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

# How do we Categorize Calls?

We split calls into two groups based on how long they lasted.

## Group 1 — Calls Longer Than 30 Seconds

If a user talked for more than 30 seconds, it usually means they were interested in what the agent was offering. We check these 4 things for such calls:

1. **Conversation Quality.** Check if the agent is having a proper conversation. Is it listening carefully without cutting the user off? If the user interrupts, does the agent stop and let them speak? Is the agent repeating the same response over and over? (It should not.)
2. **Voice Quality.** The agent's voice should sound clean and natural. Watch out for the voice breaking or cutting out, words lagging or sounding slow, and words being stretched out unnaturally (gibberish).
3. **Transcription Issues.** The transcriber converts what the user says into text. If it makes mistakes, the agent misunderstands the user. Common problems are transcribing words incorrectly and missing words from the user's sentences entirely.
4. **Agent Hallucination.** This is when the agent confidently says something that is completely wrong or made up, and presents it as a fact. Why does this happen? The agent was not able to call the right function (tool), or did not have access to that piece of information at the time, so it made up an answer instead based on the prompt and trained data.

## Group 2 — Calls Shorter Than 30 Seconds

If a user hung up within 30 seconds, it usually means they were not interested, or something in the agent's opening put them off, or the agent's voice sounded robotic and less natural.

Along with all 4 checks from Group 1, we also check one more thing:

1. **Reason for Early Disconnect.** Why did the user hang up so quickly? Common reasons: the agent's opening lines were too heavy or overwhelming, the agent used difficult or confusing words right at the start, or the intro/greeting was too long.