Cet article est issu du data challenge Management & Data Science sur le thème « Service client, innovation et mobilité » sponsorisé par BIP&GO (un des leaders européens du télépéage) en partenariat avec l’Adetem dans le cadre du programme Share Marketing.

Introduction

The explosion of big data and analytics provided marketers and managers with the ability to better understand and evaluate their business, in order to empower them in their decision-making and thus enhance the performances of their company (Brynjolfsson & McAfee, 2012). The development of data sources, statistical analysis algorithms and computing power made this revolution possible in the management and strategy fields, leading companies to develop competencies and facilities to analyze massive data through which competitive advantage can be reached (Provost & Fawcett, 2013).

This opportunity is all the more important for the marketing department, which can now develop customized products and targeted relationships with their customers to maximize the relevance of their actions. However, no study has been conducted so far to determine if those departments are capable of taking full advantage of the opportunity offered by the advent of massive client data.

Some authors highlighted the lack of budget and skilled technicians in the marketing departments (Brady, Saren, & Tzokas, 1999; Leeflang, Verhoef, Dahlström, & Freundt, 2014) but there is a gap to fill in studying how marketing and organization should evolve to get the best out of their data combined with advanced analytical methods. This lack of research is a major issue since Marketing is currently shifting to a data era, leading to huge changes in the practices of the company.

By studying a customer database provided by the marketing department of a French company, this study aims at grasping the difficulties encountered by marketing executives in the data management and analytical process in order to suggest possible ways of improvement and food for thoughts in this particular field.

1. Definition and problematization

As a first step, the literature related to the use of data in business and companies was extensively studied, to understand the main existing topics in research. Then a focus was made on the marketing literature to see to what extent data issues are analyzed by researchers in the context of marketing activities.

1.1. The existing literature on data issues in companies

It is not new for companies to use data for their statistical analysis to take better decisions (Agarwal & Dhar, 2014). This data-driven strategy phenomenon can be found in the literature with the concept of « Business Intelligence » (BI) since the late 80s, which refers broadly to the process of gathering and analyzing data to generate relevant information for decision making (Chee et al., 2009). The term BI also designates any tool and technology that makes these analyses possible, and the products or results from these analyses (Chee et al., 2009). Thus, data analysis as a decision tool has been studied for a long time in the literature, but in recent years major technological changes have unveiled new opportunities for companies, along with new challenges and problems for research to study. For example, the development of social media allowed access to larger amounts of data, faster than before (Zhao, Fan, & Hu, 2014). Better algorithms and computing power enabled to conduct more and more complex analyses (Agarwal & Dhar, 2014). Several authors highlighted the enormous power that data analysis represents given the newly available means (computing power, new algorithms, machine learning, bigger storage capacity…) (Jagadish et al., 2014). Analyses are becoming more and more complex and powerful thanks to these new means, and companies can better measure and understand their business and thus make better decisions (Brynjolfsson & McAfee, 2012).

This unprecedented access to large quantities of data and technological innovations introduced new opportunities, leading companies to rely more than ever on BI tools to analyze data and make decisions. These new practices raised many technological questions opening new interests for researchers (Brynjolfsson & McAfee, 2012). As a result, an increasingly large literature is available, particularly in the Information Systems field, to address the technical issues faced by companies like the process of cleaning data (Chu, Ilyas, Krishnan, & Wang, 2016), storing data, structuring data (Agarwal & Dhar, 2014), and finding the relevant analyses (Chaudhuri & Dayal, 1997) for example.

1.2. The study of Data-related issues in the Marketing literature

The opportunity coming with the explosion of data and new analytical capabilities is particularly interesting for some strategic departments like marketing. Indeed, these revolutions have fostered the development of personalized markets and products, and the establishment of closer and more relevant relationships with clients (Wedel & Kannan, 2016). Marketing analytics allow to provide the consumer with more value and are becoming essential to keep a competitive advantage.

As highlighted by Brady et al. (1999), marketing is by nature a field meant to evolve and adapt a lot. The paradigm shift from traditional marketing (marketing mix) to relational marketing has already brought new tools and competencies to companies like CRM tools for examples (Brady et al., 1999). Today, marketing is getting into the data era and must tackle new challenges and think about the changes needed in companies and marketing departments, to address this huge shift.

If some articles began to give extensive lists of marketing analyses (Fan, Lau, & Zhao, 2015) they do not expose the potential practical difficulties encountered by the marketing department in implementing these analyses. It is surprising how rarely these challenges are addressed in the marketing literature. In the late 90s some authors started to emphasize the importance of integrating IT in marketing practices (Brady et al., 1999; Brady, Saren, & Tzokas, 2002), but these researches were limited to the usage of IT in routine tasks such as emailing campaigns (Brady et al., 2002) or CRM tools (Reid & Catterall, 2005).

