Final Paper for the Collaborative Innovation Networks (COIN) WS 09/10 Seeding a Viral Marketing Campaign

Anita Etrati Leshek Gruska Raoof Mansoor Fabian Stein Drazen Uzelac

Table of Contents List of Figures ................................................................................................................. IV List of Tables .................................................................................................................... V 1. Introduction ...................................................................................................................1 2. Theory ...........................................................................................................................2 2.1 SNT, Social Capital and Organizational Performance .............................................2 2.2 Venture Capitalists and Social Capital .....................................................................4 2.3 Product Diversity and Performance .........................................................................4 3. Research Material and Methods ....................................................................................5 3.1 Crunchbase ...............................................................................................................5 3.1.1

Data Collection ...............................................................................................6

3.1.2

Networks .........................................................................................................7

3.1.3

Centrality Measures ........................................................................................9

3.2 Performance Measures ...........................................................................................10 3.3 Hypothesis ..............................................................................................................11 3.4 Linear Regression ...................................................................................................12 4. Results .........................................................................................................................12 5. Discussion and Conclusions .......................................................................................17 6. Further Research .........................................................................................................19 6.1 Variations to our Research .....................................................................................19 6.2 Further Possible Research Fields ...........................................................................20 References........................................................................................................................21 Appendix I .......................................................................................................................24 getCB company.php ......................................................................................................24 getCB company offices aqui.php ..................................................................................30 getCB people.php ..........................................................................................................33 getCB fincancial organizations.php ..............................................................................36

Erklärung .........................................................................................................................40

0. Abstract With the growth of the new technology Internet electronic communication have become an important issue. There is no faster and more manipulable way to spread messages than within a social network. The Messages spread like wildfire around the global network. This can have either beneficial or harmful effects for reputation of marketing advertisement. This paper is going to analyze the strategies for spreading a message in a social networks and reaching a snowball effect, i.e. virus effect. An experiment will be conducted to investigate two seeding strategies, central and decentral. A description of experimental design will be presented. Due to failed campaign there were only a few results, which make it hard to draw solid conclusion, but critical success factors for a viral marketing campaign are discussed, and it is analyzed in what extent these factors were met by our project´s campaign.

1. Introduction A number of marketing campaigns are conducted through a viral marketing, achieving the benefits such as low budget and great reach of message. Tools and methods of social network analysis are being used to reach better results, i.e. to amplify the effect of performing viral marketing. But not every viral marketing project is successful. Deciding on the appropriate strategy is one of the critical steps and can be a problematic decision, since it will affect the whole later project. Related to the strategy, many authors have different suggestions, i.e. there are both suggestions for conducting centralized and decentralized seeding as a strategy for spreading a viral campaign. Choosing the wrong strategy however means failure or even obtaining results contradistinctive to the ones desired. Appropriate strategy would diminish risk of such incidents, since providing the right strategy at the beginning of the project means preventing the erroneous and futile effort as well as the costs for conducting such campaign. How to decide which strategy is the better one, which would result in more successful viral campaign? Unfortunately, at the present there is a scarcity of empirical studies on this subject, with authors basing their works mostly on theoretical models and simulations. The specific centrality issue mentioned above, as well as an extensive framework, is rarely scrutinized. Elaborating the impact of applying centralized or decentralized seeding in the viral campaign and providing empirical data supporting the results could contribute to reduction of uncertainty considering the strategy choice appropriate for viral campaign and reducing the risk of failure occurring. This seminar paper has a goal to address issues concerning strategies and approaches of a viral marketing and to suggest possible answers to questions considering those. To be able to do so, after defining central terms of the paper, the literature on the subject will be analyzed and the resulting questions presented. Our data will be generated and gathered using experimental design and analyzing results of experiments conducted in different social networks on internet, such as XING. Conducted experiments will also support already existing viral marketing project conducted by a marketing agency. Based on the literature

and the experiments results, the conclusion and suggestions considering virtual marketing strategy will then be presented.

