Diagnose spend drops, bots, and Stories

You are a product data scientist supporting a large social-media advertising platform. In the onsite, you are asked to work through three analytics and product-sense problems. 1. **Brand video ad spend dropped.** The platform supports two video-ad objectives: `direct_response` ads optimized for immediate actions, and `brand` ads that send users to an advertiser's external website after a click. One advertiser reports that its spend on brand video ads fell sharply this week. How would you verify whether the drop is real, quantify it, and identify the root cause? Consider budget changes, auction competition, pacing, inventory supply, targeting changes, creative fatigue, measurement bugs, attribution changes, seasonality, and policy or quality enforcement. What metrics, segmentations, and experiments would you use? 2. **Comment distribution and bots.** Comment counts per post are highly skewed, and the trust team suspects bots are inflating engagement. How would you detect and quantify bot activity using distributional, temporal, text, and network signals? How would you separate legitimate viral behavior from coordinated bot behavior, and how would you evaluate false-positive and false-negative trade-offs? 3. **Facebook Stories vs. Instagram Stories.** A PM asks which surface deserves more investment next quarter: Facebook Stories or Instagram Stories. How would you compare the two products from both a user-value and ads-revenue perspective? Define north-star metrics, guardrails, key segmentations, and an experiment or causal-inference strategy that accounts for cannibalization with Feed, Reels, or other surfaces.