Monetisation of information : an explorative research on data driven business models in the start-up world

(1)

Monetisation of information: An explorative research on data

driven business models in the start-up world

Thesis Supervisor: Prof. Dr. Peter Van Baalen

Antoine Korulski | Student number: 11422394 Master Thesis Digital Business | MSc. Business Studies | Amsterdam Business School | University of Amsterdam January 2018

(2)

Statement of Originality

This document is written by student Antoine Korulski who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

While the data economy and data driven business models have raised attention amongst practitioners, research has not developed sufficiently on the topic (Otto & Aier, 2013). A study performed on business leaders by Kart et al. (2011) suggests that there is an issue in capturing the value of data and to monetise on this asset. Companies able to seek strategical positioning around big data through business remodelling are able to capture a significant competitive advantage; such as Wal-Mart, which optimises its supply chain through analytics (Davenport, 2006). This research seeks better understanding of data driven business models through an explorative research based on a multiple case study. The research is built on the foundation of a business model framework of five dimensions, respectively: revenue model, source of data, data cost structure, customer segment and value proposition; derived from literature. The primary data is collected using a qualitative approach of semi-structured interviews with founders and C-executives of leading data driven start-ups in Amsterdam, Shanghai and London. The in-depth analysis and comparison of the study-cases defined attributes of successful data driven business models, which revenue model is based on personalised pricing and consultancy services. Moreover, the findings define the best sources of data, which are predominately acquired freely through customers and offer competitive advantage to the company.

(4)

Acknowledgments

First of all, a large thank you to my supervisor Peter Van Baalen, for his patience and support in the completion of this project.

Furthermore, I would like to take the opportunity to express my gratitude to the respondents, who took the time to answer my questions and form the foundation of this research. Their kindness and availability to share insights about their ventures is greatly appreciated.

Finally, to the most important persons: Parents, Brother and Friends who inspired me from the beginning and offered a continuous support, a gigantic thank you.

(5)

Index of Tables and Figures

Table 1a: Information on Selected study cases………29

Table 1b: Information on Selected study cases………30

Table 2: Interview questions linked to the propositions………..……....31

Table 3: Revenue model (P1) ………...33

Table 4: Data source (P2) ………....34

Table 5: Data cost structure (P3) ……….35

Table 6: Customer segment (P4) ……….35

Table 7: Value proposition (P5) ………..36

Table 8: DDBM framework applied to Hatch………..40

Table 9: DDBM framework applied to FuelUp………...……43

Table 10: DDBM framework applied to Ultra IoT………..45

Table 11: DDBM framework applied to Drcom………..48

Table 12: DDBM framework applied to Class of 2020………...50

Table 13: DDBM framework applied to Stockchain………...52

Table 14: References and sources matched with the dimensions & features of the framework………54

Table 15: Revenue models applied in the case studies analysed……….55

Table 16: Data sources applied in the case studies analysed………...57

Table 17: Data cost structure applied in the case studies analysed………..58

Table 18: Customer segment applied in the case studies analysed………..59

Table 19: Value proposition applied in the case studies analysed………...60

Figure 1: The Business Model Canvas (Osterwalder & Pigneur, 2010) ……….19

Figure 2: Conceptual Framework……….22

(7)

List of Acronyms

API Application Interface Programming

AWS Amazon Web Service

B2B Business to Business

B2C Business to Consumer

C2C Consumer to Consumer

CEO Chief Executive Officer

CLM Customer Lifecycle Management

CLV Customer Lifetime Value

COO Chief Operation Officer

CTO Chief Technology Officer

CRM Customer Relationship Management

DDBM Data Driven Business Model

DFC Discounted Cash Flow

ERP Enterprise Resource planning

GM General Manager

HCP Health Care Professional

IoT Internet of Things

NGO Non-Governmental Organization

P&G Procter & Gamble

SME Subject Matter Expert

(8)

1. Introduction

1.1. Preliminary elements

Companies collect personal information at every stage of economic transactions. This tremendous amount of data about their customers, suppliers and various transactions is collected through multiple channels such as websites, transaction books, surveys, personal files. This data, as one calls it, is the “Big Data”. It is often unstructured, meaning that there is no formatting, it is in the state it was collected (Coronel, Morris, & Rob, 2013). After the Big Data has been cleaned, structured and converted, executives can extract valuable information from it, for instance by allowing a company to make predictive analysis through data mining. In recent years, a trend that emerged is the data economy, in which business models are based on the provision of quality data (Otto & Aier, 2013). This information can be priced and sold to partners or competitors positioned at a different level in the value chain. On one hand, big data is considered as a highly valuable asset by companies that own it. On the other hand, data is hard to price because of its specificities such as the perishability of the information it contains. Some other characteristics of data assets are the fact that their veracity is hard to prove (Schroeck et al., 2012), as well as the high collecting cost versus the cheap reproduction cost.

“What’s the scope of the opportunity for companies? In 2015, the digital universe

contained 4.4 zettabytes. (A zettabyte is one sextillion bytes, or

1,000,000,000,000,000,000,000 characters.) And that already unfathomable store will

increase ten times over by 2020, thanks to millions of devices sharing information in

the Internet of Things. But only five percent of that data is being analysed today. The

(9)

Companies, nowadays, draw competitive advantage thanks to the use of data and analytics, such as Wal-Mart, which optimises its supply chain through analytics (Davenport, 2006). The number one social media company, Facebook, makes most of its revenue by providing users’ information to advertising agencies. There are various ways to draw a business model based on data, in most cases this data must be anonymised to protect users’ identity. Then, it is important to differentiate the data by customising the information and the price (Shapiro & Varian, 1999). Otherwise, the risk is that the data is sold as an exclusivity to one company, which would alienate the possibility to sell the data to other companies.

Gupta (2009) states that GM (General Motors) and P&G (Procter & Gamble) spend over four billion USD annually on advertising. This spending shows us how much it costs to acquire new customers and grow a database. However, not all of this cost can be internalised and considered as an asset because data depreciates over time.

In the current state of the world, many start-ups gamble their valuation on data collection and value of the information collected. VCs (Venture Capitalists) and Angel investors are struggling to find the real value of these companies based on database creations. Some financial techniques such as DFC (Discounted Cash Flows) can be applied for valuation estimates (King, 2007). However, those valuations are speculative and may lead investors into confusion. One strategy used by companies to acquire data is done through mergers and acquisitions. However, this process often fails because companies overestimate revenues they will make through the acquisitions (Lewis & McKone, 2016). This is partly due to the fact that companies struggle to evaluate immaterial assets: trademark, brands, intellectual property, etc. (Reed, 2007).

