AI can crunch the numbers here, but the strategy layer still needs a human hand.

The calculation steps (CPL, ROI, underperformer flagging) are structurally repeatable, but the 'messy spreadsheet' means each run likely involves unique data-cleaning judgment calls. Optimization recommendations also shift based on client context, making this semi-repeatable at best.

Metrics like CPL and ROI have clear formulas, but 'underperforming' and 'estimated uplift' are undefined — the agent must infer thresholds and benchmarks. Without explicit success criteria, the agent can produce plausible-looking output that misses the mark.

The spreadsheet must be shared with the agent, and its messiness introduces real risk — missing values, inconsistent naming, or ambiguous columns could corrupt downstream calculations. The agent has no access to Google Ads directly, so it's working only from what's in the file.

Wrong CPL or ROI figures could lead to misallocated ad spend across 9 client accounts, which has real financial consequences. However, the output is a report, not an automated action, so a human review step before acting keeps the damage reversible.

Calculating metrics is mechanical, but generating credible optimization recommendations with estimated uplift requires knowledge of local service market benchmarks, seasonal patterns, and client-specific constraints that the agent cannot infer from a spreadsheet alone.