2. Literature Review 2.1 What is Viral Marketing Viral marketing, although it is a relatively young phenomenon, is getting increasingly popular in last few years. Nevertheless, there is already diversity in understanding what it actually is, and there are already several terms being used synonymously. So what is viral marketing? It is generally agreed that the term was coined in 1997, when it was used for illustrating phenomenon loosely defined as “network-enhanced word of mouth”, related to promotion of Hotmail.(Porter, Golan, 2006)Free email service Hotmail was promoted by simply adding a tag "Get your private, free e-mail from Hotmail at http://www.hotmail.com" to messages sent from Hotmail accounts.(Jurvetson, 2000)The recipients which red the tag and joined Hotmail would then send the same tag in their messages, thus contributing to spreading the tag like a virus would spread. Hotmail gained 12 million users in 18 months with an advertising budget of $50,000. Leaning on Hotmail example, Helm (2000) describes viral marketing as “a communication and distribution concept that relies on customers to transmit digital products via electronic mail to other potential customers in their social sphere and to animate these contacts to also transmit the products”, but reduces so the viral marketing only to digital products.(Helm, 2000)This paper is going to be oriented on definition by Wilson (2000), who says that “Viral marketing describes any strategy that encourages individuals to pass on a marketing message to others, creating the potential for exponential growth in the message's exposure and influence. Like viruses, such strategies take advantage of rapid multiplication to explode the message to thousands, to millions”.(Wilson, 2000)Another definition by Langner (2007), describes viral marketing as purposefully inducing of word-to-mouth advertising for purpose of marketing a companies or their output.(Langner, 2007) Terms as “buzz marketing”( Greg, 2004),as well as “word-of-mouth”, “leveraging the media”, “network marketing” and “creating a buzz” had been also used to refer to viral marketing.(Porter, Golan, 2006)But there are significant differences between old fashion word-of-mouth, traditional marketing and viral marketing. While traditional consumer wordof-mouth being informal communication directed at other consumers about goods and services and/or their sellers,(Westbrook, 1987)viral marketing is extensively using internet and other technologies to promote and manage such communication between consumers and between potential customers. The importance of word-to-mouth has been shown in many researches, i.e. in influencing customer’s product judgment,(Bone, 1995)for purchasing decision-making context,(Bansal, Voyer, 2000)or for reducing risk concerning customers buying decision(Murray, 1995). A Lucid Marketing survey analysis shown that 68% of individuals consulted their friends and relatives before buying a consumer electronic product,(Burke, 2003)one must keep in mind that word-of-mouth can be positive as well as negative for company or product. How to characterize viral marketing? Trying to determine the principles of viral marketing, Wilson states six following basic elements, of which not all need to be implemented, but all contribute to powerful strategy: (1) giving away product or service (for example free email service),

(2) providing for effortless transfer to others (with instruments like emails, websites, internet forums and blogs) (3) scaling easily from small to very large (having infrastructure prepared for exponential demand growth, because if demand is not met, the growth might die out. Product or service must be available), (4) Exploiting common motivations and behaviors (like greed, desire to be cool, popular etc.), (5) utilizing existing communication networks (such as placing message in existing communication between people), (6) Taking advantages of others’ resources (for example others’ websites, blogs etc.).(Wilson, 2000) However, these principles could be argued and extended, i.e. reasoning that for (1) complementary to giving away a product or service, marketing could implement satisfying needs and having useful effects (this would also include marketing being funny or provocative), for (4) not only common motivations and behaviors should be considered, but even more the preferences of the specific target group which would be most promising in terms of spreading effect of viral message. Other than these principles, viral marketing message should be entertaining, unique and extraordinary (not all of these characteristics are necessary for success, but they all contribute to it).(Langner, 2007)Even for a product without a “wow” factor, a buzz can be created if the content of message is provocative enough.(Kirby, 2004) Success factors will be addressed again in detail relating an existing viral campaign in chapter 5. There are many benefits that viral marketing yields. Maybe the most obvious is the financial one, concerning the very small expense of implementing viral marketing compared to traditional marketing, because once started it is spreading on “costs” of the individuals passing the message. Other than that, related to diffusion of message, when passed voluntarily, it may be viewed more favorably by the recipient. This can results in higher acceptance of message. Those forwarding the message are also more effective in targeting people that might be interested in the marketing message, because of knowing the taste and interests of their friends and colleagues.(Dobele et al., 2005)The viral nature of the dissemination can result in exponential spreading, reaching a great number of customers, and this in relatively short time (as seen in example of Hotmail). Besides these benefits, viral marketing is related to considerable risks as well. After initializing the viral marketing champagne, there is a certain loss of control over it. The individuals passing the messages are in position to influence the word-of-mouth to have a negative connotation, thus severely damage the campaign goal. After the message is initiated, it is often hard to control in what content will it going to be forwarded, because the information passed by the recipients might get filtered, incomplete and biased.(Helm, 2000, Skrob, 2005)If viral marketing goes wrong, the customers may perceive the messages as spam, may be annoyed by campaign itself, and sometimes there are concerns about ethical and legal questions related to viral marketing campaigns.(Kaikati, & Kaikati, 2004) A research by Verband der deutschen Internetwirtschaft, eco electronic commerce forum e.V. shows that 93% of the consumer feels disturbed by Internet commercial for instance via Email. And about 77% consumer deletes mails without opening them.(www.eco.de) Nevertheless, the interpersonal influence via viral marketing occur in technologies media

settings and is significantly different from occurring in conventional contexts in several ways.(Subramani, Rajagopalan, 2003)

2.2 Seeding Strategy in Social Networks regarding Social Network Analysis Sharing information in order to be up to date and do researches e.g. work and study, book hotel& flight and check the weather are the most common activities today. It is inconceivable to evade the growth of network telecommunication, computing and traditional media.(Lievrou, 2004) People barter their opinion, emotions and our comments openly with "free" access for everyone. For this kind of information exchange some portals are available like studi.vz, facebook, Xing. But why are these portals important? They give us the opportunities to look behind the scene. Due to that we try to look closer on network structural behavior and understand the visualisation, description and the statistical modeling.(van Duijn, Vermunt, 2006) A social network analysis (SNA) visualizes a distinct researches perspective within the social and network science.(Wasserman, Faust, 1994) A social networks are relationships between groups of individuals and they can be mapped.(Gloor, Cooper, 2007). Social network analysis is a tool to analyze the position and relationship of individuals,(Wassermann,Faust 1994, Scott 2000). A social network is an important framework for a selective spread of message to the receivers, and SNA has become an accepted tool to uncover the structure of social networks.