(10)

performed on 720 IT and business leaders by Kart et al. (2011) suggests that there is an issue in capturing the value and the monetisation of data. The research by Hartman et al. (2014) attempts to close the gap by proposing a taxonomy of Data Driven Business Models (DDBM) amongst start-ups. Their paper provides one of the first, if not the first, empirically derived taxonomy of DDMBs and is a great foundation for further research. As per stated by the authors of the research, three shortcomings appear to open the path to further studies. (1) “The diversity and sample sized.” (2) “The application of a framework derived from literature.” (3) “The

framework used for coding and clustering reduces the complexity of the companies to a limited

number of binary features. As stated by Hartman et al. (2014): “future work should also

embrace a greater foresight perspective. So far, the study has identified clusters of currently

existing data-driven business models.” This suggests that future business models are not

limited to those clusters, as the digital economy and various data driven business models are in constant innovation. In addition to that, new technologies and innovation come often first through start-up companies (Criscuolo et al. 2012).

1.2. Research question

The purpose of this study is to heel in the research gap stated above, by introducing the following research question.

What are the underlying business models, in terms of revenue streams and cost

structure, in start-ups relying on data as a key resource of growth?

Through this research question, the study aims to answer the following sub-questions: 1. How do data driven start-ups make revenue out of their data?

2. What are their customer segments?

3. What are the data sources and key partners in providing data?

(11)

5. What are the common aspects and differences in the data driven business model

among the six case start-ups?

In order to answer this research question and its sub-questions, an explorative multiple-case study of six start-ups is conducted. The start-ups have been sampled through analysis of their core competency related to data indifferently of their industry or geographical locations. To increase the validity of the study, the analysis of the business models is carried out through a single lens, which is the business model framework proposed by Osterwalder et al. (2010). This systematic analysis insures that the case studies are analysed through the same angle. Furthermore, to reinforce the credibility of the research secondary data from various sources has been gathered, with a primary focus on the research done at the University of Cambridge, by the research group Cambridge Service Alliance, on building a taxonomy and framework on data driven business models (Hartman et al., 2014). As well as primary data gathered through semi-structured interviews. The semi-structured interviews include start-ups executives from the six cases but also a manager in a digital department of a renown Dutch bank, ABN Amro (this due to the proximity of banks to data and their interest in capturing value from data), the founder of a digital consultancy company, the CTO of TomTom (observation day at TomTom HQ), a lawyer expert in data privacy for the International Committee of the Red Cross. The various sources used in this research aim to increase the validity of the findings through triangulation (Saunders & Lewis, 2012).

1.3 Academic and managerial relevance

This research contributes to the existing academic literature in multiple ways. Firstly, it presents an alternative to one of the only existing frameworks for data driven business models.

(12)

of a business model for start-ups that aim at capturing value from data. Secondly, it offers a quick overview of what is the current evolution of business models in the digital economy. This is done indifferently of the geographic position or industry of the start-ups analysed. Thirdly, the study aims at closing the research gap on defining what is a data driven business model and suggesting further implication in creating growth mainly through the resource of data. Surprisingly, scholars have published little on this subject (Chesbrough & Rosenbloom, 2002 and Hartman et al., 2014).

In terms of managerial relevance, this study aims at offering strategical insights on how to position a company when the key resource is data. For IT leaders and other executives, it may offer an overview of the possibilities provided by their companies' intangible assets (e.g. information goods) and help them identify solutions to monetise on the data or, in other words, the “gold” they are seating on. In the context of e-business and innovative companies, this study offers valuable insights on how to position the company to maximise revenue streams from a resource that might have been neglected.

1.4 Structure of the thesis

To begin with, this study explores extensively the current state of research on the concepts of information goods and data driven business models. This knowledge is then summarised in the literature review allowing a definition of the key constructs and highlighting the research gap. Furthermore, this section will use various theories which support the relevancy of the main research question of this paper. As a second step, the conceptual framework is formulated and the methodology defined. The methodology section aims at presenting the case studies, explaining the research methods, procedures and data analysis of first-hand data and relate it with the secondary data from existing research. In the following

(13)

chapter, the results from the coding procedure are presented. Then, based on the analysed data, the empirical findings are presented and discussed to propose a revisited framework. Finally, the conclusion and limitations of the study are presented, as well as recommendations for further researches will be given.

2. Literature Review

In this chapter, different definitions will be presented and put in perspective to data. The concept of business model framework will be discussed and the application of data driven business models will be established. The conceptual framework and working propositions will be explained.

2.1. Information good

The term information is used very broadly. It can be anything that can be digitised, encoded as a stream of bits (Shapiro & Varian, 1999). In the digital economy, the focus is made on information originated from personal information, for instance personal preference data, localisation of users. When talking about “information goods” it is important to realise that information has different value depending on the consumers. Some information has business value and some information has entertainment value, but regardless of the source of the value, people are willing to pay for information (Shapiro & Varian, 1999). Information has a particular characteristic, it is expensive to produce but cheap to reproduce, which makes it harder to value because all the value is hypothetical and depends on what this information is used for but the intrinsic value is quite low. Thus, information is considered as an experience good because the consumer has to experience it to value it (Shapiro & Varian, 1999).

(14)

2.2. Big data

The name Big data is derived from the circumstance that the datasets are too large to be processed by a regular database system, which are not powerful enough to capture, transform, analyse and save these datasets (Manyika et al., 2011). It is hard to estimate a dataset that can be categorised as Big data but usually the range is from a few dozen terabytes to many petabytes, depending on the business industry and the software used (Manyika et al., 2011). The three types of data identified by Purcell (2014) are:

• Unstructured data: In the format in which they were collected, no transformation or formatting has been performed (Coronel, Morris & Rob, 2013). Some examples are PDF, documents, emails (Baltzan, 2012).

• Structured data: This data has been formatted according to the software requirements to allow use, storage and process of information (Coronel, Morris, & Rob, 2013). Transactional databases used by most vendors (e.g. SAP) store and operate with structured data (Manyika et al., 2011).

