Making engagement profitable: Relationship between user engagement and advertising on Facebook

(1)

Making engagement profitable:

Relationship between user engagement

and advertising on Facebook

24 June 2016

(2)

- 2 -

1. Introduction

Click. You like your friend’s status. Click. You share that interesting article you read. Click. Why did you click on it? It is an advertisement! But you did not even notice… It became so normal to see sponsored content amongst your News Feed that we do not even notice it is advertising. If it catches our attention, we engage. Or do we?

Engagement has become part of essential keywords for companies, marketers and advertisers. They compete for user’s attention every second they spend online. However, it is not only attention that they want. They also need users to participate and engage. From business perspective, user engagement is defined as “an assessment of an individual's response to some type of offering, such as a product, a service or a website” (Rouse n. pag.) and it is a crucial component in the online business environment. In some ways, it defines whether business is successful or not (Meares, n. pag). Exploiting the promise of engagement, large businesses like Facebook convince marketers that if they just pay for the advertising, user engagement will come instantly. However, to ensure profitability, social media platforms worked very hard that it would be hard to promote social media content without paying for it: algorithms structure visibility on interfaces and it is difficult for businesses to reach users organically. This is why it is relevant to analyse engagement within the context of advertising. As the largest social network for my research I have chosen Facebook and its advertising interface. It is vital to scrutinize Facebook as a platform as it holds influence over almost 1.7 billion1 monthly users’ and 1.09 billion2 daily users’ lives on social media. As put by Gillespie (2015): “recognizing that social media platforms shape the social dynamics that depend on them allows us to draw connections between the design (technical, economic, and political) of platforms and the contours of the public discourse they host” (2). Gillespie demonstrates the importance of critically looking at the social media and justifies looking into its structure, which is code and affordances like design, and the discourse they use, which is what I will do in this thesis as well.

When it comes to engagement, it is not always clear how to evaluate whether users are engaged. Why does Facebook analytics have a section on post engagement? And what does it mean when it states that “10 people were engaged”? Is it possible to measure engagement online in quantitive form? These are just a few questions that are crucial to look

1_{www.newsroom.fb.com/company-info/}

(5)

- 5 -

into when it comes to the relation between engagement and advertising. Therefore, I will attempt to answer the question which encompasses the key elements of the platform itself and the decision it makes in relation to user engagement:

How does Facebook define, capture and rank user engagement through advertising interface?

To answer this question this research will look closely into Facebook’s advertising interface3 which will be the gateway into accessing the software behind Facebook advertising and will allow observing how Facebook defines and captures engagement. Using platform studies perspective, I will look into the essential elements of advertising interface while running a realtime advertising campaign. The process from creating advert to analysing its results will be outlined for examination. It will be possible to analyse how Facebook presents engagement and quantifies it, as well as to see user engagement patterns and how they are influenced by advertising. The research will also develop metrics critique (in terms of metrification of the interface) and will attempt to contrast engagement defined by Facebook to existent engagement models. Looking at the particular case of an advertising campaign it will be useful to explore the grammatization of advert creation, quantifications of interaction and comparing the money invested in it – all in relation to the advert campaign and user engagement with it.

Many academics look into the software of platforms, yet not enough turn to it in relation to engagement and advertising. Engagement, in particular, has been mostly analysed in social studies or business studies yet it is uncommon to find literature on measures of engagement and connections to platform studies theory as has been done in this research. To add, a lot is written about participation and activism as engagement but I will focus on user engagement with sponsored content. The research places itself strongly in the field of platform economics as it analyses how the social interactions are given monetary value and critiques this model; it also analyses the relationship between advertising and engagement.

This thesis will show how the data users provide is captured and translated into numbers and how an outcome of the algorithm equals engagement. Although not transparent, this

3

The Facebook Interface used for research was captured in 2016 April and May. Facebook does make minor changes frequently, so it is likely that some details have changed, yet algorithms behind it will likely have stayed the same.

(6)

- 6 -

blackboxed software is responsible of governing social interactions and ranking them for company’s profit. This is the main problem of the “intransparency” of the algorithm as manipulating the algorithm for the use of advertising, the network makes money. It is crucial to always be aware that the social network is a business with a clear aim at profit; which is why the fact that software is biased and human constructed is so important. It means that engagement, however measured or translated into numbers, is part of the moneymaking machine. I will look into how Facebook evaluates engagement monetarily and ranks it depending on the kind of engagement it is. Conducting the advertising campaign I will examine the notion of how algorithm is programmed to bring the best results, and will delve into findings to outline that although Facebook implies that larger budget will bring better campaign results, that was not necessarily the case. Analysing the kinds of engagement on Facebook, I will go on to suggest an alternative for measuring engagement while questioning if engagement should be measured at al. Observing the process of setting, managing and measuring advertising campaign, I will conclude that for Facebook engagement is an algorithmically constructed number, contributing to the platform ecology.

For Facebook, both business and personal account holder are users. Personal account holders are the users that that paid content is targeted at. Therefore, I will be using the notion of “user” referring to both businesses and personal users. This understanding is different from traditional market space where there is a “business” and a “consumer” yet in the case social media, the consumer can be both: the business that buys advertising and the regular user who uses Facebook as a product. However, I would argue that at best, we are users of the platform; we are not consumers as we voluntarily provide our data to be collected and to be turned into profit. I will then return to the notion of labour and worker exploitation (Terranova 2000, Fuchs 2010) in the work.

I will start by reviewing literature and engaging with different theoretical perspectives, then I will move onto outline the methodology for the research. I will use methodology to carry out empirical analysis and to analyse findings. Finally, I will analyse findings in the context of theory and will present the conclusions I come to.

(7)

- 7 -

2. Literature review:

Facebook is the largest social network and the increasing “platformization of the web” (Helmond 2015) gives the context for Facebook to be analysed through platform studies perspective. Understanding the software behind the platform will help to better outline the relation between advertising and user engagement. Theoretical framework of this research will go through literature of user engagement and engagement theory towards platform studies. The literature of platform studies will involve the background of platforms and platform economics, not forgetting data economy. I will then move on to software studies, primarily focusing on critical algorithm studies. Lastly, literature on metrics and value creation will be reviewed. Much has been written about Facebook especially in terms of ranking algorithm (see Bucher 2012b, Birkbak and Carlsen 2016), or in terms of the sociality of it as a social network (Bechmann 2013, Berg 2014).

