Level 2 exam — Microsoft Copilot 365

Exam — fresh questions

12 original questions: recall (25%) + application (50%) + arbitration (25%) on Copilot in Excel.
Answer everything, then click Check: your score and explanations will appear.

2.1 · Recall

1. Why does Copilot stay “greyed out” or silent in Excel?

The workbook contains too many tabs The file is too heavy for the AI The data sits in a free range, not in a formal Table (Ctrl + L) The workbook font is too small

Copilot requires a named Table (Ctrl + L): that is 90% of blockages.

2.1 · Application

2. Hicham prepares his export: floating title, merges, blank rows. What does he do before Ctrl + L?

He merges even more cells He sends the file as is to Copilot He converts first, he will clean up afterwards He cleans up: title removed, one header row, one value per cell

Clean Table = unique headers, zero blank rows, one value per cell — then Ctrl + L + name.

2.2 · Recall

3. What should you keep in mind when facing a correlation flagged by Copilot?

Correlation is not causation: you interpret It is false on principle It replaces human analysis It always proves a cause

Copilot flags that two columns move together; interpretation remains human.

2.2 · Application

4. Hicham receives a “historic record” in May. Which reflex before presenting it?

Put it as the chart title right away Check column, range and source row (duplicate? entry error?) Delete the row without looking Change the colour of the chart

An anomaly is a lead to confirm at the source, not a verdict.

2.2 · Application

5. Which question does Hicham ask to make reliable a Copilot figure reused in a decision?

“Can you send it by email?” “Can you translate it into English?” “Can you make it bold?” “Which column and which range are you relying on?”

Asking back for the “how” avoids turning a wrong analysis into a decision.

2.3 · Recall

6. What must you give Copilot to get a correct formula?

The exact column names, the logic and the expected result The file password The printer name The cell colours

Precise English (columns, logic, expected) makes an accurate formula.

2.3 · Application

7. Hicham receives =SI([@Montant]>10000;[@Montant]0,05;[@Montant]0,03). What does he do before adopting it?

He adopts it as is, it looks good He asks “explain” then tests on a row at exactly 10 000 He recopies it by hand row by row He changes the commas to dots

“Explain” + test on a known case including the edge case: plausible does not mean correct.

2.3 · Arbitration

8. A colleague adopts all generated formulas without testing them “to go fast”. Hicham's verdict?

Replace them with hand-typed values Keep them: Copilot is never wrong Reject them: threshold, range or logic may be wrong — explain + test each formula Keep them except for IFs

A formula error spreads to the whole Table: testing on a known case is non-negotiable.

2.4 · Application

9. Hicham wants to show whether the North is slipping. How does he ask for the chart?

“Put blue everywhere” “Copy the neighbour's chart” “Make a stacked bar chart series 3” “Show the monthly sales trend by region, North highlighted”

Describe the message, not the type: Copilot chooses the suited visualisation.

2.4 · Arbitration

10. The proposed chart shows a spectacular “slippage” of the North, axis starting at 8 000. Verdict?

Reject as is: misleading truncated axis — restart from zero and judge honesty Distribute it with more colours Replace it with a 12-slice pie chart Distribute it as is: it is spectacular

A chart must tell the truth: zero-based axis, non-misleading scale, readable slices.

2.5 · Application

11. Hicham wants a budget/sales regression without knowing how to code. What does he ask for?

“Convert the file to PDF” “Run a linear regression sales ~ budget, scatter plot, trend line and R² via Python in Excel” “Guess next year's budget” “Print the workbook in A3”

Python in Excel + Copilot: advanced analysis described in English, live result in the workbook.

2.5 · Arbitration

12. The R² comes out at 0.72 and a colleague wants to vote the budget right away, with sensitive HR data in the workbook. Verdict?

Reject because Python is useless Vote for it: 0.72 proves causation Frame it: R² explained and judged (correlation ≠ causation) + cloud policy checked before anything sensitive Vote for it adding a 10% margin

Copilot calculates, the human interprets; the cloud is framed with the data policy.

What's next?

Score ≥ 80%: move on to Level 3 — Copilot in advanced PowerPoint and Word, building on your Excel reflexes (Table, verification, testing).
Score 60-79%: revisit the failed sections (2.1 Ctrl + L, 2.2 column/range, 2.3 explain + test, 2.4 honesty, 2.5 R² and cloud) and redo the 48h challenges.
Score < 60%: rebuild Hicham's project step by step (Ventes2025 as a Table, 3 questions, 1 tested formula, 1 honest chart, 1 explained Python analysis), then retake the test.
In all cases: keep the reflexes — Table first, correlation ≠ causation, tested formula, honest chart, judged model.