• Semi-structured data: This hybrid data has been partially processed (Coronel, Morris, & Rob, 2013). For instance, the Hypertext Markup Language (HTML) or the Extensible Markup Language (XML), both used for web development (Manyika et al., 2011).

The commonly used definition of big data by practitioners is the one given by Gartner (2012): “a high-volume, high-velocity and high-variety information assets that demand cost-effective,

innovative forms of information processing for enhanced insight and decision making.” Or

commonly called the 3Vs of big data: Volume, Velocity and Variety. To this we can add a fourth dimension, which addresses the uncertainty of data: Veracity (Schroeck et al., 2012). However, it is important to note that companies using data as their main source of value do not

(15)

necessarily have to fulfil the 4Vs criteria of big data to be considered data driven businesses (Hartman et al., 2014).

As mentioned by Manyika et al. (2011), data is easily replicable and raises several legal issues because the same data can be used by many people at the same time. Those are unique characteristics of data compared to conventional physical assets, this point being also mentioned by Shapiro and Varian (1999). The questions of ownership and rights linked to a dataset are easily raised. So, how should a dataset be attributed to the right person for the right purpose? This study will consider this topic by answering the question of techniques to monetise data.

2.3. Monetisation of information

As in traditional industries, products can be priced in different ways based on customers' characteristics and preferences. Shapiro and Varian (1999) distinguished three types of differential pricing:

• Personalised pricing: Sold to each user/costumer at a different price.

• Versioning: Offer a product line and let users choose the version of the product most appropriate for them.

• Group pricing: Set different prices for different groups of consumers, as for the student discounts.

As for Balasubramanian et al. (2015), their research defines two pricing mechanisms for information goods. These mechanisms are selling, where up-front payment allows unrestricted use, and pay-per-use, where payments are tailored to use. A central feature of pure information goods is that they can be conveniently accessed through multiple pricing mechanisms including selling, site licensing, subscriptions, differential pricing, and pay-per-use pricing (Chuang and

(16)

Jiang et al. (2007), similarly to Balasubramanian et al. (2015), refers to two pricing techniques, which are pay-per-use and perpetual. Perpetual licensing has been the model commonly adopted by software vendors, especially for PC software. With perpetual licensing, a consumer buys the software and can use it forever. This licensing scheme is also sometimes referred to as “shrink-wrapped licensing” since the full license key is often shipped with the software package itself in a shrink-wrapped box (Jiang et al., 2007).

To cite Babaioff et al. (2012): “While selling information about viewers raises obvious privacy questions, it also raises fascinating questions of a purely economic nature. How does one

quantify the value of this information? What is the optimal (i.e. revenue-maximizing) selling

strategy for information? What are the qualitative differences between selling information and

selling physical goods and services? How do these differences influence the design of markets

for information, and the algorithmic problems underlying such markets?”.

2.4. Database valuation

A customer database is used by many businesses to keep track of customer activities, purchases, personal information, etc. A database is an organised set of data, through the help of software processing.

To link the value of a database and the firm value many financial experts and theorists use two main approaches to valuation, which are based on discounted cash flows or multiples (Gupta, 2009).

Customers have a Lifetime Value (CLV) (Gupta, 2009), which becomes an important metric in a firm valuation and cannot be explained solely by the Goodwill. All those intangible assets such as trademarks, brand awareness and proprietary processes, are financially hard to value; norms like IFRS impose strict rules on Research and Development (R&D) but hardly on other

(17)

intangible assets like databases. Despite an absence of common standards, some practices enable the valuation of a database (Reed, 2007).

Database techniques for valuation (Reed, 2007):

- Cost based: This technique is commonly used amongst practitioners in valuation. Similar to an R&D valorisation, a database can be transformed into an asset by equating the value to the cost incurred in creating the asset (in this case a database). It appears that the database is often under-valued because it does not value the potential of the customer lifetime value, which often is higher.

- Income-based: Calculations are made similarly to a Discounted Cash Flows (DCF) method. The rule of thumb stipulates that the owner of a database can ask for one-fifth to one-third of the operating incomes generated through the licensed property (database). The rationale behind this “royalty method” suggest that the price is calculated on the corresponding need of using a self-made database if the firm licensing the database was not available for use.

- Market-based: A database can also be based on a past transaction of a similar database. However, since a database is rarely sold independently from the company that owns it, there is a lack of data on this subject.

As a complement to the analysis of Reed (2007), the quantity and quality are important factors that can limit and shape the evaluator’s choice on which valuation techniques to use (King, 2007). Eventually, it will not be possible for some specific businesses to sell their database because of legal aspects. However, the gap in King’s (2007) research is that there are no suggestions on how to design a strategy, at the introduction of the database, that will allow a maximum of legal flexibility or a design of data that allows ease in financial valuation.

(18)

Current research on data offers various techniques on how to value a database. However, those researches are mainly focused on the customer lifetime value (Gupta, 2007); cost-based, market based, income based techniques (Reed, 2007); quantity, quality and relevance of data (King, 2007). Those researches are mainly stating existing valuation techniques in finance and accounting to put them in analogy with database valuation. A visible gap on how to price an intangible asset such as a database persists. Research done by Reed (2007) offers good opening points about database preparation and valuation techniques. However, as mentioned by the author, a research gap remains: there is not enough data about database transaction to use it as a technique of comparative valuation. This research thesis aims to fill this gap by understanding why data is not a common transaction merchandise and identify through the case studies the various strategies to price data.

2.5. Business model framework

2.5.1. Literature on existing business model frameworks

A business model framework allows a systematic analysis and comparison of specific features (dimensions) of a start-up or established company. In recent years, the literature has been in constant evolution on this subject, the concept being now applied to innovation, strategy and digital businesses or e-business (Zott et al., 2011). Nevertheless, the consensus around presenting a business model is still missing, especially in the rapidly evolving economy (Weill et al., 2011; Zott et al., 2011 and Burkhart et al., 2011). Chesbrough and Rosenbloom (2002) were among the first to propose a business framework with functions (Hartman et al., 2014). Thereafter, Hedman and Kalling (2003) suggested a framework based on strategy theory and business model research, their framework offers seven dimensions: (1) customers, (2) offering, (3) resources, (4) activities and organisation, (5) competitors, (6) supply of factor and production inputs, (7) longitudinal process component. Johnson et al. (2008) suggested a

(19)

framework with 4 interconnected components, namely (1) the customer value proposition, (2) the profit formula, (3) key resources, (4) key processes.