2.1.User engagement definitions

To start with, I would like to mention that there is a variety of definitions of user engagement. There is not one agreed terminology but rather multiple versions highlighting different key elements of it. O’Brien and Toms (2008, 23) argue that is a “quality of user experiences” and is characterized by various actors, while Lehman et al. (2012, 164) write that it is a “quality of user experience that emphasizes on the positive aspects of the interaction”. On the other hand, Rouse (n.pag.) claims that it is an “assessment” of a one’s reaction. We can already note that academic background influence the difference in opinions. Naturally, the understanding of what engagement is changes depending on the perspective. The meaning of the terminology is influenced whether it is concerning the side of the regular user or the side of the business.

2.1.1. Engagement theory

The classical study by O’Brien and Toms in 2008 (10-17) laid the building blocks for the processes of engagement and a model for it which is evaluated in terms of various attributes (See Appendix 1). They also provided a definition: “Engagement is a quality of user experiences with technology that is characterized by challenge, aesthetic and sensory appeal, feedback, novelty, interactivity, perceived control and time, awareness, motivation, interest, and affect.” (O’Brien and Toms 2008, 23). They specifically mention that user experience is with technology. The first stage is the point of engagement: where “engagement

(8)

- 8 -

is initiated” (O’Brien and Toms 2008, 10). The study looked into what attracted attention and caused users to be engaged and found that it could be factors such as graphics, design, layout and aesthetics just to name the visual experiences. Other participants of the study named that the information provided was something that caught their eye and that is relevant not only in the field of web search but also because the textual content on social media is trying to engage viewers just by the information provided. The second stage was the period of engagement: which is what we most commonly associate with the word engagement itself. It involves concentration over the content or subject and maintained interaction. It can be argued that this model of continuous sustained engagement is not so apparent on social media, yet it depends on the content. For example, seeing a picture and reacting to it takes very little time which hardly qualifies for more than a “point of engagement” but a video or longer text may naturally take more time. Third was the stage of disengagement, when people stopped the activity they were engaged in. O’Brien and Toms argue that in some cases the reason was external factors but often the conscious decision to stop what they were doing was caused by the loss of interest, time constraints, mere interruption or distraction. The penultimate stage was reengagement which entailed coming back to the activity done before, if the process was “broken” in any way. Platforms try to maximize rate of reengagement and for that they use notifications; for instance, Facebook notifies a user of others’ reactions on content they commented on, suggesting the user to come back to the content and if not reengage with it again then at least reply or react to comments section. Lastly, O’Brien and Toms write about nonengagement – that is a part of the engagement process when users were not actually engaged, they never reached the point of engagement. In some cases, they were not interested in the content, and in some cases they were not engaged enough – that is something Stewart (2014), who proposed interesting connection between aesthetic engagement and time, calls “distracted modes of engagement” (2014, 2). O’Brien and Toms propose a model for engagement and one of the important factors interestingly, is feedback. I will come back to the notion of feedback not only in terms of commenting on social media but also connecting it to the very specific ecology of Facebook.

In 2012, Lehman et al. expanded on their model of engagement while looking at a number of websites. Turner writes that it is “involvement-engagement” formula (2014, 34). His formula is relevant in the context of social media, because often users are just involved, not engaged yet. Finally, Cvikij and Michahelles (2013) look at engagement specific to Facebook brand pages from marketing perspective. Looking at different? Pages on Facebook,

(9)

- 9 -

authors provide recommendations what kind of contented should be shared to receive engagement. They found that entertaining and informative content received the most reactions, especially photos and it is argued that “interactivity has stronger effect over the engagement level” and “lower level vividness” content was more attractive (Cvijikij and Michahelles 2013, 14). As the authors acknowledge, because of the privacy settings on Facebook it can never be know how engagement rate correlates with friendship between fans of the page. It has to be noted that O’Brien and Toms (2008) only looked into gaming industry, online shopping, web search and learning software (7). Of course such study would really benefit from looking at social media. Some sources do not even define engagement itself, they move straight to measuring and assessing it.

2.1.2. Engagement models

Ponciano and Brasileiro (4) address that the important parts to defining and measuring online engagement are participation, type of engagement, duration and degree of engagement. From marketing perspective Chaffey (2010, n. pag.) suggests five categories how to measure online engagement:

1) how users reach content,

2) how they engage with it (using page analytics such as Adverts Manager), 3) how users “activate to business goals” (Chaffey 2010, n. pag),

4) how they participate and contribute to brand value, 5) whether users are engaging in the long-term.

Another marketing specialist, Eric Peterson (2007, n. pag.) argue that the successful formula to calculate online engagement are indices of click-depth, recency, duration, brand, feedback and interaction. This only shows the variety of models and frameworks available for analysis.

2.2.Studying the platform

Platform studies focuses on “platformization of the web” (Helmond 2015) which involves social networks turning into social media platforms along with much of the web. To make platforms come about there was one change in internet history. The web has undergone the change to Web 2.0. which has allowed interaction, and, most importantly relies on the interactions and other user-generated content (Langlois et al. 2009). Slowly, it developed to platforms, which offered engagement with them and other users. Platforms are often critiqued to be walled gardens (Gerlitz and Helmond 2013; Birkbak and Carlsen 2016) that they have

(10)

- 10 -

their own ecology. We will see how the critique of the like economy will add up to this critique that all web is interconnected though sociality which is turned into profit. Another important critique for platforms was that of ‘filter bubbles’ (Birkbak and Carlsen 2016; Van Dijck 2013, 161) as users become involved in their own circles while forgetting that the algorithm is constructing the interactions and the content online.

Tarleton Gillespie is one of the key academics in platform studies, focusing his work on platforms and critical algorithm studies, as the two are closely intertwined in the context of platformization of the web. He calls out that platforms always have their own agendas. Gillespie (2015) writes that not only social networks add to the communications but also they may take away some information posted by users (1-2). The importance of platform studies is clear: social platforms have immense power over the social discourse. Moreover, by intervening in this process with algorithms; networks are able to control what makes it to into the public sphere and what remains hidden (Gillespie 2016, n.pag.). Although this research focuses more the underlying algorithms of the platforms, it is also important to mention that APIs play an important role in platform studies. According to Bucher (2012, n.pag), “APIs are interfaces that facilitate the controlled access to the functionality and data contained by a software service or program” and they are not transparent just as the algorithms. Helmond (2015) writes that APIs are actually largely responsible for the platformization of web as well.

