Demystifying Machine Learning for Programmatic UA & Retargeting
summary
Describes concrete ML work for programmatic UA: models update daily to “remember yesterday,” run predictions and 100–300 ms bidding in real time, and use feature‑importance charts to explain outputs. Cold start needs a rule or segmenter; warns against overfitting. For iOS context loss, compares store categories, manual tagging and ELMo app embeddings to build dynamic whitelists. Creative optimization uses contextual bandits (multi‑armed bandits with user features) instead of slow A/B testing. Advises data sharing, DSP due diligence, realistic KPIs and patience for models to converge.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 2:21Machine learning isn’t complex
- 3:04Programmatic UA basics
- 7:18Visualizing ML for transparency
- 8:42Cold start solutions iOS/Android
- 10:07Simple models outperform complex
- 11:49iOS 14 attribution impact
- 12:54Store categories for targeting
- 13:35NLP for app categorization
- 14:59Creative selection bandits
- 16:25Contextual bandits targeting
- 17:30ML-based DSP tips
- 19:37Q&A
related talks
19mMachine Learning is a new age of Personalized Game Monetization
39mMachine Learning Engineering
19mSession: Re-Engagement In 2026: How To Design & Scale Winning Retargeting Campaigns
23mThe unreasonable effectiveness of data for better monetisation
19mHow to Leverage Predictive Analytics to Enhance iOS Marketing Performance
15mSession: Embracing New Technologies – A Must-Have For Ad Monetisation And UA Managers
20mWinning the Game: Leveraging Contextual Bandits for Creative Optimization
37mApplying ML in AdTech and Lifecycle of an ML project