Shaping the Digital Consumer

(1)

Shaping the Digital Consumer

The psychology behind the data-driven construction of consumer behaviour

Research Master Thesis Media Studies

(2)

1

University of Amsterdam

Graduate School of Humanities

Submitted to the Department of Media Studies at the University of Amsterdam, Faculty of Humanities, in partial fulfilment of the requirements for the degree of Master of Arts (M.A.).

E.M. (Eline) Meissen

29 Jun 2018

Study Programme: Media Studies (Research) CROHO-Code: 60832

Course: Research Master Thesis Media Studies Course ID: HMED/159414040Y

Supervisor: Dr. T. (Thomas) Poell Second Reader: Dr. B. (Bernhard) Rieder

(3)

2

Abstract

Consumer webshops like Netflix and Amazon gather data from their users which is used for

persuasive purposes by combining psychological research with digital techniques. This thesis answers the following research question: How are persuasive techniques employed by large commercial websites informed by psychological research? It addresses this question by first making an inventory of the most commonly employed digital persuasive techniques, both from a frontend as well as a backend perspective, and, subsequently, by exploring the psychological research on which these techniques build. This research aims to contribute to platform and software studies.

Personal data is used in order to create interfaces that adapt to the person using it, to optimize the chances of nudging them towards specific actions, which is called hypernudging. This research adds to the fields of platform and software studies by providing insights into how these persuasive practices are actually being employed by businesses, drawing upon psychological theories from behavioural and neurological psychology. Businesses have proven to actively engage in persuading consumers emotionally by targeting specific personal characteristics, using data on for example personality, emotion and culture. Above all, this research shows how digital media now allow a new form of hyper-personalized persuasion, which may affect the position of power of consumers.

(4)

3

1 Introduction

“We’re obsessed with creating a shopping experience that helps people easily find and purchase the items they need, an experience that caters to the unique interests of each individual shopper. We invent, build and manage features that give Amazon customers the feeling that we really know them.” (Amazon Personalization n.pag, emphasis added)

On the 10th_{of May 2016, Amazon.com made the code from their deep learning recommendation}

system DSSTNE (pronounced “destiny”) open source. This system, which is built to generate personalized shopping recommendations, was made open source because the Amazon developers hoped to get outsiders’ input to incorporate speech, language and object recognition in the fields of search and recommendation (Amzn n.pag). This system enables a highly personalized form of product recommendation through the repurposing of big data on things like events, ratings and filtering. Amazon tracks and stores all activity from individual users on their website. Every click becomes a record in a database, which DSSTNE can then use to recommend products a user might be interested in buying. A simplified example of an event-based data entry would be: “User A clicked product X description once”. Recommendation systems like Amazon’s also work with ratings. This means that they assign implicit values to different user actions, of which a value of five is the highest. A user buying or liking a specific product would be rated higher in this system than a user merely clicking on a product’s description. When repurposing this data for product recommendations, the system sees the highest rated actions as the most important and most likely to lead to a successful recommendation. Every individual user’s choices are also compared via what they call collaborative filtering: if one user for example likes products A, B, C, D and E, another user who likes products A, B, C and D is also likely to enjoy product E and thus this product gets recommended. Collaborative filtering of course builds upon the activity of many users and data points in order to make recommendations as accurate as possible. (Smith, Linden 12-18). These are only a few examples of techniques that are incorporated in consumer recommendation systems. Amazon applies these techniques to their more than 310 million active users (Statista n.pag) who are trying to find their way through millions of products, which makes it even more complicated. This is why Amazon needs a deep learning system like

DSSTNE that, built on an artificial neural network, can learn from existing data to make more accurate inferences about new data. These systems are modelled after the presumed workings of the biological nervous system as it aims to define relationships between multiple stimuli and possible responses (Schmidhuber 86). A neural network consists of many small units called neurons, grouped into

(6)

5 are considered more important than others. When one node is triggered by input, this node also

triggers those it is connected to. This way, when an Amazon user clicks on a certain product, their neural network decides which products are connected to that product. The network can then determine which other products are most important in their connection to the first product, and the user’s

interface can be personalized accordingly.

The Amazon DSSTNE deep learning system is just one example of the technology used by large online shopping companies like Amazon to be able to provide their customers with a

personalized user experience. As shopping increasingly becomes an activity that takes place online, businesses start to embrace the benefits of digital media (like social media, algorithms, interface design etc.). Making someone buy a product or service is probably the most important goal businesses have when running a consumer website, and digital media like the Amazon DSSTNE system are used by these businesses to achieve this goal. This essentially means that businesses are using technology to steer consumer behaviour, and to try to persuade them towards spending their money. In order to know how to persuade people, businesses turn to the field of psychology. This is shown for example by the growing body of academic literature that combine the fields of economics and psychology. These two come together in interdisciplinary fields like behavioural economics and neuroeconomics.

Using psychological theory for marketing related purposes is not new, Walter Dill Scott for example wrote one of the first books on this topic in 1903 called The Psychology of Advertising in

Theory and Practice. One of the most important insights of this book was that influencing consumers

works best by evoking emotions, sentimentality and sympathy, and that advertising should not be seen as a source of information but as a means for persuasion (395-408). It comes as no surprise that the combination of psychological theory and marketing related practices have also made their way into the sphere of online consumerism. Large and influential companies are rich in data and resources. One could only imagine how insightful the data of 310 million Amazon users can be in combination with psychological knowledge and digital techniques. Many theorists have written about consumer construction and steering consumer behaviour in fields like platform politics and software studies. There has been critical reflection upon the role algorithms and interfaces play in relation to persuasion, but relatively little has been written about the role psychological research plays in these practices. This thesis aims to provide an insight in how psychological research informs the development of persuasion and steering- techniques. It looks at how psychological theories are used by commercial businesses within the space of digital media. It does so by answering the following research question: How are persuasive techniques employed by large commercial websites informed by psychological research? It is crucial for people like academics, lawmakers and government leaders to have access to knowledge about the psychological theories and practices involved to be able to take political actions and

formulate realistic policies. This knowledge can play a central role in keeping power relations between consumers and companies in balance, which will make it easier to make informed decisions about what the world of online consumerism should look like.

