Business Understanding - BigMart Sales Case Study

A common mistake in analytics and data mining projects is to jump straight into analyzing the dataset structure without first examining the context that gives meaning to the available data. In real-world environments, data is part of operational, commercial, or strategic processes, so its exploration must be guided by an initial understanding of the business problem. In this sense, exploratory data analysis (EDA) not only allows us to describe variables and detect patterns, but also to translate business needs into investigable analytical questions.

Therefore, in this exercise, a brief business narrative will be used as a starting point to guide the EDA and establish:

• who generates the data;

• for what purpose it is collected;

• what each observation in the dataset represents;

• which relationships or behaviors are worth exploring;

• and which business questions can be examined through exploratory analysis.

Fictional Business Scenario

Suppose BigMart is a national retail chain with a presence in several cities and across different store formats. Some outlets are large supermarkets, others are convenience stores, and others function as mid-sized commercial locations. The company markets multiple fast-moving consumer goods categories: food, beverages, household products, perishable and non-perishable goods, among others.

During 2013, BigMart’s commercial management noticed that sales performance was not homogeneous across products or stores. Some items had very high turnover in certain outlets but mediocre performance in others. Similarly, stores with seemingly similar characteristics exhibited vastly different results in product revenues.

The company possessed operational systems that recorded transactions, product catalog attributes, and basic store characteristics. However, that information was primarily used for administrative purposes and had not been systematically integrated to support analytical decision-making.

Given this, regional management decided to request an analytics study with two initial goals:

1. better understand the structure of sales at the product-store level;
2. identify factors associated with higher commercial performance.

This request is the origin of the dataset we will be working with.

Problem Statement

The company’s concern does not arise in the abstract. In this fictional scenario, BigMart faces four operational challenges:

• Inventory Planning
  It is not always clear how much assortment to assign to each product type at each outlet. Poor estimation can lead to overstocking or stockouts.

• Differences Across Stores
  Outlets of different sizes, locations, or commercial formats appear to behave unevenly. The company needs to know if those differences are consistent and measurable.

• Promotions and Assortment Design
  Commercial management wants to distinguish which product types perform best and whether certain features, such as category or list price, are associated with higher sales.

• Lack of Integrated Evidence
  Data exists, but it has not yet been converted into actionable evidence for decision-making. Operational records need to be transformed into analytical knowledge.

Consequently, the initial business problem can be formulated as follows:

BigMart needs to understand how sales per product and per outlet are distributed and what drives them, in order to improve assortment, commercial operations, and planning decisions.

Variable Dictionary

The following table details the nature and business interpretation of each variable present in the dataset.

Category	Variable	Description and Business Interpretation
Product	Item_Identifier	Unique product identifier. Enables tracking and linking with other data sources.
Product	Item_Weight	Product weight. Relevant for logistics, packaging, and commercial presentation.
Product	Item_Fat_Content	Fat content classification. Useful for segmenting food and beverage subgroups.
Product	Item_Visibility	Percentage of shelf display area. Indicates the item’s commercial prominence in the store.
Product	Item_Type	General product category. Key for analyzing structural differences across business lines.
Product	Item_MRP	Maximum Retail Price (list price). Main variable explaining variations in monetary sales.
Outlet	Outlet_Identifier	Unique store identifier. Enables comparative performance analysis across outlets.
Outlet	Outlet_Establishment_Year	Year of opening. Serves as an indicator of store age and operational maturity.
Outlet	Outlet_Size	Store size. Associated with assortment capacity, customer traffic, and display volume.
Outlet	Outlet_Location_Type	Geographic location type. Summarizes the socioeconomic environment and demand profile.
Outlet	Outlet_Type	Commercial format (e.g., Supermarket, Convenience Store). Defines the operational logic of the point of sale.
Target	Item_Outlet_Sales	Target Variable. Total sales of the product at the store. Central axis of commercial performance analysis.

Table 20.1: Variable dictionary for the sales dataset

Business Needs Motivating the Analysis

In practice, analytical questions serve as a guide for selecting EDA tools. To achieve this, non-technical client questions must be translated into questions that allow the data team to perform effective analytical work. In this example, following discussions with BigMart, several key needs are identified.

