Detect fake accounts and measure their impact

## Fake accounts in an ads/product platform You work on an ads-enabled product where some accounts are **fake** (bots, fraud rings, scripted signups) and they distort product and revenue metrics. ### Tasks 1. **Detection approach:** Propose how you would identify fake accounts. Cover at least: - Rule-based heuristics vs. supervised/unsupervised ML approaches - What signals/features you would use (behavioral, network/device, payment, content, velocity, graph signals) - How you would obtain labels (manual review, chargebacks, user reports) and how you’d handle noisy/biased labels 2. **How to measure prevalence:** Define how you would estimate “how many fake accounts exist” and how uncertainty would be reported. 3. **Impact measurement:** Describe how you would quantify the impact of fake accounts on key metrics (e.g., DAU/MAU, CTR/CVR, revenue, advertiser ROI, user experience). Include: - Primary metric(s), diagnostic metrics, and guardrail metrics - An experiment or quasi-experiment design to estimate impact of removing/limiting fakes - Key confounders and how you would mitigate them (selection bias, feedback loops, seasonality, delayed effects) 4. **Tradeoffs:** Discuss pros/cons and operational risks (false positives, adversarial adaptation, user friction, fairness) and how you’d monitor the system after launch.