(7)

6 The thesis starts with a theoretical discussion of the research on consumer persuasion and construction as it has been developed by academics from the field of software studies and platform politics. In doing so, the thesis discusses different ways in which digital media can play a role in altering behaviour and shows how this has already been theorized in these fields of research. This theoretical discussion features some of the key studies on this topic. While these studies provide valuable insights, they rarely get into the psychological foundation of these techno-commercial practices. After a brief methodological subchapter, this thesis will present examples of how large commercial websites steer consumer behaviour, and will then trace back some of the most important psychological theories they build on. This is done by analyzing literature, for example marketing handbooks or blogs, marketing psychology scientific publications and publications by businesses themselves. Amazon and video streaming platform Netflix for example, both run a blog that discusses technological developments regarding the websites. These industry blogs and publications, in

combination with examples from actual consumer websites, are then discussed in relation to the platform politics and software studies oriented theoretical discussion. By looking into how these theories are actually employed in combination with digital media, from a frontend as well as a backend perspective, this thesis will discuss the implications of the research on how digital techniques for persuasion might affect the consumer.

(8)

7

2 Personalizing the digital space: data-driven persuasion

The rise of digital media introduced new forms of industrial competition (Turow 26): using digital media means leaving a digital footprint, what you do on the internet is captured into data, data which companies can gather and sell as a commodity. This practice effects online as well as offline shopping. Imagine walking into your nearest supermarket: The facial recognition system notices you entering the store through footage from the security cameras. It matches your identity to data that was bought by this supermarket from a data company. This data reveals that at this time you are generally hungry and you often spend an above average amount on groceries. The nearest supermarket employee gets notified about your presence and your potential value to the store, and he comes to greet you and offer you shopping assistance. Physical as well as online stores can and are using the data you leave behind while being online in order to determine whether you are valuable enough to receive certain forms of service, as well as the things they need to do in order to make you spend your money (ibid).

José van Dijck argues that the high level of tolerance against the practices of businesses routinely accessing citizens’ personal information is remarkable. According to her, there is an ideology of dataism, which entails a widespread belief in the objective quantification and potential tracking of all kinds of human behaviour and sociality through digital media technologies (198). She references Mayer-Schönberger and Cukier, who discuss datafication as a new paradigm in science and society that has gradually become normalized over the past decades. Citizens have generally started accepting that platforms share personal information in exchange for services like social network sites or

applications (van Dijck 197). Data, in other words, appear to have become a form of currency, that citizens knowingly or unknowingly give away in order to get access to technology (198). In general, datafication is known as a technological trend which involves turning aspects of human lives into computerised data. This data, and the insights that are gained because of it, are then turned into new forms of value (Schutt, O’Neill 406). One reason why datafication has introduced new forms of competition amongst businesses is because it allows for predictive analytics. It helps businesses “foresee events before they happen” (Mayer-Schönberger, Cukier 79). Datafication enables businesses to make informed and profitable decisions based on actual consumer data; Schönberger and Cukier for example note how search inquiries from consumers allow businesses to make “words become data” (83), which is useful for businesses to get an understanding of what it is that their consumers need. Van Dijck discusses how datafication enables analyzing and predicting citizens’ behaviour, and her argument is focused around the practice of repurposing social data: “Quantified social interactions were subsequently made accessible to third parties, be it fellow users, companies, government

(9)

8 prowess on the value of data and metadata” (199). Metadata is data that describes or gives information about other data, which until recently were considered to be worthless by-products of platform

mediated services. But metadata has gradually started to be considered as a treasured resource that can be mined and repurposed (ibid). Twitter for example, can supposedly be seen as a platform that, through datafication, captures people’s sentiments, thoughts and gut feelings. According to Mayer-Schönberger and Cukier, Twitter enables metadata to be “collected passively without much effort or even awareness on the part of those being recorded” (101). Van Dijck mentions once that social media platforms are able to manipulate online human behaviour because of datafication (199), and she argues that “Scientists, government agencies and corporations, each for different reasons, have a vested interest in datafied relationships and in the development of methods that allow for prediction as well as manipulation of behavior” (200). But besides this, she does not explore the topic of manipulating behaviour any further. What she also does not get into, and neither do Mayer-Schönberger and Cukier, is how individuals can be manipulated with repurposed data. Although they acknowledge that

repurposed data can be a powerful tool for third parties to influence individuals, they do not explore how this actually happens.

Before touching upon the psychological theory that businesses draw from when they engage in steering consumer behaviour online, this theoretical discussion about persuasion and the construction of behaviour in relation to digital media will first of all get into the digital techniques that make this practice possible in the first place. The authors mentioned above discuss datafication, which is largely made possible by the technical details that constitute the backend of digital media and are not directly visible to someone using a website or application. Code, and more specifically; algorithms, are written and built to mine, analyze and repurpose data. The first subchapter, Algorithms, personalization and

persuasion, will feature theory that discusses these techniques and how they can contribute to

producing behaviour. Much has also been written about how the frontend of digital media, the interface, contributes to the production of user behaviour. These ideas, which are rooted in design theory but which also play an important role in the field of platform politics, are discussed in the second subchapter called Persuasive interface design. It then becomes clear what the theoretical discussion is lacking and how this research can contribute to gain more insight into the practices of online persuasion and steering consumer behaviour. The final subchapter Research methodology revolves around the practicalities of the research and shows how it has come together

(10)

9

2.1 Algorithms, personalization and persuasion

Algorithms are task-accomplishing methods, written in code but independent of programming

language and independent of the machines that are used to execute them (Goffey 15). Andrew Goffey points out how algorithms have material effects on their end users, and one of the examples he gives is that businesses are using data mining techniques to predict and alter consumers’ shopping preferences and behaviour. However, he argues that algorithmic effects go beyond that, and that they should be taken seriously as they play an important role in how contemporary life is shaped socially, politically and culturally (ibid). Important here is that Goffey seems to refer to algorithms as having agency to some extent: “Algorithms act, but they do so as part of an ill- defined network of actions upon actions, part of a complex of power- knowledge relations, in which unintended consequences, like the side effects of a program’s behaviour, can become critically important” (19). In 2011, internet activist and political scientist Eli Pariser makes clear what these consequences can be and why this is important. He put algorithmic personalization at the centre of public concern with his critical book about the filter

bubble. Pariser points out that with every digital interaction, people leave behind bits of data that

reveal personal information, such as their interests, politics, education and even dating history (“The Troubling Future of Internet Search” 6). The data is then repurposed, for example by Google, to present their users with personalized search results. This means that different individual users are shown different search results based upon what the company knows about the user through their digital trail of information (ibid). But Google is just one example of a company engaging in

algorithmic personalization. As discussed before, social media platforms like Facebook personalize their user’s News Feed, and webshops like Amazon algorithmically select which products are shown to consumers based on their personal data. Pariser is critical of this development and argues that algorithmic personalization leads to users being isolated within their own personal filter bubbles:

