Multi-touch Attribution: What am I training for?
summary
Compares Markov, Shapley and rule baselines inside a simulated marketing environment built from Skyscanner traces, trains and stress-tests methods, and pushes the winning model to production as a Spark bidding pipeline. Claims the chosen approach hits 90% accuracy on simulated channel payoffs, yields about a 10% lift over last-touch, and shows even-touch beats last-touch by 2.69%. Implementation uses Python for simulation and Apache Spark for production scoring and tuning.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:30What is multi-touch attribution
- 6:29Multi-touch attribution in marketing
- 11:28Designing attribution solution
- 13:55Training data challenge
- 18:50Simulated training environment
- 23:05Results and accuracy
- 28:55Stress testing production
- 30:39Conclusion
related talks
46mRNNs for data-driven multi-touch attribution at Booking.com
21mHow to Get Ahead in Advertising
38mHow to Enhance Customer Targeting in Marketing
24mUnlocking revenue: newcomer classification for boosted profits
46mFrom correlation to causation: measuring ad effectiveness at scale
31mLarge Scale CTR Prediction Lessons Learned
25mHow we bring Performance Marketing to linear TV by using machine learning
21mBuild, Measure, and Machine Learn with Angular