08

Non-Personalized

Messages 10 4.740 .943

Note: N=19

Table 4: Mean brand evaluations of incremental theorists (hypothesis 2b)

N Mean SD Brand Evaluations Personalized Messages 9 4.289 .937

Non-Personalized

Messages 8 4.925 .555

Note: N=17

(23)

(F=.244, p=0.624). Also, there was no interaction effect (F=.006, p=0.937). Table 5 summarizes our ANOVA results.

Table 5: Factorial ANOVA results

Message Personalization

Implicit Theory

Message Personalization*Implicit theory

F P

3.593 .067 .244 .624 .006 .937

This is an indication that our second hypotheses may have to be rejected. In order to formulate a more definitive answer, we have to compare the groups. To compare the brand evaluations of our four experimental groups, we ran a simple effect analysis. This enabled us to provide an answer to our second hypotheses. For hypothesis 2a, the simple effect analysis revealed no differences between brand evaluations of entity theorists, after they have been exposed to either a personalized-, or a non-personalized message (Mean Difference = 0.584, p= 0.196). In other words, entity theorists do not evaluate the brand less when they are exposed to a personalized message.

For hypothesis 2b, we analyzed the effects for incremental theorists. We proposed that there should be no difference in brand evaluations dependent on the personalization of the message. We tested this hypothesis the same way as the previous hypothesis, using the simple effect analysis. The analysis showed no differences in brand evaluations for incremental theorists based on the personalization of the message (Mean Difference=0.636, p=0.183). This is in line with our predictions, and therefore hypothesis 2b should be accepted. Table 6 summarizes these results.

Table 6: Simple effect analysis results

difference Mean P

Entity Theorists Personalized message - Non-personalized Message .584 .196

Incremental

Theorists Personalized message - Non-personalized Message 0.636 .183

Note: N=36, Dependent variable = Brand evaluations

(24)

6. Discussion

The previous section addressed the results of this study. In this section, we will discuss the results in a critical fashion. We will commence by summarizing the results, and how they relate to extant literature. After this, we will critique these results based on methodological shortcomings. Finally, we will propose some ideas for future research on this topic.

6.1 Summary of study results

6.1.1 The effects of personalized replies on brand evaluations

This study examined the predicted negative effect of personalized responses of companies to negative eWOM on the brand evaluations of the customer. With more specificity, we tried to answer the following research question: Do personalized replies by a humanized company to negative eWOM have a detrimental effect on the brand evaluations of the customer?

In order to find an answer to this research question, we formulated the following hypothesis: In a company’s replies to negative eWOM of the customer, companies who use personalized messages will be evaluated less favorably than their counterparts using non-personalized messages. Based on our results, this proposition can be accepted. We found some evidence that customers will devaluate the brand, if they respond in a personalized manor to their negative eWOM. This in contrast to previous ideas of van Noort and Willemsen (2011). These authors stated that the concept of conversational human voice would personalize the corporate image. This would in turn help to humanize the brand, which they recognized as a helpful technique in restoring the customers damaged perceptions of the brand. Also, the findings of Delbaere et al. (2011) are contradicted by this study. They found that personalization made a positive emotional response to the brand more probable relative to what a non-personalized metaphor could achieve. This study found that the non-personalized company was evaluated better by the respondents than its personalized counterpart.

However, our findings are in line with the study by Puzakova et al. (2013), who identified detrimental effects of brand humanization. These authors found that customers punish humanized brands more for their wrongdoings than they would devaluate a non-humanized brand. Our results can complement the findings of Puzakova and colleagues. Just like the situation in the Puzakova study, our research described a situation of brand wrongdoing. This brand wrongdoing can result in negative eWOM about the company. We found that the company who signals its humanized nature by using personalized messages

(25)

was actually evaluated worse than its non-personalized counterpart. Therefore our findings suggest that as a reply to negative eWOM, personalized messages may not always be the best option. This is in contrast with current practice, but in line with the study of Puzakova.

