Ads Performance
by
A weekly paid media review for ecommerce teams across Meta, Google Ads and Shopping, TikTok and Pinterest: which campaigns to cut, scale or refresh, judged on Polar's first-party numbers instead of what the platforms report.
Every Monday someone opens Meta, Google, TikTok and Pinterest side by side, sorts by ROAS and makes the same mistakes: scaling brand search because it shows 12x, comparing retargeting with prospecting, and trusting conversions each platform claims for itself. This skill runs the review the way a senior media buyer would. It reads your Meta, Google Ads and Shopping, TikTok and Pinterest data through Polar, files every campaign into a segment (prospecting, retargeting, brand, non-brand, Shopping), judges it only against its own segment, checks platform ROAS against Polar's pixel attribution and new-customer CAC, and prices every recommendation in euros. It recommends; it never touches your ad accounts.
What It Does
- Splits campaigns by platform and segment: Meta, TikTok and Pinterest prospecting and retargeting, Google brand, non-brand and Shopping, and benchmarks each one inside its own segment.
- Puts platform-reported ROAS next to Polar pixel ROAS (First Click, Cash by default) and new-customer CAC, and flags campaigns the platform over-reports.
- Calls Cut, Scale, Watch or Hold per campaign against a break-even ROAS floor, with minimum sample sizes so two orders never trigger a cut.
- Treats brand search apart: Protect or Trim, never Scale, and catches broad brand campaigns cannibalising your exact-match ones.
- Reads creative fatigue per ad over four weeks: CTR decay against CPM, so you know which asset to refresh first.
- Prices every move: net monthly gain of each cut and the revenue a reallocation adds at current ROAS.
Use Cases
- Run the Monday paid media check in two minutes instead of a morning in four ad managers.
- Find the prospecting campaigns burning budget without new customers before the month closes.
- Stop scaling brand search on a ROAS that platform attribution inflates.
- Escalate campaigns where the platform claims sales Polar can't see, instead of cutting them blind.
Example Session
/polar-ads-performance Run my ads review for last week.
Ads review · last week vs previous week
Data health: OK, through yesterday · Meta, Google, TikTok, Pinterest · Attribution: Polar pixel, First Click · Cash, vs platform-reported
• Meta reports 1.6x ROAS; Polar's pixel attributes 0.5x. Most of the gap sits in prospecting.
• The whole non-brand Google segment is below break-even on Polar numbers: the problem is the segment, not one campaign.
• Biggest move: cut ASC US and PMax US, about +€3.8k a month. No campaign clears the Scale bar, so hold that budget.
- Segment | Campaign | Verdict | Platform | Polar | € / month
- Meta prospecting | ASC · US | Cut | 0.3x | 0.0x | +€3.1k
- Google non-brand | PMax · US | Cut | 0.5x | 0.2x | +€0.7k
- Google brand | Brand broad · DE | Trim | 1.5x | 0.6x | +€1.0k
- TikTok prospecting | Spark Ads · FR | Refresh | 1.9x | 1.1x | n/a
- Pinterest prospecting | Catalog · US | Watch | 1.2x | 0.9x | n/a
- Meta retargeting | Retargeting · ROW | Flag for support | 4.4x | 0.0x | n/a
- Google brand | Brand exact · DE | Protect | 9.8x | 4.6x | n/a
Brand exact looks like your best campaign at 9.8x, but Polar attributes it 4.6x: that demand was already yours, so protect it, don't scale it. Brand broad pays 2.4x the CPC for the same searches and returns 0.6x: it's cannibalising the exact campaign.
TikTok Spark Ads still clears break-even on Polar numbers, but CTR is down 38% in four weeks at a flat CPM: refresh the creative before touching the budget. Pinterest Catalog sits just under break-even on 41 orders, below the sample size for a cut, so it stays on Watch.
⚠ Retargeting ROW: the platform claims 4.4x, Polar sees no attributed orders for two weeks. Worth a tracking check before any cut.
