The Problem
A content agency with 3 editorial staff produces 4 blog posts a month for B2B clients. Each article takes 5 hours on average: 2 hours of research, 1.5 hours writing the first draft, 1.5 hours of revision and finalization.
The problem: the editorial team is the bottleneck. Clients want more content, but research and first drafts eat up too much valuable editorial time — time better spent on quality editing and strategic planning.
- 4 posts/month — capacity limit reached, clients want more
- 5h per post — 2h of it research, which an AI agent could handle
- First drafts are rough in quality — costing 1.5h the editors would rather spend elsewhere
- Author voice is decisive — clients pay for a specific tone of voice, not for generic text
The Solution
A three-stage AI pipeline that automates research and first drafts while editors keep creative control:
- 1BriefingEditor defines topic, audience, core message and source preferences (5 min.)
- 2AI researchResearch agent gathers facts, studies and quotes, builds a structured research document (automatic, 10 min.)
- 3AI draftWriting agent produces a first draft following the given outline and tone-of-voice guidelines (automatic, 5 min.)
- 4Author reviewEditor revises, refines and adds a personal perspective (2h instead of 5h)
- 5PublishFinalized post uploaded to the CMS, SEO metadata added (15 min.)
Stack: Claude API for the research agent and writing agent, Google Docs API for workflow integration, WordPress REST API for automatic draft upload.
Results
| Metric | Before | After | Improvement |
|---|---|---|---|
| Time per post (research) | 2 hours | 15 minutes | 87% less |
| Time per post (draft) | 1.5 hours | 20 minutes | 78% less |
| Time per post (total) | 5 hours | 2.5 hours | 50% less |
| Output volume/month | 4 posts | 8 posts | ×2 with the same team |
| Client quality rating | 8.2/10 | 8.7/10 | +6% higher |
Learnings & Best Practices
- Tone-of-voice guidelines are critical: Without detailed style specs the writing agent produces generic text. With a one-page tone-of-voice document per client it nails the voice about 80% of the time.
- Editor as quality control, not as a typewriter: The shift from "I write" to "I shape and improve" is mentally demanding. Plan for a training phase.
- AI research needs source control: The research agent always outputs source links. The editor manually checks 2–3 critical facts. Zero-error policy.
- Quality rises when the load drops: When editors do less repetitive work, their energy flows into creativity — and it shows in the client rating.
The goal was never to replace editors. The goal was to free them from the research treadmill — so they can do what they truly do best: tell stories.
This case study is illustrative. The metrics shown are based on typical implementation experience and industry averages. It is not a real customer project and uses no real customer data.