6.1.2 The moderating effect of implicit theory

Our second proposition stated that the implicit theory of the customer would influence their brand evaluations. In order to examine this, we proposed two hypotheses. Hypothesis 2a was as follows: When the company’s reply to a customer’s negative eWOM is personalized, entity theorists will devaluate the brand more than when the reply is not personalized. We found no difference in the brand evaluations of the company using personalized messages versus the company using non-personalized messages for entity theorists. This would be in contrast with the findings of Puzakova et al. (2013) who suggested that entity theorists would devaluate the brand. Their motivation for this was that entity theorists see the wrongdoing of the brand as a signal of underlying badness of the company.

However, we feel that our results are influenced by a couple of factors. These points will be discussed in the section about shortcomings of this study. These shortcomings make us believe, that our results are not strong enough to really contradict the findings of Puzakova et al. (2013).

Our second hypothesis regarding the moderating effect of implicit theory was about the incremental theorists. This hypothesis was formulated as follows: For incremental theorists, personalized, or non-personalized replies to the customer’s negative eWOM will not lead to differences in brand evaluations. Our results showed no difference between brand evaluations after the manipulation of the humanization of the brand. Even though this is in line with the research stream of Puzakova et al. we are not convinced that this study gives a good representation of the phenomenon.

6.2 Methodological Critique

The methodology we chose to use in this study was not perfect. After we conducted the study, some flaws in the design can be identified.

6.2.1 The company

The company we chose to manipulate in this study was the NS. Beforehand, we had some good reasoning behind this choice. We chose for NS as our manipulation, because they seem to use both non-personalized, and personalized messages when communicating with customers via social media. We thought this would improve the reality of the manipulations, and thereby the reliability of this study. However, the feedback from respondents about the 25

(26)

company of choice was not as good as we expected. Some respondents noted that their image of this company had already formed in their heads. This, because most respondents were already very familiar with the company. For this reason, one single event, the manipulations of the company, couldn’t influence their perceptions as much. For this reason, these respondents retained their prior evaluations of the brand. This problem could be solved by using a fictional company instead.

Another point of critique on this part, is that the link between the manipulations and the questions in the survey, were not very clear to some respondents. A number of respondents noted that when they finished reading the manipulations, they continued the questionnaire without thinking about the manipulations again. To solve this, we could have made this clearer in the questionnaire. A better formulation of the brand evaluations question could have been: after reading this comments and responses by NS, what are your opinions regarding this company? By linking the manipulations to the questions more explicitly, better results could have been obtained.

6.2.2 Central tendency & Sample

One major critique on our data is the central tendency that characterizes most of our variables. For example, brand evaluations had an overall mean of 4.1 with a standard deviation of 1.23. This indicates that on our seven point Likert scale, this data is very centrally distributed. For brand evaluations, this is not the biggest of problems, since we interpret it as continuous data. However, on the implicit theory scales, this central tendency caused problems. Here, it prevented us to create two very distinctive groups of theorists. Since most respondents indicated that they didn’t feel like neither an entity theorist nor as an incremental theorist, most of our sample could not be used in the analysis.

To solve this central tendency, a number of solutions can be proposed. For example, we could have chosen a forced response strategy. In this case, we would choose an even number Likert scale, so respondents have to indicate a direction instead of choosing the middle option. Another solution to solve the central tendency would be to formulate the statements in a more provocative fashion, so respondents are triggered to formulate a more clear opinion about the statement.

Our objective was to work with a sample of about 120 respondents. For our main analysis (hypothesis 1) this was the case. However, for the moderating effect of implicit theory our sample shrunk. We ended up with only 36 suitable respondents. These also had to be divided into 4 categories, which yielded roughly about 9 respondents per group. This is not

(27)

closely enough to assume normality. For this reason, our results are not reliable. Therefore, we think our results are not strong enough to contradict the findings of Puzakova et al. (2013).

6.3 Theoretical Implications

Since our results are quite biased, and the conclusions regarding our second hypotheses are based on a very small sample, the implications of this study are limited. Our first finding however, can be considered as quite interesting. This result suggests that the humanization of the brand could actually harm the brand. This is an important theoretical implication since not much research about this phenomenon has been conducted yet.

