The impact of AI on branding elements : Opportunities and challenges as seen by branding and IT specialists

(1)

The impact of AI on

branding elements

Opportunities and challenges as seen by

branding and IT specialists

MASTER THESIS WITHIN: General Management NUMBER OF CREDITS: 15 Credits

PROGRAMME OF STUDY: Engineering Management AUTHOR: Alfedaa Sabbar and Lina Nygren Gustafsson JÖNKÖPING May 2021

(2)

Master Thesis in General Management

Title: The impact of AI on branding elements: opportunities and challenges as seen by branding and IT specialists.

Authors: Alfedaa Sabbar and Lina Nygren Gustafsson Tutor: Dinara Tokbaeva

Date: 2021-05-24

Key terms: AI, ML, GANs, DNN, Branding, Logo, Visual brand elements, Non-visual brand elements

Abstract

Background: The usage of AI is becoming increasingly necessary in almost every industry, including marketing and branding. AI can help managers, marketers and designers in the marketing and branding sectors to overcome realistic and practical challenges by providing data-driven results. These results could be used in making decisions. Nevertheless, implementing AI systems and the acceptance of it varies widely across different industries, with building brands is still behind.

Purpose: This research aims to develop a deeper understanding of why AI systems are not yet commonly used in the branding industry with emphasis on how it could be useful. As a result, the main opportunities and threats to the usage of AI in branding as seen by branding- and IT specialists are explored and expressed.

Method: To achieve the purpose of this study, a qualitative study was conducted. Semi-structured interviews were used as means to collect primary data and in total 15 interviews with branding and IT specialists were carried out. The data was transcribed and analyzed according to thematic analysis which emerged in four main themes.

Conclusion: The results show that AI is capable of creating brand elements, with limitations to mostly non-visual brand elements due to the lack of creativity and emotions in AI solutions. The findings indicate that the perceived possibilities of implementing AI in branding mostly are cost- and time-related since AI tends to be capable of solving tasks which are cost- and time-consuming. Furthermore, the perceived threats mainly involve i) losing a job or ii) intrude on the roles of branding professionals.

(3)

Acknowledgements

We, Alfedaa Sabbar and Lina Nygren Gustafsson, would like to direct our sincere appreciation to everyone who has motivated and helped us during the course of writing this research paper. First, we are especially thankful to our supervisor Dinara Tokbaeva, who throughout the process of writing this paper has supported, inspired, motivated, and guided us in the right directions. Dinara’s strong commitment to her students has motivated us to continue and strive for good results throughout the research.

Secondly, we would like to express our gratitude and appreciation to all individuals that agreed to participate in this research and dedicated their time to help us fulfil the research purpose. All individuals have provided us with great insights and valuable findings. Thank you: Ali Alhasan, Ali Latif, Andreas Lindén, Elias Stråle, Khaled Elmadawi, Luis Calebe Polano Paglioza,

Magnus Isenberg, Magnus Perman, Mattias Falkendal, Moe Selwaye, Rolando Ramirez, and the rest of individuals who preferred to be anonymous.

Lastly, a highly and sincere thank you to the students of our seminar group. You have provided us with valuable and important insights throughout the progress of this report.

(4)

List of figures

Figure 1: Disposition of the study ... 6

Figure 2: Suggested drawing ... 18

Figure 3: Background masking ... 18

Figure 4: Coloring sketches ... 18

Figure 5: Coloring palette ... 19

Figure 6: Object transfiguration ... 19

Figure 7: Generated faces with the use of GANs ... 20

Figure 8: Generated pictures using text input ... 21

Figure 9: The tree metaphor ... 22

Figure 10: Summary of chosen methods and approaches ... 25

Figure 11: Overview of interview choice and how it was executed ... 27

Figure 12: The 10 key principles in research ethics ... 33

Figure 13: Codes and pattern within the theme "brand perceptions" ... 35

Figure 14: Codes and patterns within the theme "perceptions of opportunities"... 37

Figure 15: Codes and patterns within the theme "perceptions of threats" ... 39

Figure 16: Codes and pattern within the theme "integration of AI" ... 42

List of tables

Table 1: Conclusion of mentioned brand elements... 13

Table 2: Overview of the conducted interviews ... 28

Table 3: Overview of the themes and how they originated ... 34

(7)

Introduction

Have you ever searched for a product on the internet, and a day after you start to see ads of the same product on YouTube while you are watching a video of a totally different topic? Did you receive similar ads on Instagram and Facebook as well? Do not worry, there is no one spying on you. Well, not a human being at least. This is an automated system, built by using specific algorithms aiming to understand the human beings and act accordingly to their mission. In this example, the search engine used sophisticated Google ranking system to generate the results you looked for the first day. These results were also generated automatically by using bidding machines, which helped the other platforms to target you with specific ads depending upon your previous searches. Same way used by YouTube, the largest video content website and the second largest search engine in the world. YouTube uses AI to suggest videos that are related to previously watched or searched videos, the suggestions are predicted by AI believed to catch the user's attention. These processes are a few of many examples of the automated processes that takes less than a second in order to make its own decision based on your input. This automated system is called: Artificial intelligence or in abbreviation, AI (Ma & Sun, 2020).

AI has been used in many intelligent solutions in different sectors and industries, such as marketing, designing, and analyzing. This research aims to investigate the ability of AI in creating branding solutions, such solutions might aid and support brand building and designing. In the following chapters, a background of branding, brand elements, and AI are presented. Then, a discussion about the possibility of using the AI technology in terms of creating brands.

(8)

1. Introduction

______________________________________________________________________________________________

This chapter starts by presenting the background to this research followed by the problem statement, purpose, and formulated research questions. The chapter ends with a description of the overall structure for the following content of work.

______________________________________________________________________________ 1.1 Background

This era is the era of technology, and the usage of AI in different industrial sectors are increasing rapidly. There are more than 4.6 billion smartphone users around the world, and around two billion people are a daily internet user (Lasse Shulterbraucks, 2017).

The study of how to build computers that are as intelligent or capable of solving problems as good as humans is known as artificial intelligence (Sterne Jim, 2017). Learning, watching, listening, understanding, communicating, reasoning, creativity, and problem-solving are examples of those problems (Sterne Jim, 2017). Almost 63% of companies are already using artificial intelligence tools, sometimes even without recognizing that. Likewise, around 47% of customers use AI bots that support them communicate with the company when buying products or services online (Dimitrieska et al., 2018).