Moreover the social network enables to investigate the interconnected people on how they influence each other with the information they have and hence their decision. Building a relationships and collecting informations are high-priced businesses, because interaction and influences either used to be very diffuse.(Valente, 1995) Just only one or few nodes can impulse a cascade within a network. However it is also possible that this one node or these few nodes could be the activator on the connectivity of the network. They can be the reason for an subsequent cascade failure which lately they could switch of other nodes in the network. Adilson E. Motter has shown that the size of the cascade can be drastically reduced with the intentional removals of nodes having small load and/or edges having large excess of load.(Motter, 2004) In Addition Duncan J. Watts says that the successes or the failure is not addicted of the number of characteristics of the innovators themselves than on the structure of the community of early adopters. So it is important to have a number of early adopters, because "the more early adopters exists in the network, the more likely it is that an innovation will spread". On the other hand, the number of early adopters are not only dependent. How they are connected to each other is also an important aspect to focus on. It is possible that the early and late majority can also influence indirectly the early adopters. It depends on the largeness of the network. Recently "the network of interpersonal influences is sufficiently sparse the diffusion of cascades is limited by the global connectivity of the network".(Watts, 2002) How does SNA help us finding individuals we should focus on? Customer value is usually determined as expected profit from sales to a customer, and depicts how much is worth

spending to acquire a particular customer. In terms of viral marketing, it would make more sense to orientate on customer network value, an expected increase in sales to others that results from marketing to that customer”.(Domingos, 2005)That way we would target individuals who influence others in buying products, because reaching them would pay off many times in sales to other customers. This means that in order to have an optimal result, viral marketing has to reach the most influential individuals in social network. But how to determine who is influential? Centrality measures can be used to determine the individuals in the network which are in position to reach a large number of customers, and who also have a high likelihood of forwarding a message.(Kiss, Bichler, 2008)Depending on centrality measure according to which the set of initial customers was chosen, the number of reached customers will vary. Research has shown that the success of centrality measure depends on the characteristics of the network, but there are some measures that generally provide relatively good results, one of which is out-degree centrality (Degree of a node is the number of nodes adjacent to it. For network with directional edges between nodes, outdegree is the number of edges from other nodes to the regarded node).(Kiss, Bichler, 2008)Nevertheless, an analysis of data from real person-to-person recommendation network shows that although highly-connected individuals play important role, there are limits to how influential they are.(Leskovec et al., 2007)For example, after certain number of recommendations for a product, success will decline, implying that influence is present over a few of friends, but not everybody in their network. There is also a certain level of saturation whit repeated recommendations, after which the probability of “infection” with message is unlikely.

2.3 Different Approaches for Seeding a Sozial Network

Term “seeding” can be used for praxis in traditional marketing which is referring to providing a free product samples to some of the customers. In this paper “seeding” is understood as directly addressing initial set of individuals with the marketing message, intending for them to spread the message in their network. As seen above, in order to achieve optimal results in spreading, one should consider certain criteria on which to choose individuals in the network (nodes in graph) which are going to be seeded. Number of researches investigates how to determine optimal initial set of customers for seeding. Domingos and Richardson (2001) created optimization model based on the network value of a customer.(Domingos, Richardson, 2001) They argument that not only high connectivity of a node in network is beneficial, but other node characteristics as well, such as liking of the product promoted, or their chain of influence (describes what influential effect nodes have even beyond their immediate acquaintances, because these have influence on other nodes, and so on).(Domingos, 2005) Analyzing information cascades on a network of interacting agents, Watts describes how connectivity of network and stability of nodes affect the emergence of large cascades(Watts, 2002)Since during information cascades individuals make their decisions based on the actions of other individuals, it is related to viral marketing to understand how best to influence such decisions and provoke a large cascade, i.e. viral spread of marketing message. In Watt’s model each agent (node) is assigned random number of neighbors – defining his connectivity, and threshold for switching his states from 0 (starting state) to 1defining his stability. When threshold fraction of his neighbors is reached, the agent would switch to state 1. Analyses of conducted simulations have shown that large cascades can occur in both sparse and dense networks. If the network is sparse, with highly skewed degree distribution over nodes, the propagation of cascades is constrained by connectivity of the network. In this case, the most connected nodes are most likely to trigger large cascade,

even thou such nodes are unlikely to easily change their state. If the network is dense, having node degree distribution highly peaked, propagation of cascades is limited by nodes’ stability. Nodes here are connected to many neighbors, and higher number of neighbors is required to change their state. In this case the nodes with average connectivity are more likely to serve as a trigger for a large cascade. As already stated above, the research by Leskovec et al. (2007) shown that although degree centrality of individual is important for spreading the message, it does not guarantee the success, i.e. the viral spread. Furthermore, it seems that high-centrality individuals have influence only on few neighbors, not on everyone in their network. These researches show that there is interest in how centrality measures contribute to finding the optimal set of individuals for seeding the network, but there are not many empirical studies to support the results of simulations on theoretical models. The degree centrality occurs as a promising measure, and seams that when applied for searching individuals with interest in product it could result in successful seeding set. But the responsiveness of highlyconnected nodes (here because of their high degree centrality referred as “central nodes”) could present an obstacle for a success of such approach. On the other hand, an approach such as searching for nodes with low degree centrality (here referred as “decentral nodes”) in a sparse network and seeding them could result in better accepting and forwarding of the message. Thus the following research question appears:

Research question 1: Will a marketing campaign that is centrally seeded in a social network be more successful than one with a decentral strategy? Related Hypothesis: A decentral approach will be more successful than a central.

