What is the first step in diagnosing a metric drop at Meta?
Segmentation (user groups, geography, device type, ad type, cohorts).
What is localization in diagnostics?
Narrowing down where the issue lives: upstream, core, or downstream; supply vs demand; targeting vs ranking.
Why compare cohorts?
To see if the issue is widespread, isolated, or tied to a specific population or treatment group.
What are typical hypotheses for quality or engagement issues?
Bad supply, degraded ranking, signal loss, integrity enforcement, creative fatigue, seasonal effects.
What is validation?
Use data, logs, A/B tests, or timeline analysis to confirm or reject hypotheses before taking action.