Using modern technologies such as AI makes trend predictions of the market and the customers much easier and more quickly, as well as analyzing the way customers make purchasing decisions (Dimitrieska et al., 2018). Data can be analyzed in company and management for a variety of reasons, including computer log analysis, analysis of feedback and views posted on social media, risk evaluation, customer retention, marketing, sales management, and so on (Tjepkema Lindsay, 2017). Companies alter their management practices to become more responsive, productive, profitable, and competitive. Technological advancements have given opportunities for businesses in general, especially in marketing. The industries are focusing more on branding and improving how they communicate with their customers. Advertising and marketing departments in many companies know about customers more than what customers know about those companies (Gentsch, 2019).

Artificial intelligence is now gaining momentum in business and marketing analysis (Gentsch, 2019). Marketing managers are already using AI technologies in terms of completing tasks by making them automated. Generally, in market analysis, 80 % of the tasks are considered

(9)

as time-consuming tasks, such as sampling, data extraction and data analysis, leaving just 20% for important tasks. This method can be automated using advanced big data and AI processes, giving market analysts more time to focus on truly value-added tasks like interpreting analysis findings and developing suggestions and behavior (Gentsch, 2019).

The other part of this topic is the brand and brand elements. Brand is the term or phrase connected with at least one product line item, which represents the qualifications and values that describes the item’s character (Kotler, 2000). Including marketing and brands, the use of AI would influence the world's future (Dimitrieska et al., 2018).

1.2 Problem statement

The importance of branding has increased tremendously and is still increasing due to the wide range of competitors created in the market. Nowadays, for a brand to be profitable and successful in the market, it is very important that the products are analyzed, and the competitors are considered by AI algorithms (Gentsch, 2019). The data volume of the world doubles every two years, resulting in a challenge that a human being cannot cope with it without the help of computers. Fortunately, modern technology provides memory capacities and data storages that are capable of saving all these massive amounts of data. Technologies are also capable of measuring, evaluating, and analyzing these data (Gentsch, 2019).

Besides the creativity issues, technologies and machines have no symptoms of emotional decision-making, instead, decisions are made entirely on analyzing facts and statistics (Vishnoi et al., 2019). While AI algorithms can predict customers' decisions by analyzing the pattern of their previous decisions, AI is limited in its ability to analyze why a customer made a particular decision (Gentsch, 2019). The human beings' brain learns new things as "one-shot" learning, while AI leaning method is called "deep learning", which requires a lot of data and algorithms that define the rules for AI (Teng, 2019). Building an AI system, requires three different development parts (Pradeep et al., 2019):

1. Database: Which includes information obtained from human experts, as well as a collection of rules and specifications of how this information is processed.

2. Inference Engine: Which is an automatic logic method that compares a problem submitted to the database. An inference engine can also have debugging and interpretation

(10)

capabilities. The interpretation function will explain how the inference engine arrived at a particular conclusion to the user.

3. User Interface: Which is a way of communicating between human and the software in human language, in order to complete an operation, ask a question, or send a problem to the database.

Implementing AI systems can be difficult and challenging. Aside from the time and money required to improve AI functionality, there are more possible problems with finding data sources, finding high-quality and accurate training data, also with the concerns of user privacy (Thomas & Fowler, 2020). In order for algorithms to work faster and make more accurate predictions, it is necessary to provide accurate and optimized data. Thus, companies must purify the data by cleansing the collected data. This process requires correcting or deleting all irrelevant, corrupt, missing, duplicate, or worthless data from the data set. This process is often very costly, it is also time-consuming, and it also has some security concerns, because data must be shared in order to be cleansed (Pradeep et al., 2019).

It is very difficult to find experts work through all of this data, not to forget the high cost to recruit experts for the amount of hours required to supply and analyze all the data. In addition, to write the algorithm, companies might also need to employ a mathematician or a computer data scientist (Pradeep et al., 2019). In addition to that, debugging, design, diagnosis, instruction and guidance, interpretation, tracking, monitoring, preparation, prediction, and repair are all required after implementing an AI system (Pradeep et al., 2019). Because of the cost of creating and developing an artificial intelligence system is so high, in most situations companies would still prefer humans (Teng, 2019).

Another challenging aspect is that the increased usage of AI technology, it might also lead to job losses. Companies need to consider the risks of these technologies before implementing them (Gentsch, 2019). In general, using AI, in any industry, has the same challenges, besides the extra challenges that are more related to the tasks and data needed to be analyzed and progressed. Branding for example, requires many data as we will present in the later sections. With all these challenges, is using AI supporting good enough that companies could think of taking the risk of implementing it and depend upon them in solving tasks? Is it worth the time, cost, and efforts?

(11)

There are very few literature papers talking about the impact of AI in creating brands, and most of them are covering the aspects of analyzing brands and helping in marketing them, not on creating them or finding the opportunities and challenges in implementing AI.

1.3 Purpose and research questions

The use of AI in branding has rarely been studied in existing literature. This research aims to explore the opportunities and threats for branding professionals when using AI for the purpose of creating unique brands and brand elements. Consequently, by discovering the capabilities of AI and the need of it for branding professionals. The following research questions has been created to investigate the existing gap of AI’s impact on brand and brand elements:

RQ1: What is the impact of AI on visual and non-visual branding elements?

RQ2: What are the opportunities and threats of using AI in branding from the point of view of branding- and IT specialists?

Responding to these research questions will contribute to the limited existing literature of AI and branding and provide a more comprehensive understanding of how these two separate fields can be integrated. Through a deeper understanding and insights of how AI could be integrated in brand creating, the findings will provide valuable insights of how this could be adapted into reality and ease for branding professionals.

1.4 Disposition

The disposition of the study along with an explanation of the overall content of each chapter can be found below in figure 1.

(12)

(13)

2. Theoretical framework

______________________________________________________________________________________________

This chapter presents relevant literature and concepts in relation to the research topic. The procedure for conducting and finding relevant literature is first clarified followed by the literature review in the field of branding as well as in the field of AI are introduced.

______________________________________________________________________________ 2.1 Literature review procedure

In order to find relevant and accurate literature connected to this topic, a deep search was initiated on different online libraries, such as Google Scholar, Jönköping university’s library, uppsatser.se, and researchgate.net. Since literatures in these resources goes through a pre-selection process that requires high standards, such as impact or peer-review to verify that the publications are consistently of high quality, trustworthiness, and relevance. The research scope was split into three sections to classify related studies and articles: Artificial intelligence, branding, and then the intersection of both.

