Ad Measurement

Stage 5 of the AI ad pipeline: Strategy -> Generate -> Assemble & QA -> Distribute -> Measure & Learn -> (loop). This skill closes the loop. It decides what to measure and how to learn, then drives the tools that produce the actual numbers. It does not compute attribution itself and it does not store campaign data.

This skill owns the measurement policy and drives the tools. For the actual numbers, drive analytics-reporter and the One-Click oneclick repull + ranker. This skill defines what to measure and how to learn from it.

Two laws sit above everything:

• Measure incrementality, not last-click. The question is "did this ad CAUSE more business than would have happened anyway", not "what was the last thing clicked". Last-click over-credits whoever ran the final retargeting impression.
• Real data only. Never fabricate a metric. If a number is not measured yet, write "not measured" and say what it would take to measure it. A made-up CTR, a guessed ROAS, or an invented "winner" off zero signal is a hard fail.

What this skill owns

1. Incrementality, not last-click (the measurement model).
2. The CPM-fatigue tripwire (when to refresh the creative pool).
3. The testing policy tiers (what kind of test the data volume justifies).
4. The learning loop that closes back to ad-strategy.

Everything else is delegated. The composition map is at the bottom.