🚀 Consumer Behavior Analytics & Predictive "Whale" Scoring

📌 Executive Summary

This project moves beyond standard reporting to answer the critical question: Who are our most valuable customers, and how do we find more of them?

Using a dataset of 1,000 ecommerce transactions, we performed a deep-dive analysis to identify "Whale" customers, debug the underperforming Loyalty Program, and build a Random Forest Classifier that predicts high-value potential in new users.

🚨 Critical Findings

1. The Loyalty Paradox

Non-Members are surprisingly more valuable ($2,020 CLV) than Loyalty Members ($1,797 CLV). This counterintuitive finding suggests the loyalty program may need restructuring to deliver genuine value.

2. The "Device Myth" Busted

Mobile users spend nearly identical amounts per cart as Desktop users. Do not downgrade the mobile experience — both channels are equally critical to revenue.

3. Age is Irrelevant

There is zero correlation (R² = 0.0003) between age and spending. A 20-year-old is just as likely to drop $500 as a 50-year-old. Age-based segmentation is not recommended for this customer base.

---

🔍 Key Strategic Insights

1. The Loyalty Paradox 🚨

Our analysis revealed a counter-intuitive trend: Non-Members are more valuable than Loyalty Members.

• Insight: Loyalty members use discounts 54% of the time (vs 49% for non-members) and have a lower average basket size.

• Business Impact: The current program incentivizes "discount seeking" rather than high-value purchasing.

| Metric | Loyalty Members | Non-Members |

| --- | --- | --- |

| Avg Lifetime Value (CLV) | $1,797 | $2,020 🏆 |

| Discount Usage Freq | 54% | 49% |

| Avg Spend per Order | $261 | $288 |

2. "Whale" Profiling 🐋

1. The "Deep Dive" Whale (Top 10%)

• Used In: The initial Exploratory Data Analysis (EDA) and Dashboards
• Definition: The absolute top 10% of spenders
• Purpose: To profile the "Elite" customer
• Key Traits: Female, Aged 30-35, buying Electronics & Baby Products

2. The "Predictive" Whale (Top 25%)

• Used In: The Random Forest Model & Consumer Dictionary
• Definition: The top 25% of spenders (Spend > ~$418)
• Purpose: To train the Machine Learning model (ML models need more "positive" examples to learn effectively)
• Output: Any customer with a Propensity Score > 0.60 is flagged for "VIP Outreach"

3. Myth Busting: Age & Income

• Age vs. Spend: There is zero correlation (R² = 0.0003) between age and spending power. A 20-year-old is just as likely to spend $500 as a 50-year-old.

• The Device Gap: Myth busted. Mobile users spend nearly identical amounts per cart as Desktop users.

---

🤖 The Predictive Model

We engineered a machine learning pipeline to predict high-value potential based on demographic traits, allowing marketing to target users before they make a purchase.

• Model: Random Forest Classifier (n_estimators=100)

• Target: Top 25% of Spenders (Threshold: >$418/order)

• Features: Age, Gender, Income_Level, Education_Level

• Performance: Strong positive correlation (R² = 0.81) between predicted propensity scores and actual spending.

Interactive Analysis

The project includes a Plotly Interactive Scatterplot allowing stakeholders to hover over customers to see real-time predictive scores.

(Note: Correlation between Actual Spend and Predicted Whale Propensity)

---

📂 The "Consumer Dictionary"

Instead of a static CSV, the final output of this project is a structured JSON-style Intelligence Profile for every customer. This structure is designed to be fed directly into a CRM or Marketing API.

Sample Entry:

"37-611-6911": {

    "Profile": {

        "Age": 22,

        "Gender": "Female",

        "Location": "Évry"

    },

    "Metrics": {

        "Top_Cat": "Gardening & Outdoors",

        "Spend": 333.80,

        "Loyalty_Level": 5

    },

    "Predictions": {

        "Whale_Propensity": 0.892,

        "Recommended_Action": "VIP Outreach"

    }

}

---

🛠️ Technical Implementation

Data Pipeline

1. Cleaning: Removed currency symbols ($), handled nulls, and converted data types.

2. Feature Engineering: Created CLV_Score (Spend × Frequency) and 5-Year Age Bins.

3. Encoding: Applied One-Hot Encoding to categorical variables (Gender, Education).

Libraries Used

• Pandas: Data manipulation and aggregation.

• Seaborn: Static statistical visualizations (Boxplots, Heatmaps).

• Plotly Express: Interactive, hover-able web charts.

• Scikit-Learn: Random Forest implementation.

---

🚀 How to Run

1. Install Requirements

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import plotly.express as px

2. Launch the Notebook

jupyter notebook Consumer_Behavior_Project.ipynb

3. Download the CSV Used in This Project

This can be found here (Kaggle account may be required)

---

📈 Future Recommendations

Based on the data, the following business actions are recommended:

1. Restructure Loyalty: Move from "Points per Visit" to "Points per Dollar" to reverse the negative CLV trend.

2. Target the 30-35 Demographic: Shift ad spend for Baby Products/Electronics to this high-conversion age group.

3. Deploy the "Consumer Dict": Integrate the JSON output into the email marketing platform to trigger automatic "VIP Outreach" emails when a user's Propensity Score > 0.6.

---

Author: Daniel Ohebshalom