Abstract
Customer Relationship Management (CRM) systems, which place the client at the center of business operations, play an essential role in the automobile marketing field. By helping corporations better understand customer requirements, CRM improves relationships, enhances satisfaction, and advances competitive strength. The application of Knowledge Discovery in Databases (KDD) technology allows CRM systems to evolve into intelligent tools that forecast developmental trends, identify new customers, and retain existing ones. This article analyzes the requirements of KDD in the automotive sector and establishes a potential customer model using a CRM database from an auto sales company to test its applicability.
Introduction: The "Data-Rich, Information-Poor" Challenge
The automotive industry is undergoing a fundamental shift from a "product-centric" model, focused on vehicle specifications, to a "customer-centric" model focused on buyer needs. While modern dealerships collect massive amounts of data through CRM systems, they often struggle to utilize it effectively. This creates a "data-rich, information-poor" gap: companies possess extensive records but lack the analytical tools to "disentomb" new customers or predict future buying trends. This research bridges this gap by applying KDD techniques to move beyond simple data storage toward intelligent, automated marketing.
Methodology: Decision Trees and Information Theory
The researchers utilize a standard KDD workflow: collecting data from marketing activities, organizing it into a data warehouse, and applying data mining algorithms to build a predictive model. The core technical approach is the Decision Tree algorithm, which employs a "divide and conquer" strategy to split datasets into subsets based on specific attributes.
To determine the most significant attributes, the study uses two key metrics:
- Information Entropy: A measure of the "disorder" or uncertainty within the data.
- Information Gain: A calculation that determines how much "certainty" is gained about a customer's intent to buy by analyzing a specific attribute.
By calculating the information gain for various factors, the model ranks which customer traits are most influential in the sales process.
Datasets
The study uses a sample dataset of 200 customer records extracted from the CRM system of an automobile sales company. The data includes several key attributes used to predict the target variable, "Whether to buy."
The attributes analyzed include:
- Demographics: Sex and Age (categorized as ≤35, 36–50, or >50).
- Socioeconomic Factors: Education level (e.g., Master/Doctor, Undergraduate, Below technical college) and Income level (High, Middle).
Results: Identifying the Hierarchy of Influence
The researchers calculated the Information Gain for four primary attributes to determine which best reduced the uncertainty of the "Whether to buy" variable. Starting with an initial entropy of 0.988, the results were:
- Customer Age: 0.529 (Highest impact)
- Income: 0.232
- Education: 0.109
- Sex: 0.034 (Lowest impact)
The quantitative ranking reveals that Customer Age is the most powerful predictor of purchase behavior in this dataset, while Sex is the least influential. This suggests that marketing efforts should be segmented primarily by age to maximize predictive value.
Conclusion and Future Work
The paper highlights that CRM in the Chinese market has moved from a conceptual phase to a pilot phase, where the focus is now on developing custom systems capable of "knowledge mining." By replacing a salesperson's "gut feeling" with mathematical precision, enterprises can more effectively target specific customer segments.
However, the study notes limitations, such as the small sample size of 200 records. In a real-world "Big Data" environment, these algorithms must be tested against much larger datasets. Future work should also explore more advanced algorithms, such as Random Forests or Neural Networks, and focus on longitudinal analysis to enhance the long-term value of individual customers.
Summary updated