Open the CSV in Excel, or import it into Google Sheets and choose comma-separated values. Save your working copy in a controlled location. The CSV contains the rows below and space for your notes; it has no macros, formulas or live account connection. The separate Excel workbook contains editable example inputs and formulas that update the forecast.
How to use the template
Install the skill
Copy the SKILL.md text into a file called SKILL.md inside a folder called meta-ads-audit. In Claude, switch on code execution, zip the folder and upload it under Customize > Skills > Create skill > Upload a skill. In Claude Code, save it as ~/.claude/skills/meta-ads-audit/SKILL.md.
Export one date range at ad level
In Ads Manager's Ads tab, add amount spent, impressions, link clicks, purchases, purchase value, frequency and 3-second video plays, then export. Use the same date range every week.
Ask for the audit
Attach the export and ask Claude to audit your Meta ads. Give it your average order value, contribution margin and the profit you want to keep. It calculates with code and returns the account grade, every ad's grade and three actions.
Check it against the workbook
Paste the same export into the workbook's Ads tab. It runs the identical rubric with formulas, so the grades should match Claude's exactly. Act on F ads first, then scale the A ads one step.
Preview the worksheet
Every row shown here is included in the download. Scroll the table horizontally on a small screen.
| Rule | How it works | Default |
|---|---|---|
| Break-even ROAS | 1 / contribution margin before ads | From your margin |
| Target ROAS | 1 / (margin - profit you keep) | From your margin |
| Data rule | Don't grade an ad until it has spent N x target CPA | N = 3 |
| Hard stop | Zero purchases after N x break-even CPA | N = 3 |
| Profit (40 points) | ROAS vs target, break-even and 80% of break-even | 40 / 25 / 10 / 0 |
| Conversion (15 points) | Purchases per link click vs the account | 1.2x / 0.9x / 0.6x = 15 / 10 / 5 |
| Click (15 points) | Link CTR vs the account | Same steps |
| Hook (15 points) | 3-second plays / impressions vs the account's video ads | Same steps; video only |
| Freshness (15 points) | Frequency | 2.5 or less = 15, up to 3.5 = 8, above = 0 |
| Grade bands | Score out of 100 | A 85+, B 70, C 55, D 40, F below |
| Profit cap | Below break-even ROAS can't score above D | Always on |
| Account grade | Spend-weighted average score of graded ads; capped at D if total ROAS is below break-even | Always on |
| Actions | A scale ~20% then wait 3 days, B keep, C fix weakest part, D cut or replace, F pause, N/A wait | Starting points |
Worked review example
Fictional store: $60 order value, 40% margin, keeping 10% profit, so break-even ROAS 2.50, target ROAS 3.33 and target CPA $18. Ten ads spent $5,045 and returned $15,366 (ROAS 3.05). Three ads scored A, one B, one C, one D, two F, and two hadn't spent the $54 needed to be graded.
The account grades C (score 68): profitable, but 22.9% of spend ($1,155) sits in D and F ads. The 20% off banner scored 47 points on clicks and freshness but its 2.02 ROAS is under break-even, so the profit cap holds it at D. The three actions: pause the two F ads, raise the three A ads about 20% each, and replace the discount banner.
What this template does and does not establish
The skill and workbook grade what the export shows. Purchases and revenue are Meta's attribution, not your bank account. Every threshold (3x target CPA, the 1.2x/0.9x/0.6x steps, frequency 2.5 and 3.5, the grade bands, the 20% scale step) is a starting point we chose, not a Meta rule or industry benchmark. The export is sent to Claude like any attachment; the skill makes no other connection and cannot change your ad account. Example data is fictional.
Primary references
- Anthropic: Equipping agents for the real world with Agent Skills
- Agent Skills specification
- The Decoder: Anthropic publishes Agent Skills as an open standard
- Claude Help Center: Use skills in Claude
- Claude Code docs: Skills
