The effect of different levels of personalization on ad effectiveness
Martine Eisinga – 10354379
Amsterdam, June 23, 2017
Final Version Master’s Thesis
MSc. in Business Administration – Digital Business Track
Supervisor: Dr. H. Gungor
Academic year: 2016 – 2017
Semester 2, Block 3
Statement of Originality
This document is written by Martine Eisinga 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.
Abstract
Online advertising is booming. Due to the information overload, consumer empowerment and ad blocking software there is a pressing need to make advertising relevant again. The use of personalization as a marketing technique is seen as the future of advertising. However, still many brands lack knowledge in personalization and do not sufficiently succeed in its usage. Especially the lack of knowledge of the effects of privacy concerns may be a possible explanation for inadequate personalization efforts. Understanding what level of
personalization is most effective in terms of consumer behaviour is crucial in order to design successful personalized advertisements. Besides privacy concerns, consumer behaviour also appears to depend on the type of device that is used to process an advertisement. Therefore, this study tested the effects of different levels of personalization of email advertisements on click-through and conversion, and the moderating effects of privacy concerns and type of device. In a field experiment 7200 participants were assigned to one of four personalization levels: non-personalized, slightly personalized, moderately personalized or highly
personalized. First, it was investigated whether these four levels of personalization induced different consumer behaviour and whether the type of device on which the ad was exposed had an impact on the relationship between personalization and consumer behaviour. The results indicated that personalization is always more effective than non-personalization of ads. Furthermore, the two highest levels of personalization generated the most favourable
outcomes compared to non-personalized and slightly personalized ads. The field experiment also showed that the type of device has an impact on the relationship between levels of personalization and ad effectiveness. Second, the moderating effect of privacy concerns was tested in an online experimental vignette study in which 153 participants were analysed. The results indicated that privacy concerns do moderate the relationship between levels of personalization and ad effectiveness.
Keywords: personalization, advertisements, privacy concerns, device, ad effectiveness, click-through, conversion
Table of Contents
1. Introduction p.6
2. Literature Review p.11
2.1 The Evolution of Advertising p.11
2.2 Personalization of Advertisements p.13
2.3 Advertising Effectiveness and Personalization p.14
2.4 Different Levels of Personalization p.18
2.5 Personalization-Privacy Paradox p.21
2.6 Different Types of Devices p.24
2.7 Review Summary p.25 3. Conceptual Framework p.26 4. Methodology p.27 4.1 Research Design p.27 4.2 Sampling Methodology p.29 4.2 Experimental Design p.30
4.3 Online Experiment – vignette study p.32
5. Results p.33
5.1 Data Preparation p.33
5.2 Data Analysis p.34
6. Discussion & Conclusion p.46
6.1 Discussion p.46
6.2 Managerial Implications p.50
6.3 Limitations and Future Research p.52
7. Conclusion p.54
8. Bibliography p.56
9. Appendix p.65
9.1 Email Advertisements p.65
1. Introduction
“Advertising is not dead, just bad advertising” was the bold statement of Brad Jakeman in Cannes (2016). Due to the consumer empowerment, information overload, and ad blocking software there is a pressing need of making advertising relevant again (Edwards, 2016; Eppler & Megis, 2004; Resnik & Albert, 2014). Advertising is still seen as a major contributor to the customer experience and as a determinant in staying on top of consumers’ minds. While banner advertising, sponsored search, email advertising and video pre-rolls remain dominant marketing tactics, the effectiveness of such marketing efforts has become in question because of staggeringly low response rates (Yoo, 2009). One of the biggest concerns for brands is how to gain attention within a rapidly moving technology landscape and counterattack against a perceived overload of advertising (Perlberg, 2016). Perceived relevance and customer experience have become essential to keep consumers satisfied and loyal. Fortunately, digital developments have also presented opportunities for brands to avoid bad advertising and make advertising relevant again in such a way that it contributes to the customer experience. One of these digital developments that is seen as a capability that will be most important to marketing in the future is personalization (Abramovich, 2015). Personalized advertising provides
opportunities to exploit advertising in an effective way (Davis, 2016). But despite the ability to track what type of advertising is successful and what not, it still seems unknown to which extent brands should personalize their advertisements to be most effective (Sable, 2017). Even while most researchers argue that personalization of ads results in higher ad effectiveness, Sable (2017) found that it is not effective to personalize too much. This is where privacy concerns come in. Many researchers argue that when ads are personalized, people are worried about their privacy and consequently the positive effect of personalization will backlash (Kalyanaraman & Sundar, 2006; Kang & Sundar, 2016; Sundar & Marathe, 2010). It appears
that personalization is still not fully understood. Because CMOs are spending big amounts of money when it comes to personalization (Diorio, 2016), it is very important to attain
knowledge concerning the level of personalization resulting in the best outcomes. When done right, advertising can even become a differentiator for brands while more than 90% of brands believe they are ineffective at advertising nowadays (Abramovich, 2015).
In the past, traditional media empowered companies to send one-way information to the customer. Advertising proved to be an important interaction with customers and prospects due to its positive influence on brand equity in the long run (Jedidi, Mela, & Gupta, 1999). However, the use of the online world as an advertising medium (Bright & Daugherty, 2012) has increased the competitive pressure in nearly every market while also empowering consumers (Davis, 2016). Now consumers are dictating the nature, extent, and context of marketing exchanges due to the interactivity and dynamic nature of the Web (Hanna, Rohm, & Crittenden, 2011; Bright & Daugherty, 2012). So while the growth of the Internet and technological innovations have provided endless opportunities for firms to interact with customers (Bright & Daugherty, 2012; Singh & Potdar, 2009), the Internet has simultaneously revealed itself to be an outlet where consumers are in more control of their media
consumption than ever before (Singh & Potdar, 2009).
Besides the consumer empowerment, consumers now have the possibility to use ad-blocking software to make their Internet surfing more pleasant (Verlegh, Fransen, & Krimani, 2015; Singh & Potdar, 2009). With the enormous growth of online advertising bombarded at customers every day, consumers’ overall response to these advertisements has become increasingly negative as they have become more savvy and sceptical about the values of such advertisements (Cho & Cheon, 2004; Coulter, Zaltman, & Coulter, 2001). Advertising messages may irritate consumers to the extent that it even makes them to block
(e.g. advertisements, promotions) is shown that does not correspond with their interests (Abramovich, 2015). Thus, along with the growth in online advertising comes an increasing need for more effective advertising (Danaher & Mullarkey, 2003).
