Innovation strategy and firm performance

(1)

Innovation Strategy and Firm

Performance - The Role of Data

Analytics Explained

Master thesis

(2)

The Role of Data Analytics Explained

June 2018

Executive Programme in Management Studies Master of Digital Business

University of Amsterdam - Amsterdam Business School

Author

Marco Tiberius

Student number: 11197269 Thesis version: 1.0 (final)

Date of submission: June 29, 2018

Master thesis supervisor

(3)

Statement of Originality

This document is written by Student Marco Tiberius 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.

Name: Marco Tiberius

(4)

Preface

Dear reader,

With great pride I can inform you that you are in the possession of my master’s thesis: the crown jewel of my part-time master study. Succeeding did not go lightly, but it was sure made a lot more comfortable by some people, in particular my partner Caroline and my family. My special thanks goes out to them, for supporting me throughout the process of the study. Many thanks goes out to my supervisor Ed Peelen as well, who has endlessly supported me. With his feedback and persistency he made sure that I would get the most out of my research.

Because of my own interest in digital innovation and the big data analytics environment, this research was very fun and brought me many relevant learnings. I hope you enjoy reading this research as much as I did writing it.

All the best,

(5)

Listing of figures

Figure 1: Conceptual model ... 27

Figure 2: Interaction line graph (n = 143) ... 40

(7)

Listing of tables

Table 1: Definitions Resource Based Theory ... 19

Table 2: Sample results ... 33

Table 3: Scale reliability check ... 34

Table 4: Sample’s descriptive statistics for normality... 35

Table 5: Sample frequencies - age... 36

Table 6: Sample frequencies - employees ... 36

Table 7: Sample frequencies - Functional layer ... 36

Table 8: Sample frequencies - industries... 36

Table 9: Correlation table – exploration & second order constructs ... 37

Table 10: Correlation matrix ... 38

Table 11: Regression analysis – interaction effect ... 39

Table 12: Matrix of exploration and exploitation levels regressed ... 41

(8)

Abstract

Nowadays there are countless of innovation initiatives involving big data, but multiple researches report that the maturity of data analytics capabilities in organizations is relatively low. This research has answered the question if data analytics capabilities have a significant moderating effect on the relationship between ambidexterity and firm performance. The lack of theoretical research in the data analytics domain, and absence of research covering the relationship of data analytics and innovation, have been the reasons for choosing this topic.

To answer the research question, this research has made use of the survey methodology in a mono-method performing a quantitative research. Based on a prior research by Gupta et al. (2016) a measurement construct was issued for identifying data analytics capabilities in organizations (M. Gupta & George, 2016). By surveying experts in the field of marketing data analytics, evidence was found that data analytics capabilities have a significant positive moderating effect on the relationship, meaning that it increases firm performance. Also, higher data analytics capabilities show to increase performance more than low capabilities. Evidence was also found that data analytics capabilities have a stronger effect in an explorative-minded organization, than in an organization that has an exploitative culture.

From this research a set of takeaways appear that an organization should consider when incorporating data analytics capabilities successfully: (1) adopt an innovative company culture that wants to thrive on data, (2) be prepared and be able to invest in longer periods of unknown growth potential, and (3) continuously look for ways to improve the capabilities by looking beyond the current environment of the organization.

(9)

1. Introduction

Today’s world is increasingly subject to disruptions. New and rapidly advancing technology is driving these disruptions and the business environment is changing at high velocity. Our cars are driving themselves, we can monitor our home temperature from all around the world, and with our smartphones we can pay for groceries, hotels or taxis in a few seconds. Products and services are being digitized to match the need of tomorrow’s customers. Designing and developing these products is deemed to be fast and offer a seamless user experience (Ries, 2011).

In this digital age it becomes harder for organizations to keep up with this development pace, let alone innovate their products to create a long term competitive advantage. Technology and business innovations are increasingly being fueled by big data, making it critically important for organizations’ innovation success to know how to process big data and transform it into business value. The technology behind autonomous driving; artificial intelligence (AI) is also purely based on big data, and AI is said to potentially increase the labor productivity of some developed countries with up to 37% (Purdy & Daugherty, 2016). Organizations all over the world are now investing in big data initiatives to be that data-driven organization, but how important is it that they build the right capabilities, and how will it interact with their innovation strategy?

In order to remain relevant in this era of digital transition it is important, if not mandatory, for organizations to embrace innovation and merge it in their organizational behavior without affecting the ongoing operation. They must secure the growth and profitability of the organization’s current business, while simultaneously innovate to grow new business that is independent from its core business and current clientele. In literature these concepts are called: exploitation and exploration. The concept of exploitation is described as: focusing on efficiency by processes, short term development of the current business and incremental innovation in a stable environment to increase turnover and profitability. Its counterpart, exploration, entails flexible and experimental behavior, discovering new business

(10)

environments and managing radical innovation in an uncertain environment to secure long term growth (Duncan, 1976; Gibson & Birkinshaw, 2004; March, 1991; Tushman & O’Reilly, 1996).

The act of organizations balancing these two contradicting challenges is referred to as organizational ambidexterity (Levinthal & March, 1993; March, 1991; Tushman & O’Reilly, 1996). Organizational ambidexterity first appeared when Duncan (1976) used the term in his article. He wrote about the organization’s focus to design a dual structure to stimulate innovation, whereby he argued that the dual structure was meant to coordinate mainly two things; [1] initiate innovation and [2] implement innovation (Duncan, 1976). It was in 1991 when March (1991) refined and explained the ambidextrous tradeoff as; ‘The relation between exploration of new possibilities and the exploitation of old certainties’ (March, 1991). In his article ‘Exploration and Exploitation in organizational learning’ March spoke of two ‘types of learning’ that refer to what is now known as organizational ambidexterity. The closed loop learning (knowledge contained within organization and its individuals) describes exploitation, and the mutual learning (knowledge transferred beyond the organization-employee relationship) describes exploration (March, 1991).

Life case example

The importance of organizational ambidexterity is shown in the early days of Apple. Apple was founded in 1976 by Steve Jobs, Steve Wozniak and Ronald Wayne and primarily focused their business on personal computers and software development (Isaacson, 2015). In 1997, Apple almost went bankrupt but was saved by an investment of Microsoft. Shortly after, in 2001, Apple launched the Apple iPod and the iTunes Store which resulted in huge growth for the organization (Gilbert, Eyring, & Foster, 2012). The iPod and the iTunes Store were both disrupters to the industry at that time. There was no other music library where consumers could purchase digital songs, and so, the iTunes Store changed Apples revenue model entirely. Next to that, the iPod was a huge success because it was, and still is, extremely innovative in the segment of portable music players. Releasing these products altered the revenue model and fueled Apple with a welcoming turnover.

