1
Digital Marketing Challenges:
A taxonomical approach
Stijn Abbink
Supervised by prof. dr. P.S.H. Leeflang
2
Digital Marketing Challenges
A Taxonomical approach
Research Master Thesis
Stijn Abbink
University of Groningen
Prof. dr. P.S.H. Leeflang (1
stsupervisor)
Prof. dr. P.C. Verhoef (2
ndsupervisor)
3
ABSTRACT
With the explosion of digital tools for marketing, some firms have a difficult time to adapt. In this study we investigate the most pressing challenges that firms are facing in light of these digital developments. The challenges that are most important, are in the domains of (1) online interactions, (2) organizational challenges and (3) challenges associated with data usage. We make a taxonomy of challenges which co-occur and describe the firms facing these configurations of challenges. Our analysis shows that there are four broad taxonomical groups, the troubled analyzers, the traditionalists, the immovable giants and the organizationally challenged, each with a unique set of challenges and characteristics. We end with establishing possible sources of these challenges in the mismatch between firms’ resources and its environment.
4
ACKNOWLEDGEMENTS
5
INTRODUCTION
The complexity of the marketplace is increasing and the gap between the marketplace and the capabilities of firms to deal with its complexity is widening (Day 2011). Organizations are facing severe challenges in dealing with market complexity. Most of these challenges are specifically related to digital marketing developments. Developments in the digital realm are causing market to fragment, fueled by the ongoing revolution in the interconnectedness of consumers and businesses (Day 2011). As of yet there is no sign that the end of this revolution is in sight, implying that the challenges that firms are facing as a result of this will remain to be influential (Hagel, Brown and Davidson 2009).
A study by the IBM institute for business value (2011) gives an illustration of the capabilities gap. It establishes the most pressing challenges that CMO’s are facing. The most important challenges all seem to be emanating from the increasing market complexity. The challenges identified include the dramatic increase in consumer and market data available to firms, also known as ‘big data’, and the exponential increase in word-of-mouth and customer-generated information about firms and their respective products.
While all firms are confronted with such changes in the marketplace, not all are affected equally. Some will stumble upon unresolvable challenges while other firms may see opportunities and use them to create competitive advantages over their competitors. This view on such differences between firms, follows from one of the classic approaches to strategy. As organizations are faced with their environment, those that are capable to match their resources to the environment may create sustainable competitive advantages over other firms (Andrews 1971; Wernerfelt 2006). Since the resources of firms to deal with the challenges are heterogeneous and only limitedly mobile between firms (Barney 1991), large differences in the nature and extent of experienced challenges are expected.
6
may be related to challenges of measuring online marketing effectiveness, which as of yet have not been conceptually linked together. In this article, we will not focus on the selective issues, but attempt to find a set of commonly occurring configurations of challenges. In this approach, we unravel which challenges often occur together and attempt to find the relevant resources and external factors that contribute to these challenges.
To make a contribution to unraveling the links between several important challenges brought on by the evolving markets, we will use a taxonomical approach. Taxonomical approaches are frequently used in business analysis to study overall patterns of phenomena instead of the study of selective issues (Homburg, Jensen and Krohmer 2008). This methodology has been used for example in finding configurations of how marketing contributes to business strategy (Slater and Olson 2001) and the configurations of the interface between marketing and sales (Homburg, Jensen and Krohmer 2008).
Our specific research goals are as follows:
• To empirically identify common configurations of challenges and develop a taxonomy • To make a description of taxonomical configurations in terms of their associated
externalities and resources
By developing a taxonomy of the most important marketing challenges caused by the digital environmental revolution, we endeavor to shed some light on how configurations of challenges occur together and whether specific environmental and resource characteristics can be identified to be associated with these specific challenge configurations. With this we will gain insights into how different digital challenges co-occur, what the potential sets of mismatches between the environment and the available resources are and with this we endeavor to provide a solid basis for further research into the subject.
7
a qualitative study to identify the most important digital marketing challenges that firms are facing. To do so, interviews were performed with leading marketing scientists, business experts and leading firms related to the McKinsey Company. From these interviews the challenges associated with the increasing complexity of the marketplace were identified. Subsequently, a questionnaire was issued among 777 CMO’s around the globe to help establish the relative importance of the challenges.
Leeflang et al. (working paper) identified the 10 most important digital marketing challenges. From the challenges identified, six were very prevalent among firms (>23% of the firms identified these challenges as being one of their most important challenges). These challenges, sorted by their relevant domain are summarized in table 1.
8
CONCEPTUAL BACKGROUND
The challenges that were deemed most important (see table 1) by the CMO’s in our sample of fall into three broad categories; (1) Challenges stemming from online interactions, (2) Challenges associated with data collection, analysis and use, and (3) organizational challenges. These three categories each have a set of unique challenges associated with them. Each challenge belonging to a specific category (table 1) will be discussed first; next the relevant external environmental characteristics and internal resources will be discussed shortly.
TABLE 1: IMPORTANT DIGITAL MARKETING CHALLENGES
Percentage of firms experiencing problem
Online Interaction
- Managing brand health and reputation is more challenging in a marketing environment where social media plays an important role
52.8 %
Organizational challenges
- The pervasiveness of marketing activities within companies is causing organizational challenges (e.g., role ambiguity, unclear accountability and incentives).
23.8 %
- The increasing prevalence of digital tools and technologies is threatening existing business models
24.2 %
Data Challenges
- Generating and leveraging rich and actionable customer insights is becoming a necessity to compete
55.5 % - Assessing the effectiveness of digital marketing is difficult, since online and
traditional metrics are not readily comparable
63.2 % - Marketing and related departments are facing a significant talent gap in
analytical capabilities
31.8 %
Less important challenges
- An overreliance on data and 'hard facts' is stifling creativity and breakthrough innovation.
11,5 % - Online price comparison tools are impeding companies’ ability to set optimal
prices.
11,7 % - Service automation and efforts to migrate customer interactions online are
creating customer dissatisfaction and destroying value.
10,0 % - Too often, digital marketing efforts are targeting only younger customer
segments.