As mentioned earlier, the consumer empowerment and adblocking software are reasons for brands to think about ways to advertise more effective. In addition, the
effectiveness of online advertising has even become more important due to the information overload consumers are exposed to in the Web 2.0 environment (Eppler & Mengis, 2004). According to Eppler and Mengis (2004) information overload can have impact on decision quality, decision time and on the actual number of information items that can be processed in a typical purchase situation. Since exposure to digital advertising leads to increased
advertisement awareness, brand awareness, purchase intention, and site visits (Manchanda, Dubé, Goh, & Chintagunta, 2006), ads could foster an easier prioritization of information. Considering the effectiveness, Ha and Janda (2013) argued that firms can add value by providing personalized information to simplify the decision process. Since people typically pay most attention to messages directly relevant to themselves, personalized communications have the potential to reduce information overload and better aid customer decisions (Ha & Janda, 2013).
Marketers are continuously exposed to the challenge of how to deliver the most relevant marketing message and experience to their customers that fits the needs and
behaviour of the consumer (Edwards, 2016). As just outlined, the empowerment of consumers and the information overload are advancing the importance of personalization of customer experience, including advertising. Ho and Tam (2005) agree that there is a strong need of personalization of advertising, in-store experiences and e-commerce to develop loyalty and retention and to attract new consumers, claiming that personalization has a high commercial potential in advertising (Ho & Tam, 2005). Sundar and Marathe (2010) found that
personalization delivers greater value to customers by understanding their needs and
continually monitoring their changing tastes and preferences at the individual consumer level. Other research in digital media shows that personalization can generate more favourable consumer responses because it increases the personal relevance of an ad (Anand & Shachar, 2009; Arora et al., 2008; Kalyanaraman & Sundar, 2006). Edwards (2016) has declared that personalization is the way of the future for creating customer experience in many aspects. Personalization is a marketing technique that could be used for many different types of advertising, including email advertising. While most people still prefer to receive advertising communication by email, it is sensible to make email advertising as relevant as possible in order to be most effective (Aufreiter, Boudet & Weng, 2014).
In today’s business world customers are expecting a personal and unique experience (Newman, 2015). Even while on the one hand most research in the field of online
personalization indicates that consumers have mostly positive reactions to personalization, on the other hand there are several adverse effects related to privacy concerns (Kalyanaraman & Sundar, 2006; Kang & Sundar, 2016; Sundar & Marathe, 2010). Although customers are demanding more customized services, users are increasingly concerned about privacy violations and how online vendors are using their information.
In summary, personalization is seen as a capability that will be essential to marketing in the future for advertising effectiveness, while dealing with consumer empowerment, ad blocking software and information overload. Marketers who are making use of
personalization in their customer experiences and who are able to quantify the improvement report a 19% uplift in sales (Abramovich, 2015). For email advertising, the use of
personalization even results in transaction rates that are six times higher (Abramovich, 2015). However, still more than 90% of the brands recognize they are ineffective at personalization
(Abramovich, 2015). This supposed ineffectiveness might be partially caused by occurring privacy concerns when making use of personalization (ibid).
So even while personalized advertising seems to be an effective and relevant technique for brands to shape consumers’ attitudes and behaviours positively, there are also some
adverse effects of personalized advertising, such as resistance due to privacy concerns. Existing research lacks understanding whether different degrees of personalization of advertisements have a different impact on consumer responses, which calls for extended research. To gain more extended knowledge about personalized advertising, the direct effects of different levels of personalization of advertisements on advertising effectiveness, and particularly click-through and conversion, should be examined. This leads to the following problem statement of this thesis: “What is the effect of different levels of personalization of an advertisement on advertising effectiveness and do privacy concerns and type of device on which the ads are exposed have an influence on this relationship?”
The answer to this research question has both academic and managerial impact. It provides new insights in personalized advertising. The study includes levels of
personalization, which have not been investigated before. To measure advertising effectiveness, consumer behaviour is tested instead of attitudes. Also research on email advertising is needed whereas 70% of the brands fail to use this type of advertising and because it has shown decreasing response rates (Abramovich, 2015). So there is strong need to make this type of advertising effective again. This paper contributes to marketing practices with insights for marketers of both big brands and advertising agencies. Is it really worth the money to invest in software that enables personalization of advertising and in collecting and keeping data? The findings support and guide efficient allocation of marketing budgets.
2. Literature review
The primary focus in this study is on the effect of different levels of personalization of email advertising on consumer behaviour - and specifically click through and conversion- in the digital environment of the Internet. In this section, the existing literature related to the
research question will be addressed in order to define the research topic more precisely and to set up hypotheses that will subsequently be tested. First, the literature about the evolution of advertising will be discussed. Thereafter, personalization of advertisements and its importance will be highlighted. Then the focus will shift towards advertising effectiveness, which
includes consumer behaviour, and particularly click through and conversion. What follows is a brief overview of different forms and levels of personalization of advertisements. This is followed by an extensive overview of privacy concerns that come along with the
personalization of advertisements and their impact on the effectiveness of personalized advertising. Then another factor that could affect the relation between personalization of email advertising and consumer behaviour will be discussed, namely the type of device on which the advertising is exposed. Finally, review summary of past research will be given.
2.1 The Evolution of Advertising
Nowadays traditional advertising is more and more replaced by the interactive way of advertising (Bezjian-Avery, Calder, & Iacobucci, 1998). Traditional advertising is characterized by linear exposure of advertising and passive demonstration of product information to consumers. In the early days consumers did not have a possibility to interact with companies, as one-way media – such as print and TV - was mainstream or even the only way of marketing communication (Bezjian-Avery et al., 1998). Now, in the Web 2.0
customer, provoking information from both parties, and trying to align interests and possibilities (ibid). Dynamic advertisements are more likely to produce positive consumer responses compared to static ads and are thus found to be more effective (Coyle & Thorson, 2001; Daugherty, 2009). The customer’s input enables the customization of subsequent information to the customers’ interests and counteracts irrelevant communications. However, due to the empowerment of customers (Singh & Potdar, 2009), they can choose what they would like to see and what they do not want to see.
Consumers are increasingly irritated by advertisements, which even make them to block ads. Brands and companies should overcome this form of consumer resistance.
According to Singh and Potdar (2009) it is not a problem to place ads within the pages as long as they do not become annoying. Consequently, consumer experience has gained a lot of attention within online marketing because in the Web 2.0 environment it is of crucial importance to find a way of advertising that enhances the customer experience and the effectiveness of the firm’s advertising and marketing dollar by making them to buy from brands.
Experience marketing is a new concept that is currently of interest to marketers. To differentiate a brand it is essential to understand how to provide tempting brand experiences for customers (Schmitt, 2010). The research on customer journeys of Rawson, Duncan and Jones (2013) even found that companies that have the competences to fully manage the entire experience have the potential to gain enormous rewards: greater customer satisfaction,
reduced churn and increased conversion - and subsequent revenues. Customers have the desire to interact with a brand at different touch points in order to generate a complete experience (Rawson, Duncan, & Jones, 2013). Advertising should contribute to the customer journey as well. Organizations should deliver differentiated advertising experiences to customers in order to even be able to compete (The Power of Personalization, 2015).
