Anatomy of a strong analysis prompt

Free sample lesson from Prompt-Based Analytics and Reports for Business · about 10 minutes · no payment or email needed

In this lesson you learn the five parts of a clear prompt by asking a specific question of real practice data, and you see how "show the calculation" lets you check every number.

The scenario

Sales fell last month and your director asks, "What happened?" You have a small table of sales by region and product.

A weak prompt and a strong prompt

Weak prompt: Analyze my sales data.

The AI returns general statements such as "sales varied by region" that you could have written without data.

Strong prompt, in five parts:

  • Role: Act as a business analyst.
  • Context: The table shows monthly sales by region and product for the last 3 months. Our target is growth of at least 2% per month.
  • Task: Find where sales fell last month compared with the month before, and by how much.
  • Format: A table of the three biggest decreases (Region, Product, Change in dollars, Change in percent), then two sentences describing the pattern.
  • Limits: Use only the figures given. Show the calculation for each change. If the data cannot explain why, say so.

What the strong prompt returns

The AI returns specific decreases with calculations you can check, and does not invent reasons it cannot know. With the practice data in the box below, the numbers should match this: Outdoor fell from $27,520 to $24,320 (-$3,200, -11.6%) and Footwear from $21,500 to $21,070 (-$430, -2.0%), while Apparel rose 4.9%. Your AI's wording will differ, but the numbers should not.

Why it works

Each part of the prompt does a different job.

  • Role and context tell the AI what the data is and what matters.
  • A specific question gives a specific answer.
  • "Show the calculation" lets you verify every number.

The five parts of a clear prompt

A clear prompt has five parts: Role, Context, Task, Format and Limits. Together they tell the AI who it is writing as, what is true, what to produce, what shape it should take and what to avoid.

The point of the structure is simple: the AI fills any gap you leave with a confident guess, so the more you supply, the less it invents.

  • Role: whose voice the AI writes in
  • Context: the true facts it must work from
  • Task: exactly what you want produced
  • Format: length and structure
  • Limits: tone, things to avoid, and "use only the facts above"

Watch out

AI can sound confident while getting a number wrong. Recalculate at least a couple of the changes in a spreadsheet before you trust them.

Try it yourself

Paste this into ChatGPT, Claude or your organization's approved AI tool and run it. Then change the facts to your own and run it again, and compare the two drafts.

Act as a business analyst.
Context: The table below shows monthly units, revenue (US dollars), returns and average rating for three product lines, January to June 2026. Our target is growth of at least 2% per month.
Task: Find every product line whose revenue fell from May to June, and by how much.
Format: A table (Product_Line, May revenue, June revenue, Change in dollars, Change in percent), then two sentences describing the pattern.
Limits: Use only the figures given. Show the calculation for each change. If the data cannot explain why, say so.

Data:
Month,Product_Line,Units_Sold,Revenue_USD,Units_Returned,Avg_Rating
2026-01,Outdoor,310,19220,14,4.4
2026-01,Apparel,520,19760,31,4.5
2026-01,Footwear,260,22100,21,4.2
2026-02,Outdoor,340,21080,17,4.4
2026-02,Apparel,540,20520,30,4.5
2026-02,Footwear,255,21675,20,4.3
2026-03,Outdoor,395,24885,22,4.3
2026-03,Apparel,560,21280,34,4.5
2026-03,Footwear,270,23220,22,4.2
2026-04,Outdoor,450,28350,31,4.2
2026-04,Apparel,590,23010,35,4.4
2026-04,Footwear,280,24080,24,4.2
2026-05,Outdoor,430,27520,46,3.9
2026-05,Apparel,610,23790,36,4.5
2026-05,Footwear,250,21500,22,4.1
2026-06,Outdoor,380,24320,58,3.7
2026-06,Apparel,640,24960,38,4.5
2026-06,Footwear,245,21070,23,4.2