The kill/scale rule that saved us $12k in ad spend
The exact kill/scale rule we use to test ad creatives: spend gates, hook-rate and CTR floors, CPA thresholds, and how to scale winners with volume.
Kill any creative that spends one times your target CPA without a sale — or that misses your hook-rate and CTR floors — on the same day, and scale only the ones that beat CPA and hold it for two straight days. That single rule, decided before launch instead of in the heat of a bad afternoon, is what stopped us from nursing losers and saved us roughly $12k of wasted spend in one quarter.
Why you need a rule at all
The most expensive line item in most ad accounts isn't the winning creative — it's the loser you refuse to kill. We know because we did it. A video would launch flat, and instead of cutting it we'd tell ourselves "give the algorithm another day to learn." Another day became another $200. Multiply that across a dozen tests a month and the bleed is real.
The fix wasn't a better ad. It was removing the decision from the moment. A kill/scale rule is a contract you sign with yourself while you're calm, so that when a creative is dying you're not negotiating with your own sunk cost. You just read the number and act.
The rule, stated in one line
Here's the version we run, and the thresholds we personally use. Treat the numbers as a starting point to tune to your margins, not gospel:
- Kill when a creative has spent 1x your target CPA with zero purchases, or when it clears the spend gate but sits below your hook-rate and CTR floors.
- Keep testing when the leading indicators (hook rate, CTR) are strong but CPA is borderline — the ad is working, the offer or landing page might not be.
- Scale only when a creative beats target CPA and holds it for two consecutive days, then raise budget 20-30% per day, not 3x overnight.
The exact thresholds we test against
We judge every new creative on a spend gate first, then three signals. These are the floors we use on paid social for a mid-ticket ecommerce product; adjust them to your own CPA and AOV.
| Signal | Kill if | Scale signal | What it tells you |
|---|---|---|---|
| Spend gate | Spent 1x target CPA, 0 sales | Sale before hitting 1x CPA | The hard stop that caps a loser's damage |
| 3-sec hook rate | Below ~25% | Above ~35% | Whether the first 3 seconds actually stop the scroll |
| Link CTR | Below ~1% | Above ~1.5% | Whether the ad earns the click |
| CPA vs target | Over 1.5x target after the gate | At or under target, held 2 days | The bottom line — does it make money |
The spend gate matters more than any single metric. It converts "should I kill this?" into "has it spent its allowance?" — a question with a yes/no answer instead of an argument.
The three-signal diagnosis: where a creative actually dies
A dead creative isn't just "bad." It failed at a specific stage, and the signal tells you where — which is what lets you fix the next one instead of guessing.
- Low hook rate means the opening lost people. The problem is your first 3 seconds: the visual, the pattern interrupt, the first line of the script. Everything downstream is irrelevant if nobody stays.
- Good hook, low CTR means you held attention but didn't earn the click. The middle of the ad or the call to action is soft.
- Good CTR, bad CPA means the ad did its job and the sale broke somewhere else — offer, price, landing page, or audience. Don't kill the creative for a landing-page problem.
This is why we don't kill on CPA alone. CPA is the verdict; hook rate and CTR are the autopsy. Reading all three turns every dead test into a note for the next batch.
Kill fast, scale slow
The asymmetry is the whole point. Killing is cheap and reversible — you can always relaunch a variant. Scaling is where accounts blow up, because ad platforms punish sharp budget jumps with worse delivery and a reset learning phase.
- Kill same-day the moment a creative trips the gate. No "sleep on it."
- Confirm winners over 2 days before scaling — one good day is often noise.
- Raise budget 20-30% per day on a confirmed winner, and watch CPA at each step. If it slips past target, hold the budget where it was.
- Refresh before fatigue. A scaled winner decays. Have the next batch of variants queued so you're never scrambling when frequency climbs.
The rule is worthless without creative volume
Here's the catch nobody tells you: a kill/scale rule kills most of your ads. That's correct behavior — most creatives lose. But it only works if you have enough at-bats. If you make one video a week and the rule kills it Tuesday, you're dark until next Tuesday. The rule needs a pipeline feeding it.
That's the exact problem we built Polaris to solve. Paste a product image or a link and Polaris generates UGC-style video and image ads in minutes, routing each shot through multiple top models — Google Veo, Kling, Seedance, Nano Banana — so you get real variety instead of ten clones of the same idea. When a test dies, you don't wait on an editor; you ship the next five variants the same afternoon.
- Volume of angles. Generate distinct hooks and formats fast, so the rule always has fresh creatives to judge.
- One-click Auto-Captions. Word-accurate captions burned in, in TikTok-native styles (TikTok, Hormozi, Beast, Neon) — the caption style itself becomes a testable variable.
- Recreate winning ads. Paste a reference that's working and get your own version, so scaling a winner means spinning off variants, not starting from zero.
- Works inside Claude. Our MCP connector lets you brief and generate ads right in chat, so producing a test batch is a conversation, not a project.
Why teams pick Polaris
Polaris is built for one job: making ecommerce ad creative at the volume a real testing rule demands. It's not a general video tool you bend into shape — it's an AI ad studio for Shopify-style stores. You get AI video ads from an image or prompt, AI product shots, burned-in auto-captions, and one-click recreations of ads that already work, all on credit-based pricing (Pro is $49/mo for 1,500 credits, and cheaper on 3-month and 6-month terms). The point isn't to replace your kill/scale rule — it's to keep it fed so the rule can actually do its job.
Set your thresholds, then give them enough creatives to matter. Create your first ad and start feeding the pipeline, or see pricing to find the plan that matches your testing volume.