(11)

Apple’s efforts to focus on innovation parallel to their ongoing business shows a great example of pursuing ambidexterity. The continuous effort for exploration has brought the development of both the iPod and the iTunes Store and led to business growth in a new market with new customers. While doing so, they managed to develop their traditional PC business and realize growth simultaneously. These two capabilities are the core-characteristics of an ambidextrous organization.

Operating as an ambidextrous organization still is a popular innovation strategy these days, and remains to be very successful. In the meantime, big data has been cranking up the field of innovation as an important actor. Nowadays, digital innovation by big data is happening all around us, and mastering it correctly results in more sustainable growth potential for an organization (Purdy & Daugherty, 2016). To empirically generate insights in this topic, it will therefore be the focus of this thesis to examine if the development of data analytics capabilities have a strengthening effect on the relationship between ambidexterity and firm performance.

This research explores extant theoretical insights on both topics and generates new insights on how innovation and data analytics are related. It aims to offer both scholars and practitioners an insight in how the impact of innovation can be increased by developing the right capabilities of data analytics. The world is currently being overwhelmed by initiatives that involve big data, and the outcome of this research is expected to help organizations in designing and executing their roadmap to successful innovation by implementing a successful data analytics operation. It hopefully also offers a contribution to scholars who are looking to further extend the field of knowledge by researching a combination of the topics innovation and data analytics.

The importance of researching the connection between the domains of ambidexterity and data analytics is addressed in an article by Wan et al. (2017) too, in which they argue that ‘Innovation becomes critical to an organization’s survival’ (Wan, Mao, Hsieh, & Chen, 2017, p. 3). Additionally they state that data analytics has the power to support strategical decision making, but that there are only two studies (M. Gupta & George, 2016; Shuradze & Wagner, 2016) that have conceptualized data analytics capabilities

(12)

(ambidexterity not included), of which only Gupta et al. (2016) has performed an actual research (M. Gupta & George, 2016; Wan et al., 2017).

Gupta et al. (2016) also stressed the need for adding academic research to this domain, by arguing that most publications around data analytics came from technology vendors that might negatively bias the articles (M. Gupta & George, 2016). The domain of data analytics is emerging very quickly, but theoretical insights and successful data-driven initiatives are lacking (Davenport & Bean, 2018; M. Gupta & George, 2016), making it an important domain for further research (M. Gupta & George, 2016; Rialti, Marzi, Silic, & Ciappei, 2018; Wan et al., 2017). Adding to the scarce field of research is necessary, and researching how the topics innovation and data analytics relate to each other is considered the main objective throughout this research.

(13)

2. Literature review

2.1. Data analytics

2.1.1. What is Big Data?

Big data define the information that is generated by the behavior of a data subject. A data subject is considered to be an individual or a device that shows a behavior, online and offline. Due to the rise of interconnectivity of devices, also known as the Internet of Things (IOT), organizations are able to gather big data for analytics purposes (Xia, Yang, Wang, & Vinel, 2012). Consumers own wearables, smartphones, tablets, laptops, cars and smart devices for domotica. All these devices produce data that are valuable to organizations for increasing customers’ experience, enhancing competitive performance or to gain customer insights (Shuradze & Wagner, 2016). The definition of big data is defined by Gartner (2018) as:

“High-volume, high-velocity and/or high-variety information assets that demand cost-effective, innovative forms of information processing that enable enhanced insight, decision making, and process automation” (Gartner, 2018).

Gartner’s description holds the 3 vs – volume, velocity and variety – that are generally used to describe big data. More vs have been proposed in recent years, of which veracity (the dispersion and messiness of data) and value (insights derived from data) seem to be the best addition to Gartner’s phrase for describing big data (M. Gupta & George, 2016).

In their article, Gandomi and Haider (2015) mention that big data fail to achieve their full potential when they are only collected. It is the data analytics capabilities and subsequent leverage into decision making that makes the big data valuable (Gandomi & Haider, 2015). Gupta et al. confirm this in their 2016 article, stating that the clutter around big data pulls the focus away from the aspect that really contributes to performance and competitive advantage; the ability to make sense of the huge pile of raw and unstructured data and turn it into business value (M. Gupta & George, 2016). Looking at these

(14)

statements in combination with the above Gartner description of big data we can induce that processing, insights and decision making are an important part of big data (Gartner, 2018).

2.1.2. Rise of Big Data Analytics

Big Data was given the definition when it gained its popularity years ago. ‘Big’ was, and is still used, to describe the quantity and potential of the data. Over the years this ‘bigness’ of data was widely accepted, thus when addressing ‘Big Data’ in the industry these days, we only refer to it as ‘data’ or ‘the data’. Following this shift in the field, this thesis will also refer to it as; data.

Big data analytics (hereafter: data analytics) has grown in popularity in the last decade. The number of articles that is published and found on Google Scholar by using the keywords ‘Big Data Analytics’ rose from 2.050 in 2006, to 34.100 in 2016. With an increase of more than 1600% in ten years, it is safe to say that the concept of data analytics has gained extreme popularity (Google, 2018). Not entirely unjustified, because data have been empirically proven to be valuable to management for making data-driven decisions. According to Brynjolfsson et al. (2011) firm performance can be increased 5-6% if management’s decision making is data-driven (Brynjolfsson et al., 2011).

Another important aspect of data is that it is used for developing artificial intelligence (AI). Programming machines to learn, act and communicate with a human sense can increase service levels of organizations without adding workforce, or drive our vehicles without human interference. This is all done by algorithms that automatically analyze, interpret and act on data through machine learning. A rapport on AI by Accenture (2016) states that AI has the potential to grow the labor productivity in developed countries with 11% to 37% (Purdy & Daugherty, 2016).

2.1.3. Ambidexterity and Data Analytics

Despite the popularity of data analytics, however, there is little highly cited and recommended empirical research that aims for examining the role that data analytics capabilities play in ambidexterity. As also explained by Gupta et al. (2016) the literature that currently is available on data analytics capabilities is

(15)

mostly managerial, and lacks theoretical contribution (M. Gupta & George, 2016). Just recently Wan et al. (2017) have written an article in which they propose several reasons and a conceptual model for researching data analytics capabilities in the field of ambidexterity. This is one of the few articles addressing the topic of data analytics capabilities in the ambidexterity domain. The article itself, however, doesn’t contain empirical research. They solely propose a conceptual model to be researched that concentrates on the relationship between data analytics capabilities and firm performance within the ambidexterity domain (Wan et al., 2017).

2.1.4. The Ambidexterity of Data Analytics

In an article by Shuradze et al. (2016) there is a strong focus on consumer data rather than organizational data (Shuradze & Wagner, 2016). Conversely; Rialti et al. (2018) target the domain of organizational data in their article. They do so by proposing a conceptual framework for researching how business process management systems, that enable data analytics, have an effect on an organization’s agility (Rialti et al., 2018).

