What AI can and cannot do with a spreadsheet
AI assistants can clean messy columns, answer questions about a table, spot trends and draft a summary in plain language. They can also miscalculate, misread a column or invent a pattern. The workflow below keeps the speed and adds checks.
The workflow
| Step | What to do | Check |
|---|---|---|
| 1. Define the question | Write the one decision the analysis should inform | Can you state it in a sentence? |
| 2. Prepare the data | Remove confidential or personal fields before you upload | Does your organisation allow this tool for this data? |
| 3. Clean | Ask the assistant to flag duplicates, inconsistent formats and missing values | Review each proposed fix before accepting it |
| 4. Ask | Pose business questions in plain language, one at a time | Ask it to show the calculation it used |
| 5. Chart | Request a simple chart for the main pattern | Axes, labels and scale |
| 6. Explain | Ask for a short narrative of patterns and possible causes | Separate what the data shows from what the AI guesses |
| 7. Verify | Recompute key numbers yourself | Totals reconcile; spot-check rows |
A worked example
Suppose monthly sales were ₹4,20,000 in August and ₹4,62,000 in September, from 1,200 and 1,320 orders.
| Metric | August | September | Change |
|---|---|---|---|
| Revenue | ₹4,20,000 | ₹4,62,000 | +10% |
| Orders | 1,200 | 1,320 | +10% |
| Average order value | ₹350 | ₹350 | 0% |
An assistant might say 'revenue grew 12% because customers spent more'. Your recomputation shows 10% growth driven entirely by order volume, with average order value flat. Checking the numbers changed the story — which is the point of the verification step.
A verification checklist
- Recompute totals and at least two headline numbers with a simple formula
- Spot-check five random rows against the source
- Confirm the assistant used the right columns and date range
- Look for misleading charts: truncated axes, mixed scales, missing labels
- Ask the assistant what it assumed and what it could not tell from the data
- Never share confidential data with a tool your organisation has not approved
When to use proper tools instead
For large datasets, repeatable reporting or sensitive data, use spreadsheet functions, SQL or a BI tool. See our 12 SQL patterns, Power BI careers and data visualization skills.
Where this is taught
IITM Pravartak's three-month Applied Generative AI programme includes a module on business data analysis — preparing and cleaning data, conversational analysis, growth metrics, dashboards and verifying AI analysis — taught without formulas. See the programme guide.
Explore the programme
See the curriculum, fee and enrolment details on the programme page, or talk to an advisor.
Frequently Asked Questions
Can I analyse a spreadsheet with AI if I cannot code?
Yes. Many assistants accept a spreadsheet and answer questions in plain language, but you must verify the numbers they return.
How do I know the AI's numbers are right?
Recompute key totals and growth rates yourself, spot-check rows and ask the assistant to show its calculation.
Is it safe to upload company data?
Only if your organisation's policy and the tool's terms allow it. Remove confidential and personal fields where you can.
When should I use SQL or a BI tool instead?
For large datasets, repeatable reports and sensitive data, where accuracy and an audit trail matter.