Six months of campaign data is exactly what AI is built to crunch — but the strategy call still needs a human.

The structure is consistent: ingest tabular performance data, compute metrics against KPIs, rank by deviation, identify patterns. This is the same analytical workflow every month, making it highly automatable as a recurring process.

KPIs and channel definitions need to be explicitly provided — 'underperforming' is only meaningful relative to defined benchmarks. The request for '3–5 actionable opportunities with estimated impact' is somewhat open-ended; impact estimation requires assumptions the agent must make explicit or the human must supply.

Data lives in Google Sheets and ad-platform exports, which are accessible via APIs or file upload, but the agent needs granted access, a consistent schema, and clean data. Cross-platform normalization (e.g., matching Google Ads cost data to Sheets conversion rows) is a real friction point that often requires manual prep.

A flawed analysis could lead to misallocated ad spend or misguided client recommendations, which has real financial and reputational cost. However, the output is a report, not an autonomous action — a human reviews before any budget changes are made, which limits blast radius.

Pattern detection and KPI comparison are mechanical, but translating findings into 'actionable' recommendations requires knowing client risk tolerance, competitive context, and relationship history. An agent can flag what the data shows; a strategist must decide what to actually do about it.