| name | excel-analysis |
| category | kumite-office |
| description | Analyze, transform, reconcile, and visualize tabular data in Excel with M365 Copilot — formulas, pivots, charts, conditional formatting, Power Query. Use whenever the input or output is a spreadsheet (variance commentary, headcount roll-up, pipeline hygiene, budget reconciliation, cohort analysis, forecast). Triggers: "analyze this spreadsheet", "build a pivot", "reconcile", "variance", "forecast", ".xlsx". |
Excel Analysis
Excel mistakes ship to executives. Verify every formula and every total before the file leaves your hands.
Cowork skill. Drives Copilot in Excel for human analysts. For programmatic .xlsx generation, use the agents dojo.
Quick Reference
| Task | Approach |
|---|
| Summarize a sheet | Define schema first -> ask Copilot for column-by-column profile |
| Pivot / cross-tab | Name rows, columns, values, filters explicitly |
| Reconcile two sources | Key column on both sides -> XLOOKUP or merged Power Query, then exception list |
| Variance commentary | Pivot -> top-N drivers -> narrative grounded in the pivot, not vibes |
| Forecast | State assumptions, horizon, granularity; never accept hidden constants |
| Chart for a deck | Specify chart type, axes, units, sort order — never "make a nice chart" |
| Find a bug | Trace precedents/dependents; ask Copilot to explain the formula in English |
When to Use
- Variance commentary, budget vs. actual, forecast roll-ups.
- Pipeline / funnel / cohort analysis.
- Reconciling two sources (HRIS vs. payroll, CRM vs. invoicing).
- Producing a chart or pivot for a deck or memo.
- Cleaning a vendor data dump before downstream synthesis.
How to Use
- Show Copilot the schema first. Headers, data types, key/foreign-key columns. "Column A is employee ID — unique. Column F is FY26 budget in USD."
- Convert to a Table (
Ctrl+T). Tables make ranges stable, formulas readable, and Copilot dramatically more accurate.
- Ask for a formula in plain English, then read it. "SUMIFS over column F where region = 'EMEA' and quarter = 'Q1'." Read the returned formula before pasting.
- For pivots, name the dimensions and measures explicitly. "Rows: cost center. Columns: quarter. Values: sum of actual, sum of budget, variance %."
- Always sanity-check totals against a known reference — last month's reported figure, the source extract, a hand calculation on three rows.
- Charts: name axes, units, chart type, sort order. Don't accept "make a nice chart."
- Lock the workbook (protect sheets, hide intermediate tabs) before sharing externally.
Examples
| Don't: Anti-example | Do: Good example |
|---|
| "Analyze this spreadsheet." | "Build a pivot: rows = product line, cols = month, values = sum(net revenue). Then a 3-month rolling average column." |
Trusting XLOOKUP without testing the unmatched case. | Test with a deliberately missing key; confirm the if_not_found argument. |
| "Make a chart of sales." | "Clustered column, x = quarter, y = revenue (USD millions), series = region, sorted by FY26 desc." |
| Pasting numbers into chat. | Attaching the workbook so Copilot sees columns, types, and named ranges. |
Critical Rules
- Convert ranges to Tables. Named tables eliminate whole classes of formula errors.
- No volatile functions in shared workbooks (
OFFSET, INDIRECT, NOW, RAND) — recalc storms and audit pain.
- No hard-coded totals. If it should be a formula, make it a formula. Hardcodes go stale silently.
- Reconcile against a known control before publishing any number.
- Document assumptions in a dedicated
Assumptions sheet — every analyst will thank you.
- Pivot tables must be refreshed before the file leaves your hands (
Data -> Refresh All).
- Never let Copilot invent units. Currency, scale (millions vs. thousands), period — confirm explicitly.
Common Pitfalls
- Accepting Copilot's row count without filtering blanks/headers/totals embedded in data.
- Hidden rows/columns that Copilot reads but you can't see — surfaces wrong totals.
- Merged cells that break sorting, filtering, and structured references. Unmerge before analyzing.
XLOOKUP returning #N/A silently — set if_not_found to a sentinel ("MISSING") and filter for it.- Using
SUM over a column that contains text "0" instead of numeric 0.
- Time-zone or date-format drift on imported CSVs (US vs. EU date order).
- Charts where the y-axis doesn't start at zero — visually misleading, especially in finance.
- Conditional formatting that survives sharing but slows the workbook to a crawl.
Anti-Patterns
- Pasting raw spreadsheet data into a chat instead of attaching the file.
- "Just give me the answer" — no audit trail = no defensible analysis.
- Multi-sheet roll-ups that depend on cell positions (
'Q1'!B7). Use named ranges or tables.
- Letting Copilot guess at currency conversion / FX rates. Source them, cite them, lock them.
Verify Before Send
Related