9
Online Interaction
The digital revolution that has lead marketplaces to be revolutionized and to become increasingly complex, has also transformed how individual consumers interact with one another. Through the internet, people increasingly have the means to be connected to one another, regardless of the time or place. The increased connectedness of the world through internet has given rise to a complex and vast network of ways to interact and share information (Kozinets et al. 2010). The end of these developments is not in sight and more and more people are swayed to join in on the revolution (Edelman 2010). Evidence for this lies in the explosion of online communications. As an illustration, Facebook had, at December 31st 2012, over 1.06 billion active monthly users and still anticipated large growth percentages (Facebook.com annual report 2012). In addition to Facebook, individuals have an enormous range of digital means to connect and exchange ideas with other individuals. These include other general social interaction networks such as twitter, but also more purposeful media such as brand and consumer communities.
Managing brand health and reputation in a marketing environment where social media plays an important role
The implications of these developments for firms are enormous. These implications are present especially in the domain of branding, as social media has the potential, unlike any offline medium, to make the identity of brands transparent to existing and potential customers (Naylor, Lamberton and West 2012). It is estimated that over half of all social media users are using these online communication means to keeping track of brands (Williamson 2011). Firms can actively manage their own brand communications by establishing brand fan pages where they can post relevant brand information and other brand related promotions (de Vries, Gensler and Leeflang 2012).
10
Kozinets et al. (2010) have shown that online word-of-mouth does not necessarily amplify or increase marketing messages, but that the messages themselves and even the meaning of the messages that firms send out, are systematically altered by the online community beyond the control of the firm. A complicating factor is that existing word-of-mouth theories apply only limitedly to these kinds of customer to customer interactions, as face-to-face interactions draw on a wealth of social and contextual cues (Brown, Broderick and Lee 2007).
While it is possible to direct online word-of-mouth to create value for the firm (Schau, Muniz and Arnold 2009), action taken by firms to influence it are not always successful. Lack of control, its unpredictability and its potential to backfire create serious challenges for firms. 52,8% of our sample indicated that managing their brand health is one of their most pressing challenges in an environment where social media plays such a large role.
Organizational Challenges
When the technological environment is in turmoil by rapid innovations and changes, firms often have trouble keeping up (Day 2011). Besides the specific challenges related to rapidly changing technology, firms must be able to successfully integrate the solutions to these challenges within their organization. The associated difficulties with adapting the organizational processes to environmental changes has been a topic in research for decades (e.g. Miles and Snow 1978; Morgan 1997) and their related challenges seem to ever present and relevant in today’s organizations and environments. Two of such challenges were among the most important challenges associated with digital marketing developments as identified in our sample. These will be discussed in the section below.
The pervasiveness of marketing activities within companies is causing organizational challenges
11
showed too little accountability for their expenditures and too few innovations to be perceived as contributing significantly to financial performance (Verhoef and Leeflang 2009). As a result, marketing departments lost their influence and dispersed throughout firms.
While marketing is striding to reinvent and prove itself by providing an accountable way of performing their tasks (Verhoef and Leeflang 2009), the dispersion of the function has led to difficulty in the light of the new external digital developments. As the influence of marketing within the firm increases due to external contingencies, such as unpredictable market related changes (Homburg, Workman and Krohmer 1999), firms are left with the challenging situation that the marketing function has dispersed throughout the organization. The unpredictable changes in the marketplace fueled by new digital marketing developments, have led firms to experience role ambiguity and unclear accountability. In essence, it seems that many firms (23,8% of our sample, see table 2) are experiencing challenges due to a disintegration of the marketing function now that their external environment is in turmoil as a result of the developments in digital marketing.
The increasing prevalence of digital tools and technologies is threatening existing business models
Business models are (usually typical) organizational behaviors that are used to provide value for a firm (Baden-Fuller and Morgan 2010). Business models and the way that value is created by firms are under strong influence of changes in technology and other external contingencies which can make them go into decline in an alarmingly short amount of time (Hill and Rothaermel 2003).
12
markets are making services obsolete by being substituted by others. An example of this is the Royal Dutch mail company (PostNL) who are facing yearly decreasing returns as a result of e-mail and other digital communication possibilities.
Data Usage
Every two days, more data is generated in every two days than has been generated from the dawn of civilization up to 2003 (IBM 2011). With this, also the data that is available to firms has exploded over the past decade. As data is capable of providing clear competitive advantages to those who have the ability to use it (Davenport 2006), it provides valuable opportunities. And in addition, not buying in on this huge opportunity may make a firm lag behind its more pro-active competitors. To illustrate the importance of data usage, LaValle et al. (2011) find that top performing organizations make five times more use of their insights gained from data than their more poorly performing counterparts.
Unfortunately, the gathering and leveraging of data is considered to be a daunting task. Three of the six most important challenges that marketing executives are facing at the moment (see table 1) are emanating from the abundance of data available to firms to collect and analyze. We continue by discussing each of these separately.
Generating and leveraging rich and actionable customer insights is becoming a necessity to compete
Where in the past, marketing tactics mainly consisted of offering, branding and promoting a single and identical product to the masses; this strategy has been replaced by one where the customer is the center of attention and not the offered product or range of products (Rust, Moorman & Bhalla 2010). The new customer centered strategy advocates the establishment of a long term relationship between the customer and the firm.
13
customers’ needs in order not to lose them to the competition. Secondly, stemming from the increasing heterogeneity in customer demands, traditional segmentation is becoming less and less effective because the market has split up in a large number of micro-segments (Day 2011). In order to be effective in identifying profitable business endeavors, firms need to be increasingly aware of the needs and desires of its customers.
Companies who are successful in gaining insights about their customers from their data seem to benefit in terms of perceptual and objective firm performance (Reinartz, Krafft and Hoyer 2004; LaValle 2011), making the implications of extracting insights about consumers a competitive force to be reckoned with. Because of this, companies who are not able to leverage insights due to lack of data quality, lack of analytical capabilities or any other reason will be increasingly less able to compete (Leeflang and Verhoef 2009). Despite the demonstrated importance of gaining customer insights, over 55% of the firms in our sample are struggling to generate and leverage rich insights from their data.