The basic code at the heart of the new Internet is pretty simple. The new generation of Internet filters looks at the things you seem to like – the actual things you’ve done, or the things people like you like – and tries to extrapolate. They are prediction engines, constantly creating and refining a theory of who you are and what you’ll do and want next. Together, these engines create a unique universe of information for each of us – what I’ve come to call a filter bubble – which fundamentally alters the way we encounter ideas and information. (The Filter

Bubble: What the internet is hiding from you 7-8)

He explains that algorithms can be beneficial for society, for example when they are made blind to race and gender and are used to present individuals with job opportunities. However, when algorithms are designed for personalization purposes, the results instead are likely to reflect “social mores of the culture they’re processing – a regression to the social norm” (“The Troubling Future of Internet Search” 6). Eventually, this can also lead to serious issues such as discrimination, like when

individuals are declined loans from banks because their data reveals that they are friends with people who have failed to pay their debts. Algorithmic personalization is essentially a process of prediction.

(11)

10 Another example of this is Facebook’s algorithmically constructed News Feed, which is the primary place on Facebook where users are presented with content. their sorting algorithm predicts which content a user is most likely to actively engage with. In case of Amazon DSSTNE, the algorithm predicts which product a user is most likely to buy. Pariser points out that algorithms should be designed with care and acuteness as they can be used to steer behaviour. He noted that marketeers are using personal data to design more effective, personalized advertisements, but also that the United States Armed Forces have successfully used social-graph data for military recruitment. “Drawing inferences based on people like you or people linked to you is pretty good business” (ibid). Although Pariser is critical about the use of algorithms for personalization, businesses online seem to have embraced this digital technique.

Taina Bucher discusses in more detail how algorithms in relation to datafication can influence behaviour. She focusses on social media platforms rather than more straightforward consumer

websites, but her insights on the effects of algorithms are also relevant beyond the social media realm. Her article on programmed sociality shows how social media platform Facebook, with datafication and repurposed metadata, transforms the traditional notion of friendship as something created between equals and free of structural constraints. Her argument is that technology is not neutral, but rather should be seen as a productive force. In Facebook’s case, she argues, one can see that sociality is something that is programmed. Social media platforms like Facebook play an important role in how users relate to themselves as well as to others (480). As discussed before, Facebook’s algorithm determines which content does or does not get displayed in a user’s News Feed, and in which order it is presented. Although there are many factors considered important here, Facebook has in January 2018 held a closed publisher meeting, of which Ned Berke and Matt Navarra have published the presented slides. In this meeting, Facebook revealed that changes to the algorithm that shapes a user’s News Feed are now considering ‘active’ interactions (like comments and shares) to be the most important. Another important change in the algorithm is that Facebook will start prioritizing content from friends and family over public content, like content from pages. They argue that person-to-person connection is more valuable than person-to-page connection, as this will lead to users being more actively engaged (n.pag). This is a good example of how “the power of the algorithm becomes apparent in its capacity to make certain people more visible than others. The underlying software always already intervenes in the practices of friendship by selecting which friends a user should pay attention to (Bucher 484). Bucher argues that in researching friendship on social network sites, it is important to find out how relational impulses are activated. This means that it is not only necessary to acknowledge that relationships are activated online, but also how this activation happens. “by whom, for what purpose, and according to which mechanisms?” (480). Although Bucher’s discussion of steering behaviour revolves around friendship and social media, the same can be said about businesses repurposing data when steering consumer behaviour. Bucher showed how friendship can become algorithmically defined and essentially shows that behaviour is, to some extent, programmable.

(12)

11 Regarding the topic of this thesis, this leads to questions of how consumer behaviour can be

algorithmically defined as well, and in which ways businesses actually engage in ‘programming’ consumer behaviour.

Tarleton Gillespie writes about algorithms from a similar perspective, and argues that

digitizing information and relying on the algorithmic assessment of this information means subjecting human discourse and knowledge to the procedural logics underneath these processes. Algorithmic assessment of information, according to Gillespie, always entails a knowledge logic, as the process is built on presumptions about what knowledge is and what its most relevant components are. “That we are now turning to algorithms to identify what we need to know is as momentous as having relied on credentialed experts, the scientific method, common sense, or the word of God” (168). Important in Gillespie’s argument is that the central element of understanding algorithms is approaching them not as abstract technologies but rather as products of human and institutional choices (169). Online consumers are constantly interacting with algorithms, but they are likely to be largely unaware of the

choices that have been made in building the algorithms or the implications they might have on how they are shaped as a consumer. One reason for this can be what Gillespie calls the promise of algorithmic objectivity. Algorithms are often presented as “stabilizers of trust, practical and symbolic assurances that their evaluations are fair and accurate, and free from subjectivity, error or attempted influence” (179). Gillespie explains that algorithms are often presented as neutral, impartial and reliable, which makes users more easily regard the results of algorithmic processes as relevant (ibid). In the case of consumer websites, presenting algorithms this way would likely lead to consumers having an increased sense of trust towards their algorithmically constructed shopping experience, which might make them more susceptible to what the algorithms were initially designed to do: governing their user behaviour and shaping them as consumers. An example of this would be Fit Analytics, a service that provides size recommendations and fit prediction services for consumers shopping online for clothes. This service can be integrated in webshop interfaces and is used by large online businesses like ASOS, Wehkamp, Puma and Calvin Klein (“The Global Sizing Leader” n.pag). In case of ASOS, a user is shown an interface that recommends which size they should buy. The system compares data from this specific user to data from other consumers (for example on purchase and return history, but also on data provided by the

Image 1: ASOS Size Recommendation Popup (Source: www.asos.com)

(13)

12 user manually on body weight, shape and age). The user is faced with percentages and statistics. They are presented as a purely mathematical, and thus objective and reliable result, as can be seen in Image