Emerging technologies that change the way audiences consume content are virtual assistants, chatbots, virtual reality, 360-degree live video, and personalization (Papandrea, 2017).
2.2 Personalization of Advertisements
Personalization is the biggest buzzword in marketing today and is seen as the biggest potential to make advertising relevant (Papandrea, 2017). The personalization of advertising is able to achieve the top marketing and customer experience goals, like increasing customer
satisfaction, building customer loyalty, and acquiring new customers (The Power of
Personalization, 2015).
Whether a marketing message of a brand is liked, is decided very quickly. When something relevant is provided, customers will be satisfied due to the contribution to their customer experience. When a brand misses the mark, customers will be gone (Gregg, Kalaoui, Maynes, & Schuler, 2016). Customers are confronted with a lot of bad advertising and so banner blindness and the use of ad blockers are problems marketers face daily. This can be overcome by the use of personalization, to advance advertising (Resnik & Albert, 2014). In order to diminish today’s marketing ineffectiveness and catch the consumer’s attention, personalized advertising is seen as a necessity (Köster, Rüth, Hamborg, & Kaspar, 2015; Wegert, 2015). Simply put, personalization can be viewed as excellent customer service – tailoring the experience and thus making it better for the customer (Singer, 2012).
Personalization is defined differently by previous researchers (Versanen, 2007). Personalization is often used as a synonym to customization although they are not the same concept (Arora et al., 2008). Doorn and Hoekstra (2013) refer to the concept of tailoring communication to an individual’s characteristics and/or interests or tastes as customization. However, Sundar and Maranthe (2010) refer to this concept as personalization.
defines what he or she wants and needs while personalization is provided by a brand based on a consumer’s previous behaviour or on demographic characteristics (Arora et al., 2008). Personalization can best be defined as a customer-oriented marketing strategy, aimed to deliver the right content to the right person at the right time in order to increase personal relevance of the communication and maximize immediate and future business opportunities (Ho & Tam, 2006).
In terms of advertising personalization is also defined at different levels. Goldfarb and Tucker (2011) propose that personalized advertising is matching an ad’s content to the
website content the customer visits. Personalized advertising is also described as adjusting the text of the advertising message according to a specific individual (Chen & Hsieh, 2012). Bang and Wojdynski (2016) define personalized advertising as advertising that includes information about the individual, such as demographic characteristics or personally
identifying information, and shopping-related information such as purchase habit or history and brand preference. The definition of Bang and Wojdynski (2016) is used in this study.
2.3 Advertising Effectiveness and Personalization
There has been done loads of research in the area of advertising since companies spend a lot of money on advertising and so extensive knowledge in this field is of high importance (Luo & Donthu, 2001). Wanaker became famous in the area of advertising due to his quote “Half
the money I spend on advertising is wasted; the trouble is I don’t know which half”. Factors
that might have an impact on online advertising effectiveness is therefore one of the main subjects within this area of research (Danaher & Mullarkey, 2003; Robinson, Wysocka, & Hand, 2007; Park, Levine, Kingsley Westerman, Orfgen, & Foregger, 2007; Haans, Raassens, & van Hout, 2013). It appears that due to the digital data collecting, there is knowledge about which advertising is effective and which is not (Sable, 2017). The digital data collection
allows companies to transform a brand’s acquisition costs and conversion efficiency
exponentially, by exploiting the endless stream of data that all of us shed every second of the day (Sable, 2017). Brands now can spend their money with the calming and profitable knowledge that it is no longer wasted by ineffective advertising but rather directed in a meaningful and valuable digital stream towards customers who are waiting for the brand’s messages and eager to interact and engage with the highly targeted and personalized messages (ibid). But is this totally true? We know personalized messages can be of value, however still unknown is to what extent.
Advertising effectiveness is in most research seen as the outcome of advertisements, measured in either consumer attitudes or consumer behaviour (Lutz, MacKenzie, & Belch, 1986). According to Berry, Carbone and Haeckel (2002) these attitudes and behaviours – and so advertising effectiveness - can be positively influenced by value creation in the form of experience from the brand to the consumers. Mollen and Wilson (2010) found that
engagement and involvement are two constructs to assess consumer experience. Previous research argues that personalization increases advertising relevance – and thereby
involvement and engagement – and so consumer experience. However, most of the previous research has just examined the effect of personalization of ads on consumer attitudes – involvement and engagement – and not on consumer behaviour (Xu, 2007; De Keyzer, Dens, & De Pelsmacker, 2015; Tsang, Ho, & Liang, 2004).
Most studies assume that the impact of advertising can be measured along the following elements: cognitive, affective, and conative (Hutchinson & Alba, 1991; Baker & Churchill, 1977). Those elements are also related to involvement, engagement and relevance, which in turn are achieved through personalization among others. The ability of an ad to attract attention and eventually generate product knowledge is determined by cognitive measures. According to Pieters, Warlop and Wedel (2002) attention to advertisements could
be gained by developing original and familiar ads. Intrusiveness of ads is also being
mentioned as a determinant for getting attention to advertisements (Mehta, 2000). However, Chang, Rizal and Amin (2013) do not agree with this finding as they found that intrusiveness hampers the effectiveness of advertising. Another interesting finding is that Bang and
Wojdinski discovered (2016) that personalized advertisements attract significantly longer and more attention than non-personalized ads, indicating the strong attention-grabbing effect of personalization. But it is generally accepted that when a message is perceived as more personally relevant, it does not only lead to greater attention but also to greater elaboration, message processing, and ultimately persuasion (Noar, Harington, & Aldrich, 2009).
Affective measures are used to identify established or created attitudes from advertising stimuli, and attitude toward the brand serves as a regularly used measure of effectiveness (MacKenzie & Lutz, 1989). When an advertisement is high in personal relevance, the quality or cogency of the arguments presented in the ad has a much greater impact on attitudes towards the advertised product (Cacioppo & Petty, 1981; Haans, Raasens, & van Hout, 2013; Lohtia, Donthu, & Hershberger, 2003; Noar, Harington, & Aldrich, 2009). If brands could change the attitude of customers towards advertising – i.e. making
advertisements more relevant to their personal needs, wants and characteristics– the customer could be interested in the ads and not experiencing them as annoying (Singh & Potdar, 2009; Berry et al., 2002). This will increase the effectiveness of advertising, even at the behavioural level. Xu (2007) agrees with this statement as he found that personalization of advertisements is one of the most important factors in affecting consumers’ attitude toward an advertisement, concluding that there is a direct relationship between consumer attitudes and consumer intentions so we may expect that personalization also has either an indirect or a direct effect on conative measures.