Although most research aims to explore the impact of data analytics in performance of the organization (M. Gupta & George, 2016; Lin, 2016; Rialti et al., 2018; Shuradze & Wagner, 2016; Wan et al., 2017), only some attention is given to the separation of areas where data analytics is applicable to. In their article Shuradze et al. (2016) cite research that explicitly mentions marketing to potentially benefit the most from data analytics (Shuradze & Wagner, 2016). New techniques that identify changing customer behavior are of great value to organizations that thrive for personalized offerings. Data analytics can contribute to profiling customer segments, personalization, marketing automation and increasing the effect of marketing campaigns that ultimately lead to higher conversions. This marketing-type of data analytics can be labeled as externally oriented analytics.

Assessing internal core activities by using data analytics, might also enhance performance like Rialti et al. (2018) suggest (Rialti et al., 2018). The extant studies on data analytics unfortunately mostly cover analytics for business growth purposes or for supply chain improvement (Hazen, Boone, Ezell, &

(16)

Jones-Farmer, 2014; Rialti et al., 2018; Wang, Gunasekaran, Ngai, & Papadopoulos, 2016). Trkman et al. (2010) have found a significant positive relationship of data analytics on supply chain performance, and developed a measurement model for studies in supply chain (Trkman, McCormack, Paulo Valadares de Oliveira, & Bronzo Ladeira, 2010). Deriving insights from internal activities by using data analytics capabilities is shown to benefit management in their decisions to lead the organization. This type of data analytics refers to internally oriented analytics.

Following these views on data two separate schools (I.e. external data analytics vs. internal data analytics) can be identified. A form of ambidextrous data analytics thereby comes to the fore, which makes it important to address the boundaries of this research. Looking at the current innovation environment, where digital and online is developing extremely fast, data is considered a real important asset in the field of marketing. It is currently not yet been discovered by many companies as a tool for internal auditing. A survey of Deloitte reports that 70% of the surveyed companies is in the begin phase of using data for internal improvement. Within three to five years, 58% of all respondents expect to use it in 50% of their audits (Deloitte, 2018).

On the marketing side of data, almost all companies are currently involved in big data initiatives (Davenport & Bean, 2018). The practical contribution is therefore expected to be bigger in the field of externally oriented data analytics. This research will therefore be specifically focused on delivering insights to the field of externally oriented analytics (I.e. marketing data analytics).

2.1.5. Data Analytics Capabilities as moderator

Unlike organizational ambidexterity and firm performance, data analytics has not yet been empirically researched so profoundly. This is partially related to the early popularity of ambidexterity and its effect on firm performance that originates from the 90’s and onwards (Levinthal & March, 1993; March, 1991; Tushman & O’Reilly, 1996). As previously mentioned, the number of publications around big data analytics started to increase around 2008 (Google, 2018). Genuine theoretical contributions to the literature on the big data analytics discipline, however, have been scarce. Most publications have been

(17)

written by practitioners and are therefore overall focused on the technical part of big data analytics instead of the theoretical aspect. Additionally, these contributions are mostly written by technology vendors which may influence their independent rigor to the subject. The published articles targeting the theoretical side of big data analytics mostly stem from 2016 and onwards, which is confirmed by Gupta et al. (2016) by their statement about the lacking empirical research for theoretical insights (M. Gupta & George, 2016).

Wan et al. (2017) advocate organizations’ data analytics capabilities as an important research topic in the field of ambidexterity, considering their potential positive impact on innovation (Wan et al., 2017). This proposition is backed by the 2018 annual Big Data Executive Survey of NewVantage Partners (2018), in which they explain how business innovation nowadays is driven by big data and AI. This survey reports that 73,2% of the respondents achieved measurable results due to big data and AI investments. The executives that did not achieve measurable results claim it is still too soon. Furthermore, 71,8% of the interviewed executives agreed that artificial intelligence and machine learning will have the greatest impact on their organizations in the next decade, which is more than cloud computing (12,7%) and block chain (7%) (Davenport & Bean, 2018).

Data analytics capabilities were defined by Troilo et al. (2016) as the ability to capture, analyze and use customer data (Troilo, Bouchet, Urban, & Sutton, 2016). They thus suggest that the complete ‘data lifecycle’ (I.e. from the collection to actionable decisions) represents an organization’s data analytics capabilities. In contrast; Gupta et al. (2016), Rialti et al. (2018) and Shuradze et al. (2016) state it as ‘the resources an organization can turn to for data analysis (M. Gupta & George, 2016; Rialti et al., 2018; Shuradze & Wagner, 2016). In this thesis the definition is assumed that data analytics capabilities refer to the resources of an organization to perform data analysis.

Covering the complete lifecycle of data as described by Troilo et al. (2016) (I.e. capture, analyze and use customer data) in this research would definitely be interesting. But, looking at current trends in the field of data analytics, a more concentrated approach is believed to be of more relevance. According to

(18)

Harvard Business Review (2018), citing the Big Data Executive Survey of NewVantage Partners, 97% of the organizations are now investing in big data and AI initiatives. Almost all organizations in this research (98,6%) indicate they pursue a data-driven culture, yet only one third (32,4%) claims to have succeeded and this same gap is reported yearly (Davenport & Bean, 2018). Therefore, presenting how to architect the operation in-house to harness the capabilities of data analytics is seen as the cornerstone of becoming data-driven, and thus makes a valuable contribution. Explaining how these organizations can successfully adopt data analytics capabilities in order to enhance their ambidextrous innovation strategy is not only adding practical relevance to the research, but is also expected to have a great fit with the current maturity of the organizations in practice.

Almost simultaneously, Gupta et al. (2016), Shuradze and Wagner (2016) and Wamba, Gunasekaran, Akter, Ren Dubey and Childe (2017) have constructed measurement models that address data analytics capabilities (M. Gupta & George, 2016; Shuradze & Wagner, 2016; Wamba et al., 2017). The conceptual models of Shuradze et al. (2016) and Gupta et al. (2016) follow roughly the same structure for they are both based on the Resource Based Theory. This theory considers an organization as a bundle of resources that ultimately produces a firms competitive advantages, providing a good understanding of how capabilities are built within organizations. Following the Resource Based Theory, organizational capabilities are divided in three classifications: skills, assets and knowledge (Henderson & Cockburn, 1994). These three classifications are regularly found in literature and were also used by Bharadwaj (2000) in her research for the connection between IT capability and firm performance (Bharadwaj, 2000). Gupta et al. (2016) and Shuradze et al. (2016) drew their construct from this research, and incorporated it into their research. Elaborating on these classifications and Bharadwaj’s (2000) research on IT capabilities, they have further developed these classifications, and renamed the concepts, as can be seen in table 1.

