A spreadsheet rarely announces that it is wrong. Its columns align, amounts have two decimal places, and the total looks convincing. The real inspection starts when someone asks why the same order appears twice.
Small, checkable problems are a useful starting point for AI assistance. Copilot in Excel is the example here. Microsoft's help describes analysis, formulas, and editing, distinguishing editing, planning, and chat-only modes. Availability depends on subscriptions and organizational settings. Official introduction
Practice without affecting real orders
Create a workbook containing these fictional rows. Amounts are in Chinese yuan.
| Order | Quantity | Unit price | Recorded amount |
|---|---|---|---|
| A1001 | 2 | 19.90 | 39.80 |
| A1002 | 1 | 39.00 | 39.00 |
| A1002 | 1 | 39.00 | 39.00 |
| A1003 | 3 | 12.00 | 30.00 |
There are two kinds of issue: a repeated order number and an amount that does not equal quantity times price. In a real business, a repeated number could represent separate line items; a different amount could reflect a discount. AI cannot invent the governing rule.
For this exercise, specify one record per order, with no discounts, shipping charges, or refunds. Correctness is defined within those assumptions.
Ask before editing
Open Copilot in the workbook. Following the documentation used for this guide, choose “Chat only” first so the response stays in chat. If your version differs, check your account's help rather than hunting for buttons from another interface.
Analyze this order table without changing it. Each order should have one record. Amount equals quantity multiplied by unit price, with no other charges. Separately identify duplicate order numbers and inconsistent amounts. Give locations, evidence, and questions that need confirming. Do not delete records.
“Data anomalies found” is insufficient. You need to return to specific cells. If the tool immediately supplies a repaired total, ask which row it removed, which amount it changed, and why. Hidden intermediate steps make an answer harder to trust.
Check with your own formulas
With headers in row 1 and quantity and price in columns B and C, enter =B2*C2 in E2 for the expected amount. Enter =D2-E2 in F2 for the difference. Fill down. The final row's difference should be minus CNY 6.
Check the duplicate A1002 records separately. Once the exercise confirms duplicate entry, retain one and change A1003 to CNY 36. The final total is CNY 114.80: 39.80 + 39.00 + 36.00.
That is the exercise's human-written answer key, not a measured Copilot result. Cover it and calculate first if you prefer.
If answers differ, inspect the disagreement. Were order identifiers treated as numbers? Were text-formatted amounts ignored? Were both copies deleted? Was an unstated discount assumed? Finding the cause helps more than asking for the total again.
Allow one kind of change at a time
After understanding the analysis, use an editing mode for a bounded request: add expected-amount and difference columns while preserving original amounts. Check that references change correctly down the rows, the final row is included, and source data remains intact.
Treat deletion as another step and retain the original or a copy. You can then explain the transformation and reverse it if the business rules were misunderstood.
Scale up only after these steps work. One successful exercise does not cover merged cells, multiple currencies, or references between sheets.
Measure the whole task
Count time spent checking formulas, explaining exceptions, and repairing mistaken flags—not just seconds spent generating a formula. Also ask whether the exercise helps you know where to start when the next table arrives.
This workflow follows documentation; it was not performed in your workbook and does not replace business review. Explaining one duplicate and one discrepancy clearly is a useful first result.