Even while attitudinal measures are frequently used, the most widely used conative measure in advertising effectiveness research is intention to purchase (Li, Daugherty, & Biocca, 2013). The conative way of measuring advertising effectiveness could give more solid results for business practice. Conative measures are used to anticipate a response behaviour resulting from an advertising stimulus. They normally encompass some type of behaviour intention, such as searching for supplementary information or conversion (Brucks, 1985; Hoch & Ha, 1986). Behavioural intentions and actual purchase behaviours are the key to successful marketing efforts. So marketers need either to focus on factors that directly affect consumer behaviour or on ways to positively influence customer attitudes towards their advertisements, as this will ultimately affect behavioural intentions (Ha & Janda, 2013; Bang & Wojdynski, 2016). The level of involvement, engagement, relevance, and commitment directs consumer behaviour (Tucker, 2011). According to Tucker (2011) personalization increases the level of involvement and engagement of an advertisement, and so it influences consumer behaviour. Several studies (Wang 2006; Petty, Cacioppo, & Schumann, 1983) found that those two are determinants of the strengths of advertising effectiveness.
Previous research in digital environments has shown that personalization improves advertising effectiveness – mostly in terms of attitudinal measures. Personalized messages are generally more effective than non-personalized messages in terms of being more memorable, more likeable, and sparking behavioural change (Howard & Kerin, 2004). The effectiveness of advertising depends on the goals a company or brand wants to achieve with the
advertisements (e.g. brand awareness, advertising awareness, or conversion). However, the final goal of a company or brand is sales. In terms of advertising effectiveness, it would then be more preferable to measure consumer behaviour, such as click-through or sales conversion. Even while most research has argued that personalized advertising is more effective than non-personalized advertising, most research has relied on survey data and not on experimental
analysis. De Keyzer, Dens and De Pelsmacker (2015) argue that future research in
personalized ads should consider including measures of actual behaviour such as clicking. In former research, actual clicking behaviour has not often been measured. But based on
previous research we expect the following:
H1. Personalization of advertisements has a positive effect on click-through rate and
subsequently on sales conversion.
2.4 Different levels of Personalization
Previous research distinguishes different levels of personalization of advertisements. According to De Keyzer, Dens and De Pelsmacker (2015) advertising can be placed on a continuum ranging from no personalization, to rather general or slight personalization, to full personalization. Non-targeted, generic online advertising of a random product or service can be regarded as non-personalized advertising. Slightly personalized online advertising is defined as a generic advertisement based on previous surfing behaviour shown on a different website, with some unidentifiable information included. An example of highly personalized advertising would be an ad of the exact same product an individual has just viewed, placed on a different website with more personally identifiable information included.Chellappa and Sin (2005) explain different types of information that can be included when personalizing ads: anonymous information (page visits; browsing behaviour), personally unidentifiable information (birth, age, gender, occupation, income, ZIP code, hobbies), and personally identifiable information (name, email, phone number, etc..).
Baestaens (2017) also came up with three types of targeted advertising. The first type is contextual advertising. This type of advertising uses the surfing behaviour of consumers. Data from search behaviour and websites visited are a good indicator to get to know the
interests and desires of an individual. Geotargeting is another type of targeted advertising. This type of advertising focuses on a specific market, based on the geographic location of the target group. The last type Baestaens (2017) mentioned is retargeting. This form of targeted advertising shows products or services an individual has looked for in the past.
Evidence of the effects of personalization on advertising-related outcomes has been mixed (Yu & Cude, 2009). Every level of personalization has a different balance between the perceived value/relevance of personalization and the negative effects of privacy concerns: the personalization-privacy paradox (Aguirre, Roggeveen , Grewal, & Wetzels, 2016). While personalized advertising may have broad overall benefits to advertisers, its success with individual consumers may be moderated by other factors, such as the extent to which the mechanism of personalization raises privacy concerns among message recipients.
Some researchers have found that personalized messages attract users’ attention and increase message receivers’ attitude toward the message or even toward the medium because of its perceived relevance to the consumers’ self. The positive effect of perceived
personalization can almost fully be attributed to the mediating role of personal relevance. When an ad is perceived as personally relevant, attitude toward the brand and click intention will improve (De Keyzer et al., 2015). Rimer and Kreuter (2006) even argue that greater perceived personal relevance of the personalized ad is the main driver for personalized messages to generate more behavioural changes. According to Tucker (2014) it is especially important that consumers recognize personalized advertisements as personalized. Specifically, Howard and Kerin (2004) discovered that when an ad contained a viewer’s first name, an example of the high-level of personalization, the viewer was likely to have higher purchase intention for the product recommended in the ad due to the fact that the ad will become more emotionally salient which results from meaningfulness.
At the same time, brand managers should exert personalization efforts with care, because research indicates that when consumers are aware of personalization techniques, many consider this approach as a violation of their privacy (Turow et al., 2009). Privacy concerns are often costly for firms because they heighten customers’ risk perceptions and decrease their trust, which reduces their willingness to engage (Van Slyke et al., 2006). Consequently, when the perceived privacy violation is very high the effect of personalization could even backfire. These reactions reduce the overall effectiveness of advertising (Goldfarb & Tucker, 2011). According to Keyzer et al. (2015) it is likely that only the higher degrees of personalization (full personalization) would lead to negative consumer responses. The balance between the value of personalization and privacy concerns determines consumers’ responses, including behavioural ones.
In many previous studies just one level of personalization has been tested, considering distinct types of personalized advertising in isolation. It could be very interesting to combine these types of personalization to investigate the influence of the personalization on ad
effectiveness (Baestens, 2017; Chellappa & Sin, 2005; De Keyzer, Dens, & de Pelsmacker, 2015). Arora et al. (2008) argue that extended research is needed to determine the appropriate degree on the personalization continuum to get the optimal effect. De Keyzer et al. (2015) also state that future research should examine the different degrees of personalization to test whether they would have a different impact on consumer responses.
To summarize, personalization can both enhance and diminish consumer engagement with the firm and so can either support positive or negative consumer responses; it may raise privacy concerns because consumers worry about how their data are collected and used, while it can also benefit them in meaningful ways. The optimal balance of the value of
personalization and privacy concerns must be determined (Aguirre, Roggeveen, Grewal, & Wetzels, 2016). The results of Chellappa and Sin (2005) indicate that the consumers’ value of
personalization is almost two times more influential than the consumers’ concern for privacy in determining usage of personalization services, we propose that while vendors should not ignore privacy concerns, they are sure to reap benefits when the right level of personalization of an ad is applied for the right individual. So at what level of personalization is the perceived value of personalization more dominant than the perceived privacy violation? It seems unclear to what extent an ad must be personalized to attain the optimal effect. According to previous research we expect the following:
H2: Moderately personalized ads have a higher click-through rate and conversion rate than the slightly personalized ads; moderately personalized ads are more effective than slightly personalized ads.