The second theoretical implication of this study is about the implicit theory of personality. The fact that most of our respondents did not feel like either one of the two implicit theorists, makes us believe that this theorem is not as dichotomous as we expected. One reason for this could be that persons just find it hard to formulate their opinion about the malleability of personality traits. Some respondents indicated for example that they find this malleability very situation or context dependent. In this case they found it hard to formulate an answer that characterizes their overall view of this phenomenon.

Another explanation of why we found no clear division between implicit theory characteristics is that we did not manipulate the implicit theory in this study. This is a difference with the study of Puzakova et al. (2013), where they did manipulate the implicit theory beliefs by making respondents read an article about malleability of personality traits. This is an interesting difference. Where Puzakova and colleagues did find two distinctive groups using a manipulation, we couldn’t replicate this without manipulation. This may be an indication that implicit theory differences are in reality not as black and white as current literature proposes. It is quite a good possibility that the implicit theory of personality is more like a context dependent continuum. For example, the malleability of someone’s personality may depend on one’s willingness to change that specific trait. This implies that someone may belief that in general, personality traits are malleable. However, if someone doesn’t want this specific trait to change, the attribute may not be able to change. For this reason it could be that context is a very important factor in someone’s laid beliefs about the malleability of personality traits. This may be an important implication for current literature.

6.4 Managerial implications

The main managerial implication of this study is that managers should be very wary when humanizing their brands. As Puzakova et al. (2013) found, brand humanization can backfire. 27

(28)

Our results support these findings by showing message personalization may not always be beneficial for a company. For this reason, managers should be careful when choosing to humanize their corporate image using personalized messages as a response strategy to negative eWOM.

6.5 Future research

During this study, we were able to identify a number of possible future study ideas. First the most obvious idea is to replicate this study with respondents who are characterized by either one of the implicit theories. This study failed to get a clear answer to our second hypotheses regarding the moderation effect of implicit theory. This was mostly the case since we were able to identify only a few incremental and entity theorists. Future studies can follow up on this study by either finding a large enough sample of incremental and entity theorists, or by manipulating the implicit theory of respondents as in the Puzakova study (2013).

Furthermore, future studies can be conducted about the implicit theory itself. As we discussed in paragraph 6.3, implicit theories are maybe not as dichotomous as current literature proposes. Possibly, implicit theory is in practice a more continuous variable. Future studies can identify how actual customers feel about the different implicit theories. Also, contextual variables may be of influence on implicit theory beliefs. This proposition could also be tested by future research.

Another idea for researchers is to test our propositions using respondents who do really engage in eWOM themselves. Our sample lacked these characteristics, since 77 percent indicated to never engage in eWOM themselves. A replication of this study using only persons who actively engage in eWOM could help to validate the results. Another idea in this line of research would be to identify what personality traits influence a person’s eWOM engagement.

Finally, an idea would be to test for a mediation effect of the solution of the eWOM. In this way, not only the response of the company would be a part of the webcare, but also the actions a company takes to further solve the possible problems. From a theoretical point of view, this may be an important factor in the brand evaluation restoration process. Maybe the brand evaluations of the company will only be restored if the company succeeds in providing the necessary remedial action.

(29)

7. Conclusion

The aim of this study was to answer the following question: Do personalized replies by a humanized company to negative eWOM have a detrimental effect on the brand evaluations of the customer? Even though our results were not as convincing as we had hoped, they certainly point us in a certain direction. We have enough evidence to conclude that personalized replies by a humanized company are not successful in every situation. In the case of brand

wrongdoing, which leads to an expression of negative eWOM, a personalized message by a humanized brand may actually lower the brand perceptions of the customer.

However, we were not able to formulate a definitive answer on our second hypotheses. The proposed moderating effect of implicit theory of personality cannot be proved by this study. This was mainly due to the fact that we were unable to identify enough entity and incremental theorists. Due to limited time and resources, we were not able to resolve this problem in a more efficient way than we did.