The article of Shuradze et al. (2016) covers all three classifications but merely proposes a measurement model and holds no actual empirical research like Gupta et al (2016). The research of Wamba et al. (2017) looks like the research of Gupta et al. (2016) at first, but the conceptual model of Wamba et al.

(19)

(2017) is constructed differently. Gupta et al. (2016) use all three concepts (I.e. assets, skills and knowledge) to measure data analytics capabilities, in accordance to prior research concerning organization’s capabilities (Bharadwaj, 2000; Henderson & Cockburn, 1994). Wamba et al. (2017) use only two of the second order constructs (I.e. assets and skills), leaving out on the concept for the knowledge aspect of an organization (Wamba et al., 2017). Furthermore, Gupta et al.’s (2016) research uses ‘only’ 31 questions, while Wamba et al. (2017) use more than 60 in the survey (M. Gupta & George, 2016; Wamba et al., 2017).

Table 1: Definitions Resource Based Theory

Author Concepts

Henderson et al. (1994) Assets Skills Knowledge Bharadwaj (2000) Infrastructure Human Intangible Gupta et al. (2016) Tangible Human Intangible Shuradze et al. (2016) Infrastructure Personnel Relationship Wamba et al. (2018) Infrastructure flexibility Management capabilities

Personnel expertise capability

The survey of Gupta et al. (2016) for measuring data analytics capabilities reported high Cronbach’s Alpha results (α = ,87 to ,94) making it fit for reusing in this research. The research explains what exactly drives the analytics capabilities in organizations (M. Gupta & George, 2016). The central topic of their research is; “to clarify a firm’s ability to assemble, integrate and deploy the big data specific resources” (M. Gupta & George, 2016, p. 1)

2.2. Ambidexterity

2.2.1. Types of Ambidexterity

Gibson and Birkinshaw (2004) identified two separate fields in ambidexterity1_{and proposed the two}

separate terms structural ambidexterity and contextual ambidexterity. Structural ambidexterity is

1_{If ambidexterity is performed within an organization, like exploitation and exploration, it is essentially referred to as organizational}

ambidexterity. So, unless clearly stated otherwise, using ‘ambidexterity’ in this thesis will refer to ‘organizational ambidexterity’ in the context

of organization’s innovation. This is for clarity purposes, because research in ambidexterity began to shift to departmental contexts like IT and HR (Ketkar & Puri, 2017; Tai, Wang, & Wang, 2017).

(20)

described as two separate business units solely focusing on exploitation or exploration. They describe contextual ambidexterity as the ability to separately conduct exploitation and exploration behavior within a single organization or business unit. Contextual ambidexterity can be found within an organization, a business unit, department, or within the characteristics of the individual. (Gibson & Birkinshaw, 2004)

2.2.2. Risks

Organizations that fail to pursue ambidexterity risk becoming obsolete over time (Chang & Hughes, 2012). Besides the act of pursuing ambidexterity, however, the most important challenge is finding the right balance between exploitative and explorative behavior (March, 1991; He & Wong, 2004; Junni, Sarala, Taras, & Tarba, 2013). An imbalance of the two practices can lead to either a success trap or a failure trap (Auh & Menguc, 2005; Gibson & Birkinshaw, 2004; Levinthal & March, 1993; March, 1991; Tushman & O’Reilly, 1996).

The success trap relates to the phenomena of organizations that instinctively tend to focus on exploitative behavior. It arises from success in a safe and familiar business environment. This focus yields stable results and in a relatively shorter time period than exploratory efforts. Because of this ‘easy and accessible success’ it might lead an organization to oversee the long-term benefits of exploration. The short term results then clutter the organization’s vision for long term successes.

On the opposite of the success trap, is the failure trap. This is described as the increased effort to focus on explorative behavior when a firm is running below expectations (Levinthal & March, 1993). Organizations’ negative results have been discussed to cause this focus that subsequently brings increasing pressure and more investments to the organization. The structural changes for managing the increase might diminish the focus on exploitation, causing disappointing results again which puts the organization in a loop (Levinthal & March, 1993; Tushman & O’Reilly, 1996; Gibson & Birkinshaw, 2004). These two traps as a phenomena are a threat to organizations and are ‘potentially self-destructive’ (March, 1991).

(21)

2.2.3. Influential Factors

Recent decades have had much empirical and academic research to the drivers and performance outcomes of ambidexterity. The main drivers for fostering ambidexterity were identified by scholars as: culture of the organization (Beckman, 2006; Tushman & O’Reilly, 1996), leadership (Chang & Hughes, 2012; Gibson & Birkinshaw, 2004; Lubatkin, Simsek, Ling, & John, 2006), environmental context (Gibson & Birkinshaw, 2004), and prior negative sales results (He & Wong, 2004).

In their article of 2008, Raisch and Birkinshaw (2008) have aggregated prior research on ambidexterity into a comprehensive model to show what areas of ambidexterity have been researched and sort them into five key area. The model provides a clear overview what [1] organizational antecedents drive [2] organizational ambidexterity and what [3] performance outcomes organizational ambidexterity delivers. The [4] environmental factors and [5] other moderators are both placed as having a direct effect on organizational ambidexterity, or having a moderating effect on its relation with organizational antecedents and performance outcomes. This meta-analysis shows that previous research has mainly targeted the organizational antecedents like structure, context and leadership, and that research for moderation by complex variables are scarcer (Raisch & Birkinshaw, 2008).

2.2.4. Ambidexterity as independent variable

The research will be concentrating on exploitative and exploratory efforts of organizations as the unit of measurement. As explained by Gupta, Smith and Shalley in their article of 2006, referring to Benner and Tushman (2002) and He and Wong (2004), these concepts address the innovation intentions of an organization best (Benner & Tushman, 2002; A. K. Gupta, Smith, & Shalley, 2006; He & Wong, 2004). This is confirmed by Jansen, Van den Bosch and Voldebra (2005) who focus one these 2 concepts when aiming to measure innovation effects (Jansen, Bosch, & Volberda, 2005).

Although the concepts of exploitation and exploration might seem straightforward, Gupta et al. (2006) stress the need to clarify what is intended with these concepts in a particular research, because not

(22)

describing them can easily cause misinterpretation (A. K. Gupta et al., 2006). In order to prevent misalignment on the definitions that this research pursues, exploitation is defined as; ‘The extent to which an organization is encouraging the improvement of processes, by using existing knowledge and chasing developments that benefit their current client base and areas where an organization is currently active’. The concept of exploration is defined as; ‘The extent to which an organization stimulates flexible working and investing time in experiments to gain knowledge needed for developing services or products that target new customers in new industries’.

