Home › AI at work › Analyse a spreadsheet

Hand your spreadsheet to AI

Upload a spreadsheet, ask questions in plain English, and get answers, charts, new files and the summary email back. Here's how to do it well, safely, and with every number checked.

Starter 20 minutes Free · practice file included Read the privacy step first
📎 office-supplies-sales.csv · Which region sold the most this year, and is anything unusual?

East leads with £15,201, about 13% ahead of North (£13,491). The four regions are close overall, but they peak at different times: West in May, South in August.

East£15,201
West£14,329
South£13,960
North£13,491

Worth checking: October looks like a collapse, but the data stops on 2 October. Two customers, Summit Print and Lantern Studio, haven't ordered since mid-June.

sales-summary.xlsx
What you'll be able to do

A brilliant data colleague, on demand

When you need the answer rather than the formula, the AI can do the analysis itself. The skill is asking well, and checking what comes back.

  • Upload a spreadsheet and ask questions about it in plain English
  • Get tables, charts and new Excel files back to download
  • Have the AI write the summary email or slide text that goes with the numbers
  • Check the AI's answers so you never send a wrong figure to your boss
  • Know when this is the right approach and when to stay inside Excel

Good for: anyone who needs answers from data but isn't a spreadsheet person, people preparing for meetings at short notice, and managers who get sent spreadsheets and need the story behind them.

Before you start

What you need

An AI assistant that can read files and run analysisClaude, ChatGPT, Gemini, or Copilot (Microsoft 365 Copilot can do much of this inside Excel itself).
A spreadsheetStart with our practice file of 300 made-up office-supply orders.
A question you actually want answered
Step by step

Do it in six steps

  1. Decide whether your data can be uploaded

    This step matters more than any prompt. When you upload a file, it leaves your computer and goes to the AI company.

    • Practice and public data: go ahead.
    • Work data: check your company's AI policy first. Many companies provide an approved tool (Microsoft Copilot, or a business plan of ChatGPT, Claude or Gemini) where your data is protected under a company agreement. Use that, not your personal account.
    • Personal or sensitive data (customer names, contact details, salaries, health information): don't upload it to a personal account at all. Remove those columns, or replace names with codes like "Customer 1", before uploading.

    Not allowed to upload? The formula guide gets you the same answers without the data leaving your computer.

  2. Upload and get your bearings

    Attach the file and start with a broad question. This tells you whether the AI has read it correctly:

    Copy this prompt
    I've attached a spreadsheet of sales orders. Before we start, please tell me:
    - what each column contains
    - how many rows there are and the date range covered
    - any problems you notice (blanks, odd values, duplicates)
    
    Keep it short.

    If it describes something different from what you expect, correct it now: "Unit Price is in pounds, not dollars", or "each row is one order, not one customer".

  3. Ask real questions

    Now ask what you actually want to know. One question at a time works best:

    Copy this prompt
    Which region had the highest total sales (Quantity × Unit Price), and how much more than the lowest? Show me a small table of all four regions.
    Then try
    Which months were the best and worst? Is there a trend over the year?
    Then try
    Which customers ordered in the first half of the year but haven't ordered since July? List them with their last order date and total spent.
    Watch for the trap. The practice data stops on 2 October, so October looks like a disaster. A good AI will point this out. If it doesn't, you should. Part of using AI well is noticing when an answer doesn't make sense.
  4. Get charts and files back

    Copy this prompt
    Please make a bar chart of total sales by month, with £ on the axis and a clear title, that I can paste into a PowerPoint slide. Leave October out because it's only partial.
    Then try
    Please create an Excel file I can download with three tabs: "By region" (totals by region and month), "By product" (totals by product, highest first) and "Lapsed customers" (from your earlier answer).

    The AI will give you an image to download, or a link to the file. Open the file in Excel and check it looks right.

  5. Check the answers

    AI is very good at this, but it can still make mistakes: misreading a column, double-counting, or confidently stating something it guessed. Before anything leaves your hands:

    Copy this prompt
    Please double-check your figures. Show me how you calculated the regional totals, confirm the grand total of all sales, and tell me anything you assumed or weren't sure about.

    Then do one check yourself. Total the Total column in Excel (or use a quick pivot table) and compare it with the AI's grand total. If one key number matches, you can trust the rest a great deal more.

  6. Turn the numbers into words

    This is where AI saves the most time. Ask it to write what you'd otherwise spend an hour on:

    Copy this prompt
    Write a short email to my manager summarising this year's sales so far: the headline total, the best and worst regions, the trend by month, and the lapsed customers we should contact. Friendly and professional, under 150 words, with 3 bullet points. Mention that October is partial.

    Or for a meeting:

    Then try
    Turn the key findings into text for 3 PowerPoint slides: a title and 3 short bullet points each. Suggest which chart goes on which slide.

    Always read it before sending. Make sure every number in the text matches the numbers you checked.

ADVERTISEMENT · IN-ARTICLE
If it goes wrong

Quick fixes

"I can't open this file"

Save it as CSV (File › Save As › CSV) and try again. Remove password protection, and split very large files.

The answer is clearly wrong

Say so, and say why: "That can't be right. North can't be £90,000 when all sales together are about £57,000. Please recheck how you calculated it."

The AI gives different answers each time

Ask it to show its calculation step by step, then ask for the method it used. Stick with a conversation where you've checked the numbers once.

The chart looks rough

Describe what you want: colours, font size, labels on the bars, no gridlines. Or ask for the summary table and make the chart in Excel yourself.

It made up a reason

AI may guess why sales dipped ("probably seasonal"). Treat any "why" as a suggestion to check, not a fact.

Try it on other jobs

Use it on your own work

Survey results

"Summarise the free-text comments into the 5 most common themes, with how many people mentioned each."

Expenses

"Which cost categories went up most compared to last year? Flag anything over £500 without a receipt reference."

Timesheets

"Who is regularly working over 45 hours a week? Show a chart by person."

Stock

"Which products are selling faster than we're restocking? When will each run out at this rate?"

Customer feedback

"Group these complaints by product and urgency, and draft a reply template for the most common one."

Choosing the right tool

When to use this, and when to stay in Excel

Use the AI for one-off questions, quick charts, first drafts of reports, and finding patterns you didn't know to look for. Stay in Excel, with formulas, pivot tables and a dashboard, for anything you'll repeat every week, share with colleagues, or need to audit later. The best results come from using both: AI to explore, Excel to make it permanent.

Keep going

What to build next