Weekly campaign digest with anomalies
Every Monday you spend three hours pulling campaign numbers from your ad platforms and email tool into a status update nobody reads carefully. Design an agent that assembles a weekly performance digest, flags anomalies like a sudden cost spike or a collapsing click rate, and suggests next actions. Walk me through where it gets the numbers, what a human checks before the digest circulates, and how you'd verify its anomaly flags aren't noise. Then tell me why your marketing lead should care.
flag_weekly_anomalies(metric_names: list[str], current_values: list[int], baseline_weeks: list[list[int]], threshold_percent: int, min_baseline: int) → list[str][["cost","clicks"],[1500,900],[[1000,1000,1000,1000],[1000,1000,1000,1000]],30,100]out["cost=spike","clicks=ok"][["ctr_bps"],[40],[[100,100,100,100]],30,100]out["ctr_bps=drop"][["forum_signups"],[30],[[3,4,5,4]],30,100]out["forum_signups=low_volume"]Turn the fuzzy goal into a bounded agent task. Say what the agent reads, what it may do on its own versus draft for a human, and what it must never touch. Then earn the trust: how you’d verify it before rollout, and the pitch (the problem, what you built, why it matters).
Agent build: design the agent in plain language (or narrate it). The coach grades the decomposition, the guardrails, the verification plan, and the pitch.