AI Powered Uplift Modeling and Stream Forecasting in Music Industry
summary
Shows concrete wins: uplift modeling cuts cost per add-to-library by ~30% using two-model, class-transformation and dedicated uplift random forest methods on a dataset of ~10 billion daily records and 15,000+ parameters; stream forecasting uses a Facebook Prophet wrapper with 1,700 regressors (librosa audio features, metadata, marketing spend), MCMC uncertainty, change-point detection and partial dependence analysis to find spend plateaus and save 27% of marketing budget while improving accuracy ~60%.
AI-written summary · about
chapters (22)
- 0:00AI in music industry
- 1:29About speaker and Metis
- 2:56Music business and AI
- 6:12AI adoption challenges
- 8:49Music data types and scale
- 12:24AI use cases in music
- 15:46Customer lifecycle modeling
- 18:59Churn prediction challenges
- 23:19Uplift modeling concept
- 26:33Two-model uplift prediction
- 29:26Class transformation for uplift
- 31:59Uplift random forest
- 36:21Real-world uplift example
- 41:07Ad spend optimization case
- 42:51Stream forecasting
- 44:38Prophet for music streaming
- 48:11Model interpretability importance
- 50:26Marketing spend impact
- 53:45Budget optimization insights
- 54:06AI ethics and governance
- 55:11AI implementation success
- 55:54Q&A
related talks
13mRevolutionizing the Music Industry with Data Science
32mWhy start using uplift models for more efficient marketing campaigns
13mOptimizing Subscription Services with Uplift Modeling
30mPredictive Analytics in Social Media and Online Display Advertising
30mMMM and MTA: The Future of AI-Driven Marketing
24mUnlocking revenue: newcomer classification for boosted profits
24mImprove your marketing reach using ...
21mDemystifying Machine Learning for Programmatic UA & Retargeting