Gerlitz and Helmond (2013, 4-6) outline the history of Facebook Like button. Instead of commenting users could then press “like” on statuses, pictures and other content (Pearlman 2009, n.pag). Then Facebook was already counting likes and showing people who “liked” the content (Gerlitz and Helmond 2013, 5). In upcoming years the Like button was also introduced as a social plugin in the web sphere. This way the data of “liking” content online became fed-back to Facebook even if it was “outside” the social platform and this gave way to the start of what Gerlitz and Helmond call “the like economy” (2013) and which will be discussed more in the next chapter on platform economics. Interestingly, in February, 2016 (Thielman 2016) Facebook updated the Like button after many years of expressed need for dislike button (Gerlitz and Helmond 2013). Now, instead of simply liking content in their feed, users can “react” to it in emotions: ‘love’, ‘sad’, ‘angry’, ‘wow’ and ‘haha’ which also come with illustrative icons (see Figure 1). These reactions are now included in the engagement overview on the business side of Facebook, too which makes marketers job

(11)

- 11 -

easier because now they know the exact reaction or emotion that user feels when they engage with the content and whether this is the reaction that was expected.

Figure 1.| Facebook reaction buttons (Costine in Tech Crunch 2016).

2.2.1. Platform economics

Social platforms are critiqued for elaborate ways of making profit. On the user end, platforms gather as much information as possible to sell it to other businesses. This personal data is then used for targeted advertising (Bodle 2011, 331) which is later fed-back to the user all while tracking their engagement with it. Langlois and Elmer (2013) write that platforms have “economic interests in gathering as much user data as possible” (5). Although Facebook markets itself as the platform to connect people (Van Dijck 2013, 45), it is making money by doing so. Facebook’s main avenue for profit is advertising and it sells ad space in users’ News Feed for the businesses.

Part of the economy of Facebook is the Like button. As outlined in previous section, it is the key component of collecting information on social network and also outside it through its social plugin. Gerlitz and Helmond analyse this “like economy” (2013, 2) and critique its way of making all web part of it. Similarly, Bodle (2011) writes, Facebook is “colonizing online spaces of the Web with ‘Like’ buttons” (333). While Gerlitz and Helmond (2013) explicitly outline data flows in regard to the Like button and its social plugin and titles the undergoing process “cross syndication” (12), their analysis also work within the walls of the social network as well and with other forms of engagement. So, the data is collected about all types of engagement with the content and that potentially generates more engagement because people see what others like, share and comment on and get involved themselves (Gerlitz and Helmond 2013, 12). When users get involved, Bodle (2011) writes that their activity can be easily comodified by Facebook. Because of this, Facebook uses captured content to multiply it across the networks of the user. As put by Gerlitz and Helmond (2013): “the like economy still features a number of devices that seek to deploy the logic of

(12)

- 12 -

recommendation cultures in order to set in motion the multiplication of data production” (12). This notion will be particularly important when relating it to how users engage on Facebook.

When it comes to data mining, it is important to say that all captured data adds up to profit for businesses like Facebook. As Gerlitz and Helmond write (2013) “user activities are of economic value because they produce valuable user data that can enter multiple relations of exchange and are set up to multiply themselves” (13). Van Dijck (2013) also writes that “combination of content, metadata, and behavioral and profiling data that makes the resource of connectivity interesting for data analysts and marketers.” (162). These authors highlight how platform ecology is dependent on the data capture and profiling. Even more interesting argument is put by the work of Gerlitz and Helmond (2013) as they highlight how the “the underlying data mining processes foster participation by default, tracking users’ browsing behaviour, storing Like button impressions or instantly sharing app engagement to the ticker” (14). This is important in terms of News Feed engagement.

Finally, as part of the platform studies perspective I will use the concept of affordances to closer examine Facebook advertising interface. As a term, it can be used in various subjects but in context of technology it refers to what can be done and what is allowed to be done. In my research I will look into the interface of Business Manager and use the discursive interface analysis as done by Mel Stanfill. She herself wrote that:

Discursive interface analysis goes beyond function, examining affordances broadly – the features, but also what is foregrounded, how it is explained, and how technically

possible uses become more or less normative through productive constraint. (Mel

Stanfill 2014, 4; emphasis by author).

The author demonstrates here how important it is to look into affordances due to the influential power they hold over construction of behaviours and norms. Stanfill (2014, 9) connects social platforms and affordances: “These functional affordances construct fandom as consumptive, in the sense of taking in from the site rather than producing on their own, and they also work toward centering the sites through non-portable content”. We can see here that the notion of Liking is particularly important to Stanfill and she highlights that it is an activity based on consumption rather than production. I would relate this notion to the field of user engagement and the idea of participation: in the context of Facebook, we can argue that participation is minimal.

(13)

- 13 -

2.2.2. Software studies

Developed from cyberculture, internet and new media studies and rising since mid-2000s, software studies focus on processes happening in the digital culture and questioning the processes themselves and the effect they have on the wider context. Manovich (2013) writes that nothing was ever “only digital” (30), software is the key to everything. Berry (2011) writes that “software is a tangle, a knot, which ties together the physical and the ephemeral, the material and the ethereal, into a multi-linear ensemble that can be controlled and directed” (3). Manovich (2013) also stresses on a similar concern, claiming that “there is only software” (34), but more importantly, behind it, there are people. All decision constructing software were made by people, with agenda and particular goals. Central to software is code (Berry 2011, Mackenzie 2006), yet this research will focus more on algorithm and also will look into the notion of “grammars of action” (Agre 1994). Moreover, I will look into the concept of feedback to explain the process of how Facebook collects data and how it reaches the user again in form of advertising. As a concept, feedback was appropriated from nature sciences to cybernetics by Norbert Wiener. Feedback is also an important notion for user, yet not in terms of information fed back but rather as a positive or negative response in order to improve a product or service.

2.2.3. Problematics of algorithms and the News Feed algorithm

The key component for analysis in software studies is algorithms (Gillespie; Berg; Bucher). Sometimes, “algorithms need not be software” (Gillespie 2014, 1): they serve as human-designed calculations to perform programmed actions. Algorithm studies highlight the fact that the human impact is at its core, it is biased and non transparent which is the most troubling part. The fact that the algorithm ”gets to make decisions about how Facebook users are informed about their social networks” (Birkbak and Carlsen 2016, n.pag.) shows the power that software holds and the importance of opening these black boxes and questioning their intentions.