Related to acceptance of the message, it would be interesting to investigate how the receivers’ perception of the sender influences his participation in viral marketing campaign. Research question 2: Does the sender´s reputation have a significant influence on receivers’ campaign participation? Hypothesis: Seeding through a sender identified as a business company will result in higher participation then as a student group.

3. Methods 3.1 Project and campaign description Within the advanced seminar “Coincourse 2009” our project was to implement a seeding strategy for a viral marketing campaign. The Project has been conducted in cooperation with a German marketing agency which assigned for the client www.ProductSearcher.de. The business model of this B2B / C2B company is to bring prospective customers and vendors for any kind of products together.

Therefore the prospective customer describes the product he is looking for and ProductSearcher.de will forward the request to a fitting supplier. Due to the fact that our client is a young start-up company their goal was to get more attention on the Internet whenever someone searches for a vendor for any kind of products. Especially their appearance on Google searches should be increased. For Example if someone uses Google to search for printers, ProductSearcher.de wants to be listed on the first search results site. Therefore the Google page rank has to be enhanced. The more pages link to another page - and more importantly – link back to that first web page, the higher the page rank will be and the web page will be listed earlier in the search results. In order to increase the page rank and therewith the clients attention they hired the marketing agency “Cologne Marketing” which set up a viral marketing campaign that creates back-links to the clients homepage. The campaign was a lottery and worked as follows: in order to participate the lottery one had to post a short advice for founding a start-up company on his own homepage or preferably on his blog and then link it to the clients blogcomments. In this blog-comment one had so set a link back to his own blog. Giving advice for founding a start-up company was supposed to be suited because the client celebrated his first anniversary since his own start recently. Participants had the opportunity to win 500 Euros. Now our team’s task was to set up a seeding strategy for that campaign that would spread the “virus” as far as possible.

3.2 Experimental Design To answer our research questions whether a central or a decentral seeding strategy would be more successful, and if the senders reputation has any significant influence, we chose a “randomized experiment” for our experimental design. This design seemed to be most appropriate because our question addresses a cause-effect relationship (http://www.socialresearchmethods.net/kb/destypes.php), in particular whether the campaign will spread more efficiently and wider inside a certain network-component, if the component has been seeded centrally, or if it has been seeded decentrally.

The following describes the experimental design of a randomized experiment in general. First of all one distinguishes between "random selection" and "random assignment" (Trochim, 2006). Random selection signifies, how a sample is drawn from the population. (Trochim, 2006) Inherently the sample is drawn completely random form the population in a randomized experiment. In our case we chose certain people form the Xing network who were interested in topics like start-ups or in consulting those etc., and visualised connections between them via condor. So these people are considered to be our population. From that we chose our samples randomly. Random assignment signifies how these samples are assigned to different groups or treatments. Again, people were assigned randomly to the treatment group, respectively were seeded randomly. Our observation is, if a person participated the campaign or if he did not and will be marked as O1.

Our Treatment is that a person is seeded and will be marked as X. An important aspect to be aware of is internal validity. The major question here is, whether the treatment influences the observations, or if there are other possible causes. “All that internal validity means is that you have evidence that what you did in the study (i.e., the program) caused what you observed (i.e., the outcome) to happen.”(Trochim, 2006) Although the campaign respectively the lottery has been set up about 2 weeks before our work started, we can assume that the campaign had no attention on the Internet so far because no one had participated at this point and the client ProductSearcher.de is not considered to be a big player. There might be many reasons why people would not participate after they have been seeded, on which we had no influence. But we can tell that internal validity is constituted because if someone participates the campaign, it is because the person was seeded directly by us, or the person heard about it from someone in his network component seeded by us. We can tell this because condor gives us the opportunity to visualize where the participants come from and how they are linked to each other.

3.2 Application of the seeding strategies in different Networks 3.2.1 Lottery and family & friends In the beginning of our project we were under pressure to generate results (participants) as fast as possible because the client had set up a deadline for the marketing agency. This deadline forced us to get 25 participants within a week. Therefore we started seeding in lottery blogs or websites without any experimental design, but we were prepared to identify the persons who got aware of the campaign by a lottery website and to find out if they were linked to any other possible participants so that we could follow if the virus were spreading from there. The motivation for this target group was clearly the money they might win. Additionally we asked our team´s friends to participate to get some secure entries on the client´s campaign website. Another reason for seeding in lottery blogs and including friends was to get the trend going and to get some hits before addressing people interested in start-ups. The idea by the marketing agency behind that was to avoid that start-up enthusiasts would think there is nothing going on the client´s site and therefore not finding it worth to engage. Over all we seeded the viral campaign in 19 lottery blogs and websites. Plus we asked 21 of our friends with an own website or blog to participate, which some of them did.

