Overview
This project tackles a common marketing challenge: "How do you identify which email subscribers are most likely to make a purchase, so you can target them more effectively?"
Using data from an online education company with 100,000 email subscribers, the analysis applies classification models (Gradient Boosting, CatBoost, and AdaBoost) to score leads based on their likelihood of converting. Key predictive features include member ratings, event attendance (tag count), and country of origin.
The predictions are tied directly to business outcomes through a return on investment (ROI) analysis. The core strategy: classify leads as "Hot" (likely buyers targeted with sales emails) or "Cold" (nurtured with value content until they're ready to buy). This approach aims to reduce unsubscribe rates while maintaining revenue.
The project also covers the full deployment pipeline, including a FastAPI backend for serving predictions and a Streamlit frontend that lets marketing teams interact with the model and adjust thresholds based on business constraints.
Key Takeaway: By being more selective about who receives sales emails, the company could potentially increase net revenue by 16% while improving customer experience.