Market Mix Modeling: Causation and Correlation

Understanding the Impact of Marketing Strategies on Sales Performance

📌 Technologies Used: Python, Pandas, NumPy, Statsmodels, Seaborn, Matplotlib

---

📜 Project Overview

This project investigates the impact of different marketing strategies on sales performance. Using Market Mix Modeling (MMM), the analysis distinguishes causation from correlation in marketing investments, helping businesses optimize resource allocation.

The study leverages regression analysis, correlation analysis, and exploratory data analysis (EDA) to derive insights from historical campaign data.

---

📂 Table of Contents

1. Introduction
2. Data Preprocessing
3. Feature Engineering
4. Exploratory Data Analysis (EDA) & Statistical Analysis
5. Correlation & Causal Inference - Regression Analysis
6. Recommendations & Strategies
7. How to Run

---

1️⃣ Introduction

Marketing campaigns influence customer acquisition, but which strategies truly impact revenue? This project aims to determine:

• Which marketing channels drive the most revenue?
• Are campaigns causally linked to sales growth or just correlated?
• How do different customer types respond to campaigns?

This project uses OLS Regression to quantify marketing effectiveness.

---

2️⃣ Data Preprocessing

The dataset contains campaign-related variables such as:

• Campaign_Email, Campaign_Flyer, Campaign_Phone
• Sales_Contact_1, Sales_Contact_2, ..., Sales_Contact_5
• Amount_Collected (Target variable)
• Client_Type (Large, Medium, Small, Private Facility)

Preprocessing Steps

✔ Handling missing values
✔ Standardizing column names
✔ Creating time-based features (Month, Year)

---

3️⃣ Feature Engineering

• Time-based aggregation: Calendar_Month, Calendar_Year
• Customer segmentation analysis: Grouping by Client_Type
• Sales impact analysis: Summarizing key revenue drivers

---

4️⃣ Exploratory Data Analysis & Statistical Analysis

🔹 Key Insights

• Distribution of revenue (Amount_Collected) across client types
• Correlation of campaigns with sales revenue
• Visualization of marketing effectiveness

📊 Example Visualization

• Sales distribution per Client_Type
• Time-series trend of Amount_Collected
• Correlation heatmap of campaign effectiveness

---

5️⃣ Causal Inference - Regression Analysis

We performed OLS Regression to measure the impact of marketing campaigns on revenue.

🔹 Model Summary

R-squared = 0.480  # 48% variance explained
F-statistic = 342.1 (p-value = 0.00)  # Model is statistically significant
Durbin-Watson = 0.624  # Suggests autocorrelation

🔹 Significant Predictors (p < 0.05)

Feature	Coefficient	p-value	Impact on Revenue
Campaign_Flyer	3.34	0.000	🚀 Positive
Sales_Contact_1	4.24	0.000	🚀 Positive
Sales_Contact_2	3.64	0.000	🚀 Positive
Sales_Contact_3	2.34	0.000	🚀 Positive
Sales_Contact_4	10.95	0.000	🚀 Strongest Impact

❌ Not significant predictors: Campaign_Email, Campaign_Phone, Sales_Contact_5

---

6️⃣ Recommendations & Strategies

📌 Actionable Insights:

1. Increase investment in flyer campaigns & sales contacts (significant positive effect).
2. Campaign phone calls are ineffective—reallocate budget.
3. Sales Contact 4 has the highest impact—focus on optimizing this interaction.
4. Different client types respond differently—customized strategies needed.

---

7️⃣ How to Run

📥 Installation

Ensure you have the required Python libraries installed:

pip install pandas numpy seaborn statsmodels matplotlib

▶ Run the Notebook

Launch Jupyter Notebook or Google Colab and execute:

!pip install pytimetk -q
import pandas as pd, numpy as np, seaborn as sns, statsmodels.api as sm
import statsmodels.formula.api as smf

📊 Reproduce Results

1. Run all cells to process the data and generate insights.
2. Analyze the regression summary and visualizations.

---

🏆 Final Thoughts

This project provides a data-driven strategy to optimize marketing investments by identifying causal relationships, not just correlations. Future improvements include:

• Implementing A/B testing for campaign effectiveness.
• Exploring machine learning models (e.g., Random Forest, XGBoost) for better predictions.

📢 Feel free to contribute, raise issues, or suggest improvements! 🚀