🎯 Freemium User Conversion – Predictive Targeting Strategy

This project addresses a common challenge for freemium platforms: how to identify and target free users who are most likely to upgrade to a paid subscription.

Our predictive model was built on user-level behavioral data from a real marketing campaign involving 41,540 freemium users, of which only 3.7% converted. Using classification models and business-driven targeting strategy, we optimized which users should be marketed to — and achieved a projected 219% ROI.

---

🧠 Business Objective

Identify and prioritize freemium users most likely to convert to paid users
in order to maximize ROI and marketing efficiency

---

📦 Dataset

• Total records: 41,540 users
• Adopters: 1,540 (3.7%)
• Non-adopters: 40,000
• Features:
  • Demographics: age, gender, country
  • Engagement: songs listened, tracks loved, shout activity
  • Social: friend count, average friend age/gender
  • Behavioral changes: delta in listening habits

---

🧪 Modeling Approach

1. Data Preparation

• Label: adopter (binary target – did the user convert?)
• Handled imbalance with ROSE (Random Over-Sampling Examples) to balance classes
• Split into training and validation via 10-fold cross-validation

2. Models Evaluated

• ✅ Decision Tree (final model – best AUC)
• ❌ KNN
• ❌ Naive Bayes

3. Final Model: CART Decision Tree (rpart)

• Cross-validated AUC: 0.770
• Selected using caret with ROC as optimization metric
• Extracted top decision splits:
  • delta_songsListened
  • lovedTracks
  • age, shouts, avg_friend_age

---

📈 Business Impact

📊 Cumulative Response Curve (CRC)

% Targeted	% Conversions Captured
20%	45%
40% ✅	80% (Optimal)
60%	90%
100%	100%

💸 ROI Analysis

• Optimal Targeting Level: Top 40% of users
• Net Profit: $88,765
• Marketing Cost: $40,447
• ROI: 219%

Targeting beyond 40% leads to diminishing returns and negative cost-to-conversion efficiency.

---

💡 Recommendations

• 📍 Target Top 40% of scored users from the model for future campaigns
• 🧠 Segment users further for tailored content
• 📉 Avoid mass marketing to low-conversion clusters

---

🛠 Tools & Stack

• 🐍 R (rpart, caret, pROC, ROSE, ggplot2)
• 📊 Excel / PowerPoint (for business communication)
• 📁 Files:
  • homework2_final.R – Code for data cleaning, modeling, evaluation
  • HW2_Presentation-1.pptx – Final business presentation
  • data_test_output.xlsx – Output predictions

---

👨‍💼 Team

• Justin Varghese – Predictive Modeling & Evaluation
• Anshu Mehta – EDA & Model Interpretation
• Shrawani Tare – Data Cleaning & Strategy
• Tenzin Jangchup – Visual Storytelling & Business Framing
• Chenxiang Ma – Feature Engineering & Testing

---

📬 Contact

Created with 🧠 + 💼 by Justin Varghese
Reach out for collaborations in freemium modeling, customer analytics, or campaign optimization.

🔗 GitHub Profile