Incrementality Bidding and Attribution
summary
Estimates causal ad value at auction scale by injecting randomization and using instrumental variables (two‑stage least squares / control function) to recover heterogeneous treatment effects, with a practical ghost bids system that simulates winning probabilities across 10+ billion daily auctions. Models ad response as a continuous-time kernel mixture, integrates the area under the response curve to get value per impression, and combines ML regularization with Hausman‑style correction to balance bias, variance and production latency.
AI-written summary · about
chapters (9)
- 0:00Intro
- 2:05Why does incrementality matter
- 3:58The Harvard Business Review
- 8:06The Ideal Experiment
- 17:53Instrumental Variables
- 22:56Challenges
- 26:36Spikes
- 27:29Modeling Spikes
- 30:43Multitouch Attribution
related talks
46mFrom correlation to causation: measuring ad effectiveness at scale
21mHow to Get Ahead in Advertising
18mBuilding Models Using Ad Sales as an Example
28mCorrelation Sucks - Long Live Causality! Mastering Causal Inference in Data Science
18mData-driven Decision Making in Ad-Tech
24mOffline Ranking Validation - Predicting A/B Test Results
39mBuilding Causal Measurement
36mML: Predictive analytics in Ad Engineering