A two-person support team kept answering the same questions because nobody had time to write documentation. We built a weekly pipeline that turns solved tickets into ready-to-review help-center articles — lifting self-serve deflection by 34% in four months.
Typical results
74
Live help articles
30 min
Weekly review time
Figures reflect typical results for this kind of automation.
The challenge
The help center had 19 articles, most outdated, while the same twenty questions consumed the support queue week after week. Writing documentation always lost to answering tickets — with two agents covering hundreds of conversations, nobody could spare an afternoon for the knowledge base. Customers who might have self-served opened tickets instead, feeding the backlog that prevented documentation from being written.
Before and after
Help-center articles
3.9×Repetitive tickets
−33%Time per published article
−94%The automation
Every week, our automation clusters that week's resolved tickets by topic and flags clusters with no matching help-center article. OpenAI drafts an article from the actual resolutions — the real steps agents used, rewritten in the company's documentation style. Drafts land in a Slack review queue where the support lead approves, edits or rejects with one tap; approved articles publish straight to the help center, categorized and internally linked.
How it works
- 01
Tickets clustered weekly
Every Friday, the week's resolved tickets are grouped by topic and checked against existing help-center coverage.
- 02
AI drafts from real fixes
OpenAI writes an article from the actual resolution steps agents used, in the company's documentation voice.
- 03
One-tap review in Slack
The support lead approves, edits or rejects each draft from a Slack queue — no docs tool required.
- 04
Auto-publish and link
Approved articles go live in the help center, categorized and linked from related pages automatically.
- 05
Deflection measured
A monthly report tracks article views against ticket volume, showing which articles keep tickets out of the queue.
Starts from
- Resolved tickets
- Weekly topic clusters
Delivers
- Slack review queue
- Published help articles
- Deflection tracking
The impact
The help center grew from 19 to 74 articles in four months, each grounded in real resolutions. Self-serve deflection climbed 34%, repetitive tickets fell from roughly 120 to 80 a month, and the support lead spends 30 minutes a week reviewing drafts instead of three hours writing each article.
Questions about this automation
How does the flow decide which topics need a new article?
Every Friday it groups the week's resolved tickets by topic and compares each cluster with existing help-center articles. Clusters with enough tickets and no matching article become drafting candidates, and clusters that match an outdated article are flagged for an update. The minimum cluster size is a setting the support lead controls.
Will the drafts match our documentation style?
The drafting prompt includes sample published articles and a short style guide, so OpenAI writes with the same voice, structure and terminology. Each draft is built from the resolution steps agents actually used, and customer details are stripped out before drafting, so the article describes the fix without any personal data.
What does the review step look like?
Each draft arrives in a Slack queue with the source tickets linked and three options: approve, edit or reject. Approved articles publish to the help center with a category and links to related pages. A monthly report then compares article views with ticket volume per topic, showing which articles reduce repeat questions.