Skip to main content
2.3

The Analyst

Estimated time: 30 minTool: Claude Pro
After this drill, you can:

After this drill, you can upload data to Claude — CSV, PDF, images — and extract actionable insights through conversation.

Why this matters

Claude Pro's file upload capability turns it into a personal data analyst. You don't need to know statistics, Excel formulas, or SQL. You need to know how to ask the right questions about your data. This drill demonstrates the workflow: upload → understand → question → insight → action. The step most people skip is the last one: AI analysis only matters if it changes what you do.

How to do it

  1. 1

    Find a real dataset you have access to

    A CSV export from a tool you use (email newsletter, sales data, survey results, timesheet). A PDF report with tables. A screenshot of a data visualization you want to understand.

  2. 2

    Upload it to Claude and use the 'understand first' prompt

    Before asking for analysis, have Claude describe what it sees. This catches misunderstandings before they become wrong answers.

  3. 3

    Ask three increasingly specific questions

    Start with "what is the most important thing in this data?" Then get more specific based on what matters for your context.

  4. 4

    Ask for one actionable recommendation

    The test of any analysis: 'Based on this data, what should I do differently?' If the answer is vague, push harder.

The prompt

PROMPT — Upload and Understand FirstModel: Claude Pro
I'm uploading [FILE TYPE — e.g. a CSV export, a PDF report, a screenshot]. Before any analysis, please:

1. Describe what you see: what data is in this file, what columns/fields exist, and what time period or scope it covers
2. Flag any obvious data quality issues (missing values, inconsistent formats, outliers)
3. Tell me the top 3 questions this data could answer

Then I'll tell you what I specifically need to know.
PROMPT — Analysis and RecommendationsModel: Claude Pro
Based on this data:

1. What is the single most important insight — the one finding that would change how I operate?
2. What is performing significantly better or worse than average? What explains it?
3. What would you recommend I do differently in the next 30 days based on this data?

For each answer, show me the specific data point(s) that support it.

Success criteria

  • ✓You uploaded a real file and Claude correctly understood what was in it
  • ✓You asked at least three specific questions about the data
  • ✓You received at least one actionable recommendation grounded in the data
  • ✓You could articulate one thing you would do differently based on the analysis

Common mistakes

Uploading data without telling Claude what you need to know

→ After the "understand first" step, you must give Claude your context: what decision you are trying to make, what would constitute a good outcome.

Treating AI analysis as fact without checking key numbers

→ Spot-check two or three numbers against your source data. Claude processes data accurately most of the time, but mistakes happen on complex calculations.

Asking for generic analysis instead of decision-relevant questions

→ "What are the trends?" is weak. "Is the cost per acquisition increasing or decreasing, and does it vary by channel?" is actionable.