The search for literatures started by specifying the keywords that could be used to find the relevant literature that could support this research. The keywords are within the domain of impact of AI on branding and brand management. Keywords such as “artificial intelligence”, “machine learning”, “deep learning”, “neural networks”, “management”, “brand*”, “marketing”, “design*”. The results of these searches were tremendously many, therefore filters were added to narrow down the number of results and to make the research more specific. The filters chosen were “Business” and “management” to display results related to the domain of general management. After the filter was applied, the number of research became low enough to be manageable to go through them and get an overview about the topics that could be interesting, and relevant to the topic chosen. 10 to 15 research literatures were selected from each section defined previously (AI & branding), and only two related to the third section where both AI and branding were supposed to be combined, this indicates that the topic was not researched enough before. After going through the titles, abstracts, and introductions, the topic was defined, and manual filtration occurs again in order to select the most relevant articles and research.

Next search round started after the gap in the research, topic, and the research questions were defined and this search focused more on researching detailed topics. At this stage, filtering went faster and easier, as the topic was known and already narrowed. Keywords that were used at

(14)

this point were more specified: “data analysis” “generative adversarial networks”, “deep neural networks”, “brand elements”, “logo”, “brand vision”, “brand personality”, “colors”, and so on. The same filters were implied here too, and the results of this round were more accurate and related to the domain of this research. The final step involved checking the validity of the selected papers and literatures. The research studies added and used in this paper are all cited before, and they are well known studies to researchers in this topic. Number of citations varied between 50 and 2500 times, as displayed by the academic search engines used to find them.

2.2 Branding

Today, all types of organizations within all sectors want to have a brand and some persons even want to be managed as brands (Farhana, 2012). Individual customers are urged to see themselves as personal brands, deserving to grow, develop, and nurture in the same way that commercial brand do (de Vries et al., 2017). With easy access to internet and social medias, clients are more empowered than ever meaning the end for brands without identity. Brands took a significant turn in 1990s resulting in a new sense of meaning and became the most important element of the marketing mix, which consists of: product, price, place, and promotion (Doyle, 2016; Wood, 2000). Branding is an essential part of any marketing strategy, since it helps companies to communicate with their potential customers (Ivanovic & Collin, 2014). The cult of branding was fueled by a spike in advertising investment following the early 1990s’ recession. Instead of strictly pushing goods and services, businesses used ads and sponsorship to create brand sense and meaning. The goods and services of companies, which previously was center of marketing efforts, became nothing more than promotional vehicles for the brand’s overall beliefs, appearance, and experience. Apple, Nike, and The Body Shop are all good examples of companies who turned their otherwise widespread and essential items, into a lifestyle statement and preference of greater significance than just satisfying a consumer’s desire for easy consumables (Doyle, 2016).

However, the definition and measurement of a brand, distinguishes between experts (Kapferer, 2012). Adam West et al. (2018) mentioned that it is a general knowledge to academics and industries that brands embody “something greater than just a product”. Traditionally, a brand is defined as: “the name, associated with one or more items in the product line, which is used to identify the source of character of the item(s)” (Kotler, 2000). While technically, Keller (2003)

(15)

says, “whenever a marketer creates a new name, logo, or symbol for a new product, he or she has created a brand”.

There are a lot of different definitions for brands, individuals tend to see it from their own perspective. Ivanovic and Collin (2014) for example, define branding as the process of providing goods or services unique identity in form of brand names. Doyle (2016) looks at branding as a set of characteristics that distinguishes and adds value to a business, organization, products, service concept, or even an individual in comparison to its rivals, its advocates, stakeholders, and consumers. A brand is a name, phrase, logo, sign, mark, symbol, style, design or some other feature or set of features that distinguishes one company's good or service from competitors' (Ojasalo et al., 2008; Wood, 2000). It is the amount of all the emotional interactions that individuals have towards it. When buyers or customers recognize a brand and its characteristics, this is referred to as "brand recognition." A brand is a commitment to the customer and a set of qualities that someone purchases in order to be satisfied. These qualities may be actual or imaginary, logical or emotional, visible or non-visible (Ambler, 1992; Ojasalo et al., 2008).

2.2.1 The importance of branding

The brand is an overview of all the qualities connected with it, and it may improve consumers' trust in their choices in the stage of post-purchase. The selling firm is expected to get higher rates based on the strong brand they created, and in the best-case scenario, competitors will face a problem of rejection to their products (Ojasalo et al., 2008). Per year, companies in the United States alone, spend nearly $130 billion on conventional ads to develop their brands and boost their revenue (de Vries et al., 2017). It has long been acknowledged that strong brands can provide a company with a long-term strategic advantage (Hankinson, 2012).

According to Kotler and Armstrong (2010), branding is important because it reside in the minds of customers and therefore reflect a consumer's emotions and perceptions about the product and its quality. In other words, brands embody what a service or product means to a customer (Kotler & Armstrong, 2010). The well-known advantages of brand building revolved all around utility of a powerful corporate brand in building customer relationships (Ojasalo et al., 2008). Airey (2009) indicates that branding is critical because customers choose and evaluate products based on their perceived value rather than their real value. Apple announced in 2003 that they were going to change the color of Apple's logo from red to silver. Within few hours, over 200 people had

(16)

submitted an online petition asking the old logo to be reinstated. This could be explained by the strong commitment that customers have towards the brand, that is why building a strong brand is very important (Walsh et al., 2010). A strong brand may also help to advance affiliate partnerships, such as those with value-added distributors and other sales channel partners (Ojasalo et al., 2008).

Without a hesitation, brand developing, marketing and management responsibility rests on everyone in the company, each in their own roles and activities, especially in the different communication channels and conversations they have with customers (Ojasalo et al., 2008; Wood, 2000). The brand's principles and values should be correctly and regularly conveyed to everyone at all levels of the company in order for them to be communicated with external customers and all stakeholders. To improve brand success, high-level management must actively participate in the branding process (Hankinson, 2012). As a result, the elements of building and creating brands should be made available to all employees on a tangible basis (Ojasalo et al., 2008). The brand message should be delivered correctly and accurately to all internal and external stakeholders by using the marketing mix (Wood, 2000).

