Asking a support team which articles are missing always yields the same list: 5 to 10 highly visible topics, seen the previous week. The real list is longer, flatter and more expensive.
Classic gap: how do you spot a recurring topic with no article?
A topic is a gap when it returns regularly and consistently triggers a long conversation. Simple 3-dimensional criterion:
- Frequency: > 5 conversations / month over 3 consecutive months.
- Cost: average resolution time > global median × 1.3.
- Auto-resolution: 0 articles cited in agent or FIN replies.
Silent gap: why does an existing article stay unfindable?
Subtler and more frequent. The article exists, but it's poorly titled, poorly indexed, or buried in the wrong section. Symptom: agents reply with a paraphrase instead of pasting the link. The rate at which an article is cited by its own team is the best discoverability indicator.
Ghost gap: how do you detect a stale article?
Article that exists, is findable, but describes a UX from two versions ago. Detection: reopen rate of conversations that cite it. A cited article whose conversation reopens within 48h two out of three times is an article that lies.
How do you audit your knowledge base in 90 minutes a month?
- Extract recurring topics not covered in the last 90 days.
- For each existing article, compute citation rate + reopen rate.
- Rank by estimated total cost (volume × average resolution time).
- Decide: create / rewrite / deprecate. Five decisions per month max, otherwise nothing ships.
A KB isn't a project, it's a monthly discipline. Teams that invest 90 minutes a month on this loop systematically out-perform those who rewrite 30 articles in one go every 9 months.