Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/aicorsair/python-case-study-365-data-science-customer-segmentation-in-marketing
This repository contains a detailed case study on the segmentation of 365 Data Science customers using real-world data from an onboarding survey.
https://github.com/aicorsair/python-case-study-365-data-science-customer-segmentation-in-marketing
customer-segmentation data-analysis data-analytics data-cleaning data-preprocessing data-science data-visualization feature-engineering feature-selection hierarchical-clustering k-means-clustering machine-learning marketing marketing-analytics matplotlib pandas python scikit-learn social-media social-media-analysis
Last synced: 1 day ago
JSON representation
This repository contains a detailed case study on the segmentation of 365 Data Science customers using real-world data from an onboarding survey.
- Host: GitHub
- URL: https://github.com/aicorsair/python-case-study-365-data-science-customer-segmentation-in-marketing
- Owner: AiCorsair
- Created: 2025-01-27T16:48:33.000Z (15 days ago)
- Default Branch: main
- Last Pushed: 2025-02-08T21:02:10.000Z (3 days ago)
- Last Synced: 2025-02-08T22:18:41.027Z (3 days ago)
- Topics: customer-segmentation, data-analysis, data-analytics, data-cleaning, data-preprocessing, data-science, data-visualization, feature-engineering, feature-selection, hierarchical-clustering, k-means-clustering, machine-learning, marketing, marketing-analytics, matplotlib, pandas, python, scikit-learn, social-media, social-media-analysis
- Language: Jupyter Notebook
- Homepage:
- Size: 425 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Python Case Study: 365 Data Science Customer Segmentation in Marketing
## Table of Contents
- [I. Introduction](#I.-Introduction)
- [II. Key Questions](#II.-Key-Questions)
- [III. Exploratory Data Analysis](#III.-Exploratory-Data-Analysis)
- [IV. Hierarchical & K-Means Clustering](#IV.-Hierarchical-&-K-Means-Clustering)
- [V. K-Means Model Assessment](#V.-K-Means-Model-Assessment)
- [VI. Key Visualizations & Tables](#VI.-Key-Visualizations-&-Tables)
- [1. Hierarchical Dendrogram & K-Means Elbow Method](#1.-Hierarchical-Dendrogram-&-K-Means-Elbow-Method)
- [2. Correlation of CLV with Regions & Channels](#2.-Correlation-of-CLV-with-Regions-&-Channels)
- [3. Bar Plots of Minutes Watched & CLV by Cluster](#3.-Bar-Plots-of-Minutes-Watched-&-CLV-by-Cluster)
- [VII. Marketing Strategy Recommendations](#VII.-Marketing-Strategy-Recommendations)
- [1. Facebook Campaigns](#1.-Facebook-Campaigns)
- [2. YouTube Influencers Strategy](#2.-YouTube-Influencers-Strategy)
- [3. Google Campaigns & SEO](#3.-Google-Campaigns-&-SEO)
- [4. LinkedIn Content Strategy](#4.-LinkedIn-Content-Strategy)
- [5. Referral Program Enhancement](#5.-Referral-Program-Enhancement)
- [6. Alternative Channels Strategy](#6.-Alternative-Channels-Strategy)
## I. Introduction
In this project, we analyzed real-world customer data from an onboarding survey for the 365 Data Science platform to perform customer segmentation. This is critical for businesses to understand customer behavior and develop personalized marketing strategies.
The project included data cleaning, exploration, feature engineering, clustering algorithms implementation, and results interpretation. We applied two widely used clustering techniques: **K-Means** and **Hierarchical Clustering**.
The analysis segmented a sample of over `3,800` members (representing the entire population) to identify patterns in acquisition channels and geographic influences. These insights will assist the marketing team in prioritizing resources during campaigns.
The `Customer_Data.csv` file includes:
- **minutes_watched:** Total minutes a student watched since joining.
- **CLV:** Average revenue a customer generated over their lifetime (in USD or EUR, currency uncertain).
- **region:** `0` for English-speaking countries (USA, Canada, UK, Australia), `1` for Western Europe, and `2` for other countries.
- **channel:** The source through which the customer learned about the 365 program (e.g. Google, LinkedIn).
## II. Key Questions
- Are there specific geographical locations or acquisition channels where students predominantly discover the 365 Data Science platform?
- How are geographic locations and acquisition channels connected to student engagement and spending?
## III. Exploratory Data Analysis
- We loaded and previewed the customer data, generated summary statistics, displayed column data types, and counted the missing values. Null values in the ‘minutes_watched’ column were replaced with `0`, as these students purchased the program but had not watched any video content.
- We computed the correlation matrix for numerical columns and visualized it using a heatmap. Additionally, we created a scatter plot to examine the relationship between 'minutes_watched' and 'CLV'.
## IV. Hierarchical & K-Means Clustering
- We converted the categorical columns ('region' and 'channel') into dummy variables, automatically replacing the original columns. Additionally, we renamed all columns for better readability and scaled the data.
- We performed hierarchical clustering using the Ward method and plotted a dendrogram to visualize the results.
- We calculated the Within-Cluster Sum of Squares (WCSS) for various cluster counts in K-means and used the Elbow Method plot to identify the optimal number of clusters.
## V. K-Means Model Assessment
- We applied K-means with `8` clusters and assigned cluster labels to ‘customers_kmean’ (a copy of the original ‘customers’ DataFrame).