2.2.2 Brand measurement

Brands need to have good marketing and advertisement strategies to be able to deliver the brand's messages. Customers will evaluate whether this brand is providing the product that suits their necessities through the messages they receive from the advertisements (de Vries et al., 2017).

There are thousands and thousands of brands everywhere in this world, but how can one measure their strength and know which brand is the stronger one? According to Hankinson (2012), brand awareness has recently been recognized as an integrating factor for the growth of strong brands. Another measurement aspect that is used to indicate the success of a brand is how much the customer is attached emotionally to a brand (Ojasalo et al., 2008). The success of a brand could also be measured by the brand performance, and performance is of two categories, customer-based and business-based performance. Customer-based performance consists of satisfaction of customers and their awareness, image expectations, brand mentality or behavior, and brand loyalty. while business-based performance covers financial measures which includes production costs, marketing, sales, profits, and the expenses communications and training (Hankinson, 2012). According to John (2016), understanding what customers know about brands allows researchers and brand managers to better identify and measure the strength of the brands.

(17)

2.2.3 Brand elements

If there was only one specific goal for developing a brand, it should represent product-related qualities such as “trustworthiness”, “ease of use”, “awareness”, and “profitability” (Ojasalo et al., 2008).

Brands consist of different elements that companies need to create in order to have a strong brand. As mentioned before, brand qualities may be actual or imaginary, logical or emotional, visible, or non-visible as defined in this study (Ambler, 1992; Ojasalo et al., 2008). Companies have a finite amount of funds and firstly they must determine where to invest them. Each business must be able to address the customers’ question: “why should I buy from you?”. When developing a brand, simply just having a brand name consistent with the value proposition of the brand is not enough. After selecting the brand name, its different definitions and promises must be developed through brand identity work (Kotler, 2012).

A brand’s characteristics are both visible and non-visible: a name, products, services, a trademark or visual logo, credibility, brand loyalty, mental connections, business culture, and fundamental beliefs, all of which combine to create a recognizable, convincing, and meaningful brand impression in the mind of the beholder (Doyle, 2016). According to Farhana (2012), brand elements are those that can be trademarked, that help to recognize and distinguish a brand from others. Brand names, web addresses, logos, icons, characters, spokespersons, slogans, jingles, products' packages, and colors are the most important brand elements. Brand elements may be selected to raise brand recognition while also facilitating the creation of deep, beneficial, and distinct brand associations (Keller, 2003). Due to the limitation of this study, only some of the brand's elements are chosen and presented in the following text.

Brand Name - A brand name is a vital, and core symbol of the brand, serving as the foundation for

the recognition, awareness, and communications processes. Names has a strong influence on external stakeholders and target audience, that is why it needs to be managed actively. A brand name should convey the company and the products message; it should be inspiring and a believable promise to grab consumers’ attention, furthermore it must be memorable and easy to pronounce (Farhana, 2012).

(18)

Logo - A logo is a visual depiction, a graphical symbol that elicits memory associations with the

target brand. They could be images, typographs, or colored shapes (Farhana, 2012). This form or shape has a significant impact on perceptions (Walsh et al., 2010). Logos are frequently important in establishing equity, particularly in terms of brand recognition (Keller, 2003). According to Farhana (2012), the visual identification of a brand is critical to creating and retaining a foothold in the marketplace. Logos provide two essential roles for brands: identity and distinction.

Colors - Another aspect influencing brand awareness are the brand's colors. Color sense has been

thoroughly researched in recent years (Bottomley & Doyle, 2006). According to the study of Aaker et al. (2001), colors could be linked with the dimensions of brand personality. For example, red color represents love; blue represents calm and trust; green represents honesty, pink, yellow, and purple represent energy and enthusiasm.

Personality - Personality is a non-visual element of the brand. Aaker (1997) defined brand

personality as the collection of human traits identified and related to a brand. It is one of the core values of the brand. Personality is related to the emotional characteristics of customers (Ojasalo et al., 2008).

Vision - Vision is another non-visual element. Companies have important visions that they stand

for, and these visions are relevant to their target audiences. Managers must clearly and consistently define the values of the company and the principles for their brands. The vision of the company should be clear and known by its customers and stockholders in order to increase their awareness about the company (Kohli, 2014).

Beside personality and vision, a company needs to define the brand’s promise, attributes, and message. They should represent and reflect the company’s real character, it should be based on true bases, not “wished” bases (Ojasalo et al., 2008). Table one below presents an overview of the mentioned brand elements.

(19)

Brand Element Element type

Brand Name Visual

Logo Visual

Color Visual

Personality Non-visual

Vision Non-visual

Table 1: Conclusion of mentioned brand elements (own elaboration)

2.3 AI

In this subchapter a brief background regarding AI and some of its usage areas is presented, especially those related to the topic branding.

2.3.1 Background of AI

For most of the general public, artificial intelligence or "AI" means those robots that are risking us of losing our jobs and businesses. But for people who make decisions in enterprises AI means something else (FORTUNE Knowledge Group, 2017).

Generally, AI refers to a branch of science concerned with building smart machines that can perform a task that would typically require human intelligence (Dobrev, 2012). This technology has become so efficient that some individuals fear that AI can become more intelligent than humans (Thompson, 2018). This fear arises from the fact that computers generally outperform humans in rule-based tasks that require convergent intelligence. Consequently, on the perception of this fact, the question if AI can outperform humans in divergent intelligence arises (Thompson, 2018).

The above-mentioned concerns are legitimate considering the February 1996 event when IBM’s Deep Blue computers defeated Garry Kasparov, the world-renowned chess master, eventually winning a six-game series in 1997, becoming the first computer to do so (Whitney, 2017). AI has developed into an interesting phenomenon and is still developing. According to Anyoha (2017), it begins with Alan Turing’s 1950 paper on "Computer Machinery and Intelligence" that asked the question, "if humans can learn through available information to make decisions, why not machines?".

(20)

Anyoha (2017) posits that the first AI program, Logic Theorist, was developed in 1955 by Allen Newell, Cliff Shaw, and Herbert Simons. While this program could execute commands, it was greatly limited in storing information. Between 1957-1974 machine learning developed immensely with programs and hardware that were less cheap, more accessible, and could store more information. This theme carried on into the 1980s with the expansion of algorithm toolkits by John Hopsfield and David Rumelhus who popularized deep learning that allowed computers that could mimic human experts. However, between 1982 and 1990, there was a lull in AI development largely occasioned by the Japanese government’s funding of the fifth-generation computer project which was a total failure.

