Анализ взаимосвязи между продажами категорий товаров и их дистрибьюции

Determining the effects and their sizes of various marketing variables on sales has been among the most important activities in academic marketing research. There is a lot of evidence of the effects that advertising and pricing have, but the same cannot be said for the distribution.1 Hanssens et. al. say that “distribution is one of the most potent marketing contributors to sales and market share” and add that “its elasticity can be substantially greater than one”2, but empirical studies on distribution are few and far between.
The distribution of products is considered one of the key drivers of the sales growth. Several studies have shown a convex relationship between distribution and sales in cross-sectional data.3 Generally, when considering the interdependent relationship of sales and distribution, it seems obvious and logical that there is one. The more of the product available for purchase, the more it is purchased, and the more it is purchased and in demand, the broader the distribution, but the theoretical and empirical evidence of such a relationship is limited.
The distribution landscape is changing very fast, and several factors are impacting its evolution. There are the more obvious one, like e-commerce and new generations that are making up the majority of the economically active population. There are also the black swans that come out of nowhere and drastically change every industry, bringing about the new normal. That is why companies need to be able to analyze different elements of their marketing mix, and see how each one affects their sales volume, so that they can adjust it to bring maximum possible profit.
Analysis of the relevant literature on the subject of distribution helped us identify a major research gap. Mainly, it is the type of the relationship that distribution and sales have. Logically, wider distribution causes higher sales volume, but there are other extraneous factors that must be accounted for, such as prices, promotions, or even product characteristics, so everything is not as easy as it seems at first. Also, categories of the products play a role, as the literature indicates that the actual strength of the relationship between sales and distribution is moderated by the product type and category, as well as the growth stage of the category.4
In this research, the main focus will be on the relationship that distribution and sales of category products have. There is some research that shows the quantitative effect that distribution
1 Bucklin, Randolph & Siddarth, S. & Silva-Risso, Jorge. (2008). Distribution Intensity and New Car Choice. Journal of Marketing Research – J MARKET RES-CHICAGO. 45. 473-486. 10.1509/jmkr.45.4.473.
2 Hanssens, Dominique M., Leonard J. Parsons, and Randall L. Schultz (2001), Market Response Models:
Econometric and Time Series Analysis, 2nd edition, Kluwer Academic Publishers, p. 347
3 Wilbur, K. C. and Farris, P. W. (2014). Distribution and market share. Journal of Retailing, 90(2):154{167. 4 PERFORMANCE IMPACT OF DISTRIBUTION EXPANSION: A REVIEW AND RESEARCH
AGENDA (Hibbard, et al)
6
has on sales in general, but only few that focus on specific product categories, and that is why our focus will be on one specific category.
The goal of our project is to analyze the type of relationship that sales and distribution have, whether it is linear or non-linear, so that the company can better analyze the effects that changing distribution can have on sales. To reach this goal, we have formulated questions that we will need to answer, which are the following:
– Is there an interdependent relationship between sales and distribution?
– What type of relationship is there between sales and distribution, linear or non-linear?
– What differentiates the products with linear relationship between sales and distribution
from those that demonstrate non-linear relationship?
– What is the most suitable model to analyze this relationship?
First of all, we start with theoretical part. In the first chapter, we do a literature review on
distribution, category management, and the relationship between sales and distribution. We then discuss machine learning in retail, its tasks, issues, and limitations. Later, we discuss the cases of machine learning application in business, focusing on machine learning in retail, specifically its usage in Magnit, Perekrestok, Pyaterochka, and M-Video, and the results that using machine learning has yielded those companies. Afterwards, we move on to machine learning methods, and go deeper on the methods that we have chosen for our project.
In the second chapter, we present the empirical part of our project. We start with describing the company for whom this analysis is conducted, Procter & Gamble. Then describe the business task, and set more detailed goals and objectives based on this business tasks. We also create a step- by-step analysis with brief description of each step. We start our analysis with exploratory data analysis, where we familiarize ourselves with data, clean it of outliers and drop missing data, do some basic analyses, such as correlation analysis. We also calculate several new variables that will aid us in our analysis further. Then we look at some products separately to understand whether there is a linear relationship between sales and distribution, with the help of scatter plots. Afterwards, we describe our challenges, which are computational power of our machines, data, especially lack of data on other elements of marketing mix, and different time periods in which products are sold, as some of them have been selling for 5 years, and they have more data, while some of them have only been on the market for three months, for example. Then, we preprocess data, and move on to applying different machine learning models, namely, linear regression and random forest. We also apply different types of transformations in linear regression to see which ones would yield highest results. After this, we analyze the results of both models, and try to formulate what differentiates products with linear relationship between sales and distribution from those without.
7

As an expected result, we intend to find out when products demonstrate linear relationship between sales and distribution, and when they do not do that, and what product characteristics differentiate those two groups from one another.
There were two people working on this project, and responsibilities were divided in this way:
– Daria Dobrego was responsible for literature analysis of category management, exploratory data analysis, baseline linear regression and analysis of the results, and building random forest and analyzing and interpreting results of this model;
– Zamira Magomedova was responsible for literature analysis of distribution and machine learning in retail, exploratory data analysis, linear regression with transformations and analysis and interpretation of the results, and writing the text of the thesis.