H3: The highly personalized ads have a lower click-through rate and conversion rate than the moderately personalized ads; highly personalized ads are less effective than moderately personalized ads.
H4: The highly personalized ads have a lower click-through rate and conversion rate than slightly personalized ads; highly personalized ads are less effective than slightly personalized ads.
2.5 Personalization-Privacy Paradox
In the previous section was outlined that different levels of personalization should be studied to examine at which level the privacy concerns are least dominant relatively to the perceived value of personalization, as these constructs give insights in online consumer behaviour in the context of personalization (Chellappa & Sin, 2005) As shown, research has been widely published on online advertising relevance and personalization efforts to accomplish such. Subsequently, previous research has found that privacy concerns in general moderate the
relationship between personalization and consumer responses to ads (Goldfarb & Tucker, 2010). Still, researchers are not unanimous about the effectiveness of personalized advertising due to privacy concerns, which consists of consumers’ perceptions regarding exchange
relationships with marketers that gather and use personal information and the resulting behaviours (Phelps et al., 2001).The controversy across different research about the
effectiveness of personalization, taking privacy concerns into account, can be explained by a lack of knowledge regarding the optimal balance of the personalization-privacy paradox at different levels of personalization (Aguirre et al., 2016).
Research in digital media shows that personalization can generate more favourable consumer responses because it increases the personal relevance of an ad – due to increased engagement and involvement (Anand & Schachar, 2009; Arora et al., 2008; Kalyanaraman & Sundar, 2006). Consumers might see personalized ad content as more tempting and connected to their interests (Anand & Schachar, 2009). Marshall (2014) stated that even when
consumers fear the risks to their privacy of such personalized ads, some of them still enjoy getting the benefit of tailored ad messages. Baek and Morimoto (2012) argued that just highly personalized ads could increase consumer fears about losing control over personally sensitive information resulting in privacy concerns.
Privacy concerns could arguably lead to reactance, which results in resisting the ad’s appeal (White et al., 2008). Reactance is a motivational state in which consumers resist something they find intimidating by behaving in the opposite way to the one intended (Brehm, 1966; Clee & Wicklund, 1980). Reactance to personalized advertising is greatest when the information used is more unique, but is always present in the situation of
personalization (White et al., 2008). When privacy concerns are more prominent, consumers are more likely to have a prevention focus instead of a promotion focus (Goldfarb & Tucker, 2010). There is evidence that consumers are concerned that the information being used in ads
is too personal to be used in an ad without a corresponding sense of control over their data, which refers to privacy concerns (Tucker, 2013). Consumers also may see personalization as “not only creepy, but off-putting” if they feel that the firm has violated their privacy (Stone, 2010). There is evidence that this feeling of violated privacy could have negative effects on the customer behaviour in terms of purchase intention, possibly implying that customer appreciation of the convenience and uniqueness of targeted, personalized ads is moderated by privacy concerns (Goldfarb & Tucker, 2010).
However it still seems unclear when personalization is perceived as relevant and when it is seen as creepy and off-putting. Previous research claims on one hand that personalization of advertisements is valued by consumers due to its relevance. This relevance is a
consequence from the involvement and engagement that comes with the personalized ads. On the other hand, currently there is no evidence that suggests to what extent consumers find online personalization useful. Neither there is understanding of how consumers’ concern for privacy will affect their behaviours even if they find personalization to be of value (ibid). So there is an on-going discussion whether privacy concerns in every situation moderate the relationship between personalization of ads and consumer responses. As companies are more and more investing their resources in personalization of ads, it is of importance to consider that the outcomes of investments in online personalization may be rigorously weakened if consumers do not use these services due to privacy concerns, and to understand and evaluate the different values consumers may place in enjoying various types of personalization (ibid).
H5: Privacy concerns moderate the relationship between different levels of personalization of
2.6 Different Types of Devices
Nowadays firms exist that track consumer behaviour across different devices to be able to give strategic advice on how to design the customer experience. One of the waves feeding the fast growth of cross-platform advertising is the growing popularity of the mobile device (Tanner, 2015). The penetration rate (rate of circulation of a product in its market) of mobile devices is rising (Xu, 2007). While the number of mobile users is now already higher than the number of desktop users, the advertising budget is still mostly spent on desktop advertising.
Customer behaviour differs on different devices. Even while in many countries, people now use mobile devices for more than 50% of their searches, people are still more likely to buy on desktop (Tanner, 2015). These trends have resulted in the increasing use of handheld devices as advertising medium next to the use of desktop as advertising medium (Xu, 2007). Research revealed that consumers react differently to advertising on different devices. Tsang, Ho, and Liang (2004) found that consumers generally have negative attitudes toward mobile advertising. These negative attitudes – and subsequent behaviour - can be explained by the personal and intimate nature of mobile phones (ibid). Xu (2007) found that personalization of advertising is one of the most crucial factors in affecting consumers’ attitude toward mobile advertising. Research of Metrixlab (2016) confirms this as they found that mobile banners are more effective compared to desktop banners in terms of ad recognition and message recall. Even while Tsang, Ho, and Liang (2004) argue that the intimate nature of the mobile device results in the negative attitudes of consumers towards mobile advertising, the research of Metrixlab (2016) claims that this intimate nature together with the immediacy results in the stronger breakthrough of mobile ads compared to desktop. Moreover, the research of Metrixlab (2016) shows that mobile banners tend to take up a larger share of the screen, which would result in more favourable outcomes. Contrarily they found that the likeability, credibility and impact on purchase intent is weaker for mobile ads than for ads exposed on
desktop. They propose that these outcomes could partially result from the low expertise of brands in mobile, not recognizing the high importance of mobile yet.
In summary, it is of relevance to examine whether the effect of personalization is different for the different devices and subsequently if the type of device has an effect on the relationship between levels of personalization and ad effectiveness. Known is that mobile is key when it comes to targeting the millennial market, as more than 40% of this group
researched their last purchase using a smartphone (Metrixlab, 2016). However desktop is also still high in popularity when it comes to advertising. According to previous literature mobile ads generate different consumer behaviour compared to ads exposed on desktop. This is mainly due to the different characteristics of the devices and its target groups. However, no previous research has examined the possible moderating effect of the type of device on the relationship between personalization and advertising effectiveness. Based on the existing literature, the following is expected:
H6: The type of device on which the advertisement is exposed will moderate the relationship between the level of advertisement personalization and the advertising effectiveness.