3.2.2 Blogs To answer our first research question, whether the sender´s reputation has a significant influence on campaign participation, we focused on websites and blogs, that support other people in founding new stat-ups or give general advise for doing so. After being rejected by lots of people during our seeding, we were curious to find out if emails sent from a "professional" label - as ProductSearcher.de is supposed to be - would get more people convinced to participate than emails sent from our gmx.de label. Therefore we chose the randomized experiment as our experimental design because we are addressing a cause-effect relationship. Treatment R1: Firm label People who got the treatment R1 received an email from the ProductSearcher.de domain.

The page owners were addressed personally by name which we got to know from the legal information. If the owners name could not be found out the appellation was "Dear (name of website/blog)- Team". In the email itself the company shortly introduced its intention and asked if the would like to contribute a hint for founding a start-up company. (For detailed information see appendix). Treatment R2: Student label People who got the treatment R2 received an email from our gmx.de domain. Again the pages owners were addressed personally by name (or the entire team). In this treatment email we identified ourselves as students who are interested in start-up companies and asked in our name if the would like to contribute a hint to the client´s website. With condor and the factfinder tool we identified blogs and websites that showed general interest in entrepreneurship and in founding start-ups. Terms to identify "company founder" blogs:

• • • • • • • • • • •

Unternehmensgründung Gesellschafter Selbstständigkeit Handelskammer Geschäftsplan Existenzgründung Entrepreneurship IHK Geschäftsidee Start-Up E-Business

Terms to identify "consultants" blogs:

• • • •

Unternehmesberatung Unternehmensberater Existenzberatung Existenzberater

After identifying 24 of these websites we assigned the first half to treatment R1 which will be called KP-group and the second half to treatment R2 which will be called ST-group. In the ST-group 3 emails bounced back, so eventually 9 people got treatment R2. During our project we realized that the centrality of our blog networks was not correct. The Condor Blog Search Tool uses the search words as central nodes, which falsify the results.

3.2.3 Xing To answer our second research question, whether a central or decentral seeding strategy would be more successful, we conducted the following experiment.

As a result of our problems with the blogs centralities we changed our approach and decided to conduct our seeding experiment in the Xing-network. How did we get the sampling? The group of interest for this study consists of people who are involved in founding companies, especially start-ups and people who play consultatory roles in these processes. The first subgroup will be called “company founder” and the second subgroup will be called “consultants” in the course of this study. Consequently these people have an interest in entrepreneurship, company management and exchange of experience. The sample was taken randomly from the Xing-network, a social network with focus on business in a timeframe from end of October 2009 till the beginning of December 2009. The total sample size was approximately 2000 and nearly 140 were submitted for treatment. Using random selection allows us to improve external validity, so in our opinion the sample can be considered as representative for the original population. We used the CoolPeople tool to identify the persons who fitted to the campaign. CoolPeople is a tool, which allows us to crawl data from social networks, in our case from the Xingnetwork. To find the “company founder” target group, we selected “Unternehmer” and to find the “consultants” we selected “Unternehmensberater” as job positions in the CoolPeople search mask. Our next step was to import this CoolPeople data into Condor for further analysis. After importing these datasets, Condor visualized the networks and we could start our seeding experiment. The following describes our measurements and seeding strategies. Measurement in this design consists of whether an actor starts participating in the campaign or not. We will spot this through observation and call the observation before the seeding O1 and the observation after the seeding O2. Central Seeding Strategy The central seeding strategy is about identifying the central actors in the network components and to choose them for our treatment. The network components were chosen randomly.

Decentralized Seeding Strategy The decentralized seeding strategy is about identifying decentralized actors in the network components and to seed 30% of them. We chose this constant value to make sure that the results will be comparable. The outcomes were rounded. Here again, the components were chosen randomly.

The following describes our design. We conducted 4 pretest-posttest two-group experiments, seeding central and decentral in the “company founder” network and the same procedure in the “consultants” network. For each experiment we use random selection and random assignment. For central seeding we randomly chose 20 central actors to seed. Then, for the decentral strategy, we randomly chose 20 components and seeded 30% of the decentralized actors in that component, as described above.

1. Central seeding strategy in "company founder" network

R1 is a group of 20 random selected actors from the "company founder" network, which were seeded using the central strategy. This treatment is marked here with X. R2 is a control group of 20 random selected actors from the same network, which were not seeded. We want to observe, if the actor joins the campaign. O1 is the initial state, in which the actor does not participate in the campaign. At observation O2 we check if that has changed. 2. Central seeding strategy in "consultants" network

R1 is a group of 20 random selected actors from the "consultants" network, which were seeded using the central strategy. This treatment is marked here with X. R2 is a control group of 20 random selected actors from the same network, which were not seeded. We want to observe, if the actor joins the campaign. O1 is the initial state, in which the actor does not participate in the campaign. At observation O2 we check if that has changed. 3. Decentral seeding strategy in "company founder" network

R1 is a group of 20 random selected components of the "company founder" network, in which 30% of the actors were seeded using the decentral strategy. This treatment is marked here with X. R2 is a group of 20 random selected components of the "company founder" network, in which 30% of the actors were not seeded. We want to observe, if the actor joins the campaign. O1 is the initial state, in which the actor does not participate in the campaign. At observation O2 we check if that has changed.