This article helps thus filling the gap in the literature on the practical implementation of client database analysis for decision making in marketing, and also on its potential obstacles and key issues. By conducting a case study in a French company, this paper seeks to answer the following research question: What are the key challenges in processing consumers data for marketing decision making?

2. Case study

2.1. Method

This exploratory study aims at identifying the issues encountered by marketers during the conception, creation and analysis of their consumer databases, to develop customer insights. Thus, the case study method was chosen, and more specifically the « descriptive case study » which allows to draw up a state of play of a particular phenomenon or context (Baxter & Jack, 2008).

2.2. Case presentation

For this exploratory analysis, a large customer database from a French company was analyzed, which contains numerous key information on the consumption behaviors of the customers. This company was suitable for the analysis because it provides multiple services, all over Europe and thus generating a significant amount of data. The company is operating in the field of terrestrial transport auxiliary, with 1.5M customers and more than 24M€ turnover.

The database analyzed in the study contains all the 1,500,000 subscriptions to the services, along with the usage information, the transactions, and subscribers’ details.

The data comes from 12 tables spread over 2 different databases. This data is sufficient to have an overview of the marketing analyses issues arising in many B2C businesses.

2.3. Analysis and introduction of data-related challenges

By interviewing 777 marketing executives, Leeflang et al. (2014) showed that the first challenge in data analysis for the respondents was the ability to have extensive consumer insight.

Thus, to understand the challenges encountered by managers who want to go from customers database to customers insight, a typical marketing analysis has been conducted in this case study: a customer segmentation with clustering methods, whose aim is to describe and understand the customers base. Each step of the analysis has been documented and the difficulties encountered have been reported by the authors to highlight the challenges faced by analysts when dealing with this kind of analysis at a company level.

The first step of the process was to collect all customers data tables from the company to create a single merged table for the cluster analysis. On the 12 tables given by the company, 9 of them were using an « IdC » column to identify customers in an anonymized way. Each line represented an action made by the customer like a subscription to an offer or a cancellation of a subscription for example, but since a client can make different actions, he or she could appear on multiple lines, so the information had to be aggregated by merging every line for the same client under a unique line, with aggregation functions depending on the available information and the desired information. On the 3 remaining tables, there was also an « IdC » column, but it was referring to a client’s offer (a client can have several offers) and not to a client. This means that the company stores its information at a client level on some tables and an offer level on others. No matching table was available to link a client and its offers, thus the information on the 3 tables was not compatible with the information of the 9 other tables and couldn’t be used for the clustering.

Another issue that can be highlighted is about temporality. Some information was timestamped while some was semi-aggregated with non-explicit temporal markers. Thus, it was not always possible to know if data was the aggregation during a 1-year period, or since a given starting date. This made it hard if not impossible to lead time-based analysis, and to know if and when data was updated.

Some information was factually wrong, abnormal or incomplete, duplicated and not compatible between tables or even among the same table where they were supposed to be unique.

Finally, variable names were not always explicit, and documentation was not detailed enough to understand when and how variables were created and how they were measured

3. Results and implications

3.1. Explanation and causes of data-related problems

The case studied here showed every symptom of an IT system made by computer scientists for computer scientists. In many companies, the responsibility for collecting, organizing and cleaning data lies with IT departments (Wende, 2007). The goals driving the collection and processing of data are not the same for IT and marketing departments. IT departments must focus on robustness of data, efficiency of processing, scalability, accessibility and flexibility of its usage without focusing on a specific use case. On the other hand, marketing experts and analysts to a greater extent must serve some specific use cases and maximize the business impact of their analysis. There are two entities driven by different objectives working together.

In technical terms this dichotomy is reflected by two databases systems, often maintained separately and whose goals are different (Conn, 2005):

• The transactional systems, with the Online Transactional Processing (OLTP) whose goal is to log day-to-day operations in real-time. This system allows the creation of huge operational databases that track every record, every transaction or operation (Chaudhuri & Dayal, 1997).
• The analytical systems with the Online Analytical Processing (OLAP) whose goal is to provide business experts (executive, managers, analysts) with all the necessary information to make a better and faster decision on longer time horizon (Saagari, Anusha, Priyanka, & Sailaja, 2013). OLAP systems are often used inside data warehouses which are « subject-oriented, integrated, time-varying, non-volatile collection of data that is used primarily in organizational decision making » (Chaudhuri & Dayal, 1997).

In this case study, the data collected by the company is close to a transactional system and is not usable per se for marketing analysis. Moreover, it suffers from a few bad practices in term of database structure, field naming and technical issues in the implementing of the process that fed the data. As a result, the data is sometimes unclear or unreliable. The lack of documentation makes problem identification harder because there is little certainty about the intentional nature of potential errors.

3.2. Recommendations and managerial implications

Specific and technical recommendations

This case study helps to highlight some practical problems related to the analysis of clients’ databases and proposes some specific recommendations.