2.7 Review Summary
To summarize, a substantial amount of research has been done in the area of advertising effectiveness. But still, advertising relevance is lacking which results in staggering response rates from consumers to advertisements. Still, companies are spending loads of money on advertising. There is a growing desire for brands to find a way to make advertising effective again. The personalization of ads is becoming a hot topic within advertising. Expected is that personalization increases both involvement and engagement which in turn results in higher consumer experience which is expected to generate favourable consumer behaviour – higher
click through and conversion. However, taken privacy concerns into account, it is
questionable if the expected positive effects of personalized mail advertising on consumer behaviour will not be overruled by the negative effects of privacy concerns that come along with personalization of ads. So there is high need for marketers to understand which level of personalization of ads will result in the best consumer response for customers.
3. Conceptual Framework
Expected is a positive relationship between personalization of email advertisements and advertising effectiveness in terms of click-through and conversion. Personalization of email ads has a positive relation on customer experience - which is determined by involvement and engagement – and a higher customer experience leads to a higher click-through and
conversion – and so higher ad effectiveness. Privacy concerns will moderate this relationship as the balance between perceived relevance and privacy concerns should be optimal to get the best effect. The main relationship is also influenced by type of device the ad is exposed on.
4. Methodology
In this section the research design through which the hypotheses were tested is discussed. This is followed by a description of the sampling methodology of this study. Finally, the specific design of the methods used in this research, is highlighted.
4.1 Research Design
The major objective of this study was to examine the effect of different levels of ad
personalization on advertising effectiveness. Personalization can be done for many different types of advertising such as Search Engine Optimization, Search Engine Advertising, Price Comparison, email marketing, text messaging, online video advertising, and banner
advertising (Breuer & Brettel, 2012; Bang & Wojdynski, 2016). In this study we tested the effect of personalization for email advertising. Email is one of the best ways to reach your consumers. Even while it is one of the most budget-friendly forms of marketing, a lot of companies still do not get the optimal results by using this form of advertising. 91% of consumers check their email daily. Thereby, the vast majority of adults prefer to receive advertising communication via email (Aufreiter, Boudet, & Weng, 2014). Personalized emails deliver six times higher conversion rates, but 70% of brands fail in using them (Abramovich, 2015).A Forrester Research found that most people unsubscribe from email lists due to irrelevance. Another reason for using email advertising in this study is that it is not allowed to use the highest level of personalization, which includes personally identifiable information, in for example banner advertising yet. By focussing on email advertising, all levels of
personalization could be tested.
A deductive approach was applied to explain the causal relationship between
field experiment and an online experimental vignette study, both quantitative research methods, was used to collect data. The application of an experiment is the best method to research causal effects. As mentioned before, the independent variable in this research was the personalization level of advertisements, distinguishing the levels non-personalized, slightly personalized, moderately personalized, and highly personalized. The dependent variable was the advertising effectiveness, measured by click through and sales conversion. The field experiment gave the opportunity to measure the effect of the level of ad
personalization on ad effectiveness. According to Boeije, ‘t Hard and Hox (2009) the
variation in the dependent variable can be attributed completely to changes in the independent variable in this case.
This study also analysed the moderating effects of privacy concerns and type of device on the relationship between the level of personalization of advertisements and advertising effectiveness. The field experiment provided the data to analyse the moderating effect of the type of device the advertisement was exposed on, on the main relationship. The additional online experimental vignette study was used to obtain data about privacy concerns and its effect on the relationship between the personalization of ads and ad effectiveness. A vignette study uses a short description or situation that is usually shown to respondents within surveys in order to elicit their judgements about the scenarios (Atzmüller & Steiner, 2010). The advantages that come along with this online experimental vignette study in form of a survey are the fast acquirement of respondents and that participants can fill in the survey in a natural environment, which results in the fact that there are no laboratorial effects on the outcome. The influence of the researcher is minimal.
4.2 Sampling Methodology
The sample group of the field experiment was the customer database of Vinylify. Vinylify’s customer database exists of people who have created a vinyl record in the past and people who have bought a vinyl record in the past. The email advertisement was send to the total customer database of Vinylify.
Ideally, the questions regarding privacy concerns should have been answered by this sample group. This, however, could not be executed. Vinylify did not allow for this part of the study, recognizing that the questionnaire itself could raise privacy concerns in its customer base. Therefore, we adopted an alternative approach to assess the moderating effect of privacy concerns on the main relationship. An online experimental vignette study was used, because this method gave the possibility to describe the situation of the field experiment. The
respondents could elicit their judgements accordingly. The use of the vignette study enabled the connection between the data of the online experimental vignette study and the data of the field experiment.
Thus, the field experiment and the online experimental vignette study made use of different samples. The online experimental vignette study in form of a survey was held among a group of random people. The respondents of this online experiment were reached through non-probability convenience sampling (Dooley, 2001), through personal contacts, e-mail, and via social media. There were as many respondents collected as possible, which contributed to large analysis of the data and to the chance of having a representative sample. A
representative sample provides the ability to generalize conclusions over the population (Dooley, 2001).
4.3 Experimental Design
First of all a field experiment was conducted. This experiment examined how many customers opened the mail, clicked on the link, or even bought the product advertised for. Since the purpose was to determine whether different levels of personalization predict customer
behaviours at a statistically significant level, a quantitative research in form of an experiment, was used. A field experiment was most suitable in this situation to overcome the privacy paradox. In cases where privacy comes in, the behaviour of people is different from people’s intentions (Norberg, Horne, & Horne, 2007). The design to test the effect of different levels of personalization of advertisements on advertising effectiveness can be characterized as a single factor between-subject design. The field experiment was conducted with data from the
company “Vinylify”, which sells customized vinyl records. An intervention in the real world was experimentally examined. The customers in Vinylify’s database were randomly allocated to four levels of personalization. The levels applied in this study are based on the levels used in the studies of Baestens (2017) and De Keyzer, Dens and De Pelsmacker (2016).
The following levels of personalization were consequently distinguished: • Non-personalized, which is a generic advertisement without any sign of
personalization.
• Slightly personalized, which is a generic advertisement with anonymous information, which contains information about past surfing behaviour (contextual).
• Moderately personalized, which is a contextual ad with unidentifiable information such as the age or geographical location (geotargeted).
• Highly personalized, which is a contextual, geotargeted ad that contains personally identifiable information, such as the name of the individual.
According to these different levels of personalization, and the difference between customers that have created and have bought a vinyl in the past, seven different emails were created (see Appendix):
Non-personalized mail: A very generic email advertisement without any signs of
personalization.
Slightly personalized mail: This email advertisement included a sentence about past
surfing behaviour. So for customers who had bought a vinyl record in the past, the sentence “because you have bought a vinyl record in the past …” was included. For customers who had created a vinyl record in the past the sentence “because you have once started creating your own vinyl…” was included.
Moderately personalized mail: This mail advertisement both covered the past surfing
behaviour sentence as well as the place of residence of that specific customer. Again, different mails for the customers who bought and for the ones who created a vinyl record in the past, were set up.
Highly personalized mail: This mail advertisement covered the past surfing behaviour
sentence, the place of residence and the first name of the customer. Again two different mails were used for the customers who bought and who created a vinyl record in the past.
