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

# Prompting Tips for Better Data Analysis

> Practical prompting techniques to get faster, more accurate analytics answers from Hazel — with real examples from hundreds of conversations.

## How to get great answers from Hazel

Hazel understands natural language, so you don't need to learn a special syntax. But a few small habits will dramatically improve the quality of your answers.

### Be specific about what you want

Vague questions get vague answers. Specific questions get specific answers.

<Columns cols={2}>
  <Card title="Too vague" icon="circle-xmark">
    "How are we doing?"
  </Card>

  <Card title="Much better" icon="circle-check">
    "What was our revenue last week compared to the week before? Break down by new vs returning customers."
  </Card>
</Columns>

### Give Hazel context about your business

Hazel knows ecommerce deeply, but it doesn't know your specific constraints. Tell it what matters.

> "We're a seasonal apparel brand. We can't launch new products in less than 8 weeks from design to shipping. Keep that in mind when making recommendations."

> "Our COGS are roughly MSRP divided by 4. Factor that in when analyzing margins."

> "We run deep promotional events twice a year where prices drop to near cost — exclude any sale price more than 30% below the median when analyzing pricing."

### Ask for the format you want

If you want a table, say so. If you want a chart, ask for it. If you want bullet points for a leadership update, tell Hazel.

> "Show this as a chart."

> "Give me a table with product name, revenue, units sold, and return rate."

> "Summarize this in 3 bullet points I can share with my team."

### Start with the baseline

When asking for analysis, tell Hazel to anchor on your overall numbers first. This makes the insights much more useful.

> "For reference, start by showing our overall LTV and repeat purchase rate. Then show me which products beat those numbers."

### Ask for enough detail

Don't settle for high-level summaries when you need granularity. Tell Hazel how deep to go.

> "Show me at least 25 products, not just the top 5."

> "Break this down by variant if there are meaningful pricing differences between colors or sizes."

> "Don't just give me categories — give me specific product recommendations."

### Use follow-ups

Hazel remembers everything within a thread. Your best answers often come from follow-up questions.

1. Start broad: "What's driving the change in our AOV this month?"
2. Drill in: "Show me the specific products contributing to that shift."
3. Get actionable: "Which of those should I feature in next week's email campaign?"

### Use Memory for things you repeat

If you find yourself adding the same context to every prompt — metric definitions, product exclusions, fiscal year start dates — put it in [Memory](/docs/features/memory) once and it applies everywhere.

Good candidates for Memory:

* **Metric definitions** — "Net revenue means gross revenue minus refunds and discounts"
* **Exclusions** — "Always exclude gift cards and free samples from revenue calculations"
* **Business rules** — "We consider a customer 'new' if they've placed exactly one order"
* **Historical context** — "We migrated platforms in May 2025, so order dates before that may show the migration date instead of the original order date"

### Start a new thread for new topics

Hazel works best when a thread stays focused on one topic. If you're switching from "marketing performance" to "subscription churn," start a fresh thread. This keeps Hazel focused and avoids hitting conversation length limits.

<Tip>
  **When in doubt, just ask.** You can't break anything. If Hazel's first answer isn't quite right, give it more context and ask again — it learns within the conversation.
</Tip>

***

**Next up:** Learn the difference between [Ask and Deep Thinking](/docs/features/ask-vs-deep-thinking) to pick the right mode for your question.