How I automated weekly client reporting for a 3-person agency

The Problem

A 3-person digital marketing agency was spending every Friday afternoon building client reports. The process:

1. Export Google Analytics data for each client (manual, 15 min/client)
2. Pull revenue numbers from Stripe (manual, 10 min/client)
3. Check project tracker for deliverables completed (manual, 10 min/client)
4. Copy-paste into a Google Doc template (manual, 20 min/client)
5. Review, fix formatting, send (15 min/client)

With 5 clients, that's roughly 4 hours every Friday. One person's entire afternoon, every single week.

The Solution

We built a Claude Code automation with a single CLAUDE.md that:

1. Reads CSV exports from all three sources (dropped into a folder)
2. Matches data across sources by client name
3. Generates a formatted markdown report per client
4. Flags any metrics that changed more than 20% week-over-week

The folder structure

~/client-reports/
  CLAUDE.md
  /data/
    analytics_export.csv
    stripe_export.csv
    project_tracker.csv
  /output/
    (reports appear here)

The workflow now

1. Friday morning: export the 3 CSVs (still manual, ~5 min total)
2. Drop them in the /data/ folder
3. Run claude in the terminal and say "Generate this week's client reports"
4. Review the 5 reports in /output/
5. Send to clients

Total time: ~15 minutes instead of 4 hours.

What Made It Work

The key was being very specific in the CLAUDE.md about:

• Exact column mappings ("revenue is in the 'amount' column in Stripe")
• The report format the clients expected
• What counts as a "flag" worth highlighting
• File naming conventions

Results

• Time saved: 3.75 hours/week (195 hours/year)
• Consistency: Reports look identical every week
• Catch rate: The 20% change flag caught a tracking error in week 2 that would have gone unnoticed