1. When users want to find out more about how this system works, they are referred to the website of

Fit Analytics, which on their homepage present their system with the following text: “Data is destiny. Fit Analytics' sizing platform is based on the biggest dataset in the industry and powers billions of

unbeatably accurate sizing recommendations” (“Data is Destiny” n.pag, emphasis added), and “Our

platform uses the most advanced machine-learning algorithms in the business to solve sizing at scale and produce truly data-driven results” (ibid, emphasis added). Like Gillespie argued, the algorithms are positioned as the best, and thus most reliable algorithms in the business. Objectivity and

impartiality are promised by their acclaimed ‘biggest dataset of the industry’ which supposedly make their ‘truly data-driven results’ unbeatably accurate. No further information can be found on how the system actually works. In order to understand algorithms as tools of knowledge and discourse, Gillespie highlights six dimensions that may help uncover the political ramifications of algorithms. The second dimension, the cycle of anticipation, is most relevant for this thesis as it entails finding out what the implications are of algorithm providers’ attempts to predict and construct their users, as well as uncovering how conclusions are drawn (168). When, as argued before, businesses draw from psychological theory to predict and construct consumer behaviour, they are also likely to build their algorithms with this knowledge in mind. Uncovering what these theories are can help get a more complete understanding of how businesses use algorithms for persuasion.

Bernhard Rieder shows the importance of critically researching the technical aspects of software. He focuses here on algorithmic techniques, which according to him are techniques that serve as a “middle ground between concrete, implemented algorithms and the broader study and theorization of software” (1). Algorithmic techniques are not necessarily new, and when looking at what has changed in relation to our current situation, Rieder argues that there are four elements to consider important. First of all, datasets of ‘unusual’ data have become widely available (13). Second, there has been a rise of powerful analytical techniques (including machine learning). Third, integrated

environments for instant application of algorithmic decisions have become implemented by companies (14). And finally, there has been an extension of market settings where fast, yet informed decisions are rewarded. Rieder argues that these developments reinforce each other, which makes our current situation regarding algorithmic techniques stand out in contrast to former periods (15). The scope of this thesis does not allow for an in-depth analysis of actual algorithms. Rather, I would like to argue that algorithmic techniques are the driving forces behind enabling consumer behaviour to be steered by businesses. One of the techniques that plays a central role in this thesis is dynamic, algorithmic classification. In making this argument, this thesis is written from the perspective of John Cheney-Lippold. He writes about power in relation to online marketing in which his main focus is on how digital techniques have made it possible for people to become dynamically categorized based upon inferred behaviour rather than self-declared behaviour. Businesses can use data to classify their users

(14)

13 (or potential users) probabilistically and dynamically into categories. These are then used to determine the content that is presented to the user, and likely also the ways in which a user is able to or allowed to move through a website. "Users are not categorized according to one-off census survey data but through a process of continual interaction with, and modification of, the categories through which biopolitics works" (173). Because of the possibility for dynamic algorithmic categorization, businesses online gain a considerable advantage over their (potential) customers. Every person visiting the Amazon website for example, sees the website differently. It is filled with content based on what Amazon thinks they should see based upon the data that was gathered and the outcome of the algorithmic categorization (Smith and Linden 12). Algorithmic classification has allowed the digital aisle to become completely personalized, which enables online businesses to persuade and target their users more precisely. Because the inferred categorizations are dynamically compiled and applied, the characteristics of the categories themselves are prone to change as the practices of users change (Rieder “Controls.” n.pag). When all this data gets used in combination with psychological theory and experimentation, which Cheney-Lippold does not get into, companies are likely to gain an even larger advantage over consumers in which the business is always one step ahead. Cheney-Lippold then argues that because of “categorical behaviours that are statistically defined through a cybernetics of purchasing and research that marketers have deemed valuable for identification and categorization […] database marketers are able to classify their users in a manner which becomes wholly embedded within the logic of consumption” (171).

Two of the factors that play a central role in algorithmic categorization are big data and algorithmic analytics, which are connected to each other in various ways. Websites, and specifically social platforms such as Facebook, generate and harvest immense amounts of unusual data which go far beyond the classic demographics. Nowadays, data on all sorts of personal information is harvested through data mining. Think for example of social interactions and relationships, affect, sentiment, intentions, cultural and political preferences etcetera. Big data, a term that seems to have become widely used over the past decade, entails both these types of data as well as the infrastructures and techniques around them which take care of capturing, storing and analyzing the data as well as repurposing the data for business (Rieder “Platforms.” n.pag). Online, businesses often have

algorithms which mine these data to produce insights. An algorithm can for example mine data about the colours of product images a user tends to click on more often. When the algorithm analyzes the data and notices that a user often clicks on images with a lot of blue in them, it can then rearrange the products shown to this user on the website by featuring more items with images that have blue in them. What is happening here is that consumers are manipulated psychologically, with the help of algorithms. Businesses can mine their own data, buy data from an external company, or work with a combination of both. Data mining is at the root of techniques that enable consumer persuasion, as without data, personalization and algorithmic classification become impossible. Rieder writes that one could argue that data mining techniques can essentially embody forms of cognition:

(15)

14 On the level of signification, data mining techniques attribute meaning to every variable in relation to a purpose; on the level of performativity, the move to increasingly integrated digital infrastructures means that every classificatory decision can be pushed back into the world instantly, showing a specific ad, hiding a specific post, refusing a loan to a specific applicant, setting the price of a product to a specific level, and so forth. No data point remains innocent. (12)

What is happening here, is that the algorithmic technique behind this does not in itself directly determine how results are used, but it does serve as a middle ground that allows for the production of new forms of knowledge that shape decision making and affect how power can be operationalized (ibid). An example of how this can become more problematic is when algorithms analyze the popularity of models on fashion websites in relation to skin colour. If an algorithm, through data analysis, notices that product pictures with models with certain skin colours get less clicks than others, then they might also be featured less prominently. Fashion websites can own data about which models work better than others for persuading customers to make a purchase, and this can impact job

opportunities for models that do not fit within that ‘popular’ category. Examples like this show that algorithms and big data can lead to issues with serious consequences such as (indirect) discrimination. The technologies discussed above show that digital media can be used by businesses online in order to steer their users’ behaviour, which essentially makes these businesses better equipped for persuading then they were ever before. Communications professor Joseph Turow argues that the real centre of power in advertising is in media buying, which is the idea of choosing which channels to reach people with. This also includes then the people who create the software around media buyers. These people, according to Turow, can actively shape people’s identities as they determine what individuals see online, which coupons they receive and which products or services they are offered. “Into the twenty-first century the media-buying system's strategy of social discrimination will

increasingly define how we as individuals relate to society—not only how much we pay but what we see and when and how we see it” (4). What happens when we, as a society, begin to define ourselves by how businesses begin to define us, based upon ideas about us that advertisers and marketers know, but about which we have no clue? Turow argues that we start to live in an age where the advertising industry is defining our identity and our worth (“Joseph Turow, The Daily You.” n.pag), and that “marketers and media intrude in—and shape—our lives. Every day most if not all Americans who use the internet, along with hundreds of millions of other users from all over the planet, are being quietly peeked at, poked, analyzed, and tagged as they move through the online world” (1). All this is being done by businesses because they aim to unravel the ways in which individuals can be activated to spend their money, so that they can persuade people on a more individual, and more efficient level than ever before (ibid).