Further analyzing the extant research, a foundation of organizational learning is clearly visible as the basis for ambidexterity. March (1991) has defined ambidexterity through two types of learning in his article and a lot of ambidexterity research has been further built on this article (March, 1991). Auh et al. (2005) have even used ‘organizational learning’ as a variable in their conceptual model and Danneels (2002) has examined the moderating effect of organization’s learning competences on the relationship between product innovation and the organization’s renewal (Auh & Menguc, 2005). Danneels (2002) conceptualized the theories on firm competences that are concerned with product innovation and firm renewal. He argues that in order to engage in exploration you have to be capable of ‘learning-to-learn’ (Danneels, 2002). This summarizes the ability of an organization to obtain knowledge and new competences in order to realize innovation. Danneels (2002) states: “Drawing on organizational learning theory, new product projects are depicted as serving to further develop existing competences or as vehicles for the firm to learn new domains of activity” (Danneels, 2002, p. 1097)

2.3. Firm Performance

2.3.1. Firm Performance as dependent variable

The extant relevant research on ambidexterity has targeted firm performance through different constructs of measurement. These constructs can be divided in objective performance and perceptual performance. Perceptual performance can be subdivided into relative performance (benchmarked with competitors) or absolute performance (financial measurement) (Junni et al., 2013).

(23)

Due to mixed results of research on the relationship of ambidexterity and firm performance, Junni, Sarala, Taras and Tarba (2013) have constructed a meta-analysis that systematically examines the extant literature to find an explanation why results have different outcomes. They have found that the relationships are moderated by contextual factors and methodological choices. The relationship between ambidexterity and firm performance was highly positive in non-manufacturing industries and at higher level of analysis (I.e. firm level vs. individual level). Also, the use of perceptual measurement in a cross-sectional, multi-method design study was more likely to show positive result (Junni et al., 2013). It is recommended by Raisch and Birkinshaw (2008) and Junni et al. (2013) to use multi-dimensional indicators to measure firm performance (e.g. ROI, sales growth, profit growth, market share) and in a longitudinal research (Junni et al., 2013; Raisch & Birkinshaw, 2008). This is congruent with March’s (1991) motivation that exploratory behavior is only measurable in the long run (March, 1991). Junni et al. (2013) also argue that firm performance is best measured through a combination of absolute performance, relative performance and perceptual performance. Apart from the moderating effects shown by Junni et al. (2013), they too have found an overall significant positive effect of ambidexterity on firm performance, as did most former research. (Gibson & Birkinshaw, 2004; Junni et al., 2013; Lubatkin et al., 2006)

2.3.2. Firm Performance measures

In effort to address the challenge of dispersed measurement methods in their research, He et al. (2004) argue that managers should consider looking at performance in the extension of exploration and exploitation. They say: “While existing innovation management practices have been largely founded on established typologies with corresponding resource allocation and performance benchmark metrics (e.g., percentage allocation of R&D expenditure into basic versus applied research or product versus process innovation), senior managers may need to consider introducing new metrics to prioritize resource allocation and benchmark performance along the explorative versus exploitative innovation dimensions” (He & Wong, 2004, p. 492). He et al. (2004) thus suggest looking at the desired achievement of the investment, instead of just the relative spend of the whole. It is not important that 10% of the R&D goes to the Marketing department, it is important to measure how it is divided between

(24)

the exploitation and exploration purposes as well as the subsequent performance outcomes. Performance measures, according to He et al. (2004), should therefore include a statement of the separation between the exploitation and exploration dimensions (He & Wong, 2004).

These dispersed ways of measuring performance have been examined by Junni et al. (2013) in their meta-analysis on the relationship between organizational ambidexterity and performance. Their research showed no significant evidence for the moderating effects of objective performance measures related to organizational ambidexterity and firm performance, unless performance is split into profit and growth. Separation between the two shows significant positive effects of respectively exploitation efforts to profit, and exploration efforts to growth (Junni et al., 2013). This finding explains that it is important to identify the right performance measures, because it can influence the outcome of an analysis if an effect is only measured in profit or only in growth.

2.4. Research Gaps and Hypotheses

2.4.1. Research Question

The extant research on how data analytics capabilities can increase the business performance of an ambidextrous organization is limited. Given the lack of the academic research in this topic, this thesis is dedicated to bridge that gap by researching the performance results of combining data analytics capabilities with an innovation strategy. This will ultimately help organizations to make the right decisions when thriving for a data-driven culture. The research question this thesis will continuously try to answer is:

To what extent does the development of an organization’s data analytics capabilities have a strengthening effect on the relationship between ambidexterity and firm performance?

2.4.2. Hypotheses

According to Gupta’s et al. research (2016), data analytics capabilities show a positive direct effect on firm performance (M. Gupta & George, 2016). Because the research includes explaining the role data

(25)

analytics plays in addition to an ambidextrous innovation strategy, it is reasonable to also test the direct effect of data analytics capabilities on firm performance and compare it with the effect in conjunction with ambidexterity. Since ambidexterity is the foundation of innovation, it is assumed that this will ultimately strengthen the relationship. It is therefore hypothesized that:

H1: The indirect effect of data analytics capabilities on firm performance through ambidexterity

is stronger than the direct effect of data analytics capabilities on firm performance.

To subsequently test for a significant moderating effect of data analytics capabilities (moderating variable) on the relationship between ambidexterity (independent variable) and firm performance (dependent variable), the following hypotheses are proposed:

H2A: There is a significant positive moderating effect of data analytics capabilities on the

ambidexterity – firm performance relationship.

H2B: The strength of the moderating effect is dependent on the level of data analytics

capabilities.

Shuradze et al. (2016) suggest that marketing is most likely to benefit the most from data analytics capabilities due to personalization of marketing. The obtained insights in customer behavior is expected to lead to better matching commercial offerings, realizing more turnover in the existing business. This implies a stronger relationship between data analytics capabilities and exploitation. (Shuradze & Wagner, 2016).

This proposition is contradicted by Wan et al. (2017) in their article, stating that the relationship is stronger between performance and exploration. Supposedly this has to do with the risk-averse stance that exploitative-minded organizations tend to adopt, contrary to the adoption of radical ideas to explore novel areas of growth where explorative-minded organizations are continuously looking for (Wan et al., 2017). Since neither of them have performed an actual research to find out, this research will clarify on

(26)

this topic. Because data is mostly collected, and therefore expected to contribute, in familiar areas rather than unfamiliar area’s it is hypothesized that:

H3: Data analytics capabilities have a stronger moderating effect on the

(27)

3. Research Design

3.1. Conceptual Model

This research is dedicated to finding the contribution of data analytics capabilities to the ambidexterity- firm performance relationship. It is designed to reveal if data analytics capabilities have a significant interaction effect in the model. Because the independent variable (ambidexterity) has a causal relationship with the dependent variable (firm performance) without the presence of a third variable (data analytics capabilities) the conceptual model places data analytics capabilities as a moderating variable. Data analytics capabilities are not built and developed by ambidexterity, which excludes a mediation conceptual model. The identified hypotheses have been aggregated in a conceptual model accordingly that is found in figure 1.

