Building Models Using Ad Sales as an Example
summary
The talk addresses building predictive models to estimate relationships between metrics (for example in ad sales) when it’s impossible to run all experiments. It discusses how to design and organize experiments to get high-quality models with minimal resource cost, focusing on trade-offs when many options exist but testing every variant is too expensive.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:09Multiple test variants problem
- 3:01Modeling approach
- 5:58Efficient experiment points
- 7:45Handling model errors
- 9:12Metric discrepancies
- 10:16Traffic losses impact
- 11:37Scaling experiment results
- 13:02Q&A
related talks
35mWhat Failure Taught Me About Building High-Stakes Models
27mAI and analytics for better marketing performance
36mML: Predictive analytics in Ad Engineering
46mGarbage data in, garbage models out
24mHard things about Sales - and how A/B testing can help
46mFrom correlation to causation: measuring ad effectiveness at scale
29mAI and ML for Computational Advertising
40mApplying Data Science and Analytics in Marketing