According to Anyoha (2017), this trend changed between 1990 and 2000. With limited government financing and less hype, AI technology development spurred between this period; its heights were marked by the previously mentioned defeat of Garry Kasparov in 1997. From then forward, AI development has increased storage size, speed, and execution in line with Moore Law’s estimates that suggest that computer speeds double after every year.

The majority of AI research has concentrated on machine learning approaches since then. Mitchell (1997) provides the most often used description of Machine Learning (ML): “A computer program is said to learn from experience E with respect to some class of tasks T and performance test P, if its performance at tasks in T, as calculated by P, increases with experience E.” Despite its distinct evolution, machine learning has become the dominant model in AI science and is commonly regarded as a subfield of AI (Goodfellow et al., 2016).

The market for artificial intelligence (AI) technologies is thriving very fast. Beyond the hype and the heightened media attention, the numerous startups and the internet giants racing to acquire them. There is a vital financial increase in investment and adoption by enterprises. A Narrative Science survey in 2016 found that 38% of enterprises were already using AI, growing to 62% by 2018. Forrester Research observed an increase of 300% in investment in artificial intelligence in 2017 compared to that in 2016. IDC estimated that the market of AI will grow from $8 billion in 2016 to more than $47 billion in 2020 (Press, 2017).

The world’s biggest companies such as Apple, Facebook, Amazon, Tencent, and Alibaba, have all invested heavily in artificial intelligence to improve their entire product ecosystem. Google, the world's largest search engine and the most visited website, has restructured its business model to apply artificial intelligence to its products and services. It has grown into a service for

(21)

over 1 billion users and uses AI in a wide range of areas from advertising, marketing, product design, customer service, data analysis and analytics to data science and machine learning (Harries, 2018).

2.3.2 The usages of AI

Profit maximization is an important aspect of most businesses; consequently, marketing plays a vital role in the business since it helps the business owners to reach their target audience. With globalization, traditional marketing strategies such as using billboards and face-to-face interactions have been rendered obsolete since consumers “of a business” product no longer live in the same locality, thus, the increased need for Artificial Intelligence Marketing (AIM) (Thiraviyam, 2018). According to Thiraviyam (2018), the big data capabilities, and the increased speed and efficiency of machine learning helps market teams such as those of big corporations like Google and Facebook to connect with their target audience at ease. Thiraviyam (2018) argues that, while the core of marketing has not changed, the way humans communicate has changed marketing, and AI has consequently taken advantage of this phenomenon to provide businesses with a competitive edge (Thiraviyam, 2018).

AI can be used in marketing in the following ways; first, machine learning algorithms can conduct market research that helps the business to identify potential threats and opportunities in the market (Jarek & Mazurek, 2019). Its big data capabilities aid in collecting data from its target audience’s computing activity and automatically generating AI-powered personalized marketing. Secondly, since most businesses today embrace computing, the need for computer security becomes vital. AI can help keep business computer systems free of cyberattacks so that systems can run efficiently (Hacioglu & Sevgilioglu, 2019). This is made possible since AI can prioritize risks and instantly spots malware on networks; thus, avoiding intrusion of a business’ network systems.

There are different digital asset management platforms that are using the technologies of AI and Machine Learning to enable marketers and creatives to manipulate, distribute and analyze brand security as a single source of truth. This promotes knowledge about brand management beyond the competition (Roetzer, 2020).

In addition, AI algorithms can generate personalized marketing emails hence, helping businesses reach specific audiences. This is because these algorithms can map an internet user’s website experience and collate an individual's interactions with online content, eventually,

(22)

analyzing the data to generate a personalized curated email (Pearson, 2019). Thus, automatically sending tailored marketing emails to would-be consumers. Also, tailored AI marketing can be used to give personalized marketing insights. AI algorithms can monitor internet user's purchase behavior, onsite internet interactions, and online past communications. It then uses its predictive analysis to suggest hyper-personalized experiences (Shen, 2014). This can be evidenced by websites’ suggestions of geo-specific events.

AI is making technological breakthroughs in the marketing industry, and by using predictive analytics powered by advanced machine learning algorithms, companies will be able to deliver better products that are of interest to their potential customers. By automating the acquisition of customer insights, artificial intelligence is changing how digital marketing is used (Qureshi, 2017).

Moreover, AI helps improving the customers experience by providing faster and more efficient solutions. For example, understanding human emotions and predicting consumer behavior contributes significantly to improving customer experiences. On the other hand, it could mean tagging an AI - an enhanced experience as such, so that consumers understand their behavior and benefits, rather than manually checking if they do not like what they see (Kopanakis, 2019; LaBarre, 2018).

Separate from marketing, branding provides a business with a specific image different from other businesses of the same type. Thus, a business brand entails a business identity. The role of AI in branding mirrors the question posited by the authors at the beginning of the paper, ‘can machine learning out-perform humans in terms of divergent intelligence?’ Stevens Jr and Zabelina (2020) carried out research to determine AI’s potential to engage in a creative process. The study involved measuring the brainpower of individuals using computers to extract electroencephalogram (EEG) features and categorizing human brain activity as either more intelligent or less intelligent. The study was based on the assumption that if machines could accurately classify human brainpower, then they could learn and execute more intelligent and less intelligent activities at will.

Participants in the study were required to think about normal things such as bricks or bread to determine less intelligence. On the other hand, to measure higher brain activities, the study required the participants to think about uncommon things. When the computer results were compared, less intelligent mental activities had a mean of 63.9%, while more intelligent brain

(23)

activities recorded 82.3% (Stevens Jr & Zabelina, 2020). This study therefore evidence’s AI’s potential to engage in creative thought. The quintessence of this assertion was evidenced in 2016 when IBM Watson cognitive platform was used to create the first-ever AI-created movie trailer for the 20th Century Fox horror movie, “Morgan” (IBM, n.d.).

2.3.3 GANs and creativity

The development of AI in creativity is still at a less advanced stage, quoting John Smith, manager of Multimedia and Vision at IBM, “it is easy for AI to come up with something novel just randomly, but it is very hard to come up with something novel and unexpected and useful" (IBM, n.d.). However, AI features such as generative adversarial networks (GAN) and deep neural networks (DNN) have been utilized in numerous creative business activities such as cancer diagnosis, self-driven cars, and facial recognition software (Dickson, 2018).

