Ad Performance Lab

Ads Manager shows every number and decides nothing. Small advertisers end up calling winners after two days, keeping fatigued ads running, and planning the next creative from gut feel.

This Claude Code plugin reads the CSV you export from Meta, TikTok, Pinterest or another Ads Manager. The report breaks it down by campaign, ad set or ad, with spend, CPM, CTR, CPC, conversions, CPA, ROAS and frequency. It flags what's beating or missing your target CPA or ROAS, what's too early to judge, creatives with a CTR under half the account's median, and frequency over 3. The A/B test checks whether a difference in CTR or conversion rate is real, with a two-proportion z-test, the confidence interval, and how many more impressions or clicks each variant needs if it isn't yet. The test log checks every recorded result against the statistics, so a "win" with p = 0.76 gets caught, and it collects your learnings by variable. Budget suggestions scale winners and cut losers in 20 to 30% steps, evened back to your daily total. The brief command turns all of it into one clear hypothesis, three concepts and the sample size to decide it.

Inside: 4 commands (/ads-report, /ab-test, /test-log, /creative-brief), a skill with the rules and how to read a test, the helper (Node.js, no packages), a brief template, and a sample export, test log and brief with hand-checked outputs.

For shop owners, freelancers and social media managers running paid social on small budgets.

Limitations: it doesn't connect to ad platforms or change budgets. Conversions are as the platform attributes them. Suggestions follow stated rules and aren't forecasts.