How do I set a kill threshold for an underperforming ad

What this step is

Setting the kill threshold dynamically is step 50 (Part VIII), marked moderate difficulty, almost nobody does this. A static threshold is the default because it's simpler to write down once; it's also wrong by the time order value moves.

Why weekly, not fixed

A static threshold goes stale the moment order value moves: too generous when order value drops, too tight when it climbs. Recomputing weekly as trailing seven-day day-one order value multiplied by a target return keeps the threshold tracking reality rather than a number written once at launch.

The overnight receipt

Raising the target return from 0.5 to 0.8 tightened what the account could pay for a purchase from about A$45 to about A$28, overnight, with nothing else changed. Almost nothing cleared the new bar immediately; the machine culled hard and rode the one proven winner, correct behaviour for a front end that has to pay for itself, and genuinely uncomfortable to watch for a fortnight.

The age floor and the probability table

Nothing dies before seventy-two hours, whatever the spend column says, because conversion reporting settles over twenty-four to seventy-two hours and a day-one kill is a decision on incomplete data. Two times target spend with zero purchases carries roughly a 13.5% chance the creative is fine; three times target with zero purchases drops that to about 5%. Record the spend-at-kill multiple every time something is killed, and read the distribution after a month: a median below two times target means the account is flinching rather than testing.

Where it breaks

Applying the age floor to a whole campaign instead of each individual ad, so new creative inherits an old ad set's clock. Killing on a day that contained a launch, a budget move over about 20%, a tracking change, or a public holiday, any of which can produce a false signal that looks like a genuine cost spike.