2.5.2. Osterwalder et al. (2010) business model framework

In the start-up world, as well as the corporate world, one of the most prominent name for business model practitioners is Osterwalder (Stuckenberg et al., 2011). Osterwalder et al. (2010) business model canvas is one of the most applied and cited by Google scholars, to be exact 1001 times in 2014 (Hartman et al., 2014). The framework offers 9 dimensions: (1) key activities, (2) customer segment, (3) key partners, (4) key resources, (5) revenue streams, (6) cost structure, (7) distribution channel, (8) value proposition and finally (9) customer relationship. This one-page canvas aims at offering a clear understanding of the company values and business activities.

(20)

2.6. Data driven business model (DDBM)

In the current world, data is largely described as the new “gold” or the new “oil”. Its value has been collectively acknowledged (Amit & Zott, 2012). The digitalisation of the current economy in all the aspects of commercial and non-commercial activities is now undeniable. Therefore, building and implementing a data-driven business model has become an unavoidable area of study and application (Amit & Zott, 2012 and Hartman et al., 2014).

While a data driven business model stricto sensu has not been clearly defined in the scholarly literature, it is universally used amongst practitioners (Svrluga, 2012 and Diebold, 2012). Surprisingly, scholars have published little on this subject (Chesbrough & Rosenbloom, 2002 and Hartman et al., 2014). Furthermore, the term can be contemplated as being on the verge of establishment, as a grant from the British Research Council was introduced for a research on “New Economic Models in Digital Economy” (Hartman et al., 2014). The work of Hartman et al. (2014) is one of the unique researches on building a framework and building a taxonomy for data driven start-ups. In this aspect, the research was completed by Brownlow et al. (2015) on building a blueprint for innovation for data-driven business models.

A data driven business model applies to companies that rely on data as a major resource for their growth; this definition has three implications as stated by Hartman et al. (2014):

• The company does not have to solely focus on analytics but can limit itself to collecting and/or aggregating this data.

• The company is not limited to selling information or data but can as well develop other products in relation to data. An example of this could be Fitbit, whose bracelets track and monitor people's daily activity.

• Any company relies on data to a certain extent, even a small shop has a cashier machine that registers all transactions, but the prime focus in a DDBM is that data is the key resource and the main driver of the company’s activities.

(21)

“Data-driven businesses have been demonstrated to have an output and productivity that is 5–

6 per cent higher than similar organizations who are not utilizing data-driven processes.”

(Brownlow et al., 2015 via Brynjolfsson et al., 2011).

Banks being one of the industry at the forefront of data innovation with 71 percent of banks stating that data enables them to build competitive advantage (Turner, Schroeck & Shockley, 2013).

2.7. Research propositions and conceptual framework

The literature review in the previous sections, demonstrates that scholars have started to research on data driven business models but have not reached an agreement on how to systematically analyse business models in data driven industries. The global consensus is that the researches on this topic are too shallow and the evolution of business models in the digital economy is swift (Otto & Aier, 2013; Hartman et al., 2014; Criscuolo et al. 2012; Chesbrough & Rosenbloom, 2002 and Hartman et al., 2014).

In consideration of the stated gap in the literature review and here above, this study aims to formulate its research question as follows:

What are the underlying business models, in terms of revenue streams and cost

structure, in start-ups relying on data as a key resource of growth?

In order to answer this research question and building on research already done on DDBMs and business model frameworks, this research presents the subsequent conceptual model in Figure 2 aiming at answering the following sub-questions.

(22)

2. What are their customer segments?

3. What are the data sources and key partners in providing data?

4. What are the value propositions of data driven start-ups?

5. What are the common aspects and differences in the data driven business model

among the six case start-ups?

Figure 2: Conceptual Framework

Source: Author

Based on the review of existing business model frameworks, notably Osterwalder et al. (2010) canvas and Hartman et al. (2014) DDBM framework, this research formulates five dimensions of research or propositions:

• (P1) The revenue model: In order to survive in the long run, each company has to focus on at least one revenue stream (Hartman et al., 2014).

DDBM P1. REVENUE MODEL P.2 SOURCE OF DATA P.3 COST STRUCTURE P.4 CUSTOMER SEGMENT P.5 VALUE PROPOSITION

(23)

• (P2) Data source: The two main sources are internal and external such as third parties. Internal data comes from IT systems within the company (e.g. ERPs) and external data can be acquired freely on the Web or via data providers (Hartman et al., 2014).

• (P3) Cost Structure: At each step of business activities there are cost incurred such as labour or assets (Osterwalder, 2004). This study focuses on the advantage of cost structures when dealing with data. For example, a company like Tesla, which gathers data directly from the car it sells, has a cost advantage in acquiring data compared to another car producer.

• (P4) Target consumer or user: The end users who beneficiate from the data offer. The most generic classification will be used in this research, which divides consumers into other businesses (B2B) and consumers (B2C) (Morris et al., 2005).

• (P5) Value proposition: The central dimension of all business model frameworks; it is the expression of what the customer values and seeks from its supplier to create value (Johnson et al., 2008; Chesbrough & Rosenbloom, 2002 and Barnes et al., 2009).

The scope of this study will be mainly based on the systematic analysis of those five dimensions in the case studies. Some aspects on the key partners of the companies analysed will be discussed. However, to keep the focus on the monetisation of information in data driven companies, the scope is limited deliberately.

In the next chapter on methodology, the five dimensions stated above and the choice of case studies will be developed extensively.

(24)

3. Research Methodology and Data

The purpose of this research is to offer an in-depth explorative vision on how start-ups with data as a key resource generate revenue. As well as suggest a framework inspired from literature to systematically analyse the various dimensions of the case studies. This chapter presents the methodology of this study, which leads to the model presented in figure 3.