According to Isola et al. (2018), GANs are designed to produce outcomes for a given task using data that the AI has already been trained on; this qualifies GANs for image-to-image translation. Zhu et al. (2020) stated that image-to-image translation is a type of vision and graphics problem in which the aim is to learn the matching between an input source image and an output resulting image by using a training collection of matched pairs of images. In other words, GANs help in translating a source image into another new image by using saved and analyzed data. Automatic image-to-image translation is characterized similarly to automatic language translation, by providing appropriate and sufficient training data, the task of converting or translating one potential representation of a scene into another gets more accurate (Isola et al., 2018).

GANs has been developed in different aspects and are used in producing and generating different outputs. In image translating, GANs has been used in suggesting drawings, for example face swaps between people in pictures and video frames, masking the background from a picture, coloring sketches, and even in finding color palettes (Isola et al., 2018). The following figures represent some of the samples that are generated by using GANs:

(24)

Figure 2: Suggested drawing (Isola et al., 2018)

Figure 3: Background masking (Isola et al., 2018)

(25)

Figure 5: Coloring palette (Isola et al., 2018)

Figure 6: Object transfiguration (Zhu et al., 2020)

The findings on object transfiguration between horses and zebras are presented in the top two rows in the above picture. These come from training on 939 images from the wild horse class and 1177 images from the zebra class. The system also trained on 996 images for apples and 1020 images of navel orange, the results are presented in the images in the bottom two rows (Zhu et al., 2020). To make it clearer, the following figure presents another example.

(26)

Figure 7: Generated faces with the use of GANs (Karras et al., 2019)

The figure above shows two sets of sources, A and B, used as input. With power of GANs, the AI algorithms analyzes the sources and produces new generated faces which does not exist in real life (Karras et al., 2019).

Another usage of GANs which could also represent the creativity that AI has reached, is translating text-to-image. This is another way of generating images but by using text instead of other images as the previous examples showed. These images are generated by developed algorithms based on GANs and AI. Despite that GANs has produced promising results in terms of creating sharper images, it is still difficult for GANs to produce high-resolution (e.g., 256x256) images. Several solutions have been proposed to enhance the training and improve the image qualities (Zhang et al., 2018).

(27)

Figure 8: Generated pictures using text input (Zhang et al., 2018)

The figure above shows the text description (input) of the required images and how GANs generated these pictures in different resolutions and qualities. All these birds are generated using AI algorithms and a big set of data, with long-term training the AI can generate new birds as a combination of different birds (Zhang et al., 2018).

To summarize it, GANs are originally composed of a generator and a discriminator that are trained with different targets. GANs generator is trained to create samples that are close to the actual data distribution in order to deceive the discriminator, while the discriminator is trained to differentiate between actual samples from the true data distribution and false samples produced by the generator. GANs have recently shown considerable promise in simulating dynamic data distributions such as those of messages, images, and videos (Zhang et al., 2018).

(28)

3. Method

______________________________________________________________________________________________

The following chapter describes the research method and its connection to the purpose of the study. By that, it addresses the research philosophy and explains the choice of research approach. Furthermore, the research design and method used to collect data as well as the analysis of the data is described. The chapter ends with an evaluation of the research quality and specifies ethical issues.

______________________________________________________________________________ To describe how the research process makes its progress Easterby-Smith et al. (2018) use the metaphor of a tree. This study will follow the concept of the tree metaphor proposed by Easterby-Smith et al. (2018) illustrated in figure 9.

Figure 9: The tree metaphor (own elaboration based on the concept of Easterby-Smith et al., 2018)

3.1 Research philosophy

The inner ring shown in figure 9, and core of this metaphor, represents ontology which is the philosophical study about the nature of reality. In other words, ontology can be translated into the existence of everything knowable or how one view and make assumptions regarding the reality (Byrne, 2016; Easterby-Smith et al., 2018; Saunders et al., 2019). Situated on a continuum, there are four different ontological positions reaching from realism to nominalism. The realism position, which also consist of several varieties, assumes that the world exists regardless of experience or

(29)

observation. The nominalism position, on the other hand, argues that the reality cannot exist without the involvement of perception or observation (Easterby-Smith et al., 2018). The second ring shown in figure 9, epistemology, represents the philosophical study about the nature of knowledge and concerns ways of examining “how we know what we know” (Byrne, 2016; Easterby-Smith et al., 2018). The opposing views of positivism and social constructionism constitutes the epistemological philosophy. The positivist viewpoint holds that the world exists externally, and objective approaches should be used to assess its properties. Adopters of social constructionism believes that people’s interactions with one another, rather than objective and external facts, add meaning to the reality (Easterby-Smith et al., 2018).

This research is based on the ontological position of relativism and the epistemological view of social constructionism. The research topic implies both an understanding of to what extent it is possible to use AI for this purpose and what the outcome could be for professionals using AI in branding. By that, this study involves social actors and perspectives of individuals and due to this, various interpretations of situations related to the subject will occur. As a result of their own perspective of the world, individuals can interpret different situations in various ways. It is reasonable to argue that different observers of this topic will have different viewpoints depending on background, interests, age, education, and knowledge. Furthermore, the role of the researcher requires to comprehend the individual’s subjective reality in order to make sense and understand their motives and intentions in a meaningful way. Thereby, the ontological view of relativism and the epistemological position of social constructionism are believed to better suit the purpose of this research.

3.2 Research approach

Placed at the third ring in figure 9, we find methodology: a collection of techniques used to create a coherent content (Easterby-Smith et al., 2018). Saunders et al. (2012) suggests 3 main approaches to consider: deductive, inductive, and abductive. The deductive approach involves drawing logic conclusions from an established theory, if the formed premises appoint to be true one can deduce the conclusion to be true. Deduction has several distinguishing characteristics such as the quest for explanations of casual interactions between variables and theories, and the need of facts to be measurable, often quantitatively. The inductive reasoning argues that the conclusion is supported by observations instead of derived upon them. While a large sample is commonly used in the

(30)

deductive approach, the emphasis of induction is often on a limited sample size and the approach is particularly interested with the context in which events occur. Lastly, the abductive approach determines possible premises to conclusions made from a deriving ‘surprising fact’ observed. These surprises can happen at any point during the process, even while composing the project report which causes a back-and-forth approach similarly to a combination of the deductive and inductive approaches. Abduction often entails gathering data, explaining themes and patterns, and incorporating it into an overall conceptual framework, which can then be tested against current and new data (Saunders et al., 2012).