Assessing the effectiveness of digital marketing is difficult, since online and traditional metrics are not readily comparable
One of the fundamental activities of any marketing department is identifying the best metrics to measure their performance (Ambler 2003). Effective firms use metrics-based quantitative research to assess the contribution of marketing activities to the performance of the firm (Lehmann 2004; Srinivasan, VanHuele and Pauwels 2010). The customer-centered approach, which has replaced the product centered approach, requires different ways to analyze marketing effectiveness, shifting the balance from overall measures of sales to individual measures such as satisfaction (Rust, Moorman and Bhalla 2010).
14
For instance, social media outcomes are still often being measured with offline metrics such as ROI of online expenses and brand awareness (Kumar et al. 2013), whilst the outcomes of such activities may be far more complex than can be measured with these offline metrics. Hoffmann and Fodor (2010) acknowledge these difficulties in stating that social media channels cannot only be measured by classic offline metrics but need, in order to be effective, also be measured in other forms, such as changes customer behaviors that have relevant long term effects on performance.
To make matters even more problematic, researchers and practitioners do not yet seem to be converging into a state of general acceptance of best practice online metrics (Shankar et al. 2012), making the identification of best practices difficult at best.
Due to these difficulties, a staggering 63.2 percent of the CMO’s in our sample name the difficulties with assessing the effectiveness of their digital marketing efforts as one of their most prominent challenges at the moment.
Marketing and related departments are facing a significant talent gap in analytical capabilities
A prerequisite for using digital marketing tools, such as generating customer data, leveraging this data and the development and use of online metrics is having sufficient analytical capability within the firm. It is critical for organizations who want to use collected data to their advantage to have analytically talented employees who understand the marketing department’s requests for analyses and insights regarding the performance of the marketing activities (Verhoef and Lemon 2013).
15
from data, many of these do not have a background in marketing. This is however a critical prerequisite for providing effective insights and analyses to the marketing department (Verhoef and Lemon 2013). Davenport and Patil (2012) illustrate this by their observation that the ideal ‘data scientist’ (as they call it) not only has analytical capabilities but has these capabilities in combination with visual and verbal skills in ‘story-telling’ to get the buy-in from the managers that they are advising.
16
Taxonomies and the resource based-view
The challenges associated with the digital technological development of the marketplace do not occur in isolation of each other. Some occur in one firm but not in another. In addition some challenges seem to be occurring simultaneously more often than others. One of the key goals of this paper is to investigate whether there are sets of related challenges and what constitutes the existence of these sets. We theorize that there may be some underlying mechanism on which we will elaborate first.
As a theoretical framework to base our analysis on, we use the resource based view of the firm. This organizational strategic theory was introduced by Wernerfelt in 1984 and built on the contemporary research of that time. It provides a theory of how firms operate in their environment by matching this environment using their specific set of resources and how this set of resources are the basis of competitive advantages and performance (Barney et al. 2011). While it was originally theorized to be useful in strategy literature, the application of the theory for marketing has been quite fruitful (Wernerfelt 2013).
The resources that are used to operate in the environment are considered very broadly. It is anything that firms can consider to be a strength or weakness, including tangible assets such as machinery and capital but also intangibles such as skills and capabilities (Wernerfelt, 1984).
17
As discussed, the internal capabilities and the external environment should be optimally matched for firms to be successful. Where the two match less well, challenges for the firm arise. While the focus of our research is relatively new, there is a body of literature that deals with matching internal and external environments in more general terms, and may provide insights into the findings of our analysis of digital challenges and a set of internal and environmental factors.
Day (2011) identifies two important reasons why firms’ marketing departments cannot respond well to the changes in their external environment. The first of these are organizational rigidities. There are three types of rigidities; the first of these rigidities is path-dependency and lock-in, where organizations are stuck in the processes that they learned and mastered in the past but which are now becoming increasingly obsolete. The second, very much related rigidity, is inertia and complacency. Here, the organization fails to look forward and experiment with forming new capabilities to addressing environmental challenges and stick to their obsolete capabilities. The final of the three rigidities is named structural insularity. This means that the structure of the organization does not allow for cross functional dialogue and exchange of ideas and information.
The second important reason why firms do not respond well to their environments identified by Day (2011) is that of lagging reaction. The speed by which the organization can identify and act on changes in the environment is key to the effectiveness of it response. Taking too long to notice the environment changing, and adapting the organization accordingly, will mean that the environment has changed so much by the time the organizational changes are implemented, that it will be again obsolete. Next to organizational rigidity, the speed by which changes are implemented is, because of these reasons, very important. This speed is strongly dependent on the decision making processes that allow firms to act on the information they have gathered about the market, as classic managerial decision making is slow and often flawed.
18 Organizational resources
There are a number of organizational resources that are of specific importance when diagnosing a configuration of challenges.
Firm size
The size of a firm is an important factor. Larger firms tend to have a much broader resource base, leading them to be more able to respond to changes in the external environment (Darnall et al. 2010). On the other hand, with rapid developing environments firms may, despite their resources have a difficult time. Smaller firms tend to be more innovative and less challenged with the discussed organizational rigidities that allow them to adjust to their environment quickly. Hence, they are more responsive to emerging digital tensions (Frambach & Schillewaert 2002).
Customer data
The extent to which firms have access to marketing data regarding their customers, is a key resource influencing marketplace performance (Srivastava, Shervani and Fahey 1999). This ability leads data to be a potentially important competitive advantage (Davenport 2006; LaValle et al. 2010). Those that are able to generate and leverage insights about their customers can for instance hugely increase the effectiveness of their segmentation and targeting in an environment where this is becoming increasingly difficult (Day 2011). An important prerequisite for analyzing the data and being able to find valuable insights is the quality of the data. As the market has split into a large number of micro segments, detailed customer data is an important resource to be able to leverage the behavioral detail to target these segments and as a result to remain competitive.
Decision making procedures
19
for the disregard of the insights procured from data. Formal decision making procedures using the insights collected from data can therefore be an important organizational resource.
External environment
In addition to the relevant resources that a firm possesses, the other part of the match is the external environment. The most relevant factors of the environment for the digital marketing tensions are listed below.
