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

# Knowledge Retrieval

> How the AI Analyst answers definition, schema, and conceptual questions

For questions about metrics, tables, and how SourceMedium works, the AI Analyst uses **Knowledge Retrieval** — providing instant answers from documentation and metadata without querying your warehouse.

***

## When Knowledge Retrieval Is Used

Knowledge Retrieval handles questions about **definitions, concepts, and schema** — anything that can be answered from SourceMedium documentation rather than your warehouse data.

### Examples

```
What is LTV?
How is ROAS calculated?
What does conversion rate mean?
Which table has order data?
What columns are in the customers table?
How does attribution work?
What's the difference between first-click and last-click?
How do I set up UTM tracking?
```

These questions share key characteristics:

* **"What is" / "How does"** phrasing
* **Metric definitions** (LTV, ROAS, AOV)
* **Schema questions** (tables, columns)
* **How SourceMedium works**

***

## How It Works

<Steps>
  <Step title="Classify Question">
    The AI recognizes that your question is about concepts or schema, not about querying data.
  </Step>

  <Step title="Search Knowledge Base">
    The question is matched against SourceMedium documentation and metadata.
  </Step>

  <Step title="Generate Answer">
    The AI synthesizes a clear answer from the relevant sources, formatted for Slack.
  </Step>
</Steps>

***

## What You'll See

Knowledge Retrieval responses include:

| Component   | Description                                |
| ----------- | ------------------------------------------ |
| **Answer**  | Clear explanation of the concept or term   |
| **Source**  | Links to relevant documentation pages      |
| **Context** | Related concepts you might want to explore |

<Note>
  No SQL is generated for Knowledge Retrieval questions. The answer comes directly from documentation and metadata.
</Note>

***

## Question Categories

### Metric Definitions

```
What is LTV?
How is ROAS calculated?
What does AOV mean?
How is conversion rate defined?
```

The AI explains what the metric measures, how it's calculated, and where you can find it in your data.

### Schema & Tables

```
Which table has order data?
What columns are in obt_orders?
Where do I find customer information?
What's the difference between dim_orders and obt_orders?
```

The AI describes table purposes, key columns, and when to use each table.

### SourceMedium Concepts

```
How does attribution work?
What's the difference between first-click and last-click?
How do I set up UTM tracking?
What is the attribution source hierarchy?
```

The AI explains SourceMedium-specific methodology and configuration.

### Data Structures

```
What are the possible values for order_type?
How is sales_channel determined?
What does sm_default_channel mean?
```

The AI explains dimension values, enums, and how data is categorized.

***

## Tips for Knowledge Questions

<AccordionGroup>
  <Accordion title="Ask 'what is' for definitions">
    "What is LTV?" triggers Knowledge Retrieval. "What was our LTV last month?" triggers Standard Analysis.
  </Accordion>

  <Accordion title="Ask about 'which table' for schema">
    Schema questions are answered from documentation, not by querying metadata.
  </Accordion>

  <Accordion title="Follow up for more detail">
    After getting a definition, you can ask follow-up questions in the same thread: "How do I calculate that in SQL?" or "Show me an example."

    In channels, mention the bot with `@SourceMedium` for follow-up analysis.
  </Accordion>
</AccordionGroup>

***

## From Knowledge to Analysis

After understanding a concept, you can immediately apply it:

**Knowledge:** "What is ROAS?"

> ROAS (Return on Ad Spend) is calculated as attributed revenue divided by ad spend...

**Follow-up:** "What was our ROAS by channel last month?"

> The AI now runs a Standard Analysis query to get your actual ROAS data.

<Tip>
  The AI retains context in the same thread. If you ask about a concept and then want to see your data, it can connect the two in that thread.
</Tip>

***

## Knowledge Sources

Knowledge Retrieval draws from:

* **Data Dictionary** — Metrics and dimensions
* **Table Schemas** — Column definitions and relationships
* **Core Concepts** — Attribution, data transformations
* **Integration Guides** — Platform-specific information
* **FAQs** — Common questions and troubleshooting

***

## Related

<CardGroup cols={2}>
  <Card title="Standard Analysis" icon="bolt" href="/ai-analyst/workflows/standard">
    How specific data queries are handled.
  </Card>

  <Card title="Deep Analysis" icon="brain" href="/ai-analyst/workflows/deep-analysis">
    How open-ended questions trigger multi-perspective analysis.
  </Card>
</CardGroup>

<Card title="Thread continuity and follow-ups" icon="message-lines" href="/ai-analyst/thread-continuity-follow-ups">
  Learn what carries forward in the same thread and when to start a new thread.
</Card>

***

## Summary and Opinion Follow-Ups

Some follow-ups do not require Knowledge Retrieval or fresh SQL. If you ask for a summary, rewrite, or recommendation after an analysis, the AI Analyst uses the retained thread context.

```
Summarize that in three bullets
What is your recommendation?
Rewrite this for leadership
What should we investigate next?
```

<Note>
  These responses are grounded in the prior thread. If there is no prior analysis context, the AI Analyst will ask you to run an analysis first.
</Note>

***

## Explore the Knowledge Base

These are the sources the AI draws from for Knowledge Retrieval:

<CardGroup cols={2}>
  <Card title="Glossary" icon="book-open" href="/help-center/glossary">
    Definitions of common terms and acronyms.
  </Card>

  <Card title="Attribution Concepts" icon="route" href="/help-center/core-concepts/attribution/attribution-in-sourcemedium">
    How SourceMedium attributes orders to channels.
  </Card>
</CardGroup>