Fill in before sending
[ad account][minimum results]
Use my Opus Growth connection for this. If you cannot see the tool you need, search among the Opus Growth tools (with search_tools if it is in your list); if you cannot reach the tools at all, do not guess, tell me.
Audit the edits made in Meta ad account [ad account] over the last 30 days and measure their effect on learning and performance.
1. From the change history, list every significant change with who made it and when. Include:
- budget changes (flag any single step above 20%)
- bid or cost cap changes
- targeting edits
- optimization event changes
- new ads added to live ad sets
- ad sets paused for more than 7 days and then restarted (derive this from the status changes)
Meta does not report a learning reset directly. Mark "likely learning reset" from the edit type and size: optimization event changes, targeting edits, bid strategy changes, new ads, budget steps above 20% and restarts after a long pause count as likely; small budget steps and renames do not.
2. For each significant edit, pull the ad set's daily results and cost per result for the 7 days before and the 7 days after the edit, using the action that matches its optimization event.
3. Data quality checks before comparing:
- Confounded: another significant edit on the same ad set (or its campaign budget) inside either 7-day window. Mark it confounded and treat the comparison as indicative only.
- Incomplete: the edit is less than 7 days old, so the after window is short. Mark it and compare on the days available.
- Low volume: fewer than [minimum results] results in either window. Mark it and do not draw a conclusion.
Deliverable:
- A table with these columns: date, person, ad set, change (old value to new value), likely learning reset (yes / no), results before, results after, cost per result before, cost per result after, data quality flag.
- A summary by team member: number of likely resetting edits and their average change in cost per result, calculated on clean comparisons only. Show how many edits were left out as confounded, incomplete or low volume.
- 4 to 6 house rules for this account based on what the data shows, such as maximum budget step, batching edits into one change, minimum days between edits, and when to duplicate instead of editing.
This is read-only. Do not modify anything in the account.