Even though the results of this study are not as strong as we anticipated, this study still offers value. This study offers for example opportunities for future researchers to continue on this line of research. With some necessary improvements, this study could be replicated to formulate a more definitive answer on our hypotheses. Furthermore, some previously accepted ideas in literature are put up for discussion by this study.

To conclude this paper, we would like to end with an advice for the management of companies who have to deal with negative eWOM. As the example in the beginning of this paper indicated, eWOM can be pretty harsh. Remember the complaint on T-Mobile’s Facebook page: “What a couple of scammers work on your sales division. I’m a customer for 10 years now and want to renew my contract. They sent me a completely different offer per e-mail, than we have agreed upon on the phone!” For now, as a reply to this message, with the results of this study in the back of our heads, would indeed be a more non-personalized approach.

(30)

8. Bibliography

Cheung, C. M., & Lee, M. K. (2008). Online consumer reviews: Does negative electronic word-of-mouth hurt more?.

Dweck, C. S., Chiu, C. Y., & Hong, Y. Y. (1995). Implicit theories and their role in

judgments and reactions: A word from two perspectives. Psychological inquiry, 6(4), 267-285.

Hennig‐Thurau, T., Gwinner, K. P., Walsh, G., & Gremler, D. D. (2004). Electronic word‐of‐mouth via consumer‐opinion platforms: What motivates consumers to

articulate themselves on the Internet?. Journal of interactive marketing, 18(1), 38-52.

Kelleher, T. (2009). Conversational voice, communicated commitment, and public relations outcomes in interactive online communication. Journal of Communication, 59(1), 172- 188.

Kelleher, T., & Miller, B. M. (2006). Organizational blogs and the human voice: Relational strategies and relational outcomes. Journal of Computer‐Mediated

Communication, 11(2), 395-414.

Kotler, P., & Keller, K.L. (2012). Marketing Management. 14th Edition. Prentice Hall. Pearson Education.

Levy, S. R., Stroessner, S. J., & Dweck, C. S. (1998). Stereotype formation and endorsement: The role of implicit theories. Journal of Personality and Social Psychology, 74(6), 1421.

Van Noort, G., & Willemsen, L. M. (2012). Online damage control: The effects of proactive versus reactive webcare interventions in consumer-generated and brand-generated platforms. Journal of Interactive Marketing, 26(3), 131-140.

Park, C., & Lee, T. M. (2009). Information direction, website reputation and eWOM effect: A 30

(31)

moderating role of product type. Journal of Business Research, 62(1), 61-67.

Puzakova, M., Kwak, H., & Rocereto, J. F. (2013). When Humanizing Brands Goes Wrong: The Detrimental Effect of Brand Anthropomorphization Amid Product

wrongdoings. Journal of Marketing, 77(3), 81-100.

Richins, M. L. (1983). Negative word-of-mouth by dissatisfied consumers: a pilot study. The Journal of Marketing, 68-78.

Richins, M. L. (1984). Word of Mouth communication as negative information. Advances in consumer research, 11(1).

Shin, D., Song, J. H., & Biswas, A. (2013). Electronic word-of-mouth (eWOM) generation in new media platforms: The role of regulatory focus and collective

dissonance. Marketing Letters, 1-13.

Saunders, M., Lewis, P., Thornhill, A. (2012) Research Methods for Business Students. 6th edition. Pearson Education.

Ward, J. C., & Ostrom, A. L. (2006). Complaining to the masses: The role of protest framing in customer‐created complaint web sites. Journal of Consumer Research, 33(2), 220- 230.

Wilkie, W., (1986). Consumer Behavior. New York: John Wiley & Sons, Inc.

Willemsen, L., Neijens, P. C., & Bronner, F. A. (2013). Webcare as Customer Relationship and Reputation Management? Motives for Negative Electronic Word of Mouth and Their Effect on Webcare Receptiveness. In Advances in Advertising Research (Vol. IV) (pp. 55-69). Springer Fachmedien Wiesbaden.

When humanizing brands goes wrong: the detrimental effects of personalization amid negative eWOM handlings