Firstly, algorithm is created and programmed by humans (Gillespie 2014; 2016). This entails that the decisions made by algorithm were actually designed by programmers. Manovich (2013, 31) also highlights that software developers made choices developing products. A recent example is that Facebook’s News Feed algorithm is being implemented into daughtercompany Instagram (Instagram blog). This shows the management decision to expose users to the ranked content according to what they engage most with and not by mere

(14)

- 14 -

recency as it was till April 2016 (Instagram blog). Bucher (2012b) also writes extensively about visibility of the News Feed algorithm (previously called Edgerank) and shows the influence of software as some content never even reaches the user. In this case, certain information makes it on the News Feed while the remaining does not and it only depends on the algorithm, and in that, on its designers. Gillespie (2016, n.pag.) also demonstrates how Facebook’s Trending algorithm is run by humans in recently surfaced issues regarding the matter. Naturally, all design choices were also made by people. For this reason, we can see where the argument on the bias of the algorithm comes in. Mackenzie (2006) stresses software is not neutral, it participates. boyd also argues that “all systems are biased” (2016, n.pag). The problem arises that we, as users trust the machines and rely too much on algorithms (Gillespie 2014, 2016) and we believe they are fair in providing information. Gillespie (2014) writes that “A sociological analysis must not conceive of algorithms as abstract, technical achievements, but must unpack the warm human and institutional choices that lie behind these cold mechanisms” (3). Gillespie’s statement fits well in the context of this research and, just as the author recommends, I will look into the choices made by humans.

Drawing on internal Facebook’s research Birkbak and Carlsen (2016, n.pag) argue that “Facebook’s ordering device … affords newness (new ideas, new products, current events) and engagement (sharing discussing)”. This will later be relevant looking at advertising interface which will be focused on newness as well. Berg (2014, n.pag) who explores the algorithm from a user perspective highlights Facebook’s power over visibility. We can see here how paid content comes in: what companies pay for is for their content to make it on relevant users’ News Feed. Post boost is a clever way for Facebook to make profit as it ensures the visibility. The News Feed algorithm puts preference on the paid content which can explain also why some updates from Facebook friends will not appear on the News Feed while advertisements will. Again, we can critique the algorithm as its recommendations might not be as relevant to users as Facebook claims (Easy and effective

Fcebook adverts, 2016). Gillespie (2015) writes that platforms “pick and choose” while Van

Dijck (2013) argues that “Algorithms that promote interlinking are not just securing a “frictionless online experience,” but also making that experience manipulable and salable.” (157). It is also important that “users do not have to search for content, but content is presented to them though the multiple recommendation features built into the platform” (Gerlitz and Helmond 2013, 13). These academics demonstrate how software influences user

(15)

- 15 -

interaction with the platform. On the other hand, Van Dijck says that users can ”steer trends” (159) and shows how users can go against the code. She writes that “actual users may actively try to modify or resist the roles inscribed in coding technologies.” (160). However, her argument would be more relevant in the context of social activism. I would argue that although it is possible to train the algorithm to what one likes, it is hard to actually have influence when it comes to sponsored content.

2.2.4. Grammatization of Sociality

Finally, Phillip Agre (1994) developed the idea of “capture” (744) as a model of privacy which is central to computer based activities. Comparing this with the surveillance model, he writes that capture model uses language rather than what is observed in view, it focuses on apparatus in place for altering present actions rather than continuous process of gathering information, it re-constructs multiple captured data points rather than spying on users, it is more local rather than centralized and lastly, it is more philosophical than political. Highlighting the linguistic part of the capture model, Agre (1994) uses the term “grammars of action” (745) which means that actions made with technology are much like grammar: pre-set collection of rules, or in case of computers, algorithms. These rules limit the available alternatives of making a choice and when an option is selected, it is also captured. The collected data will influence the availability of options, and the grammar itself. With this model any activity can be grammatized, even what looked a difficult action to be captured. Once it is recorded, any information can undergo grammatization which results in data points out of user activity. Agre (1994) writes that “what matters … is not the sequences of “inputs” to or “outputs” from a given machine, but rather the ways in which human activities have been structured (746). He outlines five stages of the process of grammatization: analysis (indentifying the key elements of an activity), articulation (sensibly arranging the units), imposition (“grammar is then given a normative force” (Agre 1994, 746), instrumentation (giving the means to keep the analysis ongoing), and elaboration (integrating the results into parts of other processes of information gathering). Agre also uses different examples to illustrate how it works: from telemarketers’ script to toll-roads and waitressing service. Van Dijck (2012, 161) uses a similar perspective to look at social media and states that “preformatted entries and home page layouts ... force users to submit uniform content”. What is more, the surveillance model that Agre writes about is also important. In essence, it is the fact that all data is ready to be recorded and captured already signifies surveillance. To put in the context of research, the algorithms capture the data input but it is the system in place

(16)

- 16 -

which allows algorithms to capture it – because it is already being surveilled. This research will use Agre’s idea of capture to show how Facebook algorithmically collects data. I will especially focus on grammatization looking analysing the Business Manager interface and its properties, as the process is highly defined by rules and only allow certain actions. The fact that particularly those actions are allowed is crucial signifier of grammars of action in place.

2.3. On metrification and value creation

Research will be based on literature on quantification and metrification, also social media metrics and finally will touch upon value creation. There is an obvious relationship between Facebook and economy as it makes a large profit from advertising and, in this way, from user engagement. The Facebook business model is constructed on this “attention economy” (Terranova 2014) which is the key element in advertising, as it is consumers “eyeballs” that are being sold to businesses. I have already explored the relationship between reactions on Facebook and their conversion to monetary value. However, in this section I would like to delve deeper into the value creation and how do these reactions translate into monetary terms. To do so, I will explore literature on metrification and value creation.