Figure 1: Conceptual model

3.2. Variables and Measurement

In the research’s conceptual model three main variables are defined: ambidexterity, data analytics capabilities and firm performance. The goal of this research is to explore the effect of data analytics capabilities on the ambidexterity - firm performance relationship.

This research will dig into the important aspects of building the desired capabilities, offering practical relevance to organizations that currently are – or will be – performing data analytics. To provide this

(28)

relevance to organizations the level of analysis throughout the research will be the organization, and the measurement of the variables will be done through a survey with Likert scale (1 – 5).

3.2.1. Data Analytics Capabilities

Since Gupta et al. (2016) have performed prior research on building data analytics capabilities, this thesis uses their scale as a basis (M. Gupta & George, 2016). Some modifications have been made in order to level the number of questions per variable, and to follow the latest developments in the field of data analytics. After interviewing some experts from the field of data analytics and in accordance to KPMG’s report (2018) on data trust, some questions have been added to the survey (Erwin, 2018).

3.2.2. Ambidexterity

To describe ambidexterity, scholars have defined the concepts of exploitation and exploration in several different ways in their researches. It has – among others – been referred to as the trade-off between: closed loop learning and mutual learning (March, 1991) evolutionary change and revolutionary change (Tushman & O’Reilly, 1996), process management and responsiveness to market (Benner & Tushman, 2003), alignment and adaptability (Gibson & Birkinshaw, 2004), efficiency and flexibility (Ebben & Johnson, 2005), and defending and prospecting (Auh & Menguc, 2005), and doubtless some others.

Gibson et al.’s (2004) survey design concerning alignment and adaptability turned out not to measure in accordance to this research. They concentrate their research on the architecture of management systems, and not so much on the innovation part of the organization (Gibson & Birkinshaw, 2004). The research of Ebben and Johnson (2005) is another example of different measures than this research intends to adopt. They aim to measure the organization’s flexibility and efficiency strategy and have designed their survey based on the manufacturing industry (Ebben & Johnson, 2005).

The independent variable ambidexterity is measured through the exploitative and explorative behavior of organizations. To measure ambidexterity as an independent variable, the research lends its questions from the research of Lubatkin et al. (2006). Where the surveys of He et al. (2004) and Kyriakopoulosa

(29)

et al (2004) were found to be too generic and only targeting temporary innovation projects, Lubatkin et al. (2006) was found to fit best to this research (He & Wong, 2004; Kyriakopoulos & Moorman, 2004; Lubatkin et al., 2006). Their research has been highly cited and the survey has been re-used multiple times in the ambidexterity domain. Most importantly, this survey will be reused because it focuses explicitly on the marketing domain of ambidexterity. It therefore makes sense to reuse this survey over other the surveys.

3.2.3. Firm performance

Gibson et al. (2004) have used perceptual absolute measurements of a business unit (Gibson & Birkinshaw, 2004), while Ebben et al. (2005) have used the objective measurements of metrics like: return on investment capital, return on assets and return on equity (Ebben & Johnson, 2005; Gibson & Birkinshaw, 2004). Auh et al. (2005) have used efficiency and effectiveness as variables to measure firm performance (Auh & Menguc, 2005), and Lubatkin et al. (2006) asked CEO’s for the relative profitability and growth performance, compared to their industry’s biggest competitors (Auh & Menguc, 2005; Lubatkin et al., 2006).

The dependent variable firm performance will be measured through absolute performance and relative performance as recommended by Junni et al. (2013) (Junni et al., 2013). The scale of Gibson et al. (2004) is used for the absolute performance questions and for the relative performance measurement some questions of Gupta et al. (2016) were added to the survey (Gibson & Birkinshaw, 2004; M. Gupta & George, 2016).

3.2.4. Control variables

The research holds four control variables in total, namely: age of the organization, the number of employees of the organization, the function level of the respondent and the industry where the organization is active in.

(30)

3.2.5. Constructs of variables

The structure of this research’s survey is built up in different layers called ‘constructs’. These constructs are formed of bundles from underlying variables. The answers respondents give to the questions are aggregated to form the variables: exploitation, exploration, data, technology, basic resources, technology skills, managerial skills, data-driven culture, organizational learning and performance. Working with multiple layers enables a deeper analysis of what exactly produces a given result. To provide some clarification on the design of these scales, the structure of ambidexterity and the structure of data analytics capabilities are visually represented in appendix 1 and 2.

(31)

4. Data and Methodology

4.1. Data collection

In the length of this research there is only one planned data collection point, making the horizon is cross-sectional of nature. In order to collect this primary data, the thesis makes use of the survey method. Considering the deductive approach of this research, a survey is well suited to test the assumptions that arise from the theoretical framework. Using the survey method is consistent with prior researches, enabling prior validated survey constructs to be reused.

The questionnaire used by Gupta et al. (2016) and Shuradze et al. (2016) have shown a good fit with explaining the research question of this research, are derived from prior validated research and were tested high for reliability (M. Gupta & George, 2016; Shuradze & Wagner, 2016). Aiming to extend the current field of knowledge in a similar format, the survey method enables easier comparison of the results to prior researches. Adding to that, the survey method lends itself for an effective distribution in comparison to other methods like interviews, case study or experiment. Because the survey is the only data collection strategy, the overall research design is a quantitative mono-method research2_.

4.2. Population

The defined population consists of individuals that have experience in the data and analytics environment of their organization. Since the population is quite large the decision was made to take a sample from the population. A potential participant to the survey could be a data analyst, a marketing manager, an online marketer or the head of Digital. The population is not limited to the Netherlands, because data analytics is performed globally. The sampling, however, will most likely consist of people that come from the Netherlands since the samples is taken from the Netherlands.

2_{The interviews that took place with field experts were to discuss the design of the questionnaire and not to collect}

(32)

4.3. Survey distribution

For the distribution of the survey this research has used the services of a market research agency that has spread the questionnaire among their panel members. The panel members that fit the description of the population are selected as a sample on the basis of their background and expertise known at the agency. This information is checked by the agency every two months to make sure the information is up to date when selecting the sample. The distribution of this questionnaire was issued for the sample-criteria: age (18-65), having a paid job (YES) and the department in the organization the participant is responsible for (MARKETING).

