Skip to main content

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.

The AI Analyst uses different workflows depending on what you’re asking. Each workflow is optimized for its type of question — from quick knowledge lookups to comprehensive multi-perspective analyses.

​
How Questions Are Routed

When you send a message, the AI Analyst classifies it by intent:
Question TypeWorkflow
Specific metric questionsStandard Analysis
Raw rows or fieldsRaw Data Pull
Open-ended / strategicDeep Analysis
Definitions & conceptsKnowledge Retrieval
Summaries and opinionsSummaries & Opinions
The classification happens automatically based on the question’s structure and intent.

​
Thread continuity behavior

The AI keeps follow-up context in the same thread.
  • In DM, you can ask follow-ups naturally.
  • In channels, mention the bot with @SourceMedium for follow-ups.
  • To reset context, start a new thread.

Thread continuity and follow-ups

Learn what carries forward in the same thread, what does not, and how to reset cleanly.

​
Workflow Overview

Standard Analysis

Single-path SQL analysis for specific metrics and bounded time ranges.

Raw Data Pull

Single-path SQL retrieval for rows, fields, and exports. Returns a CSV automatically.

Deep Analysis

Multi-perspective analysis for strategic questions. Runs multiple Standard Analyses in parallel and synthesizes insights.

Knowledge Retrieval

Instant answers about metrics, tables, and concepts. Draws from SourceMedium documentation and metadata.

Follow-Up Summary & Opinion

Thread-aware summaries, rewrites, recommendations, and follow-up analysis.

​
Choosing the Right Approach

​
For Data Answers

Ask specific, bounded questions: 
What was our revenue last week?
Top 10 products by units sold this month
What's our ROAS by channel for the past 30 days?
These trigger Standard Analysis for quick results.

​
For Raw Data Pulls

Ask for rows, fields, or exports: 
Pull order ID, date, UTM source, and revenue for last week
Export customers acquired in March with email and first order date
Show line items for SKU ABC-123 from the past 30 days
These trigger Raw Data Pull — single-path retrieval that skips Deep Analysis and attaches a CSV.

​
For Strategic Insights

Ask open-ended, exploratory questions: 
How can we improve our marketing performance?
What trends should we be watching?
Why are our metrics declining?
These trigger Deep Analysis for comprehensive multi-perspective analysis.

​
For Definitions & Concepts

Ask about metrics, tables, or how things work: 
What is LTV?
How is ROAS calculated?
Which table has order data?
These trigger Knowledge Retrieval for instant documentation-backed answers.

​
For Follow-Ups

Reply in the same thread after an answer: 
Summarize this for leadership
What is your recommendation?
Break that down by channel
These trigger Summaries & Opinions — no SQL, uses retained thread context. It runs fresh SQL when your follow-up asks for a new computation.

​
Workflow Comparison

FeatureStandardData PullDeep AnalysisKnowledgeFollow-Up Summary/Opinion
Data SourceBigQueryBigQueryBigQueryDocs + MetadataThread context
SQL Queries112–40 (usually)0 unless new computation is requested
Parallel ExecutionNoNoYesNoNo
ChartsYes, when usefulNoYesNoNo
Best ForSpecific metricsRows and fieldsStrategic questionsLearning conceptsReframing prior results

​
How Classification Works

The AI uses contextual signals to determine the right workflow: Standard Analysis triggers:
  • Specific time ranges (“last week”, “in January”)
  • Named metrics with data requests (“revenue”, “orders”, “ROAS”)
  • Rankings and comparisons (“top 10”, “vs last month”)
  • Computed results (“by channel”, “trend over time”)
Data Pull triggers:
  • Pull/export language (“pull data”, “export rows”, “give me raw orders”)
  • Specific requested fields (“order ID, customer email, UTM source”)
  • Specific entity lookups (“show me line items for order 123”)
Deep Thinking triggers:
  • “How can we improve…” questions
  • “What should we focus on…” questions
  • Trend analysis without specific bounds
  • Strategy and optimization requests
Knowledge Retrieval triggers:
  • “What is…” or “What does…mean”
  • “How is X calculated?”
  • “Which table has…” or “What columns…”
  • Data health and attribution health questions
Summary and opinion triggers:
  • “Summarize this”
  • “What should we do next?”
  • “What is your recommendation?”
  • “Rewrite that for leadership”
If you want to override the classification, be explicit. “Just give me the revenue numbers” will use Standard Analysis even if your phrasing might otherwise trigger Deep Thinking.

Understanding Results

How to read responses, charts, and SQL output.

Example Questions

See examples across all domains and workflows.