> ## Documentation Index
> Fetch the complete documentation index at: https://docs.poly.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Overview
> Categorize calls using a custom LLM prompt and category set.
Go to **Analytics > Agent Analysis** to categorize calls at scale – identify why calls are being transferred, which outcomes are most common, or how agent quality is trending over time. This feature is also known as **Call Categorization**.
**Agent Analysis** uses an LLM to analyze and categorize calls based on configurable prompts and lists of categories, across groups of up to **250** calls. You can configure up to **10 analysis tasks** per project, each with its own prompt, categories, and color-coding.
For exploratory analysis and pattern-finding across conversations, consider using [Smart Analyst](/smart-analyst/introduction) instead. Smart Analyst's deep sampling feature analyzes up to 500 conversations per query using natural-language questions, and is the preferred path for most analytical use cases. Use Agent Analysis to experiment with structured, repeatable categorization with a fixed set of categories.
## Overview
Each project can have up to **10 analysis tasks**, each with its own prompt, categories, and color-coding. Once configured, batches of up to 250 calls can be grouped and processed per task, returning one category per call. Results are displayed in a visual breakdown and can be reviewed or adjusted manually.
Because only one category is assigned per call, design your categories to be **mutually exclusive**. If a call could fit multiple categories, prioritize the most actionable or specific one. Overlapping categories reduce the usefulness of your results.
## How it works
Go to **Analytics > Agent Analysis** and open **Analysis configuration**. The configuration includes:
* A **Prompt** describing the task, e.g. *"You are a call center manager assessing agent performance."*
* A **Question** to guide the LLM's judgment, e.g. *"Was this call handled successfully?"*
* A set of **Categories** – each with a name, description, and color
Click **Create group**. This opens the [Conversations](/analytics/conversations/introduction) view, where calls can be filtered and selected.
Once the group is created, it appears on the **Agent Analysis** page. Each group shows a visual breakdown of categories and a list of processed calls.
Click **Open** to review individual calls and access [transcripts](/call-data/studio-transcripts) and metadata.
Category assignments can be edited manually. These changes persist unless the call is reprocessed in a new batch.
## Analysis over time
Agent Analysis can run automatically as a **one time analysis**, or on a **daily or weekly schedule**.
**How scheduled batches work:**
* Set up an **automated schedule** from the Agent Analysis dashboard
* Choose **daily** or **weekly** runs to continuously analyze the latest call data
* Each batch appears as a new entry, with trend visualization across multiple runs
* Review progress using **bar charts and filters** that compare results from recent batches
## Related pages
Explore patterns across conversations using natural-language questions and deep sampling.
Review detailed transcripts and metadata for individual conversations.
Monitor safety metrics and risky interactions across your agent.
Export and archive call transcripts for compliance and training.