Region
The geographical region in which an organization operates might be one of the most important determinants of the external environment. Firstly, in most regions, there are still plenty of opportunities for Internet penetration (Brashear et al. 2009). Developing countries, have notable lower Internet penetration rates (more than double) compared to Europe, Oceania/Australia, and North America (“World Internet Usage Statistics,” 2011). As the digital advance differs between developing countries and developed countries, different resources are needed to address the difficulty at the stage that the development is in.
Secondly, digital marketing effort are finding its way into emerging markets at a rapid speed. For example, Kumar, Sunder and Ramaseshan (2011), find that firms in the Asia-Pacific region are rapidly imitating data driven customer relation management practices from firms in more developed regions such as Europe and the US. As such resources for digital marketing practices are adapted to a very different environment, their effect (of which very little research exists) may be very different than in developed regions.
Business categories
20
leading on the one hand to more internalization of relevant resources for this capability, and on the other a seemingly more receptive environment to this approach (Sun 2006).
The second important dimension of the business category is the buyer. Especially, there is an important difference between business-to-business environments and business-to-consumer environments. B2B markets differ in the number of customers and how the customers react to a firm’s offering (Vargo and Lusch 2011) For example there is a large distinction in how the internet is used to communicate with customers in a B2C setting compared to a B2B setting (Pasuraman and Zinkhan 2002) making marketing practices driven by the digital revolution very different. The resources needed to create the optimal match between these resources and these different external environments may therefore differ greatly. While this is the case, the use of digital tools in B2B environments, is gaining ground because its potential is being increasingly investigated and confirmed (e.g. Brennan and Croft 2012; Voss, Godfrey and Seiders 2010)
Online Sales
Whether or not, and the extent to which firms engage in selling their products online determines their environment. As opposed to the other external environmental factors, firms do not only have to account for the online environment, but can also be active in shaping this environment to their advantage.
21
METHODOLOGY
To arrive at a taxonomical classification of the challenges that firms face as a result of digital tensions we will perform a cluster analysis. More precisely, we will use a mixture model approach to separate the mixture of different problem distributions in the data, which will be explained in more detail later. We start by explaining the method of data collection and provide a description of the sample. Secondly, we discuss the operationalization of the variables that we employ to describe our separate clusters. Finally, we discuss the details of the modeling approach that we employ to arrive at our empirically based conclusions.
Sample and Data Collection
The data that we used has been collected in through an online survey by the McKinsey Company in cooperation with the University of Groningen. The survey was distributed in the end of 2011 and the data collection finished in the beginning of 2012. The sample consisted of a random selection of readers of McKinsey quarterly, a business magazine focusing on business readers, with articles predominately on management and organizational theory. As compensation for participation, responding firms were offered access to a special issue of the magazine.
The sample that we use in this study consists of a total of 632 respondents. To arrive at this sample we had to exclude 145 respondents who did not completely finish the survey making the inclusion of these observations problematic due to missing data. The sample contains observations from a selection of different industries, regions and a primary focus on both business-to-business and business-to-consumer interactions. Tables 2, 3, and 4 provide cross-tabulations of these firm characteristics and the number of associated observations.
TABLE 2: SAMPLE DESCRIPTION - REGION x FOCUS
Region Total Other US and Europe
Business to Consumer 52 157 209
Business to Business 80 343 423
22
TABLE 3: SAMPLE DESCRIPTION - REGION x INDUSTRY
Region Total Other US and Europe
Services 25 100 125
Financial 12 46 58
High Tech/ Telecomm 24 92 116
Manufacturing 22 94 116
Other 49 168 217
Total 132 500 632
TABLE 4: SAMPLE DESCRIPTION - INDUSTRY x FOCUS Industry
Services Financial High Tech/ Telecomm
Manufacturing Other Total Business to Consumer 21 23 26 58 81 209 Business to Business 105 35 90 58 135 423 Total 126 58 116 116 209 632 Variables
The survey was developed to assess, amongst others, dimensions of company and environmental characteristics, the level of customer insights and online business orientation. Furthermore, the most important marketing challenges in a digital era were recorded and each firm indicated whether or not this problem belonged to their most pressing concerns.
23
General introduction to configurative approaches
There are two broad approaches to arriving at a taxonomy using a clustering approach. The two approaches are derived from Ketchen and Shook (1996) who provide an overview of approaches to clustering methods in business research. These authors suggest that in advance of performing any sort of clustering procedure, authors need to be very clear and explicit about the purpose of the analysis, the match between the study and the method of analysis, and the variables included in the analysis. As a guidance tool they describe two broad approaches. The first approach is the deductive approach, where the choice of variables, the number of variables used in the analysis, as well as the expected number of taxonomical groups are strongly based upon and predicted from existing theory. The second approach of Ketchen and Shook (2006) is the inductive approach. Here, the goal is an exploratory classification of observations on the basis of a wide range of variables, increasing the likelihood of finding meaningful classifications in the structure of the data. One of the most important prerequisites here is that the researcher (and the associated literature) has little specific a priori expectations of the nature of results.
Table 5: SUMMARY OF MEASURED VARIABLES Measurement
scale
Description Organizational Resource
variables
Size Ordinal Number of Employees: 0-50 (1), 50-500 (2), 500-10.000 (3) and >500-10.000 (4)
Level of consumer data Ordinal Aggregated sales data (1), Individual sales data (2), Basic customer information (3), Detailed customer information (4)
Decision making reliance Nominal Decision making relies primarily on expertise (0), decision making relies primarily on data and facts (1)
Environmental variables
Region Nominal Europe and US (0), other (1)
Industry Nominal Services (1), Financial (2), Telecom/High tech (3), Manufacturing (4) and Other industries (5)
Revenue from online sales Ordinal Less than 2% (1), Between 2% and 10% (2), Over 10% (3)
Focus Nominal Business-to-business (0), Business-to consumer (1) Social media usage Continuous
(5-point scale)
24
In our research we will use the second approach. As we discussed in the introduction, the limited knowledge on this field, brought by the novel nature of this topic has led us to follow an exploratory approach, in search of a meaningful classification of challenges and firms to further our knowledge regarding the topic.
