Skip to content

meta-account-doctor: the account audit

Run the week Cowork-ready

Audits the whole Meta ad account and tells you where it is leaking money, in order. It reads four layers through the official Meta Ads connector:

  1. Signal: event match quality, deduplication, whether the account can be trusted to measure itself at all.
  2. Structure: learning-phase floors, budget caps, exclusions, anything fighting delivery.
  3. Creative: similarity clustering, fatigue, net new reach, and a kill list.
  4. Spend discipline, cross-checked against your actual backend revenue rather than Ads Manager’s version of it.

Every finding cites a dated decision rule, and every dollar figure shows its arithmetic. You can argue with the reasoning because the reasoning is on the page.

Read-only. It proposes, it never writes.

  • Weekly, as the diagnostic half of your account routine.
  • Taking over an inherited account and needing to know what you just signed up for.
  • Performance is fine but you suspect it should be better.
  • Before a client renewal conversation, so you know what they are about to ask.
Cowork ✓ Desktop ✓

Both. It needs the Meta Ads connector authorized.

“Audit my Meta account for examplebrand

/um-toolkit:meta-account-doctor examplebrand, full checkup”

  • The ad account, and it will confirm the name Meta reports before it grades anything.
  • The pixel or dataset id for the signal layer.
  • Backend revenue where you have it. Without it, spend discipline is graded against platform-reported numbers and says so.

A leak-ranked action list: biggest recoverable waste first, not a checklist in file order. Each item carries what is wrong, the rule it violates with that rule’s date, the arithmetic behind the dollar figure, and the specific fix.

Plus the four layer verdicts, so you can see whether the problem is measurement, structure, creative, or spend.

You: “Where is my ad account leaking?”

Top finding: Structure. Six ad sets are splitting a $9k weekly budget so thinly that four never exit learning. At the account’s current CPA, the rule’s 50-conversions-per-week floor needs roughly $2,900 per ad set, so consolidating six to three recovers the exercise. Arithmetic shown. Rule cited with its date. Next: three consolidation nominations, awaiting your approval.

  • Signal findings come first for a reason. If the account cannot measure itself, every downstream verdict is built on sand. Fix EMQ and dedup before you touch structure.
  • NO-DATA is not a pass. If the dataset was unreachable, the signal layer reports that it could not evaluate, and the account does not get graded clean. A middle-band score on an audit that evaluated nothing is exactly the failure this design refuses.
  • It grades against backend revenue when it can reach it. Ads Manager ROAS is not the scoreboard.
  • Run it weekly, not monthly. Creative fatigue and learning resets move faster than a monthly cadence catches.

Runs inside weekly-account-os. Its signal layer is signal-surgeon’s territory in depth. Changes it proposes execute through meta-scaling-copilot. For a sudden drop specifically, use drop-triage.