Individuals’ experiences, perceptions and perspectives are relevant for this topic and the emphasis is on sense-making and understanding of the gathered data to obtain new insights. The purpose of this study does not include to logically deduce assumptions from well-establish hypotheses, neither does it involve testing an overall conceptual framework or theory. The aim, on the other hand, is to investigate the topic, define themes and patterns, and generalize the findings from specific to the general. Therefore, this research will embrace the inductive research approach. This decision is also supported by the choice of a qualitative research which will be addressed in detail in the next section.

3.3 Research design

The outer ring in figure 9 represents the range of methods and techniques available at the disposal of the researcher (Easterby-Smith et al., 2018). One can either choose to follow a qualitative, quantitative, or mixed methods research design. Quantitative research often refers to a data collection technique that uses or generates numerical data in contrast to qualitative research that uses or generates non-numerical data. Generally, quantitative research distinguishes between experiments and surveys while qualitative research considers a broader interval of research strategies (Saunders et al., 2019).

To answer the stated research questions, this study applies a qualitative research design and semi-structured interviews are used to meet the expectations of a qualitative study. The use of semi-structured interviews as a means of data collection can be beneficial in a variety of circumstances, for example when questions are nuanced or open-ended, or when it is important to establish personal contact (Saunders et al., 2012). As this topic concerns the impact of AI technologies on branding elements in organizations, interviews of semi-structured characteristics

(31)

give the interviewees more opportunities to express themselves and allows for a discussion to take place, hence making connections between AI, branding and branding elements easier to understand. Considering this, interviews of semi-structured characteristics were thought as the best suit for this study. A general summary of the methodological approaches and techniques used for conducting research in this study is provided below in Figure 10.

Figure 10: Summary of chosen methods and approaches (own elaboration)

3.4 Data collection

Data can be collected by using primary, secondary, or a combination between both types of data. Primary data refers to entirely new information that the researcher has gathered on their own (Easterby-Smith et al., 2018) whereas secondary data sources include information that has already been gathered for another purpose (Saunders et al., 2012). Both primary and secondary sources are taken into consideration to support the research objectives of this study. Although appropriate secondary data of AI and branding combined is vague, it is important to include existing literature regarding the separate fields to support the primary data and research issue.

3.4.1 Sampling selection

There are two types of sampling methods available to use: i) probability or representative sampling, usually associated with surveys, and ii) non-probability sampling. These sampling methods then consists of a range of sampling techniques and the choice of technique depends on the study (Saunders et al., 2012). This study uses a non-probability sampling method characterized by the heterogeneous purposive sampling technique. The study has a clear focus in terms of two key themes (AI and branding) and aims for a large variation in terms of participants with diverse characteristics. Saunders et al., (2012) describes heterogeneous purposive sampling as a beneficial

(32)

technique for small sample sizes and to receive maximum variations in the collected data. To some extent this study also applied the snowballing sampling technique by asking the participants to recommend additional individuals who might be relevant for the study (Flick, 2007; Saunders et al., 2012).

Sample size was set to approximately 12-15 persons. This estimation was based partly on the minimum non-probability sample size for semi-structured interviews suggested by Saunders et al. (2012), which range between 5-25 individuals, and partly in consideration to meet the requirements from Jönköping University of approximately 10-15 interview hours. Professionals with various geographical background were approached to assure a full focus of the topic regardless of country setting. All individuals were approached either via LinkedIn, company websites, direct e-mail or through personal networks.

3.4.2 Semi-structured interviews

The first step towards conducting the interviews involved creating an interview guide, Flick (2007) indicates that this is typical for semi-structured interviews. To allow for a discussion and two-way communication to take place, semi-structured interviews with non-standardized characteristics were selected. Interview questions were grouped into three categories since the topic concerns two separate industries: branding, industry, and AI. Branding questions were highly focused on branding and branding elements whereas the industry questions were more focused on branding in different industries, contexts, and regions. The questions categorized in “AI” focused on AI and its capabilities. The choice of these categories and questions was based upon how well they were believed to cover all the aspects of the research topic.

Questions were approached in a flexible manner depending on the respondent’s background and knowledge. According to Saunders et al. (2012) the use of pre-defined questions in semi-structured interviews may vary depending on the interactions characteristics between the researcher and respondent(s). All interviews in this study were conducted on a one-to-one basis, involving only the researchers and a single participant. Saunders et al. (2012) claims that these types of interviews usually take place ‘face to face’, however, due to the circumstances caused by the covid-19 situation all interviews were conducted online via Microsoft teams and zoom. Figure 11 presents a brief overview of the choice of interview and how they were conducted.

(33)

Figure 11: Overview of interview choice and how it was executed (own elaboration)

Before any interviews were held, all participants were given relevant information about the research topic in order to give them the chance to prepare for the interview. Additionally, the respondents were introduced to the research topic once again in the beginning of the interview. All interviews were recorded and most of them were transcribed live using Microsoft Teams live transcription feature.

Saturation in data collection can be described as the creation of rich data during the data gathering period, by paying attention to the scope and replication from respondents, and thereby constructing the theoretical aspects of the gathering (Morse, 2015). Saturation is believed to be achieved when the data is adequate, rich, appropriate, and good. This can be quite confusing, however, replication from respondents can be a reasonable metric. Responses from participants will never be the exact same since details differs but overall, replication is achieved when responses from participants have essential characteristics in common (Morse, 2015. The responses from respondents started to get similar characteristics after about 14 interviews had been conducted and it was decided that one last interview should be conducted. In total 15 interviews were conducted which lasted between 40-82 minutes with a total of 14 hours and 18 minutes. Both researchers were present at all the interviews except for one which was conducted in a foreign language to one of the researchers. A summary of the conducted interviews can be found below in table 2.

(34)

Participant Position Int. length (min)

Expertise

A AI manager 60 Using AI in analyzing marketing.

B Brand consultant 60 Brand building, rebranding, brand strategy, brand marketing.

C Brand designer 45 Designing, brand building, and specialized in visual brand elements.