Model Specification
A common way to arrive at a taxonomy based on data is to assign individual observations to a specific set of observations with similar characteristics using statistical analysis (Homburg, Jensen and Krohmer 2008). There are many approaches of these statistical techniques, which each have their own advantages and disadvantages. The popular standard for years in such endeavors has been classical clustering approaches such as k-means clustering (Magidson and Vermunt 2002) where individual cases are assigned to one absolute class by means of a non-model based approach.
In our research, we use an often used approach to clustering called latent class clustering or a mixture modeling approach. In the mixture modeling approach a statistical model of the population from which the observations are drawn is postulated (Magidson and Vermunt 2002). This population is assumed to consist of several heterogeneous clusters of unknown proportions and properties that are mixed in the data (Leeflang et al. 2000). The goal of the mixture model approach is to un-mix the data in order to be able to identify the set of different clusters that are present in the data. As opposed to other clustering methods, the mixture model approach is model based instead of data based and therefore allows for statistical testing and estimation.
25
This method has several advantages over the classical approach to clustering that we will discuss next on the basis of an article by Magidson and Vermunt (2002). First, instead of providing a discrete cluster allocation for each individual observation, mixture model clustering classified observations as a function of the posterior cluster membership probabilities. This counters the absolute nature of the classical clustering approaches and gives a more realistic representation of cluster membership based on the probability of belonging to a certain cluster. Secondly, in contrast to the classical approach, since the mixture model approach is estimated using a maximum likelihood function, the outcomes provide insight into the optimal number of clusters to be extracted, conditional on the structure of the data. Thirdly, in the classical approaches, the variables that have a larger variance, have a larger impact on the cluster allocation, influencing the solution. A commonly proposed solution of this issue, standardizing the variables, does not entirely solve the problem, since clusters are unknown and within cluster variance cannot be accounted for. In the mixture model approach, no standardization of the variables is needed, removing the variance bias and keeping the maximum amount variance in the data that the model can account for.
The other model based way, next to likelihood maximization, to arrive at an estimation of mixture model parameters is by the use of Markov chain Monte Carlo simulation methods (Wedel and Kamakura 1999). Despite it is being increasingly used by researchers, there are several disadvantages making it less suitable for our goals. Firstly, the added value of MCMC methods for cluster analysis, above existing methods, has not yet been sufficiently demonstrated (Ryden 2008). Secondly, point estimates, instead of the MCMC, which produces confidence interval estimators, are sufficient for the current study. Additionally, when local optima exist in the data, the MCMC estimators tend not to converge substantially enough and making interpretations of the results very difficult (Ryden, 2008).
Model Estimation
26
within our observations. We cluster on the basis of the six nominal variables (Challenge is a pressing concern = 1, challenge is not a pressing concern = 0).
The estimation of the model is based on equation 1 (adapted from Vermunt and Magidson, 2005):
(1)
Where,
y
i = The vector of observed challenges for firm i θ = A vector of model parametersπ = The prior probability of belonging to latent class k. That is the size of class K.
As we can see from the model, the distribution of the vector of observed challenges
y
given the model parameters θ,|θ
, is assumed to be a mixture of the distributions of the challengeswithin each class
|θ
). These underlying distributions, as follows from the nominal natureof our variables, follow a multinomial logistic distribution of the six nominal digital problem variables. Assigning values to the parameter estimates is done by the estimation of a log-likelihood function optimizing the log-likelihood of the parameters given the patterns of observations in the data.
27
To select the number of classes to extract from the data we use the likelihood function. The optimal number of classes has the highest likelihood. To evaluate the likelihood we use a correction in the form of the Bayesian information criterion. This corrects for the number of observations and parameters and enforces them as penalties which are optimal for the mixture model approach (Andrews and Currim 2003).
RESULTS
We will subsequently discuss the steps we will take to arrive at our configurations of the digital marketing challenges. Firstly, we estimate a number of potential models and select the model with the optimal number of configurational categories to extract from the data. Secondly, this optimal solution will be checked for reliability and external validity. We conclude by making an in depth description of each category and attaching semantic meaning to it.
Model estimation and selection
To estimate the model we have used the Latent Gold software package by Magidson and Vermunt. This combines the EM algorithm with Newton-Rhapson, to arrive at the maximum likelihood estimation through an iterative procedure (for additional technical information see Vermunt and Van Dijk 2001).
28
For our data we consider up to 6 potential classes, more classes would lead to a violation of the interpretability criterion and to insufficient degrees of freedom for the model to have stable estimates. Table 6 shows the values of the criteria we discussed in the previous section:
TABLE 6: MODEL SELECTION CRITERIA BIC(LL) AIC3(LL) # of Parameters Classification Error AWE 1-Cluster 3728,213 3709,071 6 0 3783,3550 2-Cluster 3786,499 3693,98 29 0,1441 4358,8165 3-Cluster 3931,213 3685,821 52 0,0286 4791,9471 4-Cluster 3851,717 3691,939 75 0,0329 4432,4982 5-Cluster 4008,883 3696,233 98 0,1107 5013,6529 6-Cluster 4051,095 3665,066 121 0,0519 5435,3944
On the basis of the classification error statistic, we look to extract three or four classes from the data. From the Bayesian information criterion we can see that there is a dip for the four cluster solution which also shows up in the average weight of evidence. Due to the small differences in fit and classification error, the argument for three latent classes could also be made. We base the final decision on the interpretability of the results. The most distinct and interpretable class solution occurs in the four cluster solution, which we will now select as the most optimal solution given the outcomes of the analysis (for purposes of completeness, the semantic cluster description of the three class solution is in appendix A).
Numerical results
After selecting the optimal number of classes to extract from the data, we now estimate the final model. Table 7, provides the parameter estimates for the four class solution. Note that the nominal variables for the problem have been contrast coded, and the parameter for not considering a given challenge as important is set to 0.
29
challenges variable in the model leads to a significant increase in likelihood. All challenges proved to contribute significantly to increasing the likelihood.
TABLE 7: MIXED MODEL PARAMETER ESTIMATES Cluster 1 Cluster 2 Cluster 3 Cluster 4 Wald Statistic (p-value) Cluster Size 0,4352 0,3359 0,1165 0,1124 Social Media
4. Managing brand health and reputation is more challenging in a marketing environment where social media plays an important role.
