Vague questions get vague answers
“Analyze this spreadsheet” produces a generic summary. A specific question produces a useful answer:
This file has 2025 orders: date, region, product, units, revenue.
Which three products grew fastest in revenue between Q1 and Q4?
Show the Q1 and Q4 numbers for each.
A good data question has
- A metric: revenue, count, average, conversion rate.
- A comparison: over time, between groups, against a target.
- A scope: which rows, which dates, which regions.
- A decision it supports: why you are asking.
Describe the data first
Before the question, help the assistant understand the file:
Columns: order_id, order_date (YYYY-MM-DD), region (N/S/E/W),
units, revenue (USD, excluding tax). Each row is one order line.
First, describe the data: row count, date range, and any
problems you notice. Don't analyze yet.
This “describe first” step catches problems early, such as missing months or a column the assistant misread.
Explore, then focus
Ask for a few interesting patterns to explore, then pick the ones that matter for your decision. Analysis without a decision attached tends to produce interesting trivia rather than useful insight.
Practice task
Take a spreadsheet from your work or a public dataset. Write a column description, ask the assistant to describe the data without analyzing it, then write two specific questions with a metric, comparison, scope and decision.