As indicated before, the level of personalization within the advertisement was varied to test its effect on the dependent variable, advertising effectiveness. In this study advertising
effectiveness was measured by click behaviour (opening rate and click through on the link) and conversion rates of the customers. The click and purchase behaviour was tracked through Mailchimp, which is an email marketing service. Also the device (either mobile or desktop)
on which actions were taken was tracked in the field experiment. Again, this was done through Mailchimp, providing the ability to track which action was taken on which device.
4.4 Online Experiment – Vignette Study
The additional online experimental vignette study, in the form of a survey, provided explanatory principles to make sense of the observations in the field experiment. A within-subject design was used for this experiment. The online experiment enabled to assess the effect of privacy concerns on the relation between different levels of personalization and ad effectiveness. The exact goal of the research was not pronounced before starting the online questionnaire, because this could lead to social desirable answers of respondents (DeMaio, 1984).
The questionnaire used in this study was of quantitative nature. The answers to the questions of the survey gave insight in the privacy concerns of customers in general and in their behavioural intentions (not open, open, click, buy) when they would get the four different mails in which the four levels of personalization were expressed. The questions indicating the degree of concern about privacy of a person were collected from previous research within this area (Chellappa & Sin, 2005; Malhotra, Kim & Agarwal, 2004). As such, the construct “privacy concerns” was already validated by previous research. An example of these questions is: “indicate to what extent you are concerned about threats to your personal
privacy”. Respondents could answer according to a 7-point Likert scale, ranging from
"strongly disagree" to "strongly agree". This scale had also been used in the studies of Chelleppa and Sin (2005) and Malhotra et al. (2004) about privacy concerns.
For the questions that are related to the data of the field experiment, a vignette of the field experiment was set up. These questions were included to be able to connect the results of the survey to the data gathered from the experiment. A hypothetical situation was described
about the Vinylify mail advertisements. Respondents could answer on a 4-point Likert scale ranging from “I would not open the email” to “I would fully finish the transaction and buy the product” according to the different mails received.
Before the distribution of the survey, we conducted a pre-test prior to the main study to ensure the quality of the questionnaire. The pre-test was executed with four respondents. The goal of this pilot study was to check whether the survey was working well and the questions were understandable, and whether improvements could be applied to the questionnaire (Saunders, Lewis, & Thornhill, 2012).
5. Results
This section presents the results of the data that was collected and analysed. The section starts with an explanation of how the data was prepared for data analysis. In the second part of this section the data is analysed by different means to test our hypotheses and research questions.
5.1 Data Preparation
The mail advertisements in the field experiment were send to all 7403 people in the Vinylify customer database. 7200 customers received the mail ad successfully; only these were taken into account in the analysis. 203 of the mail ads were bounced. Those 203 bounced mails were indicated as missing values in the analysis. So data was reported about 7200 customers who received the mail ad successfully.
Through the survey collection method 211 respondents took part in the survey. However not all 211 respondents fully finished the survey. The 58 respondents who did not fully finish the survey were excluded from the analysis and designated as missing data.
Ultimately data from 153 respondents was taken into the analysis about privacy affecting the relation between ad personalization and ad effectiveness. This is a representative amount of respondents if an online research is conducted (Dooley, 2001). Missing values in the data were excluded from analysis. SPSS was used to analyse the data.
5.2 Data Analysis
After preparing the data, it can be used for further analysis. The proposed hypotheses will be tested and consequently the research questions will be answered.
5.2.1 Testing hypothesis 1
First of all a general overview of how the customers responded to the email advertisement was created (see figure 2).
Figure 2. Customer behaviour according to the email ads
Of the 7200 customers who successfully received the email advertisement 1106 customers (15,4%) opened the mail. Of these 1106 customers, 133 persons clicked on the link (1,8% of 7200). Of these persons, only 49 customers finished the full transaction by also buying the
49 customers finished full transaction
by also buying the product
133 customers clicked the link 1106 customers opened the mail 7200 customers received mail successfully
product (0,7% of 7200). This click and buying behaviour is dependent on the different levels of personalization.
After the general overview of the frequencies, the hypotheses about the effect of different levels of personalization of advertisements on advertising effectiveness (opening rate/click on link/purchase behaviour) were tested. To test these hypotheses, firstly a crosstab was conducted to give an idea of the distribution of click- and purchase behaviour over the different levels of personalization.
Table 1. Crosstabs effect of different levels of personalization on click and conversion rates
Non-pers Slightly pers Moderately pers Highly pers Pearson χ2 df Sig. Opened_mail 16,1%a 13,1%b 16,2%a 16,1%a 9,241 3 0,026
Clicked_link 0,7%a 0,8%a 3,0%b 2,9%b 41,168 3 0,000
Bought_product 0,4%a 0,3%a 0,9%b 1,1%a,b 11,142 3 0,011
Table 1 indicates that for the opening rate of the email advertisement, the level of
personalization of advertisements did not make any difference. It shows that only the opening rate for the slightly personalized mail ad was significantly less than for the other levels. For the click behaviour on the link, table 1 indicates that for the two highest levels of
personalization (moderate and high personalization), the click through was significantly higher compared to the non-personalized and slightly personalized ads. Similarly, table 1 demonstrates that conversion rates for people who received the moderately or highly personalized mail ad were significantly higher than for the customers who received the non-personalized and slightly non-personalized email advertisements. So based on the outcomes of the crosstabs H1 is supported; personalized ads are more effective than non-personalized ads.
To give meaning to the outcomes of the crosstabs and to interpret the results, a binary logistic regression was conducted. Logistic regression is a multiple regression with an
(Field, 2012), and thus appropriate in our case. Since the dependent variable is of binary nature (i.e. has two categories – action not taken/action taken), a binary logistic regression was used to develop the model (ibid). Making use of the outcomes of the binary logistic regression, the interpretation is done by examining the Odds Ratio, which is the exponential of B. The O.R. indicates whether a particular exposure has an effect on a particular outcome (Field, 2012).
Before analysing the results of the binary logistic regression it is of importance to check the outcomes of the Omnibus Tests of Model Coefficients and of the Hosmer-Lemeshow Test. The Omnibus Test of Model Coefficients tests statistically whether the explained variance in a dataset is significantly greater than the unexplained variance (Field, 2012). The significance value should be less than 0,05. This indicates that the model used in the analysis outperforms the null model. The Hosmer-Lemeshow Goodness of Fit Test checks whether the data do not conflict with assumptions made by the model (Field, 2012). This should be checked to ascertain that the model assumed is correctly specified before
conclusions are drawn. The Hosmer-Lemeshow statistic should not be significant, indicating that the model adequately fits the data.
