Skip to content
Your AI and Your Data: Don't Let It Make Up the Number
Start free

How do you know if the number an AI gave you is right?

Answer four quick questions, look at an AI answer that sounds perfect, and meet the three questions you will ask of any number.

Free course. Your Maestro, an AI, .

This course is for you if you have ever uploaded a spreadsheet to ChatGPT, asked Copilot in Excel a question, or handed a CSV to Claude, and got back a number delivered with total confidence. In an hour or so you will learn to put that number to the test before it reaches your report, using three questions that work for any assistant and any table.

Before explaining anything, four quick questions. Answer with what you think today. They don't count for anything, and at the end of the course you will see them again so you can notice what changed.

Check yourself: You ask an AI for the month's total sales. It shows you the Python code it ran, with no errors, and a total. What does that tell you about the total?

Check yourself: Your sales file has one row for each product within a ticket. The AI tells you "There were 312 sales in March," and the file has 312 rows. What did it count?

Check yourself: A survey went out to 112 customers and 45 answered. The AI reports: "89% of customers are satisfied." What do you need to know before repeating that number?

Check yourself: In a column of ratings from 1 to 5, 67 of the 112 cells are empty. The AI computes the average and gets 4.4. What did it do with the empty cells?

An answer that sounds perfect

Now imagine you work at La Escuadra Hardware, a made-up chain with five branches: Centro, Chapalita, Zapopan, Tlaquepaque, and Tonalá. Your boss asks you for two March numbers by tomorrow: how many sales there were and what the average ticket was. You export the file from the system, give it to your assistant, and type: "How many sales did we have in March and what was the average ticket?"

Imagine it answers like this:

In March there were 312 sales, with an average ticket of $378.83.

This is an example, not the real answer of any assistant. But those two numbers are exactly what the most direct calculation on the course file gives: counting its rows and averaging the importe (amount) column. On page 3 you check it yourself.

Both numbers come from the file and the arithmetic is right. And still, neither one answers your boss's question. The file has one row per product, not per ticket. It includes canceled tickets. And it has a few rows repeated by an export error. By the time you finish module 2 you will know what the real average ticket is: it is more than double.

None of this is a rare flaw of one particular AI. Your question used words ("sales," "average ticket") and the AI had to turn them into a calculation on columns and rows. In that conversion it made decisions you didn't see. Has this happened to you with a number at work?

Three questions for any number

The whole course revolves around three questions. They work the same for ChatGPT, Copilot, or Claude, and for a total, an average, or a percentage:

  1. What is a row? Before counting or averaging, you need to know what each row of the table represents: a ticket, a product within a ticket, a person, an answer. If the question is about tickets and the table has products, counting rows gives you something else.
  2. What did it divide by? Every average and every percentage divides something by something. The number on the bottom, the denominator, decides what the result means.
  3. What did it leave out, or add twice? Filters it applied without telling you, rows that dropped out because they had an empty cell, repeated rows it counted twice.

This diagram is the path you will follow with every number:

  1. The AI gives you a number

  2. What is a row?

  3. What did it divide by?

  4. What did it leave out or add twice?

  5. Can you answer all three?

    • Path 1 of 2: no

      1. Ask it to show its work

      with its answer. Goes back to: What is a row?

    • Path 2 of 2: yes

      Continues at: Recompute one figure and use it

  7. Recompute one figure and use it

If any of the three questions has no clear answer, ask it to show how it calculated the number and start over with what it tells you.

Notice the last step: even when the three questions have answers, you recompute one figure yourself. You don't have to redo the whole analysis. It is enough to check one key number with your own hands, and for that the course gives you a calculator: a few lines of pandas that run right here in the browser, on the hardware store's file. You don't need to know how to code. Every line is explained on the page and most of the time you only change a word or a number.

What you will do in the course

  • Module 1: see what the AI does with your file and recompute its two numbers.
  • Module 2: what a row is, and the real average ticket.
  • Module 3: what it divided by, in a survey and in a comparison of branches.
  • Module 4: what it left out or added twice: empty cells, filters, and repeated rows.
  • Module 5: how to ask for answers that can be checked, a complete review, and when to trust a number.

Your Maestro is on every page. If one of the four questions at the start left you with doubts, ask now: I want to understand why I missed one of the questions at the start.