Figure 3: Methodology structure

Source: Author

Interpretivist paradigm research

Multiple case study _{Stat-up businesses related to data}Sampling

Data collection

Primary data Semi-structured interviews

Secondary data Public sources and academic

Data analysis

Coding

Nvivo soft. & DDBM framework frfraframework

Validation Case studies & comparison Inductive & deductive

approach

Qualitative research design

(25)

1.1. Research design

The research design defines how a research is conducted and analysed (Van der Velde et al., 2004). This research uses a qualitative approach based on multiple-case study because it aims at exploring new theories and adding fresh insights to the field of DDBMs as there are little scholar publications on the subject (Otto & Aier, 2013; Hartman et al., 2014; Criscuolo et al. 2012; Chesbrough & Rosenbloom, 2002 and Hartman et al., 2014). Therefore, an explorative research is appropriate in those circumstances (Saunders & Lewis, 2012). Yin (2009), as well as Saunders and Lewis (2012) are formal on the fact that a qualitative research addresses the “how” or “why” questions and indicates that the objective of the study is an action or a process. In this case, the research question is formulated as how start-ups can derive revenue from their data. Moreover, the study aims at explaining the process of establishing a data driven business model for companies who have data as a key resource.

The approach of this research is a hybrid approach between inductive and deductive methods, since the business model framework is derived from previous literatures but the research ambitions as well to discover new insights among start-ups in the digital economy. In this case, no control can be exerted over the course of events and measurement would be nearly impossible, therefore an experiment design is not feasible (Eisenhardt, 1989).

This research uses multiple case studies rather than a single case study, which increases the validity and robustness of the findings (Eisenhardt, 1989). The rationale behind the adoption of a multiple case study is theoretical replication (Yin, 1998). The case studies are selected based on the hypothesis that they will produce contrasting results. The goal of this research is to point out various strategies in relation to data driven business models.

(26)

1.2. Case selection and sampling

The cases were selected based on their value proposition, which had to be based on data. To avoid any criticism of strong bias, the decision was made to do a multiple-case analysis, which is appropriate in the investigation of contemporary subject (Yin, 2014). The sample contains six case start-ups, which were sampled purposefully, in other words the cases were selected on the postulate to answer the research question of this study (Marshall, 1996). As mentioned above, the aim of this research is to demonstrate contrasting results and the selection of multiple cases follows the rational of theoretical replication (Yin, 1998).

Transposed to practice, the cases were selected partly on Angelist, which is a website for ups, angel investors and job seekers. The tool is powerful and contains a vast database of start-ups around the world, which can be selected on specific criteria. In this case, filters were applied to find data driven start-ups and preferably in Amsterdam to avoid travel costs and facilitate face-to-face interactions with the interview participants. In addition to this, one start-up case was chosen through snowball sampling (Noy, 2006). Snowball sampling can be used when accessing niche fields of study, the approach as described by Noy (2006): “generates a unique type of social knowledge - knowledge which is emergent, political and interactional”.

However, the snowball sampling, besides the one start-up selected with this method, was mainly expended to find subject matter experts in the field of data driven companies. This method led to the interview of knowledgeable people, who added depth to the research. Through the variety of sources (scholarly articles, interviews with experts and case studies) this research uses the triangulation method, which increases validity (Saunders & Lewis, 2012). During the months of May and June 2017, approximately 40 personalized emails were sent to various data driven companies. Based on the responses, only six start-ups were selected to proceed with interviews. The selection was based on the ability of the case to provide valuable information without limited factors such as non-disclosure agreements. Moreover, the cases

(27)

had to be suitable to answer the research question within the theoretical framework and the pool had to be based on a variety of companies across business industries to validate the findings. As Eisenhardt (1989) explains, this method ensures testable, novel and rich information. Moreover, by selecting a variety of cases, the approach leads to explore and rank data with the focus on illustrating existing theories or generating working propositions (Eisenhardt, 1989 & Yin 1994).

1.3. Data collection

As mentioned previously, to increase the validity and credibility of this research, three types of sources are used allowing triangulation (Saunders & Lewis, 2012). This study being a qualitative research, primary data was collected through semi-structured interviews. Secondary data was sourced from academia and publicly available reports.

Secondary data

According to Van der Velde et al. (2014), secondary data is the collection of largely available and exploitable data. Therefore, this source is widely used during the preliminary phase of the research to measure the scope and breadth of the subject. In this research, secondary data has been collected from University catalogues, books, Google scholars, reviews from professional press and publicly available reports on Start-ups such as the Angelist database to mention one. Mass media and newspapers are out of the scope as the knowledge is diffuse and may bias the research more than it could help. This research includes valuable knowledge from the Cambridge Service Alliance founded in 2010, which is a unique global partnership between the University of Cambridge, IBM and BAE Systems. This research group produces insightful and practical research on leading firms and academics on complex service systems.

(28)

Furthermore, the business model frameworks from Osterwalder (2010) and Hartman et al. (2014) have been analysed thoroughly to build the conceptual data driven business model in this study.

Primary data

According to Van der Velde et al. (2014), interviews are predominantly used to gather insightful information, opinions and facts from organisations and individuals. The advantage of in-depth interviews is the possibility to gather detailed information in a short time compared to other methods such as experiments or surveys (Yin, 1994). Furthermore, according to Eisenhardt (1989), in-depth interviews are the most commonly used method in qualitative research. One disadvantage of using interviews may be the fact that interviewees are given information they might consider the most appropriate and create incoherence (Eisenhardt, 1989). This research aims to limit the bias just mentioned by combining various data sources such as research published by external parties and interviews with experts in the field.

Scholars and researchers produce unstructured or structured interviews and everything in between (Van der Velde et al., 2004). Open questions (unstructured interviews) aim at collecting exploratory data and closed questions aim at collecting explanatory data (Van der Velde et al., 2004). This research is hybrid between explorative and explanative, therefore interviews are semi-structured. However, during the interview the freedom was given to ask spontaneous questions (limited to the main question) to develop on a subject mentioned by the respondent.

The interviews took places during the months of June, July and August 2017. The locations of the interviews were mostly at the start-ups’ head-quarters or neutral environments like parks; two interviews took place via the common communication tool called Skype. All the interviews

(29)

were recorded via a recording device and transcribed manually thereafter. In order to increase the reliability of the transcripts, the recordings were listened to three times. Moreover, if a statement was unclear, the respondent was contacted for clarification. For transparency reasons, all the transcripts are attached in the appendix. Some interviews have been anonymized and should not be published as requested by the respondent. In the table 1a and table 1b below, the case studies are described by name, industry, location, number of employees, the position of the person(s) interviewed and a description of the start-up. One case study which is not added to the table 1a or table 1b is TomTom, the reason being that no formal interviews were conducted during the observation day at TomTom headquarters in Amsterdam during May 2017.

