Which AI data analysis tools should you start with?

Start inside the spreadsheet you already use. For this first exercise, choose between Excel with Copilot and Google Sheets with Gemini according to where your records live.

Copilot answers questions about structured spreadsheet data. Gemini can analyze Sheets data and generate charts. These capabilities are described in the Microsoft instructions and Google instructions.

ChatGPT data analysis may also be on your shortlist. The sources used here do not document its features, so this guide makes no claims about its uploads, plans or limits. The walkthrough focuses on the two spreadsheet options documented above.

Before paying for anything, test your actual question with the features available in your account. Judge the result by the checking and correction work required.

What should you fix before asking for analysis?

Make a working copy and write down what a row means. An order, an item within an order and a daily total should not share an unexplained definition.

Give each column a clear job

For our hypothetical store, use order date, order ID, product, quantity, amount and status. Define amount explicitly, including whether discounts and refunds have already been applied. Keep the currency consistent within the comparison.

• Use one date convention throughout the file.
• Keep totals separate from individual records.
• Review inconsistent product names before grouping them.
• Mark missing values for investigation rather than inventing replacements.
• Check repeated order IDs before deleting anything.
• Record which statuses belong in the analysis.

If one order contains several products, keep that detail visible. Tell the assistant whether you want item counts or distinct orders. In your instructions, define distinct orders as counting each order ID once.

Choose one business measure

Decide whether the question concerns money collected, units sold or completed orders. Write the definition beside the spreadsheet. For help choosing the measure, use the guide to small business KPIs, meaning the numbers you use to track progress.

Exclude customer names, phone numbers and addresses when the question does not need them. Use an account approved for your business information. Keep a short note explaining who prepared the file and which records were left out.

Before continuing, ask a colleague to explain the table back to you. If they cannot tell what amount or status means, revise those definitions first.

How do you ask AI a data question in plain language?

Write the question as a short brief: the decision, the records, the comparison and the requested output. A prompt is simply the instruction you type into the assistant.

Microsoft recommends starting with a specific question and refining the response through follow-ups. Use that approach when drafting your own Copilot analysis request.

A useful first prompt

I need to choose which products to investigate this week. Use the Orders sheet and compare the two complete weeks I specify below. Include completed orders only. Each row represents one product within an order.

First state the date range, included row count and columns used. Then show product, earlier week amount, later week amount and difference. List missing values separately. Do not guess why a product changed.

Replace the week description with exact dates before sending. If your file has no status column, resolve that gap first. Do not ask the assistant to infer whether an order was completed.

Follow the answer with a narrower question

Try: “Show the daily totals behind the largest decline.” Then ask: “Which original records contributed to those totals?” Keep each follow-up focused on a check you can perform.

For reusable starting points, see spreadsheet prompt examples. Adapt the column names and output format rather than copying a template unchanged.

Leave recommendations until after the arithmetic review. Ask for observations and proposed explanations in separate sections, with unsupported explanations marked as questions to investigate.

How do you run the analysis in Sheets or Excel?

Use the prepared question in one tool, then inspect its first answer before requesting a chart. Keep the original file and your working copy clearly labelled.

Gemini in Google Sheets

Google specifies an eligible Workspace or Google AI plan. Open the spreadsheet, select Ask Gemini and enter your request in the side panel. The instructions recommend saving an Excel file as Google Sheets when using these features. Google's setup instructions.

1. Identify the sheet and range you want examined.
2. Submit the comparison prompt.
3. Check the reported period and included records.
4. Review the calculation against your spreadsheet.
5. Request a chart only after accepting the table.

Gemini offers analysis steps for review. Its generated chart does not respond to changes in the original dataset; an inserted chart uses data on a newly created tab. Treat refreshing the report as a separate task. Google's analysis and chart guidance.

For more setup detail, follow the dedicated Gemini in Google Sheets guide.

Copilot in Excel

Select the Copilot icon and type the question in its chat pane. The documented direct-answer feature supports table-like data, not unstructured material. Its returned static tables and image visualizations are not refreshable reports. Microsoft's feature description.

After receiving the answer, request the calculation method in plain language. Check which columns and filters were used before accepting the result. Save your final prompt alongside the reviewed output.

How can you tell whether the answer is usable?

Check the records, the arithmetic and the interpretation separately. For your first analysis, choose one product and reproduce its total directly from the original rows.

A hypothesis is a possible explanation you still need to test. In a hypothetical store, a falling product total should prompt questions about stock availability, recording gaps and the comparison period. Do not label it lost customer interest without investigating.

Repair a mismatch explicitly

Your total differs from my control total. Show the included records, excluded records and calculation rule for this product. Identify the source of the difference before rewriting the summary.

For percentage changes, ask the assistant to flag products with a zero starting value. Request an absolute difference for those cases instead. Write that exception into your reporting instructions.

My recommendation is to pause any spending decision while important totals remain unresolved. Keep a correction note so the same problem is checked during the next run.

What should your first AI business report contain?

Produce a decision note that someone can read without opening the chat history. Include the question, the period, the checked finding and the next task.

Use this report outline

• Question: what business decision are we preparing for?
• Coverage: which dates, records and exclusions were used?
• Finding: what comparison survived the checks?
• Open question: what information is still missing?
• Action: who will investigate or change something?
• Review: when will we examine the result again?

For the hypothetical store, a sensible action might be to review availability for a declining product. Present that as an investigation, not a promise of higher sales.

Schedule the first week around preparation, analysis, checking and a short team review. Record time spent correcting the answer as well as time spent generating it. Use your own experience to decide whether to repeat the workflow.

Once product reporting is reliable, consider a separate question about customer groups using the customer segmentation guide. For future planning, continue with sales forecasting with AI. Keep a forecast clearly separate from a report of past sales.