Ad Set Manager
by
Ad set decisions for ecommerce teams: judge every set on blended CAC rather than platform ROAS, tell a learning phase from a losing one, and get a scale, cut or keep-learning call with the budget delta attached.
Ad set decisions get made inside the ad platform, on the platform's own definition of a conversion, which is how a set that looks profitable in Ads Manager turns out to be buying repeat customers you already had. The daily volume makes it worse: thirty or forty ad sets, each needing a call before the day's spend compounds, and the easiest call is always to leave everything running. This skill reads spend, purchases and revenue per ad set, computes acquisition cost on your blended definition rather than the platform's, checks whether a set has enough conversions to have left the learning phase, and returns a status and a budget delta for each one.
What It Does
- Pulls spend, impressions, purchases and revenue for every active ad set.
- Computes acquisition cost on a blended definition, not the platform's attributed one.
- Flags ad sets still inside the learning phase so they are not cut early.
- Compares each set against your CAC target and against the account average.
- Detects budget starvation and delivery caps that hide a working ad set.
- Returns a status and a budget delta per set: scale, cut, or keep learning.
Use Cases
- Run a morning pass over every ad set before the day's spend compounds.
- Stop cutting ad sets that are still learning and restarting the phase from zero.
- Find the sets quietly carrying the account so budget moves toward them.
- Hand a media buyer a ranked action list instead of a dashboard export.
Example Session
Review yesterday's ad sets against a $38 CAC target and tell me what to scale, cut, or leave alone.
Eleven ad sets spent yesterday. Four need a decision today, the rest are holding.
- Ad set | Spend | Purchases | CAC | Phase | Action
- Broad-Prospecting-01 | $420 | 14 | $30 | Exited | Scale +30%
- Lookalike-3%-Video | $310 | 5 | $62 | Exited | Cut
- Retarget-30d-Carousel | $180 | 9 | $20 | Exited | Scale +20%
- Interest-Outdoor-New | $95 | 3 | $32 | Learning | Keep learning
Lookalike-3%-Video has spent past its learning window and sits 63% above target, so that is a cut rather than a patience problem. Interest-Outdoor-New is only three conversions in: its CAC reads fine but the sample does not support scaling yet, so leave the budget flat until it exits learning.
Why This Beats Prompting It Yourself
Platform ROAS vs Blended CAC: The Hidden Risk
A generic prompt answers with whatever the ad platform reported. Ask which ad sets to cut and you get a ranking built on attributed conversions, counted inside a window the platform chose, with returning customers folded in. Two ad sets showing identical ROAS can carry acquisition costs that differ by half. The model has no way to tell, because a prompt changes the phrasing of the question, not the definition of the number it is answered with.
When You Still Need a Decision Rule
The moment a budget change has to survive a finance review, the CAC behind it has to be the one the business reports. This skill states $30 against a $38 target, on 14 purchases, past the learning phase, so scaling is a rule rather than a hunch. It also refuses to call an ad set at three conversions, which is the most expensive habit in daily ad set management.
Known Limitations
It reads what your ad accounts and store expose, so an ad set with broken conversion tracking is reported as unknown rather than scored on a guess. Blended CAC needs a revenue source connected alongside the ad platform: without one the skill falls back to platform-attributed numbers and says so in the output. Learning phase detection runs on conversion counts, so very low volume accounts sit unresolved longer than the platform dashboard implies.
Install It
- Download the ZIP.
It is free and there is no account to create. - Unzip it into your agent's skills folder.
Claude Code reads ~/.claude/skills/, which is hidden by default: the command in the folder block opens it. Other agents scan their own directory, so drop the same folder there instead. - Ask your agent to use it.
Restart the agent if it was already running, then it picks the skill up with no config.
skills folder
Copy
~/.claude/skills/ad-set-manager/
SKILL.md
references/cac-rules.md
reports/
# one report per daily pass
# macOS: create the folder and open it in Finder
mkdir -p ~/.claude/skills && open ~/.claude/skills
# Windows: paste in the Explorer address bar
%USERPROFILE%\.claude\skills\Frequently asked questions
How is this different from reading Ads Manager myself?