- For the ‘customers_grouped’ variable, we grouped customers by cluster, calculated cluster size and proportion, and summarized the data with mean values.
- We renamed clusters for better interpretation (e.g. “Facebook-Centric, Highest Minutes Watched” instead of `0`) in ‘customers_grouped’ and set the index name to 'Cluster Label.'
- We created side-by-side horizontal bar plots to visualize 'Minutes Watched' and 'CLV' by cluster. Afterward, we created two scatter plots: one for original 'CLV' vs. 'Minutes Watched' and another for scaled 'CLV' vs. 'Minutes Watched' (without outliers).
## VI. Key Visualizations & Tables
### 1. Hierarchical Dendrogram & K-Means Elbow Method
- The hierarchical clustering results identified `8` distinct clusters, providing a recommended cluster count based on the data.
- The K-means Elbow Method suggested `7` clusters, differing from the `8` clusters found through hierarchical clustering.
- As a result, we chose the hierarchical clustering results and performed K-means clustering with `8` clusters.
### 2. Correlation of CLV with Regions & Channels
| Variable | Correlation with CLV |
|--------------------------------|----------------------|
| **English-Speaking Countries** | 0.7912 |
| **Western Europe** | 0.6179 |
| **Other Countries** | -0.9075 |
| **Google** | 0.0881 |
| **Facebook** | -0.2462 |
| **YouTube** | 0.1832 |
| **LinkedIn** | -0.2121 |
| **X** | 0.1832 |
| **Instagram** | 0.1832 |
| **Friend** | 0.0853 |
| **Other** | 0.1409 |
- As we can see, the correlation of CLV is highest with the region groups: English-Speaking Countries, Western Europe, and Other Countries.
- Interestingly, CLV is highly negatively correlated with countries from the rest of the world (**-0.9075**). This is likely because these countries include developing nations with lower incomes, and their students tend to seek cheaper or free data science education alternatives.
### 3. Bar Plots of Minutes Watched & CLV by Cluster
The number of minutes watched ranges from `1,165.42` to `2,767.64` (median is `1,816.31` mins), while the customer lifetime value ranges from `95.79` to `144.03` (median is `122.17`).
- **Cluster 0 (7.98%):** Students from Facebook, mainly from the rest of the world, have the highest number of minutes watched and below average CLV.
- **Cluster 1 (37.58%):** The largest cluster, with students from mixed regions, mainly from YouTube, has average minutes watched and CLV.
- **Cluster 2 (4.62%):** The smallest cluster, with students who joined through friends, mainly from English-speaking countries and Western Europe, has the highest CLV and above-average minutes watched.
- **Cluster 3 (4.93%):** Students from English-speaking countries (US, Canada, UK, Australia) and LinkedIn have the lowest minutes watched and average CLV.
- **Cluster 4 (17.5%):** Students from mixed regions who discovered the platform via Google have average minutes watched and CLV.
- **Cluster 5 (7.85%):** Students referred by friends, from the rest of the world, have the lowest CLV and below-average minutes watched.
- **Cluster 6 (12.02%):** Students mainly from the rest of the world, who joined via LinkedIn, have below-average minutes watched and CLV.
- **Cluster 7 (7.51%):** Students from mixed regions who joined through other channels have above-average minutes watched and average CLV.
## VII. Marketing Strategy Recommendations
After identifying the customer segments, we can now discuss the results with the marketing team to implement data-driven strategies. It’s noteworthy that 365 Data Science has equal spending budgets across the three region groups.
### 1. Facebook Campaigns
We recommend targeting Facebook users with personalized ads or discounts to boost CLV. Since 365 students from Facebook are primarily from outside English-speaking countries and Western Europe, it could be valuable to explore how focusing more on these regions might impact results, especially given that students from the rest of the world have lower CLV.
### 2. YouTube Influencers Strategy
To leverage 365 Data Science's popularity on YouTube, we recommend collaborating with influencers (experts in AI and data science) to create high-quality, in-demand video content, particularly targeting audiences from English-speaking countries and Western Europe. These influencers can also promote courses they create for 365 Data Science to attract new customers and boost spending.
### 3. Google Campaigns & SEO
We recommend targeting more Google searchers from English-speaking countries and Western Europe with region-specific ads that align with their preferences, while offering personalized discounts to boost CLV. Additionally, optimizing the 365 Data Science blog and free resources for SEO can attract more organic students.
### 4. LinkedIn Content Strategy
We emphasize creating LinkedIn content focused on professional development, educational growth, and addressing pain points for career switchers and students, especially from English-speaking countries and Western Europe. Positioning yourself as having the right solutions and timing special offers can attract new customers and boost spending.
### 5. Referral Program Enhancement
We recommend enhancing the referral program, with a focus on English-speaking countries and Western Europe, by offering discounts or bonuses to both referrers and new students. Highlighting successful referral stories can motivate others to participate, creating a viral effect and boosting overall spending.
### 6. Alternative Channels Strategy
If looking to tap into new opportunities and engage students on less common acquisition channels, it's essential to first understand these channels and identify which offer the highest potential ROI. Since students from outside English-speaking countries and Western Europe have lower CLV, focusing on higher-value segments and investing less in regions with lower ROI would be more effective.