After the binary logistic regression was conducted, first the Omnibus Tests of Model Coefficients was checked for hypothesis 1. The significant χ2 value of 9,507 (p=0,023) indicated that the level of personalization explained the opening rate of the mail ad. The outcome of the Hosmer and Lemeshow Test with a χ2 value of 0,000 (p=1,000) showed that the model fitted the data well. After the appropriateness of the model was confirmed, the outcomes of the binary logistic regression were analysed.
Table 2. Binary logistic regression – personalization levels*opened_mail (compared to
non-personalized)
B S.E. Wald Df Sig. O.R.
Slightly_personalized -0,236 0,095 6,206 1 0,013 0,790 Moderately_personalized 0,010 0,091 0,13 1 0,909 1,010
Highly_personalized -0,001 0,091 0,000 1 0,993 0,999
Constant -1,653 0,064 663,162 1 0,000 0,191
For the email opening rate, the level of personalization did not show much influence. The only significant finding was that the opening rate of the slightly personalized mail ad was significant less than the opening rate of the non-personalized, moderately personalized and the highly personalized mail ads.
After doing the binary logistic regression for the opening rate according to different levels of personalization, a binary logistic regression was conducted for the click-through across the different levels of personalization of the email advertisements. Table 1 already demonstrated that respondents clicked significantly more often for the moderately and the highly personalized mail ads than for non-personalized and the slightly personalized mail ads. The χ2 value indicated that the click behaviour on the link depends on the personalization level as well (p=0,000).
For the binary logistic regression the Omnibus Test and the Hosmer Lemeshow test were conducted again. The significant χ2 of 51,588 (p=0,000) of the Omnibus Test indicated that the level of personalization explained the click rate on the link in the mail ad. The χ2 value of the Hosmer and Lemeshow Test of 0,000 (p=1,000) indicated that the model used fitted the data.
Table 3. Binary logistic regression – personalization levels*clicked_link (compared to
non-personalized)
B S.E. Wald Df Sig. O.R.
Slightly_Personalized 0,221 0,389 0,325 1 0,569 1,248
Moderately_Personalized 1,526 0,321 22,616 1 0,000 4,600
Highly_Personalized 1,492 0,322 21,426 1 0,000 4,445
Constant -5,003 0,290 298,403 1 0,000 0,007
Table 3 confirms the results of the crosstabs. Only for the moderate and high levels of personalization the click through was significantly higher than for the non-personalized and the slightly personalized mail ads. The Odds Ratio of 4,600 indicates that people that received the moderately personalized ad were 4,6 times (p=0,000) more likely to click on the link than people who received the non-personalized ad. The O.R. of 4,445 indicates that people that received the highly personalized ad were 4,5 times (p=0,000) more likely to click on the link than people who received the non-personalized ad.
Finally, table 1 showed that significantly more people that were exposed to moderately or highly personalized mail ad bought the product in comparison to people that were exposed to the non-personalized or slightly-personalized mail ad. This confirms that also the purchase behaviour is dependent on the level of personalization (p=0,029).
For the binary logistic regression, that tested the effect of the level of personalization on the conversion, again the Omnibus Test and the Hosmer Lemeshow test were conducted. The significant χ2 of 11,560 (p=0,009) of the Omnibus Test indicated that the level of
personalization explained the purchase behaviour of customers who received the mail ad. The χ2 value of the Hosmer and Lemeshow Test of 0,000 (p=1,000) indicated that the model used fits the data.
Table 4. Binary logistic regression – personalization levels*Bought_Product (compared to
non-personalized)
B S.E. Wald Df Sig. O.R.
Slightly_Personalized -0,158 0,557 0,080 1 0,777 0,854 Moderately_Personalized 0,891 0,450 3,917 1 0,048 2,438
Highly_Personalized 1,008 0,443 5,169 1 0,023 2,740
Constant -5,545 0,379 214,405 1 0,000 0,004
Table 4 confirms the results of the crosstabs. The customers exposed to the
moderately (p=0,048) and highly (0,023) personalized ads are more likely to buy the product than the customers exposed to the non-personalized and slightly personalized ads. The O.R. of 2,438 indicates that people who received the moderately personalized ad are 2,4 times more likely to buy the product than people who received the non-personalized ad. The O.R. of 2,740 indicates that people who received the highly personalized ad are 2,7 times more likely to buy the product than people who received the non-personalized ad.
For both the click and buying behaviour of customers the results are significant better for moderately and highly personalized ads compared to non-personalized ads. Consequently, H1 is accepted. Personalized ads are more effective than non-personalized ads.
5.2.2 Testing hypotheses 2, 3, and 4
The logistic regression is also able to answer the hypotheses 2, 3, and 4, because of its ability to compare the outcomes of the four levels of personalization.
The second hypothesis is that the moderately personalized ad result in higher click through and conversion, and consequently in higher ad effectiveness compared to slightly personalized ads. The outcome of the binary logistic regression (table 5) indicated that the opening rate for moderately personalized ads is significantly higher than for slightly
personalized ads (O.R.=1,279; p=0,009). Also the click through (O.R.=3,686; p=0,000) and the conversion (O.R.=2,856; p=0,028) are both significantly higher for moderately
personalized ads than for slightly personalized ads. The O.R. of the click through tells that people who received the moderately personalized ad were 3,686 times more likely to click the link. Furthermore, the people who received the moderately personalized ad were 2,856 times more likely to buy the product compared to people who received the slightly personalized ad. These results support H2; moderately personalized ads are more effective than slightly personalized ads.
Table 5. Binary logistic regression moderately personalized compared to slightly personalized
B S.E. Wald Df `Sig. O.R. Omnibus
χ2 Sig. H&L χ2 Sig. Opened 0,246 0,095 6,787 1 0,009 1,279 8,601 0,014 0,000 1,000 Clicked 1,305 0,294 19,720 1 0,000 3,686 29,003 0,000 0,000 1,000 Bought 1,049 0,476 4,858 1 0,028 2,856 8,165 0,017 0,000 1,000
When comparing the outcomes of the moderately personalized mail ad and the highly personalized mail ad (table 6), we can conclude that these levels of personalization do not significantly differ in outcome in terms of opening the mail (O.R.=1,011; p=0,902), clicking the link (O.R.= 1,035; p=0,862), and buying the product (O.R.=0,890; p=0,728). Based on these outcomes, H3 is rejected.
Table 6. Binary logistic regression moderately personalized compared to highly personalized
B S.E. Wald Df `Sig. O.R. Omnibus χ2 Sig. H&L χ2 Sig. Opened 0,011 0,091 0,015 1 0,902 1,011 9,507 0,023 0,000 1,000 Clicked 0,034 0,197 0,030 1 0,862 1,035 51,588 0,000 0,000 1,000 Bought -0,117 0,336 0,121 1 0,728 0,890 11,560 0,009 0,000 1,000