Assumptions: 7 full days ending yesterday · break-even 1.0x · averages computed per segment · read-only, nothing changed in your ad accounts.
Install It
- Copy the command.
One line in your terminal. Free, no account needed. - Run it.
npx skills add pulls the skill from Polar's GitHub and installs it for the agents you pick: Claude Code, Cursor, Codex and more. - Connect Polar and ask.
Connect the Polar connector, then ask for your ads review. The skill does the rest.
skills folder
Copy
npx skills add \
https://github.com/polar-analytics/claude-skills \
--skill polar-ads-performanceFrequently asked questions
How is this different from reading ROAS in Ads Manager?
Each ad manager shows what its own platform claims. This skill puts that next to what Polar's pixel attributes to the same campaign, judges each campaign only against its own segment, and prices every move in euros. The gap between the two ROAS is usually the most useful number of the week.
Does it change anything in my ad accounts?
No. It only reads your data through Polar's MCP. Every cut, scale or trim is a recommendation you apply yourself, and the report never claims a budget was changed.
Which attribution model does it use?
Polar's pixel with First Click and cash accounting by default, the same as Polar's Attribution screen. You can ask for another model, and the model used is always stated in the report.
Why does it never scale brand search?
Brand search mostly captures people who were already looking for you, so its ROAS says little about incremental revenue. The skill rates it Protect or Trim, and flags broad brand campaigns that pay more for the same searches.
What data does it need?
Any mix of Meta, Google Ads (Search, Shopping and PMax), TikTok and Pinterest connected to Polar. The Polar pixel adds the first-party layer; without it the skill still runs on platform numbers and says so in the headline.
Can I change the thresholds?
Yes. Give Claude your target CPA, ROAS floor or break-even margin, or store them in Polar's memory, and every run uses them instead of the defaults.
An ads review is worth what your attribution is worth.
Polar joins Meta, Google, TikTok, Pinterest and Shopify with first-party pixel attribution, so the skill judges each campaign on the orders it really drove, not on what the platform claims.
<script type="application/ld+json">{"@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [{"@type": "Question", "name": "How is this different from reading ROAS in Ads Manager?", "acceptedAnswer": {"@type": "Answer", "text": "Each ad manager shows what its own platform claims. This skill puts that next to what Polar's pixel attributes to the same campaign, judges each campaign only against its own segment, and prices every move in euros. The gap between the two ROAS is usually the most useful number of the week."}}, {"@type": "Question", "name": "Does it change anything in my ad accounts?", "acceptedAnswer": {"@type": "Answer", "text": "No. It only reads your data through Polar's MCP. Every cut, scale or trim is a recommendation you apply yourself, and the report never claims a budget was changed."}}, {"@type": "Question", "name": "Which attribution model does it use?", "acceptedAnswer": {"@type": "Answer", "text": "Polar's pixel with First Click and cash accounting by default, the same as Polar's Attribution screen. You can ask for another model, and the model used is always stated in the report."}}, {"@type": "Question", "name": "Why does it never scale brand search?", "acceptedAnswer": {"@type": "Answer", "text": "Brand search mostly captures people who were already looking for you, so its ROAS says little about incremental revenue. The skill rates it Protect or Trim, and flags broad brand campaigns that pay more for the same searches."}}, {"@type": "Question", "name": "What data does it need?", "acceptedAnswer": {"@type": "Answer", "text": "Any mix of Meta, Google Ads (Search, Shopping and PMax), TikTok and Pinterest connected to Polar. The Polar pixel adds the first-party layer; without it the skill still runs on platform numbers and says so in the headline."}}, {"@type": "Question", "name": "Can I change the thresholds?", "acceptedAnswer": {"@type": "Answer", "text": "Yes. Give Claude your target CPA, ROAS floor or break-even margin, or store them in Polar's memory, and every run uses them instead of the defaults."}}]}</script>