Knowledge Gap Agent · Use case
How can we use an AI agent to find the customer questions our help articles do not answer on our own infrastructure?
A knowledge gap agent reads a period's tickets and chats, groups the questions customers asked by subject, compares each group with your help articles, and lists the questions that no article answers or that an article answers only in part, with the count and example contacts with names removed. The knowledge owner decides what to write. The agent writes and publishes no article, and its list is about questions, never about which adviser took them.
Customers keep asking.
No article answers.
- ✗ 01The questions with no article are answered again and again by hand, and nobody sees the pattern because each one sits in a different ticket.
- ✗ 02The two agents that answer customers can only be as good as the published articles, so a missing article is a question they must pass to an adviser.
Where it stops for a person
Always. The knowledge owner decides what to write.Always. It lists, it does not choose: what is worth an article, what the article says and when it is published are the knowledge owner's decisions. A question about one customer's own case is not a gap in the articles and is left out. The list is by subject and names no adviser and no customer. A subject with no gap on the list has not been cleared: the list shows what the agent matched and nothing more.
It lists the questions with no article.
The knowledge owner decides what to write.
Step · 01
The knowledge owner names the period and the channels.
Step · 02
The agent reads the tickets and chats for that period, read only, and takes the customer's question from each.
Step · 03
It groups the questions by subject and removes names and contact details from the examples.
Step · 04
It compares each group with the help articles and lists the groups with no article, or with an article that covers only part, with the count, the examples and the nearest article.
Step · 05
The knowledge owner decides which articles to write or change, and who writes them.
Reads from · connectors
Tickets and chat records for the period, from a reporting database, read only, through a view your team has prepared, or through your own APIs with the operations that change something left out; and the published help articles, as a knowledge source. The view carries the customer's question and the channel, and not the adviser's name.
Acts through · MCP servers
Nothing. It writes a list. It drafts and publishes no article and changes no ticket.
What you set up
- Prepare a read-only view of tickets and chats with the customer's question, the channel and the date, and without the adviser's name.
- Index the published help articles as a knowledge source, the same set the answering agents use.
- Build the agent in Katonic Studio and test it on a past month where you know which articles were added afterwards.
- Run it each month, or when the knowledge owner asks, with the list going to the people you name.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. Tickets and chats stay inside the company, and examples are shown with names removed.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does it write the missing articles?
No. It lists the questions with no article. The knowledge owner decides what to write, and a person writes and publishes it.
Does the list show which adviser answered?
No. It is grouped by subject, and the view it reads carries no adviser's name.
Where do the examples come from?
From the tickets and chats of the period you name, with names and contact details removed.