> ## Documentation Index
> Fetch the complete documentation index at: https://docs.raum.am/llms.txt
> Use this file to discover all available pages before exploring further.

# Detect and close knowledge gaps

> raum flags tickets where the AI had low confidence or no matching procedure. Learn to review gaps, create procedures to address them, and track resolution progress.

A knowledge gap is a topic your customers are asking about that raum doesn't yet have a procedure for — or has a procedure for but with low confidence in how well it applies. When the AI processes a ticket and can't find a strong match in your knowledge base, it flags that topic automatically. Over time, the knowledge gaps list becomes a prioritized to-do list for improving your AI's accuracy.

## How raum detects gaps

Every ticket that flows through raum is scored for confidence. The AI searches your procedures and articles using semantic similarity — it's not looking for exact keyword matches, but for meaning. When the best match falls below raum's confidence threshold, the system records the ticket as a potential knowledge gap and groups it with similar tickets under a common topic label.

You don't need to configure anything for detection to work. Gaps appear in the dashboard automatically as the AI encounters them.

<Note>
  Knowledge gaps are detected by the AI pipeline regardless of whether you're running Agent Mode or Copilot Mode. Both modes report low-confidence tickets to the same gaps list.
</Note>

## Reviewing gaps in the dashboard

Navigate to `/dashboard/knowledge-gaps` to see all detected gaps. Each gap entry shows:

* The topic label raum inferred from the pattern of tickets
* The number of tickets that triggered this gap
* Example tickets you can open to read the actual customer questions
* The current status: **open**, **addressed**, or **dismissed**

The list is sorted by ticket count by default, putting your highest-impact gaps at the top. A gap with 40 tickets behind it has a much larger effect on your AI's performance than one with 2.

<Tip>
  Click any example ticket to read the original customer message. This is the fastest way to understand what your customers are actually asking before you write a procedure.
</Tip>

## How to address a gap

Addressing a gap means giving raum the information it needs to handle that topic confidently. There are two main approaches:

<Steps>
  <Step title="Identify the gap topic">
    Open the knowledge gaps list at `/dashboard/knowledge-gaps`. Click on a gap to see its example tickets. Read 3–5 of them to understand the shape of the question — is this a billing question, a how-to, a policy question, or something else?
  </Step>

  <Step title="Decide: procedure or article">
    If the topic requires the AI to follow a specific decision path or take action (for example, processing a refund request or escalating a billing dispute), create a **procedure**. If the topic is purely informational (for example, explaining how a feature works), adding or updating a **help center article** is usually sufficient.
  </Step>

  <Step title="Create the procedure or article">
    Navigate to **Resources → Procedures** or **Resources → Articles** and create the new content. Write it with the customer questions from the gap in mind — the more closely your procedure matches the language and intent of the actual tickets, the better the AI's confidence score will be when matching.
  </Step>

  <Step title="Mark the gap as addressed">
    Return to `/dashboard/knowledge-gaps` and mark the gap as **addressed**. This moves it out of the open list and lets you track whether the fix had the intended effect. New tickets on the same topic will show up as a fresh gap if confidence remains low, giving you a feedback loop.
  </Step>
</Steps>

## Status tracking

Each gap has one of three statuses:

<CardGroup cols={3}>
  <Card title="Open" icon="circle">
    The gap has been detected and not yet acted on. These are the ones that need your attention.
  </Card>

  <Card title="Addressed" icon="circle-check">
    You've created a procedure or article to cover this topic. The AI will use the new content on future tickets.
  </Card>

  <Card title="Dismissed" icon="circle-x">
    You've reviewed the gap and decided it doesn't warrant a procedure — for example, if the tickets are out-of-scope questions or one-off edge cases.
  </Card>
</CardGroup>

## Example workflow

Here's a typical knowledge gap workflow from detection to resolution:

<Steps>
  <Step title="Weekly gap review">
    Set aside 15 minutes each week to review open knowledge gaps. Sort by ticket count and focus on the top 3–5.
  </Step>

  <Step title="Read the example tickets">
    For each gap, read the example tickets. Group them mentally — are these all variations of the same question, or are there sub-topics?
  </Step>

  <Step title="Write a procedure">
    If the question involves a decision or action, write a procedure in **Resources → Procedures**. Use the example tickets as a guide for what language and scenarios to cover.
  </Step>

  <Step title="Mark as addressed and monitor">
    Mark the gap as addressed. Over the next few days, watch whether new tickets on that topic appear in the gaps list. If they do, the procedure may need to be expanded or reworded.
  </Step>
</Steps>

<Note>
  The knowledge gaps API is available at `/api/dashboard/knowledge-gaps` if you want to build your own reporting or integrate gap data into an external workflow.
</Note>