Skip to content

drop-triage: why is performance down?

Diagnostics Cowork-ready

Answers “why did performance fall off a cliff” in about ten minutes instead of half a day, and it checks every branch rather than the one you already suspect.

It scores all four causes in a single pass:

  1. Measurement: did tracking or an attribution setting change, so the drop is in the reporting rather than the business?
  2. Learning reset: did an edit throw entities back into learning?
  3. Auction pressure: did costs rise around you?
  4. Creative fatigue: is the work simply worn out?

Plus a business-context sweep, because sometimes the answer is that a promo ended or the warehouse ran out of the hero SKU.

Scoring all four is the whole point. Most real drops have two or more causes, and a media buyer who checks their favourite one first finds a plausible story and stops looking.

  • CPA spiked overnight. ROAS tanked since Tuesday. Conversions fell off a cliff.
  • Before you touch anything. The most expensive response to a drop is a panicked edit that resets learning and creates a second, real problem on top of the first.
  • When a client asks “did Meta break, or did our ads break?”
Cowork ✓ Desktop ✓

Both. Read-only, zero risk to run.

  • The account and roughly when the drop started.
  • Backend numbers for the same window if you have them, because the first question is whether revenue actually fell or only the reporting did.
  • Anything you changed recently, including things that felt harmless.

A confidence-scored verdict card: each of the four causes scored with its evidence, the most likely combination named, the smallest safe action to take now, and an explicit do-NOT list of the tempting moves that would make it worse.

You: “CPA doubled since Friday on examplebrand.”

Verdict card. Measurement: HIGH. The attribution window changed from 7-day-click to 1-day-click on Thursday, so roughly 40% of the “drop” is definitional. Backend revenue is down only 11%. Learning: MEDIUM. Two ad sets were edited Friday and re-entered learning. Auction: LOW. CPMs flat. Fatigue: LOW. Frequency unchanged. Smallest safe action: revert the attribution setting and re-measure before anything else. Do NOT: raise budgets, kill the two ad sets in learning, or launch replacement creative this week.

  • Run it before you act, not after. The do-NOT list is the most valuable part and it only helps in advance.
  • If backend revenue is steady, the drop is probably in your reporting. That is the first branch for a reason, and it is the one buyers skip most.
  • Two causes is normal. Do not stop at the first one that scores high.
  • A cause it could not evaluate shows as NO-DATA, not “ruled out.” A branch nobody could see is not a branch that came back clean.

Fires automatically inside weekly-account-os when a sudden drop appears. If it points at measurement, go to tracking-doctor or signal-surgeon. If it points at fatigue, go to creative-testing-os.