-0,674 1,161 -0,423 -0,064 26,84 (<0,001)
Organizational Challenges
6. The pervasiveness of marketing activities within companies is causing organizational challenges (e.g., unclear accountability and incentives, changes in decision-making processes).
-1,0192 -0,9682 -0,0765 2,063 22,44 (<0,001) 7. The increasing prevalence of digital tools and
technologies is threatening existing business models.
1,330 0,692 -4,494 2,490 12,3 (0,006)
Data Usage
8. The ability to generate and leverage deep customer insights is becoming a necessity to compete effectively.
0,753 1,039 -0,188 -1,603 17,6 (<0,001) 9. Assessing the effectiveness of digital
marketing is difficult, since online and traditional metrics are not readily comparable.
0,723 1,248 1,835 -3,807 12,9 (0,005) 10. Marketing and related departments are
facing a significant talent gap in analytical capabilities.
2,498 -1,906 2,152 -2,744 14,9 (0,002)
30
to be significant on the 0,05 level except for social media which did not show significant difference between the classes.
TABLE 8: CLASS COVARIATE DESCRIPTIVES Covariates Class 1 Class 2 Class 3 Class 4 Sample Means Company Size 1 – 49 Employees 0,207 0,278 0,005 0,352 0,217 50- 500 Employees 0,216 0,186 0,097 0,173 0,191 500-10.000 Employees 0,280 0,314 0,214 0,001 0,271 > 10.000 Employees 0,297 0,222 0,685 0,474 0,322 Data Quality
Aggregated sales data. 0,125 0,257 0,315 0,001 0,184 Individual sales data 0,213 0,160 0,533 0,269 0,232 Basic customer information 0,440 0,425 0,094 0,469 0,400 Detailed customer information 0,222 0,158 0,058 0,261 0,184 Decision making relies primarily on
managerial expertise
0,075 0,202 0,145 0,245 0,135
Region Located in Europe or US 0,790 0,815 0,785 0,495 0,783
Located in other location 0,210 0,185 0,215 0,505 0,217
Focus Business to Consumer 0,292 0,416 0,240 0,374 0,334
Business to Business 0,708 0,584 0,760 0,626 0,666 Revenue from online sales Less than 2% 0,475 0,673 0,768 0,223 0,562 Between 2% and 10% 0,232 0,270 0,232 0,637 0,266 More than 10% 0,293 0,057 0,000 0,139 0,172 Industry
Business/ Legal/ Prof services 0,128 0,175 0,334 0,516 0,187
Financial 0,083 0,144 0,001 0,264 0,105
High Tech/ Telecomm 0,273 0,108 0,144 0,006 0,189
Manufacturing 0,046 0,296 0,462 0,154 0,182
Other 0,470 0,277 0,060 0,061 0,338
Social Media
usage Mean
31
Validation
Before continuing, we need to validate the cluster model that we have specified and estimated. Without validating the cluster solution, we cannot be sure that we have arrived at a meaningful and useful set of clusters that represent reality accurately (Ketchen and Shook 1998). There are two steps in the validation analysis that we equip here. The first is a reliability analysis, a necessary but not sufficient condition for validity, showing how the cluster solution is consistent over different estimations. The second is a validity procedure, known as split-half validation, in which we assess whether the model is a result of our specific dataset or if it is useful in class prediction for other sets of observations.
Reliability Analysis
To assess the reliability of the outcomes, we perform a reliability analysis. In essence, this consists of estimating our model several times and comparing the outcomes to see how stable the parameters are over the different estimations. While procedures for assessing the reliability of mixture model clustering seems to be very scarce in literature, we feel that, to be thorough, it is an essential part of the analysis. To perform the reliability analysis we follow the procedure of Dolnicar and Leisch (2010), who asses the relative merits of different clustering methods. In line with these authors we estimated the model 40 times (10 timer per identified class), keeping the algorithms used to estimate the parameters constant and varying the starting values of the iterative procedure and having increased the number of iterations of each of the estimation procedures.
The results of this reliability procedure revealed several insights that may be important for the reliability of the model and therefore the interpretation of the results. Firstly, the likelihood function of the parameters, given the data, showed several local optima. The algorithm led to the convergence of the parameters on a number of different sets of values, dependent on the starting value of the iterations. Subsequent analysis showed that the likelihood of these local optima was significantly lower (p < 0,05) than the solution that we have chosen as our final solution using the likelihood ratio test. We therefore eliminated 19 of such locally optimal solutions from our set of 24 estimations of the model.
32
variation that our estimates showed in one parameter was of 1,16 units (with a maximum of 0,65 in one direction). Because cluster solutions are by nature somewhat subject to variation between different estimations (Franke, Reisinger and Hoppe 2009), we are sufficiently confident in the validity of the results as parameters showed only limited variation. However, in the semantic interpretation of the estimates of the parameters we take into account the variation by means of assigning somewhat broader verbal labels.
Validity Analysis
For the analysis of validity, we performed a split-half cross validation. This entails splitting the data into two halves of 316 observations each. We then, estimated the parameters of our mixture model, as we did before, in both of the sets. Next, we used the estimates from the one set and predicted class membership in the other set. To be thorough we have done this for both sets. We then compared the class prediction in the set used to estimate the model with the class prediction from the other model.
33
TABLE 9: SEMANTIC CLUSTER DESCRIPTION Troubled analyzers The Tradionalists Immovable Giants The organizationally challenged Cluster Size 43,5 % 33,6 % 11,7 % 11,2 % Social Media
4. Managing brand health and reputation is more challenging in a marketing environment where social media plays an important role.
Somewhat Unimportant Important Average Importance Average Importance Organizational Challenges
6. The pervasiveness of marketing activities within companies is causing organizational challenges (e.g., unclear accountability and incentives, changes in decision-making processes).
Unimportant Unimportant Average Importance
Very Important
7. The increasing prevalence of digital tools and technologies is threatening existing business models. Important Somewhat Important Very unimportant Very important Data Usage
8. The ability to generate and leverage deep customer insights is becoming a necessity to compete effectively. Somewhat Important Important Average Importance Unimportant
9. Assessing the effectiveness of digital marketing is difficult, since online and traditional metrics are not readily comparable.
