How we bring Performance Marketing to linear TV by using machine learning
summary
Uses household-level tracking (tracking pixel + IP matching) to link TV exposure to site visits, then applies a linear attribution rule to split visits across spots. Trains LightGBM gradient-boosted trees on the most recent ~five weeks to predict cost-per-visit for the next six weeks, finds neural nets perform worse on this tabular, drifting data, and that per-customer models cut error substantially. Exposes predictions through an API and an optimizer that respects budget and timing constraints.
AI-written summary · about
chapters (4)
related talks
49mThe Impact of Machine Learning in Companies
24mAI-Assisted Media Planning
21mDemystifying Machine Learning for Programmatic UA & Retargeting
19mUnveiling the Potential: Exploring Upselling, pLTV, and ML for Operations
24mUnlocking revenue: newcomer classification for boosted profits
38mHow to Enhance Customer Targeting in Marketing
30mPredictive Analytics in Social Media and Online Display Advertising
21mHow to Get Ahead in Advertising