← All use cases

Use case B25 · Business

Three-person marketing agency generates weekly client ad reports with a Claude Code command

Chris Luna, EBCD ·

The short version

What they did

The agency built a Claude Code command that pulls each client’s Meta Ads data through the API and produces a weekly report with spend, leads, cost per lead, trend analysis and recommendations. Someone reviews it, adjusts the commentary and sends it. The post also covers prospect scraping and video ad generation. The post ends by offering to help other agencies build the same setup.

Result (as claimed)

“Client Reporting: 4 Hours to 45 Minutes.” Before: “Four hours minimum for 13 clients.” After: “Total time per client: under 4 minutes.”

Prompt

I run a marketing agency and need a weekly Meta Ads report for [client]. Their goals and reporting rules are [paste], and our model report is [attach]. Use [connected read-only ad data or attached exports] for [last week’s dates] and the four weeks before it.

1. Compare spend, leads, cost per lead and click-through rate by campaign and ad set. Show calculations and how the comparison period is defined.
2. Write a one-page client report: headline numbers, important changes and three recommendations for next week.
3. Separate observed changes from possible explanations, and include a section for numbers you couldn’t obtain.

Match the model report’s plain, professional tone. Don’t invent missing figures or state guessed causes as facts. Keep it as a draft for my review; don’t send it or change campaigns.

Do it yourself

These steps are my suggested way in, not EBCD’s setup.

Level: Beginner with exported data in a chat. Pulling the data automatically with a coding agent is Advanced.

What you need

  • A chat assistant that can read spreadsheets, such as Claude or ChatGPT.
  • A weekly export from your ad platform for each client.
  • One past report you were happy with, as a model.
  • For the full version: an AI coding agent such as Claude Code, and API access to the ad platform.

Start simple

  1. Export last week’s results and the four weeks before for one client.
  2. Upload them with your model report. Paste the prompt above.
  3. Edit the draft. Keep a list of what you changed and why.
  4. Add those changes to the prompt as rules. Run it on a second client.
  5. Time how long the review takes compared with writing it yourself.

Take it further

  • Have a coding agent build a command that pulls each client’s data and drafts the report.
  • Save one rules file per client with their goals and what they care about.
  • Add a “numbers I could not get” section that must appear in every report.

Check your work

  • Check the headline numbers against the ad platform’s own dashboard before sending.
  • Read the “why it changed” lines closely. The AI guesses at causes. Keep only what you can support.
  • Client data and ad account access are sensitive. Use read-only access, and have a person send every report.

Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Prompt adapted for this library, inspired by the case. Business size: 13 active clients, a 3-person core team. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.

More in Operations

New cases go out every Saturday in Inside AI.