4. Decentral seeding strategy in "consultants" network

R1 is a group of 20 random selected components of the "consultants" network, in which 30% of the actors were seeded using the decentral strategy. This treatment is marked here with X. R2 is a group of 20 random selected components of the "consultants" network, in which 30% of the actors were not seeded. We want to observe, if the actor joins the campaign. O1 is the initial state, in which the actor does not participate in the campaign. At observation O2 we check if that has changed. Using control groups, we can rule out all the single group threats like the history, maturation, testing, instrumentation, mortality and regression threat, but we experience multiple group threats. These threats can mostly be nullified, because our assignments were made randomly. There is a little risk, that one group accidentally consists of actors, who are more vulnerable for a viral marketing campaign than in the other group anyway. The social interaction threats like rivalry, demoralization or compensation are minimized, because of separating the groups from each other. However, it cannot be ruled out, that some actors from different groups have contact in any sort. This is caused by taking the sample from the same social network Xing. Finally, the internal validity can be seen as strong in this design.

4. Results 4.1 Results of the Xing approach Our seeding in Xing yields in nearly no results, only two actors participated in the campaign. Such a low response makes it very difficult to answer our research question. In detail, observation shows no changes after the treatments in the 1st, 2nd and 3rd experiment. The state of the control groups also stays unchanged. In the 4th experiment observation shows two participants in the treatment group and no changes in the control group.

Thus, the actors which participated in the campaign came from the “consultants” network and were seeded decentrally. As far as we can say from our experiment, it seems like the decentral seeding strategy is more efficient, but the data pool is too low to draw a conclusion. 4.2 Page rank What we can tell definitely from our analysis is, that the few links that the campaign created had no effect on the google page rank. Only one of the links shows a connection to the client´s website (Fig. 3). The links from the other pages respectively the pages themselves do not seem important enough to be recognized by the google bots.

Fig. 3 This figure shows our analysis of the linked web pages that are actually recognized by the google bots. As you can see only one page is important enough to be listed. The webpage deutsche-startups.de had already been set up before our project started.

5. Discussion As already mentioned the data basis for our results is quite thin to make clear statements to answer our research questions. In this chapter we will discuss potential reasons why the campaign participation was that low. Therefore we will once again regard critical success factors that should be fulfilled for a successful viral marketing campaign. We will analyze if these success factors were met in the campaign which the marketing agency set up for our project. Before we go into the success factors, lets point out which types of products or services are suitable for a viral marketing campaign. Therefore a distinction regarding the proprieties of the trigger of the message can be made (Langner, 2007). Infectious communication triggers can be e.g. rumours, stories or video clips. From that follows that the creation of infectious communication concepts is detached from the product an can be utilized universally.(Langner, 2007) This is important to notice because that way the power of viral marketing can be used by any company regardless what product or service it provides and in what industry it is acting. The company and the brand itself take a backseat in favour of a virulent, worthy to spread content.(Zorbach, 2004) Another kind of trigger can be an infecting product itself. The potential to recommend the product to a second person is so to speak included in the product.(Langner, 2007). An example for that could be the Apple MacBook Air. Now what are the implications of the above mentioned on the client ProductSearcher? ProductSearcher doesn´t sell own products but is a service provider for finding suitable products for their customers. As Langer suggests, a viral marketing campaign can be applied for any type of company, so in our case it is suitable. It is hard for us to tell if their service is provided in such a sophisticated way that customers would tend to recommend the website to other people. But let us assume that their service has this potential. Then the viral message should be somehow connected to what the company actually does. The marketing agency decided to set up a campaign that is detached from the service they are offering. Maybe a connection between their service an the viral message would have been better approach, but in general we can say that a viral marking campaign was suitable for their purposes, since the overall goal was to increase attention on google results, because a viral message has the potential to spread exponentially. Summarizing we can say that the service is not the subject of the campaign. It is not easy to mark down what the subject of the agency´s campaign is. Basically it is a lottery. From the campaign description in chapter 3.1 follows that there are two reasons for participating. The first one is the chance for monetary gain. The second reason is sort of intrinsic benefit. Some people feel satisfied when giving advice to others and feel happy when getting attention on the Internet.

Now to address the critical success factors, that the subject of a viral marketing campaign should comply with. As already stated above, these can be:

Free provision Fee required elements in a viral campaign can be compared to a filter, that submits spontaneous acting of a consumer. For that reason successful campaign subjects are usually for free to address a critical mass of consumers.(Langner S.) Jeffery Rayport 1996 established the saying "what is upfront is free; payment comes later". (Rayport, 2008) This success factor is completely fulfilled, since there were no payments charged for participating or only reading the client´s blog.

