ZQ Zaid Qayyum
Case 06 AI Workflow2025Sample build Representative scenario: 12-person marketing agency, US

n8n + Claude/OpenAI content ops for a marketing agency

Sample build: an end-to-end content production pipeline for a representative 12-person marketing agency. Briefs in Notion trigger AI-assisted drafts, route through human review, and publish across destinations, all orchestrated in n8n with Claude and OpenAI doing the heavy lifting.

Scenario
Representative 12-person marketing agency, US
Role
AI Workflow Engineer
Year
2025
Hero stack
n8n · Claude / OpenAI

The scenario.

A 12-person agency producing roughly 80 pieces of content a month across 14 client accounts. The bottleneck isn't writing per se; it's everything around writing: turning loose briefs into outlines, applying each client's tone of voice, moving drafts through review, formatting for different channels and tracking what shipped where. Writers spend more time on logistics than writing, and senior editors rubber-stamp work just to keep the pipeline moving.

How it's built.

  1. Notion voice cardsEach client's voice is codified as a structured "voice card": tone, banned phrases, preferred sentence shapes, examples. Stored in Notion, fed into prompts.
  2. n8n workflowThe n8n workflow: Notion brief → outline generation (Claude) → first draft (Claude or OpenAI based on content type) → tone-of-voice pass → human-review queue in Slack.
  3. Slack review queueReviewer accepts/rejects/edits in Slack threads; n8n watches for the accept reaction and moves to the publish stage.
  4. Publish routingPublish stage routes the final piece based on type: blog → CMS API, email → ESP, social → schedulers, internal → Google Drive.
  5. Human fallbackFallback logic: if any AI step times out or returns low-confidence output, the brief routes to a human writer with the partial draft attached.

What makes it tick.

  • Voice cards make AI output feel like the agency's writers wrote it, not like generic ChatGPT.
  • Cost-aware model routing: short formats use cheaper or faster models, long-form uses Claude with more context.
  • Every piece carries provenance metadata: which model, which prompt version, which reviewer approved.
  • Slack-native review interface means writers and editors don't have to learn another tool.

What it's designed to change.

This pattern is designed to roughly triple output capacity without hiring (~80 → 240 pieces per month) by moving writer time from logistics back to editorial judgment, with voice cards holding tone so quality doesn't slide as volume climbs. The commercial point: an agency stops pitching "we'll write 10 pieces a month" and starts pitching content velocity as a service. These are design targets for the modeled scenario, not measured client results.

design target 80 → 240 pieces per month
design target roughly 3x output capacity, without hiring

Tools in the build.

  • n8n
  • Anthropic API (Claude)
  • OpenAI API
  • Notion API
  • Slack API
  • Google Drive API
  • Webhooks

Got a workflow bottleneck?

If this looks like the kind of system your business needs, the fastest way to start is a short discovery call.

Book a discovery call