Here is the mistake almost everyone makes with creative testing: they either kill a hook after one bad hour of spend, or they let it run for three days out of hope because "it might just need more data." Both are guesses dressed up as discipline. Neither is a budget. By the end of this, you will have a fixed spend ceiling per hook, a set of signals that let you kill early and save money, and a set of signals that justify spending past the ceiling because the hook is actually working.
Why day counts are the wrong unit
"Test for three days" is a common rule and it is a bad one. A hook spending $5 a day for three days has had $15 to prove itself. A hook spending $80 a day for three days has had $240. Those are not comparable tests. The unit that matters is spend, not time, because spend is what buys you impressions, and impressions are what buy you a sample size large enough to trust.
The fix is to set the kill budget as a multiple of your average cost per result, not a number of days or a flat dollar figure pulled from nowhere. If your account's average cost per purchase sits at $18, a sensible kill ceiling per hook is somewhere around three to five times that figure before you make a final call, roughly $54 to $90 of spend with zero or one result. Below that spend, you do not have enough data to say the hook failed. You only have enough data to say it has not worked yet, which is a different statement.
The early-kill signals that save you the full budget
You do not have to spend the full ceiling on every hook. Some hooks announce their own death inside the first few dollars, if you are watching the right number. The metric that matters before any purchase data exists is hold rate and thumbstop rate, not cost per result.
- Thumbstop rate near the account floor. If your typical thumbstop rate runs around 25 to 30 percent and a new hook is sitting well below that after a reasonable number of impressions, the hook is not stopping the scroll. No amount of extra spend fixes a hook that fails in the first second.
- Hold rate collapsing fast. A hook can stop the scroll and still lose the viewer by second three if the payoff after the hook is weak. Watch where the drop happens in the video, not just the aggregate number, because that tells you whether the hook or the body of the video is the problem.
- CPM spiking with no corresponding engagement. This usually means the hook is getting flagged as low quality by the algorithm itself, which is its own kind of signal.
If any of these show up clearly within the first 20 to 30 percent of your planned kill budget, you can stop early. That is the whole point of watching behaviour metrics instead of waiting for cost per result: behaviour metrics arrive faster than purchases do, because purchases need the viewer to watch, click, and buy, while thumbstop only needs the viewer to watch for one second.
Worked example (hypothetical numbers)
Say your account average cost per purchase is $20. You set a kill ceiling of four times that, $80, per new hook. You launch three hooks at $15 a day each.
| Hook | Spend at checkpoint | Signal | Decision |
|---|---|---|---|
| Hook A | $22 | Thumbstop rate 12%, well under account floor | Kill now, save remaining $58 |
| Hook B | $80 | Thumbstop normal, zero purchases, cost per add to cart high | Kill at ceiling, full budget spent |
| Hook C | $45 | Thumbstop strong, two purchases, cost per purchase near account average | Pass, extend spend |
This is an illustrative example only, not a result from any specific campaign. The point is that three hooks did not cost the same to kill. Hook A died cheap because the data arrived early. Hook B had to run to the full ceiling because nothing was wrong on the surface, it simply never converted. Hook C never hit the ceiling because it passed before it got there.
The pass signals that justify spending past the ceiling
Killing is the easy half. The harder discipline is knowing when a hook has earned more budget rather than less. Two things justify extending past your kill ceiling:
- Cost per purchase trending toward or under account average, with a sample of at least two or three purchases. One cheap purchase is luck. A trend across several is a pattern.
- A front-end metric outperforming the account by a wide margin, even with few purchases. If a hook's cost per add to cart is running at half the account average, that is evidence the hook is doing its job even before the back end of the funnel has caught up.
When either of these shows up, you are no longer testing. You are scaling. That is the handoff point into budget increases, which is a separate decision with its own pacing rules.
Set the number before you launch, not after
The entire point of fixing the ceiling in advance is to remove the moment where you are staring at a dashboard mid-day, tired, deciding in real time whether to give a hook "a bit more." That decision made in the moment is almost always wrong, because it is driven by sunk cost, not by data. Write the ceiling down before you launch the hook. Write down what thumbstop rate counts as an early kill. When the spend hits the number, or the early signal fires, you act on the rule you already set, not on how you feel about the hook that afternoon.
This also protects your testing budget overall. If you are running five new hooks a week and each one is capped at four times your cost per result instead of running loose for three days regardless of signal, you know your maximum weekly exposure to failed creative before you spend a cent. That number should be a fixed, boring line item, not a surprise at the end of the week.
Set the ceiling, watch the early signals, and let the rule make the call. Guessing is what costs money, not the testing itself.