In return, for filling in these questionnaires, respondents receive credits that they can exchange for a payout. Because there is a financial incentive involved for the participants who completed the survey (n = 320), the results of the survey have been thoroughly examined for possible corruption of the data. A screen-out question3_{has also been added to the questionnaire to filter the respondents from the sample}

that have no knowledge of the data analytics activities or the data analytics capabilities of their organization.

4.4. Raw data processing

All questions (excluding control variables) have been answered through a Likert scale of 1 (strongly agree) to 5 (strongly disagree). A copy of the given questionnaire is attached as appendix 3.

Before analyzing the collected data there was some preparation needed in order to align the data. In the survey sample, there was a large amount of respondents (173 completes) that finished the questionnaire in under four minutes. Given the survey’s 58 questions to answer these responses were questionable and to secure the quality of the results, a test was separately initiated with some selected respondents in the author’s network. The results from this subsample indicated that the fastest respondent needed at least 4 minutes and 34 seconds to fill in the questionnaire. Taking into account that the respondents of the

3_{Do you have knowledge of, or insights in, your organization’s capabilities to analyze marketing-data (e.g.}

regarding; data collection, data sources, technologies used for analytics, managerial skills, technical skills and the data culture)?

(33)

agency’s sample are experienced survey participants, cut-off point was set to four minutes. All of the 173 respondents that finished within four minutes were excluded from the results, and the analyses continue with the remaining 147 completed surveys. Summary of frequencies is reported in table 2.

Table 2: Sample results

Agency's distribution

Total respondents 391 Completed surveys 320 Completed surveys (≥ 4 min.) 147

To account for errors due to missing data, the dataset has been subjected to a visual check in Excel to see where data is missing. The visual check shows that the unanswered questions throughout the dataset don’t show a pattern. Thus missing data are considered random and pose no threat to bias in the analyses. In the dataset (n=147) one question [This organization does a good job satisfying our customers] was reported missing 3 answers. All other questions that had missing values only show a maximum of two missing answers per question.

4.5. Data preparation in SPSS

4.5.1. Variable settings

To get the data prepared for analysis some steps needed to be taken. First, the measures of the variables were set to the correct format. The control variables CV3 and CV4 are set to nominal, the rest of the variables was set to scale. Since the format of the questionnaire was designed in Excel prior to importing it in Qualtrics, the questions were already given an abbreviation as name as can be seen in appendix 3. Hence, no action was needed to rename them in SPSS. Subsequently, the question labels were added, the values (1 strongly agree – 5 strongly disagree) were set to match the measurement scale and the missing values were set to display the value of 999 to make sure they are visible in the analyses.

4.5.2. Outliers

The sample (n=147) has been checked for outliers that would possibly corrupt the sample in further analyses. The scores of the variables have been standardized to Z-scores to identify outliers (variables

(34)

with Z > 3). On the basis of these scores SPSS reports a total of 6 outliers in the dataset having a Z-score of > 3. The frequency tables show that these data points are outliers and a visual check of the histograms and Q-Q plots also confirms that these data points are outliers to the sample. The 6 outliers in total come from 4 respondents of the survey that were then removed from the dataset.

4.5.3. Scale reliability

Because the individual variables are grouped into scales, the scale reliability is checked prior to the check for normality. The variables in this research are only analyzed jointly as a scale, for individually analyzing the results will not yield any insights. Before aggregating the variables, and doing incremental analyses this check for internal consistency is necessary. The Cronbach’s Alpha scale reliability check has been computed for all the survey’s scales. The results can be found in table 3.

For exploitation, technical skills, data-driven culture and organizational learning the scale reliability could be improved by removing a variable, but none of the variables that could be improved would meet the minimum requirement of the absolute increase for deleting one of the variables (α ≥ ,05). Therefore the scales remain unaltered. Overall, reliability is marked critical for organizational learning (α = ,605). Unfortunately the reliability will also not increase substantially by removing a variable and therefore analyses proceed with these reliability results.

Table 3: Scale reliability check

Scales Variables Cronbach's Alpha α if item deleted

Ambidexterity Exploitation ,681* ,730 Exploration ,774

Assets Data ,760

Technology ,750 Basic Resources ,664

Skills Technical Skills ,724* ,770 Managerial Skills ,760

Knowledge Data-Driven Culture ,685* ,702 Organizational Learning ,605* ,610 Performance Performance ,779

(35)

The reliability test also reported sufficient support for aggregating the scales of ambidexterity (α = ,805), data analytics capabilities (α = ,932) and firm performance (α = ,731).

4.5.4. Aggregating scales

The variables that form the first-order constructs are grouped by computing the mean. The mean was computed through SPSS and the variables were renamed (EXPLOIT_MEAN, EXPLOR_MEAN, ADAT_MEAN, ATEC_MEAN, ABR_MEAN, STSK_MEAN, SMSK_MEAN, KDDC_MEAN, KORL_MEAN and PERF_MEAN) to enlist them in the dataset. For ambidexterity and data analytics capabilities the variables were aggregated by computing the mean of all variables mentioned in respectively appendix 1 and appendix 2, and recoded the variables (AMB_MEAN, DAC_MEAN). Then, the labels were added to the variables, the values have been filled, and the measures have been set to ‘scale’. Aggregating these variables into scales also takes care of the missing values.

4.5.5. Normality

By reporting the descriptive statistics on skewness and kurtosis the sample was checked for normality of the distribution. The results for this test are reported in table 4. The test reported only statistics that are within boundaries (from -1 to +1) and point to a normal distribution. Furthermore, because the observations meet the limited threshold of n ≥ 30, the Central Limit Theorem applies. This allows to assume that the data are distributed normally in a given sample.

Table 4: Sample’s descriptive statistics for normality

N Skewness Kurtosis

Variables Statistic Statistic Std. Error Statistic Std. Error Exploitation 143 ,144 ,203 ,218 ,403 Exploration 143 ,256 ,203 -,441 ,403 Data 143 ,222 ,203 ,149 ,403 Technology 143 ,305 ,203 -,165 ,403 Basic Resources 143 ,283 ,203 -,234 ,403 Technical Skills 143 ,149 ,203 -,105 ,403 Managerial Skills 143 ,298 ,203 ,443 ,403 Data-Driven Culture 143 ,388 ,203 ,033 ,403 Organizational Learning 143 ,256 ,203 ,227 ,403 Performance 143 ,551 ,203 ,245 ,403

(36)

4.5.6. Final sample description

After validating the scale reliability, the frequencies for the control variables were reported to check the background of the respondents. The numerical control variables ‘company age’ and ‘employees’ have been recoded into categories using SPSS to enable easier comparison of the categories’ size. Tables 5 to 8 have been issued to present the sample.

