Optimizing Subscription Services with Uplift Modeling
summary
MO TV runs a randomized experiment offering a free 30-day trial extension and measures uplift (average treatment effect) per user, using a random forest to predict who truly needs a discount. Metric is a “gen curve” and gen (K) score comparing targeted vs random; 80/20 train/test, 5% control and 5% test cohorts, top 20% by predicted uplift get offers. Splitting methods evolve from random to registration-date stratification to a registration + Thompson sampling rollout that finally converges and yields stable, interpretable segments.
AI-written summary · about
chapters (9)
- 0:00Introduction and Overview
- 0:26Company Background and Goals
- 1:03Challenges in Subscription Modeling
- 2:19Data Collection and Feature Engineering
- 3:40Model Training and Validation
- 3:57Evaluation Metrics and Results
- 5:17Advanced Splitting Techniques
- 10:33Final Model and Future Plans
- 11:50Conclusion and Summary
related talks
32mWhy start using uplift models for more efficient marketing campaigns
38mHow to Enhance Customer Targeting in Marketing
27mPreventing churn like a bandit
19mUnveiling the Potential: Exploring Upselling, pLTV, and ML for Operations
24mUnlocking revenue: newcomer classification for boosted profits
21mSubscription fatigue: from myths to facts to opportunities
01h 00mAI Powered Uplift Modeling and Stream Forecasting in Music Industry
43mBeyond Just Prediction: Causal Thinking in Machine Learning