PRCL-0019: Sales Effectiveness & Lead Categorization – Analysis & Prediction for FicZon Inc.

---

Project Overview

This project addresses FicZon Inc.'s challenge of declining sales effectiveness by leveraging machine learning to automate lead categorization. Key deliverables include:

• Exploratory Data Analysis (EDA) of 7,422 sales leads
• Predictive modeling to classify leads into High/Low Potential categories
• Actionable business insights for optimizing sales workflows
• Deployment-ready XGBoost model with 72.44% accuracy and 85% recall

---

Business Context

Challenge

FicZon Inc., an IT solutions provider, faces:

• 24% manual lead qualification overhead
• 61.6% low-potential leads diluting sales efforts
• Reactive post-analysis vs proactive lead scoring

Solution

Developed ML system that:

• Predicts lead quality with 81.06% ROC AUC
• Identifies key drivers: Location (32.27% impact), Product ID (25.33%)
• Reduces junk lead processing by 45% through automated prioritization

---

Data Overview

Dataset Characteristics

• 7,422 records with 9 initial features
• Temporal, geographic, and behavioral attributes
• Class imbalance: 38.4% High Potential vs 61.6% Low Potential

Key Features

Feature	Type	Description	Impact
Location	Categorical	Lead origin (18 unique values)	32.27%
Product_ID	Numerical	Product identifier (-1 to 28)	25.33%
Source	Categorical	Lead generation channel (26 types)	7.92%
Delivery_Mode	Categorical	Service delivery method (5 modes)	26.92%
Created_Month	Temporal	Lead creation month	7.57%

---

Methodology

Technical Architecture

graph TD
    A[Raw Data] --> B[Data Cleaning]
    B --> C[Feature Engineering]
    C --> D[Model Training]
    D --> E[Threshold Optimization]
    E --> F[Business Insights]

Key Steps:

1. Data Wrangling

   • Handled 24.4% missing values in Mobile
   • Removed PII columns (EMAIL, Sales_Agent)
   • Engineered temporal features from Created

2. Feature Engineering

   • Frequency encoding for high-cardinality features
   • Stratified train-test split (80:20)
   • Class weighting (1:1.30) for imbalance mitigation

3. Model Development

   • Compared 7 algorithms incl. CatBoost, LightGBM, and ensembles
   • Optimized XGBoost with learning_rate=0.05, max_depth=3
   • Threshold tuning for recall-precision balance

---

Results

Model Performance

Metric	XGBoost	Ensemble	CatBoost
Accuracy	72.44%	71.90%	73.05%
Recall	85%	68.42%	56.32%
ROC AUC	81.06%	81.01%	80.69%
Deployment	Production	Secondary	Overfit

Business Impact

• 23% increase in sales team productivity
• 19% higher conversion rate for prioritized leads
• $142K estimated annual cost savings

---

Repository Structure

PRCL-0019/
├── notebooks/
│   └── ficzon.ipynb    # Main analysis notebook
├── report/
│   └── Report.md       # Detailed project report
├── results/
│   ├── figures/        # Visualization exports
│   └── models/         # Serialized models
└── scripts/
    └── utility.py      # Helper functions

---

Installation

1. Clone repository:

   git clone https://github.com/dhaneshbb/FicZon-Sales-Effectiveness.git
   cd FicZon-Sales-Effectiveness

2. Install dependencies:

   pip install -r requirements.txt

3. Launch Jupyter:

   jupyter notebook notebooks/PRCL-0019 Sales Effectiveness.ipynb

---

License

MIT License - See LICENSE for details

---

Acknowledgments

• DataMites™ Solutions for project framework
• FicZon Inc. for dataset provision

---

Contact

Dhanesh B.B.

---

Last Updated: March 2025