Table 1a: Information on Selected study cases

Company Hatch FuelUp Ultra IoT

Industry Digital marketing in

technology

Consultancy via data insights

Internet of things

Localisation London (GB),

Amsterdam (NL), Dusseldorf (DE), Kiev (UA), Portland (US) &

Bangkok (TH)

Amsterdam (NL) London (GB) &

Amsterdam (NL)

Employees 50 10 6

Respondent COO CTO CEO/Founder

Description Hatch is a marketing

technology company.

What that means is that they provide marketing

solutions like

e-commerce through

where-to-buy products.

Where-to-buy products

are a piece of scripts that brands or clients are putting on their websites that allow them to direct to online retailers from

their product page.

Therefore, users on their sites can browse their

FuelUp connects

corporations to start-ups. Their technology lets corporates find the best start-ups to work with and provides strategic insight

into the start-up

movements in their

market, giving them a competitive advantage. FuelUp developed an

advanced machine

learning algorithm, which continuously ranks and matches start-ups from their extensive global

Ultra IOT is a project based organisation, that monitors the

environment for

cities and

corporations and

project institutions,

making custom

sensor kits for

people to measure air quality, noise, whatever they want to measure. The data that is generated is

analysed and

(30)

can find where to buy it online.

ups that best fit their needs.

Source: Author

Table 1b: Information on Selected study cases

Company Drcom Class of 2020 Stockchain

Industry Health data & Digital agency

Real estate (ONG) E-commerce

Localisation Shanghai & Beijing, (CN), Ho Chi Minh City

(VN), Tokyo (JP) & Singapore (SG)

Amsterdam (NL) Amsterdam (NL)

Employees 76 9 3

Respondent General Manager China 1.Research Officer

2. Communication Officer

CEO/Founder

Description A digital agency

specialised in health care, supporting the big pharmaceutical

companies like Novartis, Roche, Pfizer in all their digital initiatives. All the different digital projects they are developing are targeting two kinds of audience; either the HCP

- the health care

professionals - or the patients. “Drcom is a boutique communication agency specialized in medical communication and a pioneer in Closed Loop Marketing (CLM) and

digital multichannel

strategy in life sciences.”

The Class of 2020 is a foundation and think thank focused on student

accommodation and

higher education issues within Europe, this can include anything from the internalisation of higher education, as well as just understanding the

different student

accommodation

real-estate markets

throughout Europe.

Their vision is for cities to attract and retain the brightest young minds, and for them to lead the way to social and economic success in return.

“A Stockchain is a network of web shops with shared inventory. A web shop in the

Stockchain is

simultaneously a point of sale and a source of new products for all other shops in the

network. Our

marketplace

management software is combining powerful

automated product

feed, order fulfilment

and drop shipping

solutions into one

platform. This enables our customers to grow

without financial

constraints through

smart collaboration

with their network.”

Source: Author

In order to collect meaningful data, the semi-structured interview was designed to link each of the five research propositions: Revenue model (P1), source of data (P2), cost structure (P3), target consumer (P4) and value proposition (P5) to a question. This aims at organising and

(31)

collecting the data in the most efficient and organised manner. Table 2 links the questions to one or many propositions.

Table 2: Interview questions linked to the propositions

Questions Other P1 P2 P3 P4 P5

Could you briefly Introduce the company? X

Could you tell me something about your position within the company and how it relates to Data?

X When you think of Data as a revenue stream, what

does it mean to you and your company?

X X

Does your company consider data as an asset and does it sell data to third parties (B2C or B2B)?

X Have you heard about The Business Model Canvas

by Osterwalder (2010) and its 9 dimensions?

X Based on that canvas, could you tell me about your

data-based revenue model? (Data-driven business model or DDBM)

X X

How do you acquire data (Data Source)? Is it internally (existing data or generated data) or externally (Free Data, customer provided or acquired)?

X X

What kind of cost do you have in acquiring data? X X

What kind of cost do you have in processing your data?

X What is your pricing method for Data (licensing,

exclusivity, other)?

X X

How did you come up with the pricing method? X X

Are you planning on changing your pricing model? If yes, why?

X X

Would it be possible to refer another executive in your company that would be able to answer those questions?

X

Is it possible to use this interview for my master Thesis and to contact you for more details if needed?

X

Source: Author

1.4. Research procedure and data analysis

As the president of COSMOS corporation, Yin (2009), presented the procedure for multiple-case study, which consists of developing theory, then multiple-case selection and finally the design of a data collection method. Data analysis in this research follows a protocol, which facilitates the

(32)

inspection of the data collected to answer the “how” and “why” questions of this research about data driven business models (Yin, 1994).

Firstly, data was collected through semi-structured interviews. Thereafter, the recordings of the interviews were transcribed manually. Finally, the transcripts followed the procedure of content analysis, which is done through coding and data classification or indexing, a technique used in qualitative researches in order to structure the data (Neuendorf, 2002). As Neuendorf (2002) mentioned, this process allows the researcher to highlight the most important features, themes and messages from semi-structured interviews. The objective is to establish relations and understanding about singularities (Ryan & Bernard, 2003).

The content from the interviews is analysed and linked to existing theories to confirm the patterns from known phenomena, as well as to explore new patterns, which is the aim of this mainly inductive research (Spiggle, 1994). The tool used for the coding process and thematic aggregation is the renowned qualitative research software called Nvivo 11.

In order to compare the case studies and make a systematic analysis, the conceptual framework mentioned in figure 2 is used in coding the interviews. This enables the findings of patterns across the case studies. The conceptual framework to analyse the data driven business is derived from existing literature, notably the Osterwalder et al. (2010) canvas and Hartman et al. (2014) DDBM framework. The proposed framework aims at analysing the business models through a holistic approach. To do this, the framework is based on 5 key dimensions (cf. Propositions P1, P2, P3, P4, P5) which are presented and fragmented in details below.

P1. Revenue Model

In the data driven start-up industry, the main strategy is to monetise information, which is given through data. For any company, at least one revenue stream is necessary to survive in the long

(33)

run. “Incorporating a revenue model into a DDBM is integral to its operational success” (Brownlow et al., 2015). Through holistic analysis of revenue streams in business model frameworks from Osterwalder et al. (2010) and Hartman et al. (2014), the below table 3 is presented with five revenue strategies, respectively: Licensing, personalised pricing, versioning, usage fee and subscription fee.