(16)

15

2.2 Persuasive interface design

Another way of approaching persuasive techniques is by looking at the frontend of digital media, which means looking at theory relating to the interface. The interface is commonly known as the shared boundary between two or more entities which can be a variety of things, for example bodies, spaces, phases, systems, concepts, human beings etcetera (“Interface” n.pag). They can govern user behaviour by offering a limited set of means for interaction, such as liking, sharing or rating. Interfaces provide the structure through which data on user engagement is captured, which is subsequently algorithmically processed. This algorithmically processed data is then returned to the user, be it more obvious in the form of metrics or in a more subtle way, for example through product recommendations. Interfaces are used to persuade users towards actions that are beneficial for online businesses, for example by making users spend money or give away more personal data. If this is done in combination with psychological theory, they can become powerful spaces of manipulation in which the consumer is shaped. As the following paragraphs will show, interface-related theory often

discusses things like constructed subjectivity and productive power. This subchapter focusses on theory from fields like platform politics, software studies and design studies, which all discuss persuasion and steering behaviour, ideas that have their roots in the field of psychology. Although these theorists all acknowledge that interfaces govern behaviour and play an important role in

constructing the user, they do not include detailed discussions of psychological theory in their work in order to explain why this is happening. Because this thesis focusses on persuasion and steering behaviour, this part of the theoretical discussion discusses computing-related interface theory where the human, or the user, is one of these entities. An interface like this, between at least one human and at least one computer, is known as the human-computer interface (HCI). From here on, when the word interface is used, it is referring to the HCI.

Johanna Drucker discusses human-computer interface design. She builds upon Brenda Laurel’s definition of the interface from 1990 as the surface where the necessary contact between interactors and tasks allowed functions to be performed (Laurel 21). Because this still says very little about the complexity of the interface, Drucker adds that the interface should be seen as a space for reading rather than as a space for looking. She uses the desktop metaphor to suggest that the interface should be seen as a space for activity, in which the icons can be thought of as objects for interaction, as is the case with actual physical desktops (9). She defines the interface then as an “artefact of complex processes and protocols, a zone in which our behaviours and actions take place. Interface is what we read and how we read combined through engagement. Interface is a provocation to cognitive experience” (ibid). Drucker explores different humanities approaches towards the interface and interface analysis. From these different approaches she concludes that the starting point of interface analysis according to her, is the notion of the constituted subject and production of meaning that are the results of relations in the electronic environment of the interface: “A constituted subject is created through the process of reading, rather than a mechanistic consumer-model of autonomous viewer” (3).

(17)

16 According to Drucker, the interface from a humanities perspective should thus be approached as a space where a subject is constructed, and that “attention to the specific relations between properties and affordances of electronic environments within a system of co-dependent relations of production will be the starting point for assumptions about interface as a space that supports interpretative events and acts of meaning production” (ibid). Matthew Fuller approaches the interface from a slightly different perspective as he writes about the interface as the point where the machinations of computers have to make themselves available to the user. But like Drucker, he also argues that it is important to ask what persona, and thus what ‘human’ is engineered by an interface (12). Interestingly, these authors both approach the interface as something between the human end user and the

machine/computer. They focus on how this technological aspect to the interface shapes the person using it, but pay little attention to those designing an interface. When the interface is seen as a space of meaning production and subject construction, this also makes it a powerful space businesses can use for persuasion and the construction of consumer behaviour. In order to be able to say something about how this is actually taking place, it is important to first look at theory related to interface design. Interface design theory discusses the effects of design in relation to behaviour. A central concept in this theoretical tradition is ‘affordances’, which builds upon the beforementioned notion of power as something that is productive. In 1988, Donald Norman defines the concept as follows:

The term affordance refers to the perceived and actual properties of the thing, primarily those fundamental properties that determine just how the thing could possibly be used […]

Affordances provide strong clues to the operations of things. Plates are for pushing. Knobs are for turning. Slots are for inserting things into. Balls are for throwing or bouncing. When affordances are taken advantage of, the user knows what to do just by looking: no picture, label, or instruction is required. Complex things may require explanation, but simple things should not. (9)

Norman discusses affordances as properties of everyday objects without mentioning interfaces, but his conceptualization remains relevant today to get a basic understanding of what affordances are. He notes how affordances provide strong clues to the operation of things, and the same can be said about aspects of interface design. When using a search bar on a website for example, a small icon of a looking glass is often featured in order to show the user that this is a place for searching. This means that the affordances tell a user how an interface should be used and this is likely to affect the user’s behaviour in relation to the object. Evans et al. relate the concept of affordances more directly to digital media: “Our conceptual definition of affordances - possibilities for action - is the multifaceted relational structure between an object/technology and the user that enables or constrains potential behavioural outcomes in a particular context” (2). Affordances can thus be used by businesses to enable or constrain behaviour, and should then be seen as tools for persuasion.

