Gaps become visible instead of inferred
The strongest signal about your documentation is not in your documentation. It is in the questions that arrived and found nothing.
Voice of the Customer
A knowledge base that improves
Most knowledge bases are written once, in a burst, by whoever had time — and then left. This is the other model: gaps identified from the questions nobody could answer, weak articles flagged by what happens after someone uses them, and a writing queue ordered by what customers actually ask.
- Gaps found from questions that had no good answer
- Weak articles flagged by the outcomes they produce
- The writing queue ordered by real demand
- Drafts built from material you already have
Nobody can name the bad articles
Every knowledge base contains articles that are wrong, out of date, or written for a version nobody runs any more. The organisation knows this and cannot name a single one, because the feedback never finds its way back.
The gaps are harder still: an article that does not exist leaves no trace at all. So content work runs on whoever complained most recently, the annual review gets deferred, and the articles people actually need stay unwritten.
Letting the queue decide what gets written
Every one of these signals is a by-product of work your team is already doing.
- Step 1
Find the questions with no answer
Issue types that keep arriving with no documented solution are the gaps. The knowledge base cannot show you what is missing from it.
- Step 2
Flag the articles that are not working
An article is judged by what happened after it was used — whether the case closed, whether the customer came back, whether the agent escalated anyway.
- Step 3
Order the queue by demand
The backlog is ordered by how often the question arrives, so the next article written removes the most work.
- Step 4
Draft from what already exists
Where the answer exists already, a draft is generated from it for a person to review. That is a far smaller job than writing from nothing.
Three signals a knowledge base cannot produce about itself
All of them live in the queue rather than in the documentation.
Articles are judged by outcomes
View counts measure whether an article was found, not whether it worked. What happened next is the signal that finds the article quietly costing you time.
Writing starts from existing material
The answer usually exists already, in a resolved case or a manual. Refining a draft is a different-sized job from authoring from scratch.
Why the usual approaches fall short
Three of these measure the wrong thing and the fourth measures nothing.
The annual content review
Scheduled, deferred, and done in a rush. It reviews what exists and cannot see what is missing.
View counts and article ratings
Measure whether an article was found, and whether somebody was annoyed enough to rate it.
Asking agents what is missing
Returns what is recently frustrating rather than what is costly, and only from the agents who reply.
Simply writing more articles
Volume is rarely the problem. Coverage of the questions that actually arrive is.
What things are called
The vocabulary that comes up when knowledge quality is being measured.
- Knowledge gap
- An issue type that keeps arriving without a documented solution behind it.
- Gap detection
- Identifying those gaps from categorisation and driver data rather than from a manual audit.
- Article outcome
- What happened after an article was used — resolved, reopened, or escalated anyway.
- Generated draft
- A starting article produced from existing material, for a person to review before publication.
- Knowledge Intelligence
- The agent that knows which article applies to which issue, and where you have no answer at all.
Frequently Asked Questions
By reading the other side of it. A gap leaves no trace in the knowledge base, but it leaves a very clear one in the queue: an issue type that keeps arriving with no documented solution attached. Categorisation and driver ranking together make those visible.
Where this fits
Knowledge Intelligence tour
Building an article end to end: templates, attributes, and generation from your own source files.
Knowledge from Slack
The capture side of the same loop: a resolved thread becoming a reviewable article.
Knowledge Intelligence Agent
The agent page: gap detection, article quality and generation across the platform.
More Voice of the Customer tours
Voice of the Customer
Find out what your documentation is missing
Give us your ticket history alongside your current knowledge base and we will show you the issue types that keep arriving with nothing to point at.
Talk to us