📌 Project Title

ChurnGuard AI – Customer Retention Capital Allocation Engine

---

🧭 Real Business Problem

Subscription-based companies often lose 15–30% of customers annually.

However:

Retaining every customer is financially inefficient

Retention campaigns involve:

• Discounts
• Loyalty credits
• Account manager effort
• Marketing outreach

Which incurs retention cost.

The real business question is NOT:

Who will churn?

But:

Which customer segments should we invest retention capital in?

This project converts churn predictions into:

✔ Revenue at Risk

✔ Retention Investment Required

✔ Recoverable Revenue

✔ Net Value Created

✔ CFO-grade Budget Allocation Strategy

---

🎯 Purpose of This Project

To build a decision-intelligence system that enables:

Finance teams to determine:

Where should retention budget be allocated
to maximise recoverable revenue?

Instead of:

Blindly running retention campaigns
across all high-risk customers

---

⚙️ Key Capabilities

• Customer churn prediction
• Risk-tier segmentation
• Revenue exposure estimation
• Retention investment simulation
• Recoverable revenue modelling
• ROI-driven intervention strategy
• CFO-ready PDF report
• Boardroom-ready PPT deck

---

🧱 Tech Stack

Layer	Tool
Inference Engine	Python
ML Model	XGBoost
Dashboard	Streamlit
Data Processing	Pandas
Reporting	ReportLab
Presentation	python-pptx

---

🚀 How to Run This Project

---

Step 1 – Clone Repository

git clone https://github.com/<your-username>/Customer_Churn_Prediction.git
cd Customer_Churn_Prediction

---

Step 2 – Create Virtual Environment

python -m venv .venv

Activate:

Mac/Linux:

source .venv/bin/activate

Windows:

.venv\Scripts\activate

---

Step 3 – Install Dependencies

pip install -r requirements.txt

---

Step 4 – Run Application

streamlit run app.py

---

Step 5 – Upload CSV

Upload customer dataset containing:

• tenure
• MonthlyCharges
• Contract
• gender
• etc.

The system will:

✔ Predict churn

✔ Segment customers

✔ Simulate retention campaign

✔ Generate CFO report

✔ Generate Boardroom PPT

---

🧪 How to Modify for Custom Dataset

1. Update training schema:

models/training_schema.json

2. Ensure uploaded CSV follows same feature structure
3. Update preprocessing rules:

src/data/preprocess.py

Dataset (Used for Training Model)

Due to size constraints, the dataset is not included in the repository.

Download it from:

https://www.kaggle.com/datasets/blastchar/telco-customer-churn

Place it in:

data/raw/

---

⚠️ Challenges Faced During Development

Challenge	Resolution
Inference schema mismatch	Implemented dynamic schema alignment
Categorical value inconsistency	Category normalization layer
Numeric dtype failures	Training-dtype enforcement
Missing engineered features	Runtime recreation (tenure_group)
Prediction failure due to OHE drift	Pipeline-schema alignment
Finance reporting mismatch	ROI-driven segment modelling

---

📚 Key Learnings

• Prediction ≠ Business Decision
• Model accuracy does not guarantee ROI
• Retention campaigns must be budget-constrained
• Financial impact modelling is essential for adoption
• Schema alignment is critical in production ML systems

---

🏢 Intended Enterprise Use

This system can assist:

• Telecom companies
• SaaS subscription platforms
• Insurance providers
• OTT services
• Fintech platforms

in:

Retention strategy planning

Budget optimisation

Audit-ready campaign justification

---

📊 Output Deliverables

• CFO Financial Impact Report (PDF)
• Boardroom Strategy Deck (PPTX)

---