Benjami Grosser is one of the key academics I will rely on, as he wrote on metrification of Facebook in particular. In his words, “within the context of Facebook, I define “metrics” as enumerations of data categories or groups that are easily obtained via typical database operations, and that represent a measurement of that data” (Grosser 2014, n.pag.). The way Grosser understands metrics supports my findings on engagement, as I also explore the idea of user actions collected by the algorithm and how they are translated into numbers. Grosser (2014, n.pag) also highlights that not all metrics are seen on the user side of Facebook and inquires which numbers are left hidden only for the platform’s usage.

In the Like Economy, Gerlitz and Helmond (2013) already outlined that “various affective reactions to web content in the form of a click on a Like button, these intensities can be transformed into a number on the Like counter and are made comparable” (11). However, we should not limit the analysis only to the reactions, because what are also turned into comparable numbers are clicks, shares and comments. It is important to look into this process of the algorithm quantifying engagement because it is exactly the value of these numbers that directly translate into profit. On the business side of Facebook, creating adverts are based on formula “pay per like” (Advert Creator interface on Advert Manager). Which means, that the more users engage with the content, the more money Facebook earns.

(17)

- 17 -

Gerlitz and Lury (2014) show that there is an increasing tendency to measure and quantify the effect on social media. It can be argued, that what social metrics service Klout calls influence is can also be called engagement: the platform uses algorithms to measure the reactions4 to published content on user’s social media through Klout. Also Gerlitz and Helmond write that “users engage in productive activities without financial reward” (14) and that resonates to what Fuchs (2012) and Terranova (2000) write about digital labour. Terranova writes that “attention is both scarce and measurable, it can become not simply a commodity like others, but a kind of capital” (Terranova 2012, 2). Arvidsson and Colleoni (2012) add to this by claiming that “value is ever more defined according to the ability to mobilize affective attention and engagement” (144). This definition of value demonstrates how understanding of engagement has changed in the context of platform economics. Terranova also highlights how measurement and attention are related in relation to profitability: “attentional assemblages of digital media enable automated forms of measurement (as in ‘clicks’, ‘downloads’, ‘likes’, ‘views’, ‘followers’, and ‘sharings’ of digital objects) open it up to marketization and financialization” ( 2-3). It has to be noted how the author focuses on the fact that measurement is automated and that it is the automation that makes monetization of attention less difficult.

This chapter has reviewed academic literature to set up theoretical framework for the research. I have looked into literature on engagement which provided more than one outlook to the concept. I have seen that models on user engagement vary, and so does the definition and understanding of engagement. I could already see that measuring engagement always takes numerical form which can be questionable if quantification is the best option. Looking into platform studies it was possible to identify key arguments in relation to Facebook: issues of platformization, data economy and intricateness of News Feed algorithm were identified. They all are all major contributors to how engagement is perceived by Facebook. I have also looked into the concept of “grammars of action” which provided me with the perspective to look at user and business side of social network in terms of data captured. Lastly, I have looked into literature on metrification, quantification and value creation which offers valuable critique of how Facebook works.

4_{“Klout takes records of activities pre-structured by a number of social media platforms, including ...}

(18)

- 18 -

3. Methodology

Consumption in general is increasingly digitalized. This does not only affect consumption itself, but also the study of consumption, inspiring novel methodologies such as netnography (Kozinets 2009, n.pag.), digital ethnography (Caliandro 2014, n.pag.) and digital methods (Rogers 2013, n.pag.). As described by Rogers, “Digital methods are techniques for the study of societal change and cultural condition with online data” (Rogers 2015, 1). The method of this research itself will be a combined look of medium-specific perspective into Facebook, which will be based on platform studies theory and also an empirical study of advertising campaign on Facebook. Literature review is also one of the methods used in the research as setting the theoretical framework is the key to guiding how the argument will be approached and structured. In terms of literature, most of the research will rely on platform studies perspective, combined with Agre’s model of capture and grammars of action. Furthermore, to better understand how data is captured and how user decisions are steered, I will use methods of affordances analysis. One of the key methods will involve discursive interface analysis which will provide knowledge into the aesthetics of the interface.

3.1.Platform studies approach

Platform studies is the key schools in terms of structuring the argument as I will use platform studies framework to look at how social network is part of the platform economy in capturing and measuring engagement and how advertising is orchestrated through software. This research will look into algorithmic part of Facebook Business Manager interface as well as how is engagement turned into collectable data. Using this approach is crucial to critically analyse and understand what lies behind the social network. Critically looking at the algorithm will help to understand how it captures user data, how it quantifies it, ranks it and feeds it to the business side of Facebook. Part of platform studies is analysis of affordances. Mel Stanfill (2014, 1-16) uses discursive interface analysis looking at different websites. She analyses the affordances of interface in terms of both function and appearance and this very approach will be appropriate and useful for this research as well. She writes that: “discursive interface analysis is a productive tactic” because it “illuminates the norms of use” (Stanfill 2014, 3). It does so by observing and noting down what is afforded by an interface, how the processes are structured and what kind of interaction occurs between the user and the platform.

(19)

- 19 -

Here we can see that much like grammars of action, discursive interface analysis, especially, what Stanfill calls cognitive affordances, are largely influential and that itself provides a reason to examine them under a microscope. As the object of study is the advertising interface of Facebook, I will use discursive interface analysis to look into the elements of the Business Manager. Captured screenshots will be heavily used to aid the argument and provide examples for the cases analysed. Visual elements are of crucial importance towards the analysis because aesthetics is also an important element for engagement (O’Brien and Toms 2008; Cvijikij and Michahelles 2013; Stewart 2014). Aesthetics are only one of the elements that Stanfill looks into. She divides her study into cognitive, functional and sensory affordances.

Research that adopts discursive interface analysis will consider site functionalities – literally, what does the site do? – as functional affordances; the uses that are made easy to find with sensory affordances like placement, sound, color, or motion; and how these features are named, classified, or explained (cognitive affordances). (Stanfill 2014, 13)

Agre’s grammars of action will be one of the key ideas used to identify data capture by Facebook. His privacy model (1994) is important as it explains how actions are grammatized. It will be looked at the advertising manager interface through the perspective of grammars of action (Agre 1994), for example, how an advert is created: software guides the user through a series of steps until the advert is ready to be sent for approval. It will be looked at the affordances of adverts manager interface: Facebook suggests to “boost” separate posts as well as run wider-applied campaigns. This view will allow the research to demonstrate how the “capture model” works in the context of social platform like Facebook. Looking at how actions are translated into data poins and then collected is important as later the algorithm translates it to engagement. All of reactions and actions taken are given a numerical value by the algorithm.