One of the more recent uses of affordance theory that is directly related to digital interface related research comes from Mel Stanfill, who argues that discourse is a particularly useful lens

(18)

17 through which power in interface design can be analyzed. She uses Foucault’s conceptualization of discourse here, which he defines as practices that systematically form the objects of which they speak (Foucault The Archaeology of Knowledge and the Discourse on Language 49). This means that through discourse, “thought and action are structured in accordance with sets of knowledge or assump-tions about what is true or correct” (Stanfill 3). This results in what Stanfill calls a methodology of discursive interface analysis in which affordances must be treated as statements, and digital spaces should be thought of as groups of statements: "Examining interfaces lets us ask: 'How is it that one particular statement appeared rather than another?' What beliefs drive design and are built in? And, ultimately, what are the consequences of these design choices?" (4). This makes looking at affordances so valuable for interface analysis; they reveal norms and ideals behind design choices and thus reveal intentions of power. In case of this research, looking at affordances used in interfaces from large commercial websites can be useful to reveal how businesses construct their user experience, and in doing so steer consumer behaviour. Finding out which psychological theory is used for this by businesses then gives a more complete understanding of why affordances can work as such. Stanfill recognizes three types of affordances: the functional, the cognitive and the sensory affordance. She defines the functional affordance as all the functionalities of an object; what can a user do with it? (5). The cognitive affordance then, entails the aspects of an object that help a user give meaning to the functionalities. An example of this is what things are called in menus or headers, or images that are used to show a user what things mean or how they can be used. “Cognitive affordances facilitate processing information, and are therefore closely tied to the social act of meaning-making” (ibid). The sensory affordance focusses on aesthetics of an object: things like font-size, colours or sounds can for example draw a users’ attention towards a certain area or functionality of the object. These affordances can be analyzed in order to find out more about the underlying assumptions and intentions behind the interfaces of websites and applications.

As discussed in Algorithms, personalization and persuasion, big data plays an important role in persuasive, algorithmic systems, as it allows for users to become dynamically categorized. This is directly related to businesses creating persuasive design, for example by creating interfaces that adapt to the person using it, which optimizes the chances of nudging users towards specific actions. Karen Yeung discusses how big data transforms personal data into something of economic value. She argues that this can be used to shape users’ individual decision-making processes (2). In order to strengthen her argument, she draws upon Boyd and Crawford, who discuss big data as an analytic phenomenon that allows industries to find patterns of information about individuals, but also about individual’s relation to others or about groups of people in datasets not capable of ordinary human assessment (662). Nudging users towards specific actions through adaptive interfaces based on big data is what Yeung calls the hypernudge. She underlines that although the algorithmic processes might be complex and sophisticated, applications and websites can use nudging, a simple design-based mechanism of influence, in order to personalize a user’s choice context (2). Big data is a powerful digital technique

(19)

18 that can be used in the creation of hypernudges because it entails streams of data from multiple sources which are algorithmically analyzed. This allows businesses to be able to gain predictive insights about things like habits and preferences. “These nudges channel user choices in directions preferred by the choice architect through processes that are subtle, unobtrusive, yet extraordinarily powerful” (ibid). An example of how consumer behaviour can be steered via the interface, is through the use of productive measures. According to David Beer, productive measures are measures which do not just represent data, but also produce new possible actions. “Data analytics do not just shape what is known, they also shape what is knowable. This is not about what is probable, but what is possible” (2). Beer takes football as an example, where data analysis keeps becoming more important in creating value of football players. When the players are valued on data of, for example, their pass completion, players are likely to play a game that focusses on pass completion and probably take less risks. With this example, Beer shows that when people are faced with personal data and metrics, they are likely to alter their behaviour in favour of these metrics (6). On a consumer this could work for example by showing the consumer how many people are currently looking at the product, how many people have bought it, what percentage of people buy which sizes of clothing, what the estimated chance of fast delivery is etcetera. By showing a hotel booking website user how many people have looked at a certain hotel room in the past hour for example, this user might be nudged towards booking this hotel room as they could fear that this room will be booked soon. “Data circulate and reshape organizations, practices, behaviours and perceptions. As this case study intimates, these forms of data are already an integrated, immanent and embedded part of the social world, of our practices and of everyday cultural

engagements” (9). As society is more and more revolving around data, it is crucial to look into the ways in which these data circulate through, reshape, alter and disrupt the configurations of power and decision making: “They shape behaviours. As people are subject to these forms of measurement they will produce different responses and outcomes, knowing, as they often will, what is coming and the way that their performance will become visible” (ibid).

As this theoretical discussion shows, interfaces provide the structure through which data on user engagement is captured, which is subsequently algorithmically processed. This algorithmically processed data is then returned to the user, be it more obvious in the form of metrics or in a more subtle way, for example through product recommendations. Although the beforementioned theorists give detailed descriptions of why technology can be used as techniques of persuasion, they do not get into how businesses are actually employing these, for example with the help of psychology. This is not something I am critiquing, as writing about the techniques from a business perspective was never their goal. However, I do argue that including this perspective can be a valuable addition to the literature. Behaviour, and particularly steering behaviour, is a recurring topic in platform politics and software studies. This is rooted in psychological theory, but none of the literature discuss this. This thesis contributes to this field of research by finding out which psychological theories are used by businesses when they engage in steering consumer behaviour. This research then aims to show the reader how

(20)

19 technology and psychology come together in the field of online marketing, and I argue that we can deepen our understanding of these persuasive techniques by looking at the psychological research that informs the development of many of these techniques. The following subchapter explains how this research was operationalized and how these insights were gained.

2.3 Research methodology

As the former chapter has shown, there is already an extensive body of literature on the topic of persuasion that mainly approaches it from the academic fields of platform politics and software studies. An essential step in this research is looking at how the industry that actually uses persuasive technology approaches the topic, which means that this research is focused on publications from the fields of marketing and advertising, but also information brought forward by businesses themselves. First, industry sources from the fields of marketing and advertising were read, including blogs, magazines and handbooks. These sources can be seen as an important link between theory and practice. Their intended audience consists of people who are likely to be working with persuasive techniques, like online marketeers, web designers and online communication specialists, which makes these types of sources influential in the field. The following list shows some the sources that were used from this category, most of which have been selected because they were on the first page of Google Search results when using the queries ‘online marketing blog’, ‘online marketing persuasion’ or ‘persuasion marketing’. In some cases, sources refer to other (similar) websites or books when they are using their content. These sources were also read and used for this research. For a full (MLA style) list of sources please consult the bibliography.

- Aaker, David A. Creating Signature Stories: Strategic Messaging That Persuades, Energizes

and Inspires (2018).

- Abercrombie, Roxanne. “Sentiment Analysis: The Next Big Thing in Business.” Business 2

Community (2016).

- Bridger, Darren. Neuro Design: Neuromarketing Insights to Boost Engagement and

Profitability (2017).

- Braingineers. “Neuromarketing Insights by Braingineers.” (2018).

- Forbes Magazine. “Neuromarketing: Companies Use Neuroscience for Consumer Insights.” (2012).

