MAX’S NOTES
How to analyze a spreadsheet with AI and check the results
AI can help you spot changes and unusual values in a spreadsheet when you ask a specific question and explain the data. Ask it to show formulas and source rows, then verify important numbers in the spreadsheet itself.
AI can help with a first pass through a spreadsheet: compare periods, flag sharp changes, and turn observations into a short list. To avoid getting a confident but unhelpful paraphrase, ask about a specific metric, state the period, and request the calculation. Then check a few numbers yourself.
1. Prepare the spreadsheet
Before you upload it, make the data easy to read:
- put column headers in the first row;
- use one row for each record;
- make column names explain what they contain;
- do not mix, for example, dates and comments in one column.
OpenAI’s data-analysis guidance recommends descriptive headers and one record per row. It also notes that available file types can depend on the model, plan, workspace settings, and account capabilities. If spreadsheet uploads are unavailable, you can paste a small anonymized excerpt or use a tool built into your spreadsheet.
2. Ask for one result
“Analyze everything” is too broad. Start with one question, such as “How did revenue change by month?” or “Which categories had the largest increase in product cost?”
Add context: the period, relevant columns, comparison method, and answer format. OpenAI’s guide recommends stating the decision the analysis should support, definitions, timeframe, and what key columns represent. A specific question makes it easier to see what the tool actually calculated.
3. Ask it to show the calculation
Example prompt:
> Compare January and February for revenue and food-cost share. Use only the “Revenue” and “Food Cost” columns. Calculate food-cost share as food cost ÷ revenue × 100%. Show the calculation in a table, include the source values, and round percentages to whole numbers. Do not explain the reasons for changes unless those reasons appear in the data.
Suppose the illustrative example has January revenue of ₽1,000,000 and food cost of ₽310,000. February has ₽1,100,000 and ₽363,000. Food-cost share is 31% and 33%, while revenue rises by 10%. This is arithmetic on illustrative data, not the result of testing a specific AI tool.
4. Check the results
Compare the source values with the spreadsheet. Recalculate at least one period. Make sure the same months and metric definitions are being compared. If the tool names a cause, ask which column supports it. Without that evidence, treat the cause as a hypothesis, not a finding.
Ask AI to separate observations from assumptions and list data limitations. For example, “Revenue fell by 8%” is an observation if the calculation is correct. “Revenue fell because of the weather” is a hypothesis if the spreadsheet contains no weather data or other evidence for the cause.
Microsoft recommends reviewing and verifying Copilot output. For analysis of large sets of text comments in Excel, Microsoft describes checking results against source data references; this applies to a specific mode and tool.
Prompt template
> I want to understand [decision or question]. Use the period [dates] and columns [names]. First check the spreadsheet for missing values and unusual entries. Then [specific comparison or calculation]. Show the source values, formula, and result as a [table/list]. Separate facts in the spreadsheet from hypotheses. Do not infer causes unless the data supports them. At the end, list analysis limitations and 2–3 numbers I should verify.
The basic rule is simple: AI suggests what may follow from the spreadsheet; you check the calculations and decide whether the data supports the conclusion.