A dashboard hands you the platform's ranking of the platform's own numbers. This recomputes acquisition cost on your blended definition, checks the learning phase before it lets anything be cut, and ends on a budget delta per ad set. You get a decision with the arithmetic attached rather than a sorted table.
Which agents does the skill run in?
Any agent that supports the open SKILL.md format: Claude Code, Cursor, Codex CLI, GitHub Copilot, Gemini CLI, Manus, Grok and others load it unmodified. The format is portable, the location is not. Each agent scans its own skills directory, so you drop the same folder into whichever one yours uses. There is no config file to edit and no API key to provision.
What does it need connected to work?
The ad account for spend and delivery, and a revenue source for orders. With both, it reports blended CAC. With only the ad account it still runs, falls back to platform-attributed conversions, and labels every number as such so nobody mistakes one for the other.
How does it decide an ad set has left the learning phase?
By conversion count over the platform's learning window, not by how long the set has been live. An ad set that has run for two weeks on four conversions is still learning, and the skill will refuse to cut or scale it. Saying nothing is cheaper than acting on four data points.
Will it ever tell me to cut a good ad set?
It can, and the output shows you why. A set below target on a small sample gets held rather than cut, and a set above target with volume gets cut whatever its creative reputation. Where the evidence is thin the recommendation is explicitly a hold, not a quiet guess.
What is included with the skill?
The skill itself, the CAC rule reference, the daily report template, and the learning phase thresholds per platform. It is free and security scanned. We re-publish the ZIP when a platform changes its delivery rules, so download it again if the phase detection starts drifting.
Scale, cut or keep learning depends on which CAC you read.
Polar defines blended CAC once across every channel, so an ad set is judged on the cost your business carries, not the one the platform reports.
<script type="application/ld+json">{"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"How is this different from reading Ads Manager myself?","acceptedAnswer":{"@type":"Answer","text":"A dashboard hands you the platform's ranking of the platform's own numbers. This recomputes acquisition cost on your blended definition, checks the learning phase before it lets anything be cut, and ends on a budget delta per ad set. You get a decision with the arithmetic attached rather than a sorted table."}},{"@type":"Question","name":"Which agents does the skill run in?","acceptedAnswer":{"@type":"Answer","text":"Any agent that supports the open SKILL.md format: Claude Code, Cursor, Codex CLI, GitHub Copilot, Gemini CLI, Manus, Grok and others load it unmodified. The format is portable, the location is not. Each agent scans its own skills directory, so you drop the same folder into whichever one yours uses. There is no config file to edit and no API key to provision."}},{"@type":"Question","name":"What does it need connected to work?","acceptedAnswer":{"@type":"Answer","text":"The ad account for spend and delivery, and a revenue source for orders. With both, it reports blended CAC. With only the ad account it still runs, falls back to platform-attributed conversions, and labels every number as such so nobody mistakes one for the other."}},{"@type":"Question","name":"How does it decide an ad set has left the learning phase?","acceptedAnswer":{"@type":"Answer","text":"By conversion count over the platform's learning window, not by how long the set has been live. An ad set that has run for two weeks on four conversions is still learning, and the skill will refuse to cut or scale it. Saying nothing is cheaper than acting on four data points."}},{"@type":"Question","name":"Will it ever tell me to cut a good ad set?","acceptedAnswer":{"@type":"Answer","text":"It can, and the output shows you why. A set below target on a small sample gets held rather than cut, and a set above target with volume gets cut whatever its creative reputation. Where the evidence is thin the recommendation is explicitly a hold, not a quiet guess."}},{"@type":"Question","name":"What is included with the skill?","acceptedAnswer":{"@type":"Answer","text":"The skill itself, the CAC rule reference, the daily report template, and the learning phase thresholds per platform. It is free and security scanned. We re-publish the ZIP when a platform changes its delivery rules, so download it again if the phase detection starts drifting."}}]}</script>