Software analysis will also be used in order to understand the algorithm. The research will observe how the algorithm behaves to try to understand how Facebook values and quantifies engagement. Of course, as I have outlined before that Facebook’s algorithms are not transparent, so we cannot simply “open” them up. Because of this issue, the research will be another example of why we need to open up the black boxes together with the managerial decisions that come into them.

(20)

- 20 -

3.2.Platform economics and metric analysis

As Facebook is not only social network but also a profitable business, this approach is important to look into how sociability is translated into monetary gain. Platform economics critique will be used to critically engage with how this platform is making profit from user information. Also metrification critique will be used to look into how Facebook perceives engagement and displays it for businesses. Gerlitz and Helmond, Gerlitz and Lury and Grosser are critical of the process of metrification on social media and their discourse will be particularly useful for the analysis of how the Business Manager quantifies the results of an advertising campaign. Furthermore, I will use the analysis of the Like button by Gerlitz and Helmond to develop the argument on reaction buttons and wider implications of ‘liking’ activity online.

3.3.User engagement

To compare to how Facebook is capturing, measuring and ranking user engagement, this research will turn to literature on engagement. This is why the research will include academics and marketing specialists who provide ideas on engagement. The theory on user engagement models will be used to compare and contrast with the way Facebook defines and captures engagement. For example, Ponciano and Brasileiro (2014), Lehman et al. (2012), Chaffey (2010) provide different engagement models. Lehman et al. focus solely on formulas of calculating and comparing engagement which is what Facebook does with its algorithm. Academics out stress on different parts of engagement and this is what I will focus on in comparison with how Facebook defines engagement. However, for my analysis, I will not use the formulas provided by academics as I try to show that quantification is not the only way to calculate and display engagement. Engagement models will aid in finding alternatives to capturing and measuring engagement.

3.4.Capturing campaign data and visualizing tools

Online tool Infogr.am was used to develop results into visualizations. Bar diagram (Figure 10) was used to show the reach and engagement in juxtaposition with amount spent for advertising. To do so I have divided the results by the amount of money spent per campaign. I have done the same with the size comparison chart of two posts (Figure 11). I chose a size comparison chart to display different campaign results on the same scale, so the sizes are more comparable than with a line graph or column chart which are scaled by the value of the largest unit. For the last graph (Figure 12) I have left the results as they were

(21)

- 21 -

collected on Business Manager interface, to show display results of two adverts, yet I have changed the icon to thumb which resembles Facebook ‘like’ icon. For all the charts, I have set different colours for each result and chose to display values. Figures were captured with a Snipping Tool, as Infogr.am only allows downloads for premium members.

This chapter has outlined the methodology for the research. After reviewing literature and defining the methodology I will move on to the empirical research starting with outlining the case study.

(22)

- 22 -

4. Empirical analysis and findings

This section will be based on empirical research and its findings. I will first introduce the case study: Facebook itself and the advertising interface in particular. I will present a realtime advertising campaign which was used to obtain insights on the advertising process on Facebook. Using developed methodology from the previous chapters, the case study will be analysed. Through the platform studies perspective the interface will be analysed looking at its design and functionality. I will employ Agre’s capture model and literature on user profiling and data economy to examine how an advert is created in a few steps: setting the audience, the budget and adding text and media to it. I will then look how sponsored content appears at the other end, for users, and how they engage, what were the results of the campaign. Lastly, I will look at how the results appear for the business.

4.1.Introduction to case study

The history of Facebook is rather well known. Emergence in 2004, (Phillips 2007, n.pag.), fast development from college platform to wide world, hitting one million users in 2004 and growing ever since, reaching close to 2 billion in recent years (Facebook

Newsroom). From the beginning Facebook declared its mission as “to give people the power

to share and make the world more open and connected” (Facebook Newsroom) and this marketed sociability has been debated and problematized among academia raising questions of distant communication, social anxiety, creation of carefully polished profiles and so on. Berg (2011) successfully identifies on the reasons why it is important to look at this platform: “people become increasingly dependent on the social network in order to participate in the main currents of online social life” (333). In fact, there is increasing occurrence of FOMO, or Fear of Missing Out (Dossey 2014, 69-73) – a social anxiety caused by being absent from what seem important social interactions and events happening on social media – which shows that often users do not have another choice but to participate and engage, otherwise they will not be involved in their social lives at all. What Berg (2011) highlights, is the importance of social media in everyday life, and because of that, we should put it under a microscope. It is worrying, how natural it is for us to perform our identity online and how we tend to select what content reaches the platform. This “programmed sociality”5

(Bucher 2012) became part of our lives yet we do not spend enough time and effort to critically examine it. The larger

5

In her PhD dissertation Bucher (2012) uses the term “programmed sociality” to describe how social relations and activity on social networks is constructed through software. She argues that influenced by various factors, our participation on platforms can be shaped, transformed and, essentially, programmed to a larger extent.

(23)

- 23 -

issue arises when, as a result of targeted advertising, sponsored content reaches user and is relevant to them which “nudges” them to engage with it. Part of why Facebook advertising is so successful is that it allows profiling users (Bodle 2011; Bechmann 2013) and this way users are exposed not to the mass generated content which they could ignore from the first glance but rather to content which is meaningful to them (as much as advertising in itself can be relevant).

Since the creation of the platform, creators had goal set as being profitable and making money from advertising which contrasts with their promoted connectivity and culture of sharing (Van Dijck 2013, 45). Facebook expanded throughout the years as business gaining more and more money from advertising and part of this platform ecology is also numbers and quantification (Grosser 2013). This is why it is important to scrutinize Facebook because users often forget that it is a business. They take its being free as a benefit while not questioning how it can remain free yet profitable. The fact that Facebook makes money profiling users and trading data is important to look at. Even when people know it is a profitable business, they still trust the software behind it: this is the reason why in his blog, Gillespie often urges users not to trust the fairness of the algorithm (2016). Langlois and Elmer (2013) write that social media platforms put sponsored content into “acts of communication” (4). In other words, any advertising that appears on one’s feed is potentially there to be engaged with, and often is. Not only that, these “acts of communication” are set to multiply (Gerlitz and Helmond 2013, 1-18) as the content user engages with appears on their friends’ feed for them to potentially engage with. Lastly, it is important to look at the social network not only from the user perspective. Businesses pay money for advertising because they want to reach consumers, yet it is not enough to reach them, in this proliferation of businesses, it is also engagement that they need and engagement that they pay for.