Somewhat Important
Important Important Very
Unimportant 10. Marketing and related departments are facing
a significant talent gap in analytical capabilities.
Very Important Unimportant Very Important Very Unimportant Covariates Company Size
1 – 49 Employees Average Average Nearly
none
Many
50- 500 Employees Many Average Few Average
500-10.000 Employees Average Many Average Nearly None
> 10.000 Employees Few Few Many Many
Data Quality
Aggregated sales data. Average Many Many Nearly none
Individual sales data Average Few Many Average
Basic customer information Average Average Few Average
Detailed customer information Average Average Few Many
Decision making
Decision making relies primarily on managerial expertise
Few Many Average Many
Region Located in other location Average Average Average Many
Located in Europe or US Average Average Average Few
Focus Business to consumer Average Many Few Average
Business to business Average Few Many Average
Revenue from online sales
Less than 2% Few Many Many Few
Between 2% and 10% Average Average Average Many
More than 10% Many Few None Average
Industry
Business/ Legal/ Prof services Average Average Many Many
Financial Average Average Nearly
none
Many
High Tech/ Telecomm Many Few Few Nearly none
Manufacturing Few Many Many average
Other Average Average Nearly
none
34
Class interpretation
To facilitate the interpretation of the results, we continue by transforming the numerical description of the classes (the parameter estimates and covariate cluster means), to a semantic table assigning verbal labels to each of the outcomes. In the top half of table 9, the relative importance of the digital marketing challenges is indicated. In the lower half, we compare the means on the descriptive variables in each class to the means of the sample. The labels provide a description of the number firms who have the described characteristic in a certain class. We will continue by naming the classes and give a short description and interpretation of the most salient properties of each class.
Class 1: Troubled analyzers
The first class is the largest of all, consisting of over 43% of the firms in our sample. The class shows to have a relatively high share of high tech firms in this class, and contains a large proportion of relatively small size firms in terms of the number of employees. The firms in this class seem to have a focus on using their data to guide their decision making. Their most pressing challenge in this matter however, is the strong experienced gap in the analytic capabilities of the marketing departments. With this analytical gap, they also experience some challenges in assessing the effectiveness of their digital marketing efforts and generating and leveraging insights from the data that they have collected. Additionally, the business model of these firms (the way in which these firms create their value) seems to be under pressure due to the prevalence of digital tools and technologies. These main challenges form a rather coherent image of the firms in this cluster. The digitization of the marketing efforts is deemed to be important but the capability to sufficiently be able to rely on data insights to guide them is as of yet missing, caused mainly due to the gap in analytical capabilities. This does not seem to be the case because of the pervasiveness of the marketing activities throughout the firm
35 Class 2: The traditionalists
The second class of firms that we identify, are the firms that seem to be traditionalists who have not yet fully made the transformation to placing the customer in the center of their focus. The class is relatively large in size consisting of approximately one third of our sample and they have a stronger focus on business-to-consumer interactions compared to the sample mean. These firms are experiencing digital challenges in two domains. Firstly, the management of the reputation and health of their brands is a large cause of concern due to the marketing environment where social media plays an important role. Secondly, this is combined with challenges in assessing the effectiveness of their digital marketing efforts due to comparability issues of online and offline metrics. Also, the ability to generate and leverage and leverage insights is seen as a challenge.
These challenges however do not seem to be emanating from the gap in analytical capabilities, as is the case for the troubled analyzers. So, while the firms perceive themselves to be sufficiently equipped to deal with the analysis of data, the primary problem lies in the online nature of metrics and social media developments. An explanation for this may lie in the description of the firms in the class. Here, we can see that the quality of data in this class is relatively low, being for relatively many firms restricted to aggregated sales data. This low data quality may stem from the lack on contact with the customer. As these firms are largely selling their products offline to consumers and many of them are manufactures of products, the connection between the customer and the firm is relatively limited, making it difficult to get first hand consumer data to quire insights from.
As the ability to generate and leverage data and measuring online performance are seen as challenging, these firms highly rely on managerial expertise to make decisions compared to the sample mean.
Class 3: Immovable giants
36
The firms in this class experience challenges throughout the range of categories that we identified. The most important challenges lie in the data usage category. They face a very large gap in analytical capabilities and have challenges assessing the effectiveness of their digital marketing efforts. They however do not perceive that their business models are threatened by these digital tools. Additional analysis of the profile of this class shows that the quality of their data is in the lower range, making the effectiveness of the firm’s digital online marketing activities difficult to assess.
The overall image seems that these are large size firms who have an absence of online interactions and social media challenges are that they are lagging behind on the digitization of their marketing efforts. No detailed data is collected and the business models stand as they do indicating that these firms may be stuck in their traditional way of working. The challenges that they experience they attribute to a lack of analytical capabilities, while their digital efforts would suggest that they are not able to collect sufficient detailed data of actually leverage insights from their customers.
Class 4: The organizationally challenged
The final of the four classes that we identify has a very clear domain of challenges. Their decision making relies on the analysis of online and offline data and they seem to be doing quite well in this regard with few challenges in the data usage category. This is reflected by their high quality of data allowing them to leverage the insights that they extract from the analysis of this data on the individual consumer level.
The challenges of these firms strongly lie in adjusting their business models to the digital age and the associated problem of the marketing activities in the firm being scattered across the different departments of the firm.
37
indeed following through on the digitization but that their business models are not well equipped to handle the associated changes.
Lack of difference in social media usage
38
DISCUSSION
The goals of the research in this paper were twofold. We based ourselves on the identification the most important digital marketing challenges (Leeflang et al. working paper) that firms are facing in these times of rapid technological change brought on by the internet and other associated digital tools. These challenges were identified using a questionnaire among the marketing managers of 777 firms across the globe. We more closely examined the six most important challenges that were identified using this sample. The most important challenges were in the domains of online consumer interactions, challenges associated with the usage of data and challenges associated with the organizational structure.
To empirically identify common configurations of challenges, we employed a taxonomical approach by employing a mixture model. In this approach, commonly occurring configurations of challenges were captured from the data. Firms belonging to each of these configurations were identified and described. We found evidence for four sets of digital marketing challenge configurations.