As it was observed in this case, the design of the database is often the most important problem to tackle, because bad design introduces many other problems like duplicated, wrong or missing information, and makes databases hard to maintain.

On relational databases, particular attention must be paid to the definition of primary keys to uniquely identify information, and to avoid duplication. Any information must be in one single table, referenced by a unique set of key to be able to enhance other tables via merging. On transactional databases it is better to add new data than to edit old data when possible, to keep a record of every transaction and limit the risk of improper overwriting. Every transaction must be time-stamped or at least the last edition time must be provided if the data is updated.

Fields must have consistent naming convention which is expressive and unique. Every field must be constrained as needed, for example with unicity constraints or boundaries to avoid abnormal or missing data. Finally, the whole process of database specification must be thoughtfully documented, table by table, column by column, constraint by constraint. The goals and stakes must be clear to better identify errors in the future.

Global and managerial recommendations

Despite the specific and technical issues, this case study highlights the important question of communication and accordance between analysts and computer scientists (cf. Figure 1) to take the most value out of data. Without a mutual understanding of both parties, the data might not be suitable to get business value out of it. Thus, it is essential to decide upstream the type and format required to make as many analyses as possible. To help establish the type of data needed, Park and Kim (2003) propose for example, a set of data to gather in order to better understand and manage a relationship with consumers, going from the acquisition of data to consumer engagement, such as personal, transactional, relational and feedback data.

This way, technical teams can better understand the data requirements for business analysis and build ad-hoc tables with clear organization and documentation to facilitate downstream analysis. The analysts must, in turn, understand the difficulties related to the collection and updating of data, and adapt their practices to maximize efficiency.

This issue of integrating the analytical along with transactional database systems have been studied in the literature through the technical perspective (Conn, 2005), but from a managerial and organizational point of view, the question has not been widely explored, in particular in the context of marketing teams. In this case study, the data held by the company is not fit for the analytical task and this is most likely due to a lack of communication between IT teams and marketing teams. From this main observation, the following recommendations can be considered to maximize the benefits of such corporations.

First of all, companies should define objectives and roadmaps that include both computer scientists and analysts to work together on joint projects. The creation of a specific position that brings together those two worlds and spread a shared vision is more than advised. It can be an « integration analyst » whose role is to organize meetings and design process for working together, or a « data scientist » who is meant to combine strong technical skills along with a business mind.

Figure 1 – Implementation of a data governance system

Wende (2007) suggests the necessity to develop a real data governance in companies to give shared data-centric philosophy and guidelines. This vision could largely benefit marketing department who will have to conduct more and more analyses to justify their strategy and expanses. Technical knowledge will be a valuable skill for marketing teams to team up with IT departments (Schlee & Harich, 2010). Indeed, many authors have shown that the lack of technical skills of marketing teams is a hurdle to efficiency in their job (Miller & Mangold, 1996; Schlee & Harich, 2010).

4. Conclusion, limits and further research

This article helps to better understand the issues of data analysis in the marketing field. There exist several studies about CRM tools or marketing analysis typologies (Wedel & Kannan, 2016), but none of them explore the practical challenges encountered by marketing executives in the implementation of a data-driven approach and in the analysis of a customer base. With this case of a French company, poor data quality and architecture has been observed, leading to difficulties in making data-driven decisions. One main conclusion of this study is the necessity for the marketing departments to communicate more with the IT department, understand better their challenges and explain their needs in a technical way.

This study holds great interest for managers, enjoining them to rethink their organization to define data quality governance models involving all the parties concerned, in order to have proper data for business analyses. This article is also designed to serve as a bridge between technical literature from information systems and marketing literature that is increasingly interested in data processing issues. This multi-field approach highlights some challenges that company are more and more facing.

This article is a first attempt in understanding how organizations and in particular marketing departments should evolve to get the best out of their data. If previous studies proposed technical solutions to technical data processing problems (the types of metrics to study, the types of statistical analyses, the issue of data cleaning and processing…), this study highlights the crucial aspect of the organizational management and data governance in the success of data analysis in marketing departments.

The major limitation of this study is common to all case studies, that is its generalizability. In qualitative case studies, it is impossible to know for sure if the studied case is representative and can be generalized (Hodkinson & Hodkinson, 2001). Nevertheless, case studies allow to explore complex questions and give practical answers to the « why » and the « how » (Hodkinson & Hodkinson, 2001). Here this case helps understanding “how” marketing departments are processing customer data and “why” they are facing some issues.

The observations from this study could be further generalized by conducting other case studies in different types of companies (different sectors or industries, different companies’ sizes…etc.). Moreover, it can also be interesting to challenge the conclusions of this article by conducting real ethnographies in companies, to effectively observe the potential issues in data organization in the company, by interviewing the different teams and understanding the potential ways for improvements, to build a real data governance and increase communication between the technical and the marketing teams.