A quick inspection of the results tells us that the respondents mostly stem from companies harnessing 1 – 50 employees, certain industries are underrepresented in the sample (Agricultural, Fast moving consumer goods and Government), and that a relatively high number of respondents are active in an industry that the list doesn’t show. Overall there are no remarkable results that would cause bias to the analysis. The main thing to consider is the limited generalizability of the results on the basis of industry.

Table 5: Sample frequencies - age Company age Frequency

1 - 5 years 19 6 - 10 years 22 11 - 20 years 40 21 - 50 years 41 > 50 years 20

Table 6: Samplefrequencies - employees Employees Frequency 1 - 50 73 51 - 100 12 101 - 500 28 501 - 2500 13 > 2500 16

Table 7: Sample frequencies - Functional layer

Table 8: Sample frequencies - industries Industry Frequency

Agricultural 1

Automotive 4

Business services 14 Fast moving consumer goods 1 Financial services 13

Government 1

Healthcare 17 Online and digital services 10

Retail 17

Telecommunications 4 Transport & logistics 9

Wholesale 8

Other 44

Functional layer Frequency

Operational level 48 Middle management level 46 Senior management level 20 C-level management 29

(37)

5. Results

5.1. Correlation

Organizational learning is discussed to be an antecedent of exploratory behavior (Levinthal & March, 1993; March, 1991; Shuradze & Wagner, 2016). To test if organizational learning influences the relationship by extreme correlations as part of the variable ‘knowledge’, the second order construct variables were tested for correlations and reported in table 9. Although all correlations with exploration are significant, the results show that knowledge (the aggregation of organizational learning and data-driven culture) does not account for the highest correlation with exploration. Instead, it is skills that show the highest significant positive correlation. This excludes risk for bias in the analyses.

Table 9: Correlation table – exploration & second order constructs

To see which variables correlate, a Pearson correlation test was performed. This test produced the results displayed in table 10 that is presented below. From the table we can see that only the scale variables show significant correlation effects (P < 0,01). No significant results were found in the relationship between the scale variables and the control variables. Therefore we failed to reject the null hypothesis and assume that the relation between the scale variables and control variables do not significantly differ from 0. In other words, the variables ambidexterity, data analytics capabilities and firm performance are not influenced by an organization’s age, size and industry, or in which functional layer of the organization the participants are working.

1 2 3 4

1. Exploration 1 2. Assets ,483** ₁

3. Skills ,564** _,815** ₁

4. Knowledge ,558** _,747** _,768** ₁

(38)

Table 10: Correlation matrix Variables Mean SD 1 2 3 4 5 6 7 1. Age of organization 28,670 31,410 _- 2. Number of employees 5624,04 51347,3 _,128 _- 3. Industry - - _,015 _,089 _- 4. Functional layer - - _-,090 _-,095 _-,118 _- 5. Ambidexterity 2,285 0,478 _-,102 _-,106 _-,067 _,036 _(0,805)

6. Data Analytics Cap. 2,536 0,504 _-,110 _-,045 _-,032 _,099 _,622** _(,932)

7. Firm Performance 2,252 0,530 _-,037 _,050 _,056 _-,055 _,606** _,606**_(,779)

**. Correlation is significant at the 0.01 level (2-tailed).

5.2. Findings

5.2.1. Effect strengths

To test if data analytics capabilities have a direct effect on firm performance, and if this effect is stronger than the moderating effect through ambidexterity, a regression analysis is performed. To enable testing of the interaction effect, the product was computed for the variables data analytics capabilities and exploitation(INT_DAC*EXPLOIT). The same was done for data analytics capabilities and exploration (INT_DAC*EXPLOR). To prevent multicollinearity among the variables and to offer better insights, the product was computed with exploitation and exploration, and not with ambidexterity.

The regression results (reported in table 11) show that both ambidexterity and data analytics capabilities have a significant positive impact on firm performance (p = <,001), where ambidexterity has a slightly bigger impact (ß = ,374) than data analytics capabilities (ß = ,373).

When the interaction variables are added in the second model the direct effect of data analytics capabilities is no longer significant (p = ,410), but the interaction models report a significant positive impact of data analytics capabilities through exploitation (p = ,046) and exploration (p = ,019). The model reports an r2_{of 0,475 which means that 47,5% of the variance in firm performance can be}

explained by the regression analysis. The reported levels of multicollinearity are also within threshold (Tolerance > 0,20 / VIF < 5). This indicated that the variables do not correlate among each other, and that if one independent variable goes up, other independent variables do not go up significantly.

(39)

The indirect effect of data analytics capabilities and ambidexterity (exploitation and exploration) accounts for bigger impact on firm performance, than the direct effect of data analytics capabilities. These result confirm H1 and it can be assumed that an organization yields more firm performance from

data analytics capabilities, when an ambidextrous innovation strategy is adopted.

Table 11: Regression analysis – interaction effect

Model

Unstandardized

Coefficients Standardized Coefficients

t Sig.

Collinearity Statistics B Std. Error Beta Tolerance VIF 1 (Constant) _,308 _,184 _1,675 _,096

Ambidexterity _,414 _,089 _,374 _4,680 _,000 _,614 _1,630 Data analytics capabilities _,393 _,084 _,373 _4,677 _,000 _,614 _1,630 2 (Constant) _1,832 _,658 _2,785 _,006

Ambidexterity _-,268 _,296 _-,242 _-,905 _,367 _,053 _18,786 Data analytics capabilities _-,224 _,271 _-,213 _-,827 _,410 _,057 _17,442 INT_DACxEXPLOIT _,129 _,064 _,496 _2,011 _,046 _,063 _15,974 INT_DACxEXPLOR _,140 _,059 _,650 _2,382 _,019 _,051 _19,567 a. Dependent Variable: Firm Performance

5.2.2. Moderating effect

In order to perform the analyses to test for a moderation effect of the sample, an additional functionality called PROCESS was added to SPSS. PROCESS is a tool that was designed to fit into SPSS providing pre-defined settings for the analysis of a wide variety of mediation and moderation based models (Hayes, 2012). In the case of this research we will use model number 1, for testing moderation effect. The option for mean centering the variables was selected before running the analysis. The moderation analysis by PROCESS reported a significant positive interaction effect of 0,2731 (p = ,0159) and r2 _{= ,475 (p < ,001)}

meaning that 47,5% of the variance in firm performance is explained by the model.

The results were also plotted in a line graph to visually represent the interaction between the variables. As can be seen in the unparalleled lines in the graph from figure 2, there is a positive interaction effect. The coefficient of the model was reported being positive (ß = 2,2113, p < ,001), indicating that the influence of data analytics capabilities on the ambidexterity - firm performance relationship is

Innovation strategy and firm performance - the role of data analytics explained