Meta & Facebook Ads
4 min read
You check 40 ad sets every Monday and still guess wrong on half of them. The fix isn't a better dashboard, it's a decision rule: every active ad set gets exactly 1 verdict each week, scale, hold, or kill, based on thresholds you set once.
We've packaged that rule into a free Claude skill called scale-hold-kill-verdict. You add it to Claude, paste your Ads Manager export, and get 1 call per ad set with the exact numbers that triggered it.
This post covers what the skill checks, the verdict logic behind it, and how to run your first weekly pass in about 10 minutes.
Key takeaways
The skill gives every active Meta ad set exactly 1 verdict per week, scale, hold, or kill, each backed by the specific numbers that triggered it.
A scale verdict requires ROAS at 1.3x breakeven for 7 consecutive days, frequency under 2.5, and spend at 75%+ of budget. 1 good day is not a streak.
Ad sets still in Meta's learning phase auto-hold, so nothing gets killed before it has enough data to be judged fairly.
Nothing pauses or scales without your yes on each item. The skill decides, you approve, then you execute.
What the scale hold kill verdict skill does
SCALE HOLD KILL
The skill runs a weekly decision pass on your Meta account. It pulls the last 7 days and the 7 days before that for every active ad set, applies fixed thresholds, and outputs 1 verdict per ad set. No verdict ships without its evidence: you see the ROAS, the streak length, and the frequency number behind every call.
The part most media buyers miss: hold is a real verdict, not indecision. Most ad sets, most weeks, should hold. The skill treats it that way, and it tells you why each hold is a hold, whether that's learning phase, inconclusive data, or stable performance at breakeven.
For every scale, it states the next budget step in increments of 20% or less. For every kill, it states what that freed spend should fund instead.
How the verdict logic works
A scale needs all 4 conditions at once: ROAS at or above 1.3x your breakeven for 7 consecutive days, CPA stable or falling week over week, frequency under 2.5, and spend already at 75% or more of budget. That last check matters because headroom has to be real. If every metric passes but the ad set isn't spending its budget, raising the budget does nothing, so the verdict flips to hold and the skill points you at creative instead.
A kill needs any 1 of 3 triggers: ROAS below breakeven for 14 days after exiting learning, CPA rising 2 consecutive weeks while below breakeven, or zero conversions past your wasted spend threshold.
Everything still in Meta's learning phase auto-holds. Meta's own guidance in the Business Help Center says an ad set needs roughly 50 optimization events in a 7 day window before delivery stabilizes, and significant edits reset that clock. Judging an ad set mid-learning means killing things that never had a fair shot, so the skill refuses to.
When the verdict is hold, the fix is creative
Underdelivering ad sets don't need more budget, they need fresh ads. Vibelets generates on-brand product ads from your store in minutes, no designer or brief required.
How to set it up in Claude
Setup takes 3 steps. First, download the skill file from the skill's Google Drive folder, free, no email gate. Second, add it to Claude: upload it as a skill under Settings, or simply attach the file to a Claude project or chat. Anthropic's Help Center guide to skills covers the upload path if you haven't used skills before.
Third, run your first pass. Export your ad sets from Ads Manager with spend, results, CPA, ROAS, frequency, and CTR for the last 7 and prior 7 days, paste it in, and ask Claude for the weekly verdicts. The skill will ask for your breakeven ROAS or target CPA once, then apply it to everything.
If you've connected Claude to your Meta account through an MCP connector, the skill pulls the numbers itself and can execute approved kills for you. But the paste workflow needs no integration at all.
Run it weekly, not daily
The skill is built as a weekly pass because the thresholds are weekly: 7 day streaks, 14 day kill windows, week over week CPA trends. Running it daily just tempts you to act on noise, which is how winners get killed and losers get scaled.
Pick a day, run the pass, approve the calls you agree with, and leave the rest alone. Holding is a decision too, and it's the one most accounts under-use.
Conclusion
1 verdict per ad set, once a week, with the numbers behind every call. That's the whole system, and the skill runs it for you. Download it, run your first pass this week, and see how many of your gut calls it overturns.
Every ad set. 1 verdict. Every week.
Free Claude skill, yours in 1 click.
Frequently asked questions
How often should I run the scale hold kill pass?
Once a week, same day each week. The verdict logic is built on 7 and 14 day windows, so a weekly cadence matches the data. Daily runs add noise, not signal.
Do I need a Meta Ads integration for this skill to work?
No. Paste mode works with a standard Ads Manager export containing spend, results, CPA, ROAS, frequency, and CTR for the last 7 and prior 7 days. A Meta Ads MCP connector is optional and only adds live pulling and 1 click execution.
What if I don't know my breakeven ROAS?
The skill asks for it once and won't guess. If you've never calculated it, the margin-truth-calculator skill from this same giveaway series computes true breakeven ROAS per product from your COGS, shipping, fees, and returns.
Can the skill pause or scale ad sets on its own?
No. Every verdict waits for your explicit yes per item. In paste mode you execute changes in Ads Manager yourself; in live mode Claude only acts on the specific items you approve.
Why did my best performing ad set get a hold verdict?
Usually because it's underdelivering: the metrics pass but spend sits below 75% of budget, which means budget isn't the constraint. More budget won't help an ad set that can't spend what it has. The fix is fresh creative, not a raise.
Sources
Make ads that sell
your Products.
Generate on-brand product ads in minutes, no designer or brief required.
Trusted by 400+ growing brands