Easy transferability Nobody likes to wait a long time for a download or is patient enough to wait long for a website to come up. Neither will someone get a special software to be able to watch a rare video format. Therefore it is important that the user can transfer or recommend a certain campaign subject easily and fast (Langner, 2007). In our opinion this is the most important criteria for a good and successful campaign and is absolutely critical. Here the agency´s campaign suffers the most shortcomings and that might very well be the main reason why the campaign had only that little participation. Consider a person which received the invitation respectively the treatment, and assume he followed the link to the client´s website. The person then had to read a full DIN A4 page of text to understand what the conditions for participating are. Assuming he did this and understood how the system works, he then had to figure out some kind of tip or advice for founding a start-up company. Even though this doesn´t have to be a "good" or deeply thought-out advice it still takes quite some time. After doing so he had to post his tip in his blog or website and was supposed to link his blog entry to the client´s website. On the client´s website he had to leave a comment with a link back to his own blog. You can imagine that not a lot of people are willing to spend that much effort and time just for the unlikely chance to win 500 Euros or for other intrinsic motivation. Some participant didn´t even seem to understand how participation worked, because they set up the links wrong. So as you can see the campaign was too complex and took too much time. It was a considerable barrier for potential participants. The whole system of how to participate the lottery should have been set up way much simpler.

Even more importantly, the message or campaign does not spread if someone only participates. The message respectively the information that "there is a nice lottery worth participating" does only get transferred if a participant tells someone else about it. The structure of the campaign completely relies on the assumption that the people in a network talk about it, without any reason for doing so, as already said related to usefulness. What could have helped out here is a "recommend to a friend" function like you have e.g. in social networks like studiVZ.de or on youtube videos.

Availability

In viral marketing the availability of the campaign subject is relevant too. It is possible that a server gets down if thousands of people try to access it at same time. So the capacity always has to be sufficient. This critical success factor was completely fulfilled since we did not notice any downtime at the client´s website.

Emotional reaction The subject of a viral marketing campaign should evoke an emotional reaction at the person who gets confronted with the message. An important aspect in viral marketing is fun. If something makes people laugh, they rather tend to spread a message. Campaign subjects with high entertaining quality - through excitement and fun or sometimes through fear and sadness - call emotions at the consumer. If he does get keen or is moved, he wants to share in with other people (vgl.Langner S.(2007)S.38, Rosen, E. (2000), The Anatomy of Buzz: How to Create Word of Mouth Marketing,S.105) In our opinion a positive emotional reaction is to be preferred. In our case we do not think that this criteria is fulfilled, at least not completely. It might be conceivable that the chance to win 500 Euros creates some sort of excitement but it doesn´t give you any "thrill", neither does it make you laugh. The satisfaction to give advice to others is as well not evoking emotional reactions, since the participant respectively the advisor gets not direct feedback which would give him a good feeling.

New and unique Innovative and unique campaign subjects are extremely important to get die consumers attention. (Langner, 2007) Especially it is important to not copy something that has been set up in a similar or identical form. (Citino, 2008). Something remarkable is worth talking about,worth paying attention to.Boring stuff quickly becomes invisible.(Godin, 2008)

Obviously a lottery is nothing innovative and is common to everybody. The same applies to giving tips for founding a new start-up company because there are already lots of consulting websites. So here the agency´s campaign clearly lacks of something innovative that picks up the potential participant´s attention.

Usefulness If a user doesn´t see any advantage in the virus he will not forward it. Next to the already mentioned entertainment factor, free versions of e.g. software programs, helpful websites or image advantages through connection to a recognized brand take in to account. Products respectively campaign subjects that get increased usefulness the more people are using it appertain as well. Instant messaging programs like ICQ for example are useless if you are the only one using it, but the more friends use it, too, the more benefiting it is.(Rosen, 2000) In the agency´s campaign for our project a user might get some useful advice and tips form other people for founding a start-up company. But he will get these tips anyways simply by reading the comment page even if he doesn´t contribute a tip by himself. And as said under point two, uniqueness, there are already lots of consulting websites and in addition to that they are specialized on that topic. So their consulting will be of a much higher quality than some casual tips form unknown people from the Internet. If the participant were actually interested in getting advice he would consult a professional. The campaign also did not provide any kind of free stuff, neither does a participant get his own image increased. Here a good way of fulfilling this success factor might have been to give away some free merchandise articles such as t-shirts with the client´s logo on it. This would even have created attention besides the Internet and the google search results.

In summary we can say that the campaign does not fulfill four of the six critical success factors, especially the fifth one. But we still do believe that social network analysis and the tools for it is very helpful to seed a viral marketing in general. Although we were not able to observe how the viral message spread in the social network, SNA enables you to identify central and decentral actors which should increase the success of spreading the message as far as possible. And if the message has started to spread you are able to see form which person it came and then you can support your seeding strategy specifically where it lacks attention. We would like to recommend further research on that topic e.g. in the next coincourse. Maybe it is possible to cooperate with a chair of marketing that designs the viral campaign subject in such a way that the critical success factors are fulfilled. A good campaign is absolutely critical to get the viral message spread and this way creating a solid data base for further analysis. It would be very interesting to see if the hypothesises are supported by sufficient data.

6. Conclusion Finally we can say that viral marketing and SNA is a sensitive issue which has to be handed carefully. We illustrated that it is important to pay attention during the first steps in planning a viral campaign. A quick start does not guarantee a rapid success. It is important to set ambitious goals and to thorough trucking's. It is also important not too to set goal high or too low, because it might not achieve as much as it could. It is also important to be careful with seeding, because people might take this message as a spam mail and delete the unopened message. It is also important to find a relationship with the receiver to build a personalize message and meet the receivers interests. Our strategy shows that we concentrate on central and decentral actors to figure out how is the one to spread out innovation.