Need 1: Understand overall sales behavior

“Before diving into details, we want to understand how sales are performing overall. Do they typically stay within a similar range, or is there high variation? Are there exceptionally high or low cases we should review?”

This is a typical way for a client or executive to express a business concern. Although it does not use statistical language, as an analytics team we can translate it into more concrete exploratory questions:

• What is the typical sales level for a product-outlet combination?
• Is the sales distribution homogeneous or does it display high dispersion?
• Are there particularly extreme high or low values?

From an EDA perspective, these questions guide the study of the sales variable using measures of location, variability, and shape, as well as the identification of potential outliers.

Need 2: Compare performance across segments

“We want to know if some product types or stores are performing better than others. Specifically, we are interested in seeing whether performance changes according to product category, outlet type, store size, or location.”

This is a standard client inquiry aimed at identifying relevant differences across groups. From the data analytics team’s perspective, this need translates into exploratory questions such as:

• Are there product categories that systematically generate higher sales?
• Do certain outlet types concentrate better results?
• Do sales change depending on store size or location?

These questions guide an analysis of heterogeneity across subgroups, using comparisons among relevant business categories and segments.

Need 3: Identify commercial concentration

“We are interested in understanding whether the business is well-diversified or if a significant portion of sales relies on very few categories or very few outlets.”

This is a common strategic concern in commercial contexts, as it allows assessing how diversified or dependent the revenue structure is. From the data analytics team’s perspective, this translates into exploratory questions such as:

• Do a few categories account for a large portion of overall revenue?
• Are there particularly dominant outlets?
• Is the commercial structure diversified or concentrated?

These questions guide a concentration analysis, focused on identifying whether sales are evenly distributed or heavily concentrated in a small set of categories or stores.

Need 4: Explore associations between attributes and sales

“We want an initial idea of which variables might be most strongly associated with sales, to better understand the business before moving on to predictive analytics.”

This is a frequent request in early analysis stages, when the client seeks an initial understanding of factors linked to sales performance. From the data analytics team’s perspective, this translates into exploratory questions such as:

• Is list price associated with higher sales?
• Does visibility appear to relate to performance?
• Do certain outlet attributes show associations with better outcomes?

These questions guide the study of correlation and bivariate relationships, aimed at detecting initial patterns between attributes and sales that can later be examined in greater depth.

From the Business Problem to Data Understanding

The transition between Business Understanding and Data Understanding occurs when business questions are converted into questions about the available data. Based on the scenario above, as an analytical team we must formulate questions such as the following:

Regarding overall dataset structure

• How many records are there?
• How many numerical and categorical variables exist?
• Are there missing values, inconsistencies, or problematic encodings?

Regarding the sales variable

• How is Item_Outlet_Sales distributed?
• How skewed is it?
• How dispersed is it relative to its center?

Regarding categorical variables

• How many product categories exist?
• How are records distributed across outlet types?
• Are there rare or sparsely represented classes?

Regarding data quality

• Are there missing values in product weight or outlet size?
• Are there inconsistent categories, such as text variants of the same class?
• Are there variables that are actually identifiers rather than analytical features?

These questions no longer belong solely to business understanding: they are Data Understanding questions and pave the way for EDA.

Once the business scenario, unit of analysis, variable dictionary, and commercial needs have been established, concrete questions can be formulated for EDA to answer. These questions form the natural bridge between the Business Understanding phase and the Data Understanding phase.

Questions about sales

• What is the typical level of Item_Outlet_Sales?
• How far do sales depart from that typical level?
• Is the distribution symmetric or right-skewed?

Questions about products

• Which product categories appear most frequently?
• Which categories concentrate higher sales volume?
• Does product price appear to play a significant role?

Questions about outlets

• Are there notable differences across outlet types?
• Does outlet size appear to matter?
• Could store age be associated with performance?

Questions about data quality and preparation

• Which variables contain missing values?
• Which categories need to be standardized?
• What transformations might be required before modeling?

These questions form the natural bridge to Data Understanding and EDA.