- Wired Magazine. “The A/B Test: Inside the Technology That's Changing the Rules of Business.” (2012).

- Jenblat, Omar. “Let's Get Emotional: The Future Of Online Marketing.” (2018).

- Marketing Week. “Make Marketing More Effective by Using the Golden Standard Big Five Personality Traits.” (2015).

Besides referring to the psychological field only, the sources mentioned above also refer to academic research from fields like marketing, economics and advertising which is often framed around

psychological theory. As is to be expected, most of these sources do not come from either an

(21)

20 like behavioural economics or neuroeconomics. Sources from these fields are relevant for this research because academically, this is where economics and psychology come together. These sources show which psychological theories are deemed relevant in economics, and the fact that these sources are referred to by industry blogs and magazines means that the psychological theory used here is likely to inform those who develop and employ digital techniques for persuasion. Some of the important sources referred to are featured in the list below, as well as other sources which are cited often

according to Google Scholar in relation to persuasion and psychology. The psychological theories that were mentioned often or seem to play a large role in the industry, are discussed in the results of this research.

- Cialdini, Robert B., and Noah J. Goldstein. “Social Influence: Compliance and Conformity.” (2004).

- Camerer, Colin, et al. “A Framework for Studying the Neurobiology of Value-Based Decision Making.” (2008)

Damasio, Antonio R. “The Somatic Marker Hypothesis and the Possible Functions of the Prefrontal Cortex.” (1998).

- Fogg, B. J., and Dean Eckles. “The Behavior Chain for Online Participation: How Successful Web Services Structure Persuasion.” (2007).

- Freedman, Jonathan L., and Scott C. Fraser. “Compliance without Pressure: The Foot-in-the-Door Technique.” (1966).

- Goldberg, Lewis R. “The Structure of Phenotypic Personality Traits.” (1993).

- Matz, S. C., et al. “Psychological Targeting as an Effective Approach to Digital Mass Persuasion.” (2017).

- Hofstede, Geert. Cultures and Organizations: Software of the Mind. (1991).

With some exceptions to the rule, businesses do not publicly share which psychological theory they draw from when engaging in the construction of consumer behaviour. Their websites do however show interesting examples of how persuasive techniques are actually used which is often directly in line with the psychological theory found in the above mentioned sources. Consumer websites were thus analyzed and examples of persuasive techniques are discussed in relation to the psychological theories to illustrate how these theories can play an important role in the development and

employment of persuasive techniques. Examples include results from interface analysis as well as analysis of the backend, which entails looking at what businesses communicate about using algorithms and data. The research aimed to include examples from a variety of sources, but as expected, not many consumer websites are very open about how they use digital media for persuasive practices. Netflix and Amazon however, do provide documentation of these practices in blog posts, which is why these platforms play a particularly important role in the results. Because these businesses are large and well known, they are likely to also influence other online businesses in how they give form to their marketing strategies. Although these companies seem to be quite open about their technology and business ideas, it is important to note that although they may seem to give a clear explanation of behind-the-scenes processes, this could very well merely be a glimpse of a real backstage process.

(22)

21 Company technology blogs should be seen as a “performed backstage” (Gillespie 180) that is carefully crafted to legitimize what their technology does. The next chapter shows the results of this research, as well as more information about why specific sources were chosen.

(23)

22

3 The Behaviour Chain Theory

The previous chapter discussed how digital media can be used for persuasive purposes, but how has persuasion been theorized in the field of psychology? One heavily cited behavioural scientist on this topic is B.J. Fogg. His work has also been featured on many industry blogs and in magazines, which in doing so draw upon his behaviour model for persuasive design theory. The well-known American business magazine Forbes which has been around since 1917 for example, uses Fogg’s work in an article that gives advise on how to win users and influence their behaviour. They introduce him by saying that “no one has perhaps been as influential on the current generation of user experience (UX) designers as Stanford researcher BJ Fogg” (Kosner n.pag). Another example of Fogg’s relevancy in business is that he has been given a place in Fortune Magazine’s top ten business guru’s currently changing the way business is done. They call Fogg “one of the most sought-after thinkers in Silicon Valley” (Reingold, Tkaczyk n.pag). In his behaviour model, Fogg argues that behaviour essentially is a product of three factors, which are motivation, ability and triggers: “for a person to perform a target behavior, he or she must (1) be sufficiently motivated, (2) have the ability to perform the behavior, and (3) be triggered to perform the behavior. These three factors must occur at the same moment, else the behavior will not happen” (1). When it comes to persuasion, Fogg notes that it is important to realize that generally, people have modest levels of motivation and ability and to persuade successfully these levels must often be manipulated, and effective persuasive technologies should be designed to boost at least one of these two factors. He uses 1-click purchasing offered by webshops as an example that increases the ability level of the consumer. This technique is used by Amazon, which they describe as follows: “When you place your first order and enter a payment method and shipping address, 1-Click ordering is automatically enabled. When you click Buy now with 1-Click on any product page, your order will be automatically charged to the payment method and shipped to the address associated with your 1-Click settings (“About 1-Click Ordering.” n.pag).

Fogg also recommends in his work with marketing professor Dean Eckels, that businesses would benefit from using their Behaviour Chain theory in order to persuade successfully. After

(24)

23 analyzing fifty commercial websites, Fogg and Eckles discovered that influence strategies of

persuasion are often employed as multiple strategies in succession. They coin the term ‘Behaviour Chain’, which can be represented in a flow chart that consists of phases aimed to achieve target behaviours: ”The Behaviour Chain consists of three Phases, each with at least two Target Behaviours. As users move along the Behaviour Chain, the Target Behaviours generally become more demanding. User behaviour in the final Phase contributes to the service's appeal to new users” (201). By

conceptualizing this, they are building on the work of Danaher et al., who argued that for online persuasion, businesses often use multiple strategies in succession, although each strategy stands on its own to help the persuader reach behavioural

and attitudinal goals (4). From this point then, new strategies can be applied. When these are used in succession they become more powerful for persuasion (Fogg, Eckles 200). Fogg and Eckles’ Behaviour Chain phases consist of 1) discovery, then 2) superficial involvement and finally 3) true commitment. Using a successful website or webservice today, almost always involves following this path (202). Image 2 is a diagram which represents these phases. The challenging part of the Behaviour Chain is getting users from being superficially involved to being truly committed. Fogg and Eckles note that websites often try to have their users sign up for something in phase two, like a free trial or a subscription to a newsletter in exchange for discounts.

