Skyrocketing Clicks with Smart Targeting
EmailOptix is a data science pipeline that optimizes email marketing campaigns by targeting users likely to click, replacing random sends with precision. Built for a case study with 100,000 email records (CTR ~2.12%), it delivers:
- Open/CTR Analysis: Computed open rate (~10.35%) and CTR (~2.12%).
- Predictive Model: Random Forest model to predict click probabilities, achieving ~40–90% CTR lift.
- Segmentation: Identified high-CTR segments (e.g., loyal buyers with 6–10 purchases, US/UK, morning sends).
- Deployment: Streamlit app and scoring pipeline for real-time targeting.
- Monitoring: Plan to track CTR and retrain the model monthly.
This project showcases end-to-end data science skills: data merging, EDA, feature engineering, modeling, evaluation, segmentation, deployment, and professional communication via PowerPoint slides.
Given a dataset of 100,000 email campaign records, the goal was to:
- Calculate open and click-through rates.
- Build a model to replace random email sends with targeted ones.
- Estimate CTR lift and propose an A/B test.
- Identify high-CTR user segments for personalized campaigns.
The dataset was imbalanced (CTR ~2.12%, max click probability ~0.0294), testing the ability to handle messy data and deliver actionable insights.
Clone the repo and set up the environment:
git clone https://github.com/your-username/EmailOptix.git
cd EmailOptix
pip install pandas numpy scikit-learn matplotlib seaborn streamlit pyngrok joblib- Data: data/output/ and data/input (e.g., targeted_users.csv, high_prob_users.csv).
- Model: models/click_prediction_model.pkl.
- Visuals: images/ (e.g., ctr_by_segment.png).
- Text: text/ (e.g., monitoring_plan.txt, slides.txt).
- app.py: Streamlit App code.
- Feature Engineering: Binned user_past_purchases into purchase_bin (‘0’, ‘1-2’, ‘3-5’, ‘6-10’, ‘11+’) and hour into hour_bin (‘Night’, ‘Morning’, etc.) for robust modeling.
- Model: Random Forest with label-encoded categories, handling low probabilities (~0.0294) via a low threshold (0.02).
- Deployment: score_users pipeline generates targeted_users.csv for high-probability users, deployed via Streamlit.
- Segmentation: Identified high-CTR segments (e.g., 4–6% CTR for ‘6-10’ purchase users, US/UK, personalized emails).
- Monitoring: Plan to track CTR, detect feature drift, and retrain monthly.
- Open Rate: ~10.35% (Phase 2).
- CTR: ~2.12% baseline, ~3–4% for targeted users (Phase 4).
- CTR Lit: ~40–90% by targeting top users (Phase 4).
- Segments: Loyal buyers (‘6-10’, ‘11+’ purchases), US/UK, and morning sends drive high CTR (Phase 5).
- Deliverables: PowerPoint slides, Streamlit app, and monitoring plan for VP-level presentation.
Low Probabilities (~0.0294): Adjusted threshold to 0.02 and used top 10% filtering to target users effectively. Imbalanced Data: Leveraged Random Forest and focused on high-CTR segments to maximize lift. Deployment: Used Ngrok for Streamlit access, with robust error handling in score_users.
This project is licensed under the MIT License.
- Built as a case study to optimize email campaigns.
- Inspired by real-world marketing analytics challenges.
- Tools: Python, Pandas, Scikit-learn, Streamlit, Ngrok.
Connect with me on LinkedIn or GitHub for collabs or inquiries!
Star this repo if you find it useful! 🌟