Analyze ad data actually retrieved from a report file or the current connection. This skill folder contains no automated analysis execution code, so report only the scope in which you read the materials and performed the necessary calculations and verification. Requests to create or change ads are routed to marketing-meta-ads-manager.
Methodology reference: AgriciDaniel/claude-ads (MIT, repository root NOTICE §1.3). Do not treat the original project's check counts, scores, or tool executions as having been performed by this skill.
Materials and permissions
- Check the actual sheets, headers, rows, period, currency, and time zone of report files such as XLSX and CSV provided by the user.
- If the official Meta MCP is connected, query with the read tools and account permissions discovered in this session. If there is no connection, perform file analysis only.
- Check the campaign objective, main conversion action, attribution settings, and the period to compare against. Do not force product category or operational philosophy surveys that are not necessary for interpretation.
- Separate materials that mix multiple ad accounts or clients by account.
Aggregation rules
- Identify the unit of the original rows. Do not double-sum ad, ad set, and campaign totals with breakdown rows such as placement and age.
- Sum additive raw figures such as cost, impressions, clicks, and conversion value within the same unit and period.
- Recompute CTR as total link clicks / total impressions, CPC as total cost / total link clicks, CPA as total cost / relevant conversions, and ROAS as total attributed conversion value / total cost. If the denominator is missing or zero, mark it as not calculable.
- Check whether click type, conversion event, attribution window, and currency differ by file. If they differ, do not compare them directly in the same table, or state the differences explicitly.
- If there are separate placement tables and age/gender tables, analyze only the marginal distributions of each table. Even when matched to the same ad and period, placement × age × gender cross values cannot be reconstructed. Cross analysis requires an original source that has the necessary dimensions together in the same row.
- Zero purchases is an observed result. Do not conclude leakage or a reason to stop without checking for low impressions, a short period, or possible tracking errors.
Analysis and recommendations
Select only the metrics and breakdown dimensions that fit the report objective. Separate observed figures, interpretive hypotheses, additional checks, and possible actions. Do not use fixed ROAS tiers, daily budgets set as multiples of product price, learning status determined solely by weekly conversion counts, or creation dates inferred from creative names. Present scenario figures as examples only when there are actual assumptions and formulas, and do not call them statistical confidence intervals.
Ad changes are not automatically executed based on analysis results. After checking the account, target, budget, and status, connect via the manager skill within the scope requested by the user.
Deliverables
Create a summary or report in the requested format. If HTML, DOCX, or PPTX is needed, perform actual generation and open verification using the installed document creation tool. Do not mark formats as complete if they have not been generated. In the default Markdown report, include data sources, period and units, calculation formulas, verified results, limitations, and items for further verification. If a descriptive review skill is installed, use it and cross-check numbers and meaning against the original.
Reference Documents
| Appendix | Contents |
|---|---|
| references/A.md | Audience and reporting purpose |
| references/B.md | Fields and aggregation units |
| references/C.md | Product and campaign context |
| references/D.md | Possible analysis items |
| references/E.md | Handling multiple files |
| references/F.md | Writing recommendations |
| references/G.md | Output format |
| references/H.md | Data consistency |
| references/I.md | Limitations and sources |
| references/J.md | Linking adjacent skills |
| references/K.md | Example data for verification |