You are running an A/B test on a product change and the primary metric is noisy, making it hard to detect small but meaningful effects. You are considering using a pre-experiment covariate to reduce variance and improve sensitivity without changing the treatment itself.
What is CUPED, and when would you use it in a product experiment? Explain how it changes the analysis, what assumptions make it useful, and when it may not help much.
Understanding of CUPED as a variance-reduction methodAbility to connect variance reduction to power and MDEJudgment about when a pre-period covariate is appropriateAwareness of experimentation pitfalls like peeking and SRM