39
The fit between the external environment and organizational resources
Having extracted and analyzed four different sets of digital marketing challenges that co-occur, we look at how these challenges could have been the result of the match between the external environment of the firm and its internal resources. In this introduction of this paper we explicated how a mismatch in this regard leads the firm to experience challenging situations that must be dealt with in order to stay competitive. We will discuss potential mismatches for each of the challenge configurations.
Class 1: The troubled analyzers
Firstly, for the troubled analyzers, we very clearly saw the mismatch between the environment and the firm. The online environment that these firms are in, as well as the high-tech industries, requires firms to be able to analyze and adapt to their digital environment, which is rapidly changing. The fast pace of the environment leads the likely culprit of their challenges to be the speed by which the organizations can respond to their environment. Day (2011) identified this to be one of the two main reasons organizations struggle to adapt their marketing practices to the environment. This point is strengthened because, for the firms in this class, the processes for data driven decision making by managers seems to be well in place, compared to the other classes in the sample.
40 Class 2: The Traditionalists
The second class of firms, the firms that are traditionalists, are experiencing a different set of challenges. The challenges these firms are experiencing are the result of two specific mismatches between the firms’ resources and its environment. Two specific environmental developments seem to have created mismatches.
The first of these is the shift from the offline business environment to the online environment. The firms in this class themselves rarely sell their products online. However, consumers, which these firms primarily focus on, have made the transition to the online realm. This seems to have led them to be somewhat out of touch with their customer. Many of the challenges that these firms are facing, emanate from being unable to form an adequate response to this shift. As the online social media environment has changed the game for maintaining a positive brand image (Williamson 2011), firms must adapt to these changes. The firms in this class seem not to have taken the necessary steps to ensure their brand health is properly managed in the online environment. In addition, the firms in the class are facing severe challenges assessing the effectiveness of their online efforts, another indication that matching the firms resources to the digitization of the environment has as of yet not been successful.
The more likely of the two causes of the mismatch Day (2011) describes are organizational rigidities. The online environment has been around for quite some time, making it less likely that the speed of adapting is the cause of their challenges. This also shows from the lack of detailed customer data which we discuss next in more detail.
41 Class 3: The immovable giants
As we discussed in the previous section, the firms in the immovable giants class, do not seem to be responding to their changing environment. The environment seems to have past these firms by, without them responding to it. Several mismatches seem to be leading them to experience challenges.
Firstly, the match between the digital environment and the digital capabilities of the firm seem to be causing challenges. The customer detail of the data, which these firms use for basing their decisions on, is very limited. These firms however, attribute the cause of their data usage challenges, to a gap in their analytical capabilities.
The lack of digital capabilities that these firms seem to be having leads them not to be active in selling their products from online platforms. This is peculiar since, as discussed, business-to-business focused firms, of which this class primarily exists, are starting to realize and capitalize on the huge potential of digital marketing as a source of competitive advantage.
Secondly, while these firms do not seem to be creating their value online and they seem not to be able to collect appropriate data from their customers and therefore are not able to leverage it, they report that the prevalence of digital tools and technologies is not threatening their business models. As we concluded, their business models are not highly adaptive to the online environment. Therefore we propose that these firms may suffer from the organizational rigidities that Day (2011) describes in dealing with their changing environment. Inertia and complacency could explain their lack of adaption to the digital environment while also accounting for our finding that the firms in this class consider their business models to be very much unthreatened. We would therefore recommend that these firms thoroughly evaluate their value creation processes and adjust them accordingly to adequately match their environment and make the transition into the digital age.
Class 4: The organizationally challenged
42
percentage of online sales is higher than the rest of the sample and they show just as many challenges with the social media as the entire sample of firms.
The primary mismatch then, occurs in the organizational domain. Imitation of data driven CRM systems by firms in more developed region may be very difficultly adapted in these organizations Kumar, Sunder and Ramaseshan (2011). One of the potential drivers of the challenges they experience in this organizational domain is the decision making practice that these firms employ. The decision making in these firms is very highly reliant on managerial judgment and expertise, even while they have this vast amount of data and insights already available within the firm. Two of Day’s (2011) causes for mismatches may be the source of such problems. Firstly, structural insularity is highly suggestible, as the marketing activities of these firms seem to be highly pervasive throughout the firm. Then, while the data and the capability to use it is present, as information is not shared (or discounted when shared) across functions, challenges occur. The second cause, lock-in and path dependence is also a potential source of the challenges. While the firms in the class seem to have invested heavily in data and analytical capabilities, a lock-in in old decision mechanisms may have caused these investments not be pay off because the new data reliant processes are not replacing old obsolete processes.
These firms then, are recommended to rigorously redesign their organizational structures to (1) ensure the free flow of data and data driven insights and (2) ensure that old processes are replaced with new ones, adapted to make use of their digital marketing capabilities to the fullest extent.
To conclude, we see that in all of the four different classifications of challenges, we can establish a link between these challenges that firms experience and a mismatch between the external environment and the internal resources and capabilities that firms possess, with each class having different mismatches.
Implications for science
43
(2011), because of the rapid developments in this field is of vital importance to track these developments as they unfold.
Secondly, to our knowledge this is the first study to consider digital marketing challenges from a taxonomical perspective. Previous studies have not looked in-depth looked into the potential of these digital challenges to occur simultaneously in firms because of the characteristics of these respective firms and the way they interacts with their environment. We have shed light on how the match between the firm and the environment potentially creates challenges. Therewith we have made a conceptual contribution to the marketing science field for other research to be based upon.
Third, it provides avenues for research and insight into digital developments. While this is not necessarily a unique contribution, the marketing science institute research agenda (MSI, 2012) specifically aims to close the gap that occurs in marketing science between the accelerating complexity of markets brought on by the rapid evolution of digital technology and the ability of organizations to respond. By illuminating this gap and specifically naming some contributions to this gap, we have taken the first steps to closing or at least to bridging the gap.
Implications for managerial practice
From the managerial perspective this study also has several important contributions. Firstly, the configurations of the different challenges can help managers to understand the challenges in the digital marketing realm within their respective firms. As we have showed, these configurations are relatively stable over different estimation and validation procedures. It is therefore likely that most firms can be classified in one of the categories that we have empirically established.