Table 3: Revenue model (P1)

Source: Author

P2. Data source

As one may expect, the key component in a data driven business framework is data. It is the “oil” and “gold” of the company and therefore considered as the key resource. The data sources are very specific to the industry and company, for instance financial services, retail or telecommunications consider self-generated data as the most substantial source (Brownlow et al., 2015). The data from those companies is generally unstructured data from transaction systems; this data needs to follow a process that is often called “data enrichment” which gives meaning to data and provides valuable information. To give an example, Kosala and Blockeel (2000) identify three main types of data sources from the Web: Web content, Web usage and Web structure data. Sources of data are extensive and practically unlimited; Singh (2010), Han

Re ve nu e M od el (P 1) Licensing Personalised pricing Versioning Usage fee subscription fee

(34)

et al. (2011), Schroeck et al. (2012), Gartner (2013) in their researches identify data sources from the Web, data warehousing, big data, and so forth. The aim in a DDBM is to consolidate those sources into streams, which can come from two main directions: externally or internally and then can be subdivided into sub-categories as presented below in table 4.

Table 4: Data source (P2)

Source: Author

P3. Cost Structure

In this research, the cost is focused and related to data. Other costs such as labour (else than data analytics), rental of work space, investments in technology are not within the scope of this study. The aim here is to identify the specific cost advantage of a company compared to the other case studies. As mentioned previously, a typical example would be Tesla manufacturing connected cars, which enables the company to capture data almost freely about driving habits and road conditions. Through the review of Osterwalder et al. (2010) framework and data analysis of the interviews, the three features related to data cost have been identified in table 5, respectively: cost of data acquisition, cost of data processing and cost of data storage.

Da ta S ou rc e (P 2) External Acquired Open data (Free) From customers Internal Generated data Existing data

(35)

Table 5: Data cost structure (P3)

Source: Author

P.4 Target Customer

A company has to identify who are the targets of its value proposition. Researchers among Morris et al. (2005) and Osterwalder (2004) identify two types of customer segments: other businesses (B2B) or consumers (B2C); presented in table 6 below. Consumer to consumer (C2C) has been suggested by Brownlow et al. (2015), where customers are used to acquire other customers but this strategy remains marginal and has not been identified in any case studies in this research.

Table 6: Customer segment (P4)

Source: Author Da ta C os t S tr uc tu re (P 3) Data acquisition Data processing Data storage Cu st om er S eg m en t ( P4 ) B2C B2B

(36)

P.5 Value proposition

The output of any data processing and analytics is an interpretation, which becomes knowledge or information. This knowledge is often mentioned in literature as the “value proposition” (Johnson & Christensen, 2008; Chesbrough & Rosenbloom, 2002 and Osterwalder et al., 2010). The value proposition is the benefit a company is offering to its customers through various services. In the case of data driven businesses, through the review of the existing literature (Hartman et al., 2014 and Brownlow et al., 2015) and case studies, the four features or services related to data have been identified in table 7; specifically: data analytics & reporting, data processing, data generation & data enhancement/enrichment and finally dashboards & visualisation.

Table 7: Value proposition (P5)

Source: Author

Based on the segmented framework presented in tables 3 to 7; the case studies are then clustered thanks to the analysis of the interviews, which are coded and organised in thematic. Cluster analysis is applicable for coded qualitative studies when the sample is small, the process is then binary and add clarity to the findings (Henry et al., 2015). Through the review of current research in data driven business models and the primary data, this study aims at producing a valid and credible framework, meaning that the research answers its purpose

Va lu e Pr op os iti on (P 5) Analytics & reporting Processing Data generation & data enrichment Dashboards & visualisation

(37)

through correct measures, which enable internal validity or causal relationship and external validity or the opportunity to generalise the results to the research domain (Yin, 2009 and Saunders et al., 2012). Furthermore, the quality of this research aims at answering the four measurements proposed by Spiggle (1994); specifically: innovation through the use of a creative and new approach to observe a phenomenon, integration through the use of materials to find untypical findings, resonance which answers the question of how informative and inspiring the research is and finally adequacy to measure how the results are linked to the study. In the next chapters the results, based on the data driven business model framework and the primary data, are presented and the shortcomings and limitations are identified.

4. Results

This chapter presents the findings from the analysis done on the semi-structured interviews through coding and thematic classification. The first section presents the within-case analysis, where the findings of each specific case are transposed to the five dimensions’ framework presented in the methodology. Thereafter, the cross-case analysis compares the six case studies and lays them in perspective to find patterns, which in the first time would support the frameworks’ propositions (Eisenhardt, 1989). In the second time, the semi-structured interviews with the subject matter experts are added to give relief and validity to the findings.

4.1. Within-case analysis

The goal in the within-case in-depth analysis is to discover patterns through the application of the DDBM framework, stipulated in the research methodology, to each case studies and identify the features of each of the five dimensions that characterise the specific case study.

(38)

4.1.1. Case study 1: Hatch

Hatch is a digital marketing start-up specialised in technology companies such as Bose, Intel, Nvidia, Asus, Philips. The company has 50 employees and is legally head-quartered in Amsterdam but has many offices throughout the world such as London, Bangkok, Portland, Kiev and Dusseldorf. Hatch provides marketing solutions such as e-commerce through where-to-buy products. Where-where-to-buy products are a piece of code scripts, that Hatch's customers put on their websites, which allows to direct to online electronic product retailers straight from the product page of the customer. In other terms, prospective end customers of product x can browse the website of Hatch’s customer such as Intel to find products and when the end consumer finds the product to purchase, the end customer can find where to buy it online. The analysis of this case study (cf. Hatch interview, appendix 1) depicted an interesting value proposition and revenue model.