4.2. Advertising interface

This research focuses on social network Facebook and its advertising interface. This section will further explore the key elements of the advertising interface: Business manager, Adverts manager, Power editor and on-page boost. In the text I refer to the interface as a whole or just by calling it Business Manager, yet at all times it includes all components of what encompasses the advertising interface with on-page boost and others. For its own convenience – in other words, to make the interface more profitable – Facebook divides the interface into different regions. There is news feed, the main section of the interface, side bar

(24)

- 24 -

for menu on the left side and side bar for notifications, recommendations and paid content on the right. Facebook differentiates between desktop feed and mobile feed as mobile screen is a lot smaller and offers less space for advertising.

The following is a snapshot of the Business Manager (Figure 2):

Figure 2. | Business Manager interface homepage includes the overview of total spending, and has links to the Advert Manager and connected pages which belong to the account.

Business Manager is a dashboard for all business related activities. It is the starting point displaying all the accounts and pages connected to a particular user. It is here where the overall spending throughout all accounts and statistics are shown. Created as a dashboard this interface provides links to managed pages, other advertising interfaces and direct link to create an advert. Interestingly, the overview section provides only two variables for per section, so by advertising it only shows total spend and total impressions, while by page overviews it displays post reach and post engagement per monthly or weekly period. Limiting access to data may be platform’s way of directing the user to use the Advert Manager, or page insights were metrics are provided in greater detail. According to Stanfill (2014, 5), this would be a cognitive affordance. It is important to look at the sensory affordances (Stanfill 2014, 10) of this interface as well. The page focuses on clear graphs and large numbers. The first thing catching user’s attention is the enlarged number (even if it’s a zero) as the amount

(25)

- 25 -

spent with a representing graph below, and next to it there is another enlarged number of impressions and the graph. It is interesting that of all the metrics, Facebook chooses to display impressions. That is likely because impressions are always a larger number then reactions, yet it also implies a form of engagement, comparing to reach. Impressions “are the number of times a post from your page is displayed, whether the post is clicked or not. People may see multiple impressions of the same post. For example, someone might see a Page update in News Feed once, and then a second time if their friend shares it” (Facebook Help 2016). This notion is critiqued by Cvijikj and Michahelles (2013, 9) as there is no action taken upon impressions. To add to the interface analysis, it also offers two graphs on page metrics, yet advertising statistics come first. On the interface, twice does the platform has options to create an advert. First, as an action button, and then part of the menu. Below the initial overview user can see pages and the advert account both of which have another graph each with some key numbers. Importantly, the three enlarged numbers representing the overview of the page display new page likes, post reach and engagement. Reach “is the number of people that have seen the person can see multiple impressions.” (Facebook Help 2016). In other words, this is the number of times Facebook’s algorithm places content in news feed or sidebar for a user. The Business Manager interface also compares metrics over the past week period. The fact that Facebook focuses on recency – all metrics are by default provided for over a week – shows that short-term campaigns are a priority for Facebook. This can be seen in regards to Facebook offering to promote a post on the page event for a day.

Adverts manager provides an overview of the campaigns and allows creating new adverts. Adverts manager is highly oriented around statistics. The user who manages adverts can see the performance of the adverts broken down in multiple sections from reach and engagement, to audience demographics and advert placement (in news feed, on mobile or even on Instagram). Adverts manager also can be used to display graphs while changing variables. For example, for demographics section, the variables are: results, amount spent, reach and impressions. Any two of these variables can be arranged into two columns for comparison. Analysing the interface it can be seen that Facebook uses its brand colour blue but also green to activate the user. On Advertising Manager interface colour green is used to show the amount spent on the campaign and also for the background of the ‘create advert’ button (which can be considered a sensory affordance, and likely a cognitive one as green colour stands out of the background and suggests to press on it). The main section of the interface is the information provided in the form of a table. It could be argued that such a

(26)

- 26 -

design appears more professional and is exactly the reason why Facebook displays the data this way. The columns are actually highly customizable, but of course with the provided metrics by Facebook – for example, there is no way to choose engagement rate for the advertising – it is only available for the page insights.

It is unclear if Facebook deliberately distinguishes between the metrics provided on the page and Advertising interface or is it because they simply do not want to offer too many options.

Figure 3.| Adverts Manager interface. Offers overview of all campaigns, results and spending.

The third part of the advertising interface is Power editor (See Appendix 2). It offers more detailed analysis, more statistics, less graphs. In terms of sensory affordances, this interface looks more complicated and detailed. There is a lot of information and options to display it available but no explanation what certain metrics mean (in comparison to the Advert Manager). Power Editor does offer to compare different campaigns on the table format which is something that was missing on the other interfaces.

On the interface of every page there is a possibility to promote content on the Page dashboard which I named ‘on-page boost’. As mentioned above, there are three main buttons two of which include word “promote” and one “boost post”. Promoting page or website result in the same advert creator as with one of the advert interfaces. Boosting a post is a slightly different action: it creates an advert for a particular post and requires fewer steps to be set in motion while offering same options. These buttons on the business side of Facebook work the

(27)

- 27 -

same way buttons such as ‘Call’, ‘Buy’ or ‘Book’ work on the user’s side. Stanfill (2014, 10) highlights the use of these imperative verbs as it shows what the platform wants you to do – one of the cognitive affordances. In the case of users, these buttons are placed by advertising – businesses can choose what button appears on their page or in their advert, so not only Facebook uses this affordance for its monetary gain on business side, it also utilizes the same technique to attract users to businesses.

What can be seen clearly from analysing the advertising interface of, is that graphs and numbers are used on every page. All elements are focuses around metrics and various sections provide different look to the same numbers. Also all measurements are used to display effectiveness in one way or another. The third key element is that many of the numbers displayed are representing money as the amount spent. This is understandable as it is the business interface and money is at the centre of business making. Yet this also shows how Facebook frames advertising in relation to engagement: all is provided in numbers, percentages and amounts of money.