References Bansal H.S., Voyer P.A. (2000) Word of mouth processes within a services purchase decision context, Journal of Service Research 3 (2) 166–177 Bone P.F. (1995) Word of mouth effects on short-term and long-term product judgements, Journal of Business Research 32 (3) 213–223 Burke K. (2003), 2003 Consumer Electronics Customer Survey (accessed on 05.02.2010) Brynjolfsson E., Smith M.D. (2000). Frictionless Commerce? A comparison of internet and conventional retailers. Management sience 2000 INFORMS, Vol.46, no4, April 2000, pp.563585 Dobele, A., Toleman, D., & Beverland, M. (2005). Controlled infection! Spreading the brand message through viral marketing. Business Horizons, 48(2), 143-149 Domingos P. (2005). Mining social networks for viral marketing. IEEE Intelligent Systems, 20(1) Domingos P., Richardson M. (2001) Mining the Network Value of Customers. Seventh International Conference on Knowledge Discovery and Data Mining Gloor P.A.,Cooper S.M. (2007) Coolhunting: Chasing Down the Next Big Thing, AMACOM Greg T. M. (2004), "Building the Buzz in the Hive Mind," Journal of Consumer Behaviour, 4 (October), 64-72. Helm S. (2000) Viral Marketing - Establishing Customer Mouse', Electronic Markets, 10 (July), 158-161

Relationships

by ‘Word-of-

Jurvetson, S. (2000), "What is Viral Marketing?" (accessed on 02.02.2010) Kaikati A. M., Kaikati, J. G. (2004). Stealth Marketing: how to reach consumers surreptitiously. California Management Review, 46(4), 6-22

Kirby J. (2004) "Getting the Bug," Brand Strategy, (July/August), 33. Kiss C., Bichler M. (2008) Identification of influencers – Measuring influence in customer networks. Decision Support Systems 46 (1), S. 233-253 Langner, S. (2007), Viral Marketing, 2. Auflage, Gabler Verlag Leskovec J., Adamic L., Huberman B. (2007), The dynamics of viral marketing, ACM Transactions on the Web, Vol. 1, No. 1, Article 5

Lievrou A. (2004) What’s changed about new media? Introduction to the fifth anniversary issue of new media & society. New Media & Society 6(1) Motter A. E. (2004). Cascade control and defense in complex networks. Physical Review Letters 93, 098701: 1–4. Murray K.B. (1991) A test of services marketing theory: consumer information acquisition activites, Journal of Marketing 55 10–25 Porter, L., & Golan, G. J. (2006). From Subservient Chickens to Brawny Men: A Comparison of Viral Advertising to Television Advertising. Journal of Interactive Advertising, 6(2), 30-38 Rosen E. (2000), The Anatomy of Buzz: How to Create Word of Mouth Marketing Skrob, J.R. (2005). Open source and viral marketing, University of Applied Science Kufstein, Austria Subraman M. R., Rajagopalan B. (2003) Knowledge-sharing and influence in online social networks via viral marketing, Communications of the ACM, v.46 n.12, December 2003 Zorbach T. (2004). Viral Clips verdrängen den guten alten Fernsehspot. io new management, 27.6.2004 Trochim W.M.K. (2006) http://www.socialresearchmethods.net/kb/random.htm (accesed on 13.12.2009) Trochim W.M.K. (2006) http://www.socialresearchmethods.net/kb/intval.php (accesed on 13.12.2009) Unknown (28.09.2001) http://www.eco.de/verband/202_3573.htm (accesed on 23.01.2010) Valente, T. W. (1995) Network Models of the Diffusion of Innovations (HamptonPress, Cresskill, NJ) van Duijn M. A. J., Vermunt J. K. (2006). What Is Special About Social Network Analysis?

Methodology, 2(1):2-6 Wasserman S., Faust C. (1994), Social Network Analysis: Methods and Applications, Cambridge University Press. Watts D. J. (2002) A Simple Model of Global Cascades in Random Networks. Proc. Natl. Acad. Sci. 99, 5766-71 Westbrook R.A. (1987) Product/consumption-based affective responses and post purchase processes, Journal of Marketing Research 24 (3) 258–270 Wilson Dr. R. F. (2000), The Six Simple Principles of Viral Marketing, Web Marketing Today, Issue 70, February 1

Eidesstattliche Erklärung Hiermit erkläre ich/erklären wir, dass ich die vorliegende Arbeit selbständig und ohne Benutzung anderer als der angegebenen Hilfsmittel angefertigt habe. Alle Stellen, die wörtlich oder sinngemäß aus veröffentlichten und nicht veröffentlichten Schriften entnommen wurden, sind als solche kenntlich gemacht. Die Arbeit ist in gleicher oder ähnlicher Form oder auszugsweise im Rahmen einer anderen Prüfung noch nicht vorgelegt worden.

Anita Etrati Drazen Uzelac Fabian Stein Leshek Gruszka Raoof Mansoor Köln den, 9.Februar 2010