(P1) Revenue model: The model of Hatch is based on two pricing methods. Firstly, a monthly subscription fee that applies to the consumer based on the country licence the customer has signed for. Secondly, a personalised pricing based on an “enterprise pack”, which is custom made for big clients. As mentioned by the COO: “It’s a recurring model. So, they will pay a monthly fee per license. And the license will be if you have a brand in a country. If I’m Bose

U.S., I will pay one license, Bose U.K. will pay another license, etc.” “We have got three

different packs and then an Enterprise one, which is more custom.” (Hatch interview, appendix

1). The COO mentioned that the company often changes the pricing method based on feedbacks from its customers and based on competition moves since the market for where-to-buy is growing and competitors are not sharing their prices (Hatch interview, appendix 1). (P2) Source of data: Hatch acquires its data solely externally mostly via its customers but also via third-party providers and open source data such as Google maps. Furthermore, web-scrapping is partially used when the data source is lacking and needs to be completed. They

(39)

will then enrich the data with their own admin interface. The COO concerning data sources: “It’s all these data that we’re getting through a third-party provider that is actually a global provider of data that we use in processing…”. On the affiliate networks: “So those affiliate

networks are actually managing the feed output to all those affiliates like us so we can connect

to them and it also provides sales data back as well.”. On enriching data: “Global location

provider gives us the data… we will enrich it with our own admin interface. So, at the end we

have our own database.” On Web-scrapping: “But really the worst-case scenario, we can

scrape. That’s where it can become a bit of data. It’s our own way of getting the data. So, we

scrape the site, we sort this information and we process it the same way as the other pieces of

data.”

(P3) Data cost structure: The main cost driver for Hatch is data processing through developers, data analysts in Ukraine and Amsterdam. As well as minor costs of infrastructure through Amazon Web Services. Usually, the company does not pay to acquire data because they receive data for free from retailors and their customers. If the company pays to acquire data, they will charge it back to its customer. Furthermore, the company stores all the data it acquires and “recycles” it for further customers. The COO on cost of data acquisition: “The model is really to reuse the same data for lots of multiple customers. So basically, our cost base is eventually

fixed.”. On data acquired from customers: “Apart from the actual processing for where-to-buy

online, it’s just about the implementation. We don’t pay to get the data. They actually give it

for free to us…” (Hatch interview, appendix 1).

(P4) Customer segment: The company solely focuses on the business to business (B2B) segment: “All our customers are those brands, Bose, Philips, etc. So, it’s B2B yes” (Hatch interview, appendix 1).

(40)

the retailor who sells Bose products and the end consumer who buys the product. All this is done through data enrichment of different data sources: “We have this data but you’re missing the block around the corner, so we might need to add it to that. And we are enriching existing

data as well.”; analytics on retailors in specific markets and finally dashboard and visualisation

it offers to its customer like Bose: “All these data we output through the API, then the results of the usage of all the users through the brands website, it’s also all gathered in an automated

dashboard so we can see the meanings, all the leads all the sales, the products per retailer,

etc.” (Hatch interview, appendix 1).

In table 8, the DDBM framework is applied to the case study of Hatch.

Table 8: DDBM framework applied to Hatch

Dimension Feature # Quotes

P1. Revenue Model Licensing 1 Personalised pricing 3 Versioning - Usage fee - Subscription fee 5 P2. Source of data External Acquired 6

Open data (free) 2

From customers 3

Internal Generated data 1

Existing data - P.3 Cost structure Data acquisition 2 Data processing 5 Data storage - P.4 Customer segment B2C - B2B 3 P.5 Value proposition Analytics/reporting 3 Processing 2

Data generation & data enrichment 6

Dashboards & visualisation 1

(41)

4.1.2. Case study 2: FuelUp

FuelUp is a start-up providing consultancy services through data insights and market research, its aim is to connect established companies with innovative start-ups. The company is based in Amsterdam and has ten employees as of today. The start-up is powered by a machine learning algorithm that ranks and matches start-ups around the globe.

The in-depth analysis of this case study (cf. FuelUp interview, appendix 2) displays a value proposition, that is a hybrid between consultancy services and reporting through data research. (P1) Revenue model: FuelUp's revenue is based on a monthly subscription fee, which gives customers access to the database and platform of the company. It is a fixed price per user or group of users, which gives access to all the data and updates as the CTO calls it: the Netflix, all you can eat model. “It is kind of a Netflix model basically, where you pay a fixed price per user, or per group of users and then you get access to all the data. That’s one of the models

that we have but that’s kind of the basic model like the “all you can eat” model.” (FuelUp

interview, appendix 2). Then FuelUp has also personalised pricing through consultancy services and specialised searches for big companies that want to connect with niche start-ups. The range is broad and varies from 10,000 EUR to more than 30,000 EUR; furthermore, the company takes a percentage if a joint-venture started between two companies that were mediated and connected through FuelUp. The result is that the company has two revenue streams, respectively: subscription fees and personalised pricing. The CTO on how the start-up comes start-up with pricing strategies: “… the pricing depends really on the scope of the customer and also what the customer is willing to pay and what in the past our customers are willing to

pay for this kind of services so that’s why I said trial and error.” On pricing model of

competitors: “I think that a lot of our competitors are also struggling with this price setting because I constantly see them changing their prices from a hundred or thousands. It is

(42)

(P2) Source of data: The start-up initially bootstrapped data from large public sets, when those platforms were free. Thereafter, FuelUp had to develop software to automate Web scrapping methods to crawl the Web and research relevant data for its value proposition. In some cases, customers bring their own data but it is a marginal source. Therefore, the company sources data solely from external sources as the CTO mentioned: “So what we did, we developed a lot of software, a lot of automation to obtain this data automatically. You have to think a bit in the

direction of crawling, external information extraction.” (FuelUp interview, appendix 2).

(P3) Data cost structure: Besides the initial software development cost to acquire data through Web crawling and scrapping, the company has no direct costs related to data acquisition. The main cost driver is data processing through manual work and automation development done by human capital. Data storage and cloud server costs are marginal as mentioned by the CTO: “Those cloud costs are relatively small compare to the cost of people. I mean it cost something

to host data but that’s not that much. I think it is a very small amount of the monthly cost, it is

basically to keep the services running. All the stuff is really in the people and in what they

develop.” (FuelUp interview, appendix 2).

(P4) Customer segment: The company is working with cross-industry customers, mostly big companies in industries such as chemical, telecom, banks. FuelUp's customer segment is business to business (B2B): “We don’t do any B2C, it is really a B2B company.” (FuelUp interview, appendix 2).

(P5) Value proposition: FuelUp's services are really about data generation and enrichment by aggregating data from various sources to produce information and knowledge on start-up markets. In order for the data to produce information, the start-up focuses on analytics to report the results of their researches. “… most of the value that we have is not in the pure data but rather in all the interpretation around it, that’s what we sell, that’s our unique selling point.”

Monetisation of information : an explorative research on data driven business models in the start-up world