Analyze and mitigate fake advertiser accounts

Your ads platform suspects there are **fake advertiser accounts** (fraudulent accounts created to scam users, evade policy, or manipulate spend). You are asked to lead the analytics approach. ### Prompt 1) Define what “fake account” means operationally and propose **metrics** to measure the problem over time. 2) Given that labels are incomplete (only a small fraction of accounts are manually reviewed), how would you **estimate prevalence** and track trends without being misled by enforcement changes? 3) Propose an approach to **evaluate** a detection/enforcement change (e.g., a new model or new threshold). Include: - offline model metrics and cost tradeoffs - online experiment design and guardrails - key failure modes (false positives, adversarial adaptation, selection bias)