The spreadsheet has several tabs, so you ask AI to find insights. It returns trends, averages and a chart. You still do not know whether the team should change anything on Monday.
Before using AI to analyse a spreadsheet, define the business question, population, period and comparison. Decide what a result would help you do. A broad request for insights leaves too much room for attractive but irrelevant calculations.
Photo by Vitaly Gariev on Unsplash.
Write a question card
Keep the request short enough to discuss with a colleague. Record the decision being considered, the rows that belong in the analysis and the meaning of the important fields.
For an illustrative service team, the question might be whether one onboarding stage is creating more waiting than another. That requires stage dates and a consistent case definition. It does not necessarily require a comparison of revenue or employee activity.
Use these fields:
Decision: what might change after the analysis?
Population: which cases are included and excluded?
Period: which dates define the comparison?
Measure: what is counted or calculated?
Baseline: what is the result being compared with?
Limitation: what cannot be concluded from this file?
Agree the unit of analysis
One spreadsheet row may represent a client, invoice, task or event. Counting rows without settling that meaning can answer the wrong question.
If a client appears five times because there are five tasks, the row count is not the client count. Explain how duplicate identifiers and reopened cases should be handled. A tool cannot settle those definitions from a column heading alone.
Ask it to describe the proposed calculation before producing a conclusion. This gives the owner a chance to catch a wrong denominator or time boundary early.
Start with one useful comparison
For the onboarding example, compare waiting time by stage using the same start and finish rules. Keep incomplete cases visible rather than silently mixing them with finished cases.
The result might suggest where to investigate. It does not establish why the delay occurred. A cluster of long waits could reflect a missing dependency, a different client mix or an inaccurate timestamp.
Microsoft's current guidance describes asking natural-language questions about spreadsheet data. The question-card method here is a business preparation step, not a claim that every AI product can inspect every workbook. Check the actual tool's supported files and your permission to use the data.
Keep calculations inspectable
Retain the source snapshot, filters and calculation method. Ask for a result that somebody can reproduce in the spreadsheet or another appropriate analytical tool.
Where AI supplies a narrative interpretation, distinguish observed differences from possible explanations. Do not let a confident paragraph turn a comparison into a causal claim.
Finish by recording the next business action, including when the sensible action is to gather better information rather than change the process.
For related practical guidance, see defining operational metrics and maintainable data-quality checks.
Hili Consulting can help frame spreadsheet analysis around a clear decision and comparison. Arrange a free 30-minute consultation and bring one example of the work you want to improve.