D Brand designer and photographer

40 Designing, brand building, brand- and market analysis.

E Designer and brand strategist

55 Brand building, brand strategy, brand marketing.

F Designer and brand strategist

70 Brand building, rebranding, brand strategy, brand marketing.

G Brand Manager and strategist

52 Brand building, rebranding, brand strategy, brand, and marketing analysis.

H AI and ML project manager

45 Analyzing data, finding AI and ML solutions, and implementing them to the market.

I AI expert 46 Analyzing and creating industrial and supply-chain solutions based on AI and ML.

J AI expert 60 Applying AI solutions to autonomous driving and robotic agricultural solutions.

K AI expert 45 Applying AI solutions in health management. L Designer and brand

strategist

80 Designing, brand building, and specialized in visual brand elements.

M Brand designer 82 Designing, market- and brand analysis. N Frontend developer 60 Developing UI and UX designs, frontend

designing, brand designing.

O Frontend developer 58 Developing UI and UX designs, frontend designing, general designing.

(35)

3.5 Data analysis procedure

The choice of research philosophy often determines the approach to how qualitative data is analyzed (Easterby-Smith et al., 2018). When data is gathered through interviews, observations, or focus groups the most prominent tools used for analysis are coding and categorizing (Flick, 2007). Virginia Braun and Victoria Clarke (2006) suggest thematic analysis as a method to analyze qualitative research. By identifying, analyzing, and documenting data patterns (themes), this approach interprets various aspects of the research topic and gives a concise overview of the data. Thematic analysis can be approached in different ways, but one must adhere to theoretical and epistemological commitments. Easterby-Smith et al. (2018) further implicates that the relationships withing the developed themes needs to be explored with the aim of comparing, analyzing, and integrating the results. Nowell et al. (2017) suggests a step-by-step procedure for conducting thematic analysis that aims to meet a study’s trustworthiness criteria. To analyze the collected data, this thesis follows the protocol proposed by Nowell et al. (2017).

All data was automatically transcribed during the interviews held via Microsoft teams due to their live transcription feature. However, for one interview held via Microsoft teams this feature did not work and the transcription had to be done manually. The interview that was held via zoom also had to be transcribed manually as well as translated since it was conducted in another language than English. During the first phase of the thematic analysis, all interviews were relistened to and the transcriptions were double checked, regardless of whether the interview was automatically transcribed. This was done to familiarize oneself with all the facets of the data and to ensure trustworthiness, notes were also taken during this phase to get a sense of what the themes could be. The data was then coded in the second phase of the analysis to identify important sections of the text. The coding was conducted by highlighting text in the word file, when finish highlighting the text, all codes were counted to see the most reoccurring codes. The third phase which involves the search for themes began when all codes were identified and counted across the data set. The codes and its context were looked at to predict what possible themes they could emerge into.

The themes were then set in relation to findings of the raw datawhich characterizes an inductive approach according to Nowell et al. (2017) and the thematic analysis is then said to be data-driven (Virginia Braun & Victoria Clarke, 2006). Saturation is a commonly used term in research analysis aimed at describing the stage of when which further observations of the data would not lead to further information. If no new themes emerge from more analysis, thematic

(36)

saturation is reached. This is possible without fully discovering relationships between all themes (Lowe et al., 2018). As the codes did not seem to merge into any new themes it was decided to stop looking for new themes.

The set themes were discussed during the fourth phase and some codes were both deleted and changed since they were believed not to be needed in the specific theme. According to Virginia Braun and Victoria Clarke (2006) it is essential for the data within the themes to cohere meaningfully. The fifth phase involved naming and defining the different themes (Nowell et al., 2017). The themes were named accordingly to what they captured and in regard to get the reader an idea of what to expect from the specific theme. Finally, in the sixth and last phase the write-up for the findings started with the aim of constructing a coherent and concise content. Figures were used to describe the themes, patterns, and codes and these can be found in the chapter: research

findings.

3.6 Research quality

The way researchers’ approach and conduct their research inevitably determines the accuracy of it. In order for research to be useful and appealing to others, it must be relevant and trustworthy. A strong qualitative research is methodological and thorough, themes are formed and the connections between them are investigated. The goal is to compare, analyze, and incorporate findings which emphasizes the importance of analysis. However, one must be able to determine if the results that emerge from the analysis are useful which highlights the significance of research quality (Easterby-Smith et al., 2018).

Most qualitative studies are incapable of being repeated or replicated since they are performed by people interacting with a specific environment at a specific time. Therefore, the value of qualitative studies often lies in its originality, not in its ability to be repeated. To summarize, qualitative studies might fail to meet the requirements of research using a quantitative approach, however this does not conclude that the research is less meaningful and important (Easterby-Smith et al., 2018). To assess the quality of this research a discussion regarding credibility, transferability, dependability, and confirmability for qualitative (naturalistic) research suggested by Guba (1981) will now take place.

(37)

3.6.1 Credibility

This criterion concerns whether the collected data can be considered trustworthy or not. Several methods can be used for this purpose and this study has taken triangulation into consideration. Triangulation refers to the use of several sources, methods, and theories to develop a thorough understanding of the research phenomenon (Guba, 1981). This research is built on a theoretical framework consisting of an extensive amount of literature with different authors. Additionally, 15 interviews have been conducted with different participants to ensure a wide set of perspectives regarding the research topic.

Easterby-Smith et al. (2018) refers to Miles et al. (2014) who states that triangulation can reduce potential bias regarding sampling strategies. However, it may also make the research analysis more difficult which entails findings to be tested against ’hard facts’, whenever it is possible. This nature of this research concern individuals’ perceptions and feelings regarding the research topic, therefore it is highly interesting to understand and interpret the perspectives of each individual. However, the study also relies on ‘hard facts’ provided by secondary data.

Lastly, an interview guide was created before interviews took place and all participants were supplied with information regarding the research topic before the interviews were conducted, to assure the credibility of the data. These steps are considered to help avoiding data quality issues according to Saunders et al. (2012).

3.6.2 Transferability

Transferability concerns to which degree qualitative research can be applied to other contexts. Both Easterby-Smith et al. (2018) and Guba (1981) agrees that qualitative studies are context bound and incapable of being replicated. This research uses a purposive sampling method aimed to increase the amount of data uncovered which according to Guba (1981) is done to ensure transferability. Guba (1981) refers to Geertz (1973) who promotes collection of “thick” descriptive data to enable possible transfer of the specific context to other similar. This research fulfils the criteria of “thick description” by providing and describing important contextual details necessary to the transferability.

The impact of AI on branding elements : Opportunities and challenges as seen by branding and IT specialists