The Insights Agent is your conversational partner in data analysis. Like any large language model, its quality scales with the quality of your prompts, think of it as a capable junior analyst: the more context and direction you give, the sharper the output.
The golden rule: context is everything
The Insights Agent doesn't intuitively know your intent, it reacts to the data and the prompts you provide. To get precise, actionable answers, include as much context as possible: the timeframe, the specific category, or the survey column you want to analyze.
💡 Iteration tip:Treat the first answer as a starting point. If it's too broad, follow up: "That's interesting,can you show me specific verbatim examples for that negative trend?"
How the agent shapes its responses
The Insights Agent doesn't use a fixed template. It dynamically selects the visualizations, charts, or summaries that best answer your specific prompt. A simple question might return a clean text summary. A broader request about "most common issues" will automatically generate a multi-layered analysis with statistical trends, sentiment breakdowns, and representative customer comments.
What to do: master the conversation
These strategies consistently produce better, more targeted answers.
| Strategy | Why it works | Example prompt |
|---|---|---|
| Onboard yourself first | Quickly understand the scope of a new project | "What data can I see in this project?" or "What are the most common themes here?" |
| Define your persona | Tailors output to your specific role and priorities | "My role is Product Manager for the web app. What feedback is relevant for me?" |
| Be ultra-specific | Prevents generic answers by targeting exact data points | "Compare sentiment on 'Ease of Use' between Q1 and Q2 for Enterprise users." |
| Explain by example | Sets a clear template for format or tone | "Summarize in three bullet points, then one Key Action Item for the dev team." |
| Check active filters | Ensures you're analyzing the right data subset | Check the system note in chat to see which dashboard filters are currently active. |
What to avoid: common pitfalls
- Blind trust: The Agent is highly accurate, but it augments your expertise rather than replacing it. If an insight contradicts your intuition, investigate it, ask for an explanation of its reasoning or cross-reference the raw data.
- Ignoring filters: If results feel "off," a dashboard filter is often still applied. Always verify active filters in the Meta Text at the start of your response.
- Expecting magic knowledge: The Agent can only analyze data uploaded to Caplena. It cannot access internal documents, Slack messages, or external market trends unless they are part of the current project.
- Vague terminology: Avoid pronouns like "it" or "that" when the conversation has covered multiple topics. Use explicit category or segment names to keep the Agent on track.
- Using it as a manual calculator: Don't ask the Agent to do arithmetic on numbers already shown in chat (e.g., converting star ratings to NPS row by row). The Agent provides pre-calculated Key Quantitative Findings from the raw dataset, manual math in a chat thread introduces errors. For value mappings or mathematical operations, use the Smart Column feature instead.