> ## 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. # Deep Analysis > How the AI Analyst handles complex, open-ended questions with multi-perspective analysis When you ask a strategic or exploratory question, the AI Analyst automatically activates **Deep Analysis** — a thorough approach that runs multiple analyses in parallel before synthesizing a comprehensive answer. *** ## When Deep Analysis Activates Deep Analysis is triggered automatically for **open-ended questions** — questions that don't have a single, specific answer and benefit from exploring multiple perspectives. ### Questions That Trigger Deep Analysis ``` How can we improve our marketing performance? What trends should we be paying attention to? Why are our metrics declining? What opportunities am I missing? How is our business performing overall? What should we focus on to grow revenue? ``` These questions require the AI to: * Consider multiple dimensions of your data * Compare across time periods, channels, and segments * Identify patterns that might not be obvious from a single query * Synthesize findings into actionable insights ### Questions That Use Standard Analysis ``` What was our revenue last week? Top 10 products by units sold How many new customers in January? What's our ROAS by channel? ``` Specific questions with clear metrics and time ranges use the faster [Standard Analysis](/ai-analyst/workflows/standard) workflow. Raw data pulls also use Standard Analysis, even when they include several fields or dimensions. If you ask to "pull," "export," or "give me rows," the AI Analyst prioritizes returning the requested data and skips Deep Analysis. *** ## How It Works Deep Analysis is essentially **multiple Standard Analyses running in parallel**, with a synthesis step at the end: The AI breaks your open-ended question into 2–3 specific analytical questions that together address your original query from different angles. **Example:** For "How can we improve marketing performance?", the AI might generate: * "Which channels have the best ROAS and how has that changed recently?" * "Which campaigns are driving new customers vs. repeat customers?" * "Where are we overspending relative to attributed revenue?" The AI runs multiple Standard Analyses simultaneously — one for your original question plus each strategic question. Each follows the full workflow: identifying tables, writing SQL, executing queries, and analyzing results. After all analyses complete, the AI synthesizes the findings into a cohesive response: * Key insights from each perspective * Patterns that emerged across the analyses * Specific recommendations backed by data * Relevant charts and data tables *** ## What You'll See When Deep Analysis is active, the AI Analyst shows progress through each phase: | Status | What's Happening | | --------------------------------- | ------------------------------------------------------------- | | 🔍 Understanding your question | Classifying and determining the analysis approach | | 🧠 Deep Analysis activated | Open-ended question detected, entering multi-perspective mode | | 🧠 Generating strategic questions | Breaking down your question into analytical angles | | 🔄 Running parallel analyses | Executing multiple SQL queries simultaneously | | ✨ Synthesizing insights | Combining results into a comprehensive answer | Deep Analysis takes longer than Standard Analysis but provides richer, more actionable insights. *** ## Example: Deep Analysis in Action **Question:** "How can we improve our marketing performance?" **What Deep Analysis Does:** 1. **Generates strategic sub-questions:** * "Which channels have the highest and lowest ROAS over the past 30 days?" * "What's the new customer acquisition cost by channel?" * "Which campaigns are underperforming relative to spend?" 2. **Runs parallel analyses** for each question, querying: * Ad performance data by channel * Customer acquisition metrics * Campaign-level ROAS and spend 3. **Synthesizes findings:** > "Your Meta campaigns are delivering 3.2x ROAS, outperforming Google Ads at 1.8x. However, Google is driving 40% of new customer acquisitions at a lower CAC ($32 vs $45). Consider reallocating 15% of Meta budget to Google's top-performing campaigns to balance immediate ROAS with customer acquisition." *** ## Tips for Better Results Deep Analysis works best when your question truly requires exploration. "What should I focus on?" is better than "Show me revenue" for triggering multi-perspective analysis. "How can we improve Q1 performance given we're launching a new product line?" gives the AI useful context for generating relevant strategic questions. After a Deep Analysis, ask follow-up questions in the same thread. The AI retains context in this thread and can dive deeper into specific findings. In channels, mention the bot with `@SourceMedium` for follow-ups. Deep Analysis takes longer but surfaces insights you might not get from a single query. The synthesis phase is where the real value emerges. *** ## Limitations * **Scope:** Currently limited to 2–3 parallel analyses to balance depth with speed * **Visualizations:** Charts are generated for the synthesized response, not each individual analysis * **Raw data pulls:** Requests for rows, fields, or exports do not use Deep Analysis If you need a quick, specific answer, phrase your question to avoid Deep Analysis. "What was our Meta ROAS last month?" will return faster than "How is Meta performing?" *** ## Related How specific data queries are handled. How definition and schema questions work. Learn what the AI remembers in the same thread, and how follow-ups work in DMs vs channels.