The weekly marketing report is where good marketers quietly lose an afternoon. Exporting from Meta, pulling Google Ads, copy-pasting into a deck, writing the same ‘here’s what happened’ narrative — every single week. Here’s the automation we use to make it run itself with n8n (Make is nearly identical if that’s your tool).
The flow at a glance
- Trigger — a Schedule node fires every Monday at 8:00am.
- Pull — Meta Ads and Google Ads nodes fetch last week’s spend, impressions, CTR, and conversions.
- Store — append a row to a Google Sheet to keep a running history for trends.
- Summarize — an AI node turns the raw numbers into a plain-English recap.
- Deliver — post to Slack and email the client/stakeholder.
Step 1 — The schedule trigger
Add a Schedule Trigger node. Set it to weekly, Monday, 08:00, in your reporting timezone. Timezone matters more than people expect — a mismatch here is the #1 reason reports show the wrong week.
Step 2 — Pull the ad data
Add the Facebook Graph API and Google Ads nodes. Query the insights endpoints for a last_7d date range and request only the fields you’ll actually report: spend, impressions, clicks, CTR, conversions, cost-per-conversion. Requesting fewer fields keeps you well under rate limits.
Step 3 — Store the history
Add a Google Sheets → Append node. One row per week per platform. This is what lets you later say ‘CTR is up 18% month-over-month’ instead of reporting a single week in a vacuum.
Step 4 — The AI summary (the part that matters)
Raw numbers don’t drive decisions — interpretation does. Add an AI node and pin it to the exact fields so it can’t hallucinate figures:
You are a performance marketer. Here is last week’s data as JSON: [DATA]. Write a 5-bullet summary for a client: what performed, what underperformed, one number that matters most, one risk, and one recommended action for next week. Use ONLY the numbers provided. No fabricated metrics.
The ‘use ONLY the numbers provided’ line is non-negotiable — it’s what keeps the model honest.
Step 5 — Deliver it
Add a Slack node to post the summary to your team channel and an email node for the client. Now the report lands before you’ve had coffee.
The gotchas nobody mentions
- Rate limits — batch requests and cache; don’t loop one call per campaign.
- Timezones — set them explicitly on the trigger AND the data query.
- Hallucinated numbers — always pin the AI to provided fields; spot-check the first runs.
- Silent failures — add an error-trigger that pings you if any node fails, so you’re not emailing a blank report.
Variations worth building
Once the core flow works, fork it: a daily spend-anomaly alert, a monthly rollup, or a version that drafts the client email in their preferred tone. Build the engine once; the variations are 20 minutes each.
Build it once, run it forever
The whole point of automation isn’t fancy — it’s boring reliability. A report that runs itself every Monday, for months, without you touching it, is worth more than a flashy dashboard you have to remember to open.
Skip the setup. The AI Marketing Automation Pack gives you 100+ tested prompts, ready-to-import n8n & Make workflows, and custom GPT & Claude agent configs — documented, with real examples. Instant download, lifetime access.