4.3.Amber’s Page promotion campaign

To observe advertisement setting and engagement the research will use a real time advertising campaign. Amber is a 19 year old singer starting her career. The advertising campaign evolved around the release of her first single, “Look Around”. The video clip was shared on her Facebook page. The advertising campaign included promoting three posts which contained video clip on her page, promoting her page for different audiences and promoting her official website. The period of advertising campaign was April 8-12. This selected period was rather short in order to maximize the engagement with the release of the record. There were two advertisings taking place from 8th to 12th of April: Page Likes advert and Website Clicks advert. The other two advertisements were two post promotions 8-9th April and both included the video clip of the song released (See Figure 4). It should be explained that Facebook considers that advertising campaign can be made up from multiple advert sets. Yet, even if the campaign consists of one advert, it is called a campaign. However, I would propose a more traditional use of the word ‘campaign’ – for all advertising run together at the same period of time, and I will refer to the campaign as a whole.

(28)

- 28 -

Figure 4.| Second promoted post on Amber’s page.

4.4.Advert creation analysis

Facebook creators made sure that businesses would be constantly offered the opportunity to promote their content. There are three interfaces providing different insights into statistics, yet leading to the same advert creator interface. There are also multiple buttons urging to ‘Promote website’, ‘Promote page’ and ‘Boost post’ which are all names for different kinds of advertising. If marketer presses ‘promote page’ they are taken to the advert creator. The step-by-step process (See Appendix 3 for examples) guides the user through creating an advert, the process is the same every time, unless one wants to ‘boost post’ then the first few steps of setting up the campaign are skipped. However, this guidance shows how the platform structures the advert creation process. Creating an advert begins with choosing the goal of the campaign and then naming the campaign (See Appendix 3). It usually follows formulated structure of “Page title – Campaign goal” which then becomes, for example, “Amber – Page Likes” campaign. In further sections I will look into three most important steps – setting the audience, budget and duration and creating text – in greater detail.

4.4.1. Setting the audience

Facebook Advert Creator allows narrowing down the target group to the smallest details. For example, depending on the post type, some posts were targeted only a county of North Brabant while others expanded to whole Netherlands and neighbouring countries. All audiences were set to be both male and female and there were two age groups: 18-65 and 35-65. Audiences varied in music related interests, age. For example, to promote the video clip, the audiences had to like one of the following: ‘Singing’, ‘Michael Bublé’, ‘Celine Dion’, ‘Norah Jones’, ‘Jammie Cullum’ or ‘Diana Krall’. Setting the audience offered variations of

(29)

- 29 -

interests, behaviours, demographics and to choose from (See Figures 5 and 6). Observing the process of setting the audience for Amber’s campaign allows us to observe how profiling on Facebook work in detail. In larger context, user profiling is a very important component of data economy: provided user data is crucial to running a platform such as Facebook. Data economy landscape is also part of platform economics landscape, because of the clear relation between information collection, its usage for advertising and profitability for the social network.

Figure 5.| Targeting audience on Facebook Advert Manager interface.

Figure 6.| Targeting audience as appears when creating advertisement on Facebook Advert Manager interface.

In both of the above figures, we can see how Facebook divides its users according to how it profiles them. The variety of options show, firstly, how much information can be

(30)

- 30 -

provided and collected from user’s profiles. The more information is provided, the easier it is for Facebook to apply targeted advertising to these users. Secondly, we can see how Facebook differentiates between demographics, interests, occupation and behaviour.

It is interesting to see how Facebook regards “family and relationships” as an interest and targets anyone who “expressed an interest in or like Pages related to” (See Figure 7) a particular hobby or activity. This information is collected and profiled over thousands of users’ actions, and as Facebook shows, a mere “interest” is enough to mark a user as a potential audience. Following this logic, let us consider an example: a user sees that their friend ‘liked’ a video about dating. The said user then thinks the video is funny and ‘likes’ it (or presses ‘haha’ for that matter). This action is captured by algorithm and may already be regarded as ‘interest’ in dating and this user may become a potential audience for the advert which is target to people who have dating as their interest. Of course, Facebook collects data continuously, so it is able to create a pretty detailed profile. Which is one of the reasons for Facebook’s success as a business (Gerlitz and Helmond 2013; Bodle 2011; Bechmann 2013). What is interesting to look at, is that Facebook distinguishes only these few categories as behaviours: Consumer Classification (followed by sub-category India), Digital activities (sub-categories of gaming, paying on Facebook), Expats, Mobile Device User (sub-categories on detailed options mobile devices), Seasonal and Events (sub-categories of sports and worldwide events), and Travel. This category is hard to define because of the variety of interests and activities involved but also because not all of these interests would be called ‘behaviours’ in a traditional sense. Also, for the occupations, Facebook offers to choose an industry, except for banking, design, personal finance and ‘online’, out of which all encompass a few categories but ‘online’: this category has ten sub-categories including social media, online marketing, display advertising and the like. This detailed categorization only shows what data Facebook considers important enough to be detailed. Of course, Facebook does not limit advertising to what it suggests suitable audiences, any word can be typed in and pages and interests will appear with an option to be selected as an interest of an audience.

(31)

- 31 -

Figure 7.| Targeting audience as appears when creating advertisement, on Facebook Advert Manager interface.

When setting the audience, Facebook already predicts engagement and the reach on the right-hand side (See Figure 8). Even when narrowing down to a particular audience group it displays the number of people who are part of that interest (See Figures 6 and 7). The more interests are added to the targeted audience – the larger is the predicted reach. The design is like a car speedometer which also shows if the audience is too specific or too broad. As a sensory affordance, Facebook uses a rather simple graphics of a pointer and colour schemes of red, green and creamy yellow, to show to signify if the audience set is too specific or too broad. The platform does not explain why it is not recommended to make the audience too specific but it is because if the amount of people is too narrow, then it is harder to predict positive engagement and larger reach for an effective advert. Basically, Facebook suggests to make the target audience wide enough that it could be displayed to enough people of out which the amount which would engage would fulfill the campaign budged and would bring profit to Facebook. Also, it should be noted that as mentioned, there are metrics of potential total reach and estimated daily reach which both show how the interface is already oriented into results and effectiveness: to make sure that the user places the advert it displays that results will be beneficial to the business and that advertising is the right choice for the user. The fact that social network is focused on performance and effectiveness before even starting the advertising reveals how Facebook presents engagement to the business – though potential numbers. This also contributes to their definition of engagement.

Making engagement profitable: Relationship between user engagement and advertising on Facebook