Registration opens up new ways of communication and brings in new user data, which makes possible new forms of persuasion. When this is achieved early on in the Behaviour Chain, it is more likely that a user can be persuaded towards phase three: true commitment. For a webshop, these steps could for example look as follows: first, it is important to attract new visitors to the website (discovery). Second, they need to be involved, for example by having them sign up to a newsletter (superficial

involvement). Then finally they will need to be converted into a paying customer (true commitment). What is crucial in order for all the phases to succeed according to Fogg and Eckels, is that users’ trust is not damaged in the process. This is why websites and webservices often focus on trust building in the first two phases. Once a user is far in phase two or already in phase three, it will be much easier to get access to personal data for repurposing without losing the user’s trust (204). A good

(25)

24 example of how consumer websites (be it consciously or unconsciously) design their website

according to Behaviour Chain principles can be found by doing an interface analysis of American insurance company VisitorsCoverage. As a first-time visitor of the website, I was in phase one of the Behaviour Chain: discovery. I had learned about the website from my Google search results and I visited the website. What happens next, and which is in line with what Fogg and Eckels suggest, is that the website encourages me to become superficially involved. In order words; it is trying to take me to phase two. The interface encourages me to interact with the website and aims to show me how their products can be helpful to me. First of all, let us look at how the website aims to make me engage with the website. The first call to action the user is presented with, is a centrally placed line in a large white font which says “Buy visitors & travel insurance in less than 5 mins” (VisitorsCoverage Travel insurance n.pag). Then, as can be seen in Image 3 it directly moves towards asking the user for data. Users are obliged to give up this data in order to see which products they can get and for what price. This makes me become superficially involved with the website, but the manner in which this is done can be considered quite radical as the website withholds crucial information (product prices) until I give away personal data. This shows how users are nudged, and not very subtly, towards engaging with the platform, which is an important technique for persuasion that will be specified upon in chapter 4.

(26)

25 As can be expected of an insurance company, trust building is an important part of the interface. What is interesting is that this is done quite excessively. The following two pages show Images 4 and 5, screenshots of the homepage interface, which had to be divided into two images as the homepage was too large to show on one page. All the aspects of the homepage that aim to build trust, for example by including experts’ knowledge, certificates or logos, or customer reviews, have been manually marked in red lines (“VisitorsCoverage Travel Insurance” n.pag).

(27)

26

(28)

27

(29)

28

4 Compliance Theory

Behind every persuasive technique that is used by commercial websites in order to steer consumer behaviour is the notion that a subject can be nudged towards specific behaviour. According to

influential behavioural psychologists Richard Thaler and Cass Sunstein, people in all areas of life can be seen as choice architects. They use the school cafeteria employee as an example. In order to get children to eat healthier, the choice architect can place fruit and vegetables strategically in order to increase the likelihood of children choosing these kinds of food. “A choice architect has the responsibility for organizing the context in which people make decisions. If you are a parent,

describing possible educational options to your son or daughter, you are a choice architect. If you are a salesperson, you are a choice architect” (3). Thaler and Sunstein stress here that choice architecture has much in common with traditional architecture, and what we should take away from this is that “there is no such thing as a ‘neutral’ design” (ibid). In digital media, programmers are the choice architects. When designing a website for example, Thaler and Sunstein would argue that every choice the programmer makes in this process eventually has an impact on what a user can do or on how their attention is focused. They use men’s bathrooms at Schiphol Airport as an example, where bathroom designers placed the image of a fly in the urinals in order to nudge the men to focus on their aim. This reduced spillage by eighty percent (4). Choice architects with access to data collection and analysis have the ability to design their nudges in a more effective and influential way, as has been shown in chapter 2 with Yeung’s conceptualization of the hypernudge. Online, consumers are often nudged towards giving away personal data. An example is American department store Target. When I visited their webshop (as a first-time user), I was shown a centrally placed request that nudged me towards their baby registry (Image 6). This can already be seen as a hypernudge as Target is likely to know already that I am a female in my late twenties, as I did not anonymize my browser settings when I was visiting the website. This hypernudge targets me as a potential pregnant person. Once a user would actually register of course, this would give the webshop more personal data to make hypernudges more specific and thus likely more successful in persuasion.

(30)

29

Image 6: Target baby registry hypernudge (“Target Homepage.”)

This has also been shown by the example of VisitorsCoverage discussed in the former chapter. The insurance company made it difficult for their users to see product specifications and prices

without the input of some basic personal data. VisitorsCoverage wants their users to become customers. In order for them to become customers, they have to sign up for insurance. This is a big request, so the website first makes a smaller request: making their users give away some personal data in return for being able to see products and prices. Once they have already given away this

information, and they have started a process of engaging with the website, it will make the step towards actually becoming a paying customer seem less big. This is commonly known as the foot-in-the-door (FITD) technique. It is a compliance theory which comes from social psychology and which was first introduced in this terminology by Jonathan Freedman and Scott Fraser in 1966. They asked a group of women, by telephone, whether they were willing to answer a few questions about their use of household products. This took only a couple of minutes. A few weeks later, they called these same women to ask whether they could send a group of six men to their home to look into their cabinets and conduct a two-hour interview. They concluded that “once someone has agreed to a small request he is more likely to comply with a larger request” (195). This has been widely discussed by academics, and has for example also been referenced by the beforementioned Fogg and Eckels. When Googling this technique, it becomes clear that this is a well-known technique in the marketing field as it has been advised as an online business strategy by beforementioned Forbes (Patel n.pag), but also for example by digital neuromarketing expert Jeremy Smith (n.pag) as well as the award winning online marketing agency Strategic Factory (“The ‘Foot in the Door’ Technique.”). Essentially, the FITD technique revolves, again, around trust building. When a website asks for something small, like in the case with VisitorsCoverage, users can get used to the business and become familiar with what it can provide. Once the bigger request is made, it is likely that the user already has become more trustful towards the business. In this case, the website also gains valuable user data in the process, which can be

repurposed to gain insights and to further steer their user’s behaviour. One could argue that users are free to choose whether or not to give up this data, but their customer experience is made very difficult when they choose not to, and it is likely that they are unaware of the algorithmic techniques behind this practise which allows for them to be dynamically categorized. This can then lead to the distribution of power between businesses and consumers to become highly imbalanced.

Shaping the Digital Consumer