--- title: "Copilot in Excel: prompts" client: "Mycronic CMT Copilot Convoy" language: en date: "2026-10-17" session_date: "2026-10-16" version: "1.0.1" canonical_url: "https://copilotinexel.jardenberg.net/prompts.html" --- 05 · Prompts to get going # Twenty-four prompts, one ladder. Understand, clean, calculate, report, visualize, verify, repeat. Copy one, change the words in brackets, press Enter. Two habits make every prompt below work better. First, name the sheet and the columns you mean; Copilot's own tips page says the same. Second, start broad and refine in the follow-up rather than writing a perfect first prompt. The conversation is the tool. Each prompt says which door it is for ([Copilot Chat or the pane inside Excel](https://copilotinexel.jardenberg.net/two-doors.html)) and which mode I would use. The square brackets are yours to fill. If a prompt asks Copilot to put things on a new sheet, that is deliberate: your source data stays untouched. Rung 1 ## Understand: what is actually in here? 01 · The workbook you did not build · Excel pane · Chat only ```text I did not build this workbook. Explain what is in it: each sheet, what each column means, how the sheets relate to each other, and anything that looks inconsistent, incomplete or hard-coded. Do not change anything. ``` 02 · The insight menu, Friday's opener improved · Copilot Chat with files added ```text What insights could I get from these files? Give me ten questions this data can answer well, and three questions it cannot answer and why. Then suggest the one question you would start with. ``` 03 · Data quality first · Excel pane · Chat only ```text Before we analyse anything, list the data quality problems on the sheet [Claims]: blank cells in key columns, duplicates, inconsistent spellings in [Account Subsidiary], dates stored as text, and obvious outliers. Show me the affected rows. Do not fix anything yet. ``` Rung 2 ## Clean: make the data trustworthy. 04 · Supervised clean-up · Excel pane · Plan or Allow editing ```text @clean-data Clean the table on [Claims]. Trim extra spaces, make the spellings in [Account Subsidiary] consistent, convert text dates in [Claim Date] to real dates, and remove exact duplicate rows. First copy the original data to a sheet called Raw. Show me each type of fix and how many cells it affects before you apply it. ``` 05 · The PDF that should have been Excel · Copilot Chat with the PDF attached ```text This PDF contains tables that I need as a spreadsheet. Extract every table into one Excel workbook, one sheet per table, keep the column headers, keep numbers as numbers and dates as dates. List the cells you were unsure about so I can check them. ``` A colleague in the group did exactly this last week and called it a minor thing. Oskar called it one of the major headaches of the last century. Both are right. 06 · Wide to long · Excel pane · Allow editing ```text The table on [Forecast] has months as columns. Turn it into a long table on a new sheet with the columns Date, Product family, Metric, Value, so I can build PivotTables on it. Keep the original sheet unchanged and explain the steps so I can repeat them next month. ``` Rung 3 ## Calculate: formulas, not pasted numbers. 07 · One new column, checkable · Excel pane · Allow editing ```text Add a column to the table on [Claims] called [Days to resolve] that calculates [the number of days between Claim Date and Closed Date, blank if not closed]. Use a formula with structured references. Explain the formula in one sentence and point me to three rows where I can check it by hand. ``` 08 · Bring data across · Excel pane · Allow editing ```text Bring [Region] from the sheet [Accounts] into the table on [Claims], matching on [Account]. Use XLOOKUP, return "not found" where there is no match, and tell me how many rows did not match and which account names they have. ``` 09 · Summary by dimension · Excel pane · Allow editing ```text Create a summary on a new sheet: [number of claims] by [Product family] and [month], with a total row and each row's share of the total. Use formulas (COUNTIFS or SUMIFS with structured references), not pasted values, so it updates when the data changes. ``` 10 · The PivotTable and the reading · Excel pane · Allow editing ```text Build a PivotTable on a new sheet with [Product family] as rows, [Account Subsidiary] as columns and the count of [Claims Title] as values, sorted by total descending. Then tell me the top three combinations and the one thing in this table you think I should ask about. ``` 11 · Assumptions you can play with · Excel pane · Plan ```text Create an Assumptions sheet with named input cells for [growth rate, unit price, warranty rate]. Link the forecast on [Forecast] to those cells so that changing an assumption updates the result. Mark every cell I am supposed to edit with a light yellow fill and leave everything else protected from accidental edits. Show me the plan before you change anything. ``` Rung 4 ## Report: from data to something a reader can use. 12 · The one-page report, Friday's block title · Excel pane · Plan ```text Create a one-page report sheet for [the aftermarket management team] covering [September 2026]: a headline sentence, the three numbers that matter most with comparison to [the previous month], one chart, and a short section called "What changed and why" that only uses facts from this workbook. Link every number to the data by formula. Show me the plan first. ``` 13 · Variance sentences without invention · Excel pane · Allow editing ```text Compare the Actual and Plan columns in the table on [Variance]. For every line where the variance is beyond plus or minus 10 percent, write one factual sentence in the Explanation column that uses only numbers from this workbook. Where the workbook does not explain the variance, write "explanation needed" instead of guessing. ``` 14 · The summary for the email · Excel pane · Chat only ```text Write a 120-word summary of the sheet [Report - 2026-09] for an email to [the division head]. Lead with the decision that is needed. Every number must come from the sheet; give me the cell reference for each number so I can check before I send. ``` Rung 5 ## Visualize: one chart, one message. 15 · A chart that says something · Excel pane · Allow editing ```text @chart-design Create one chart that shows [that claims for the X-series doubled between Q2 and Q3 while all other families were flat]. One chart, one message. Make the title say the message, not the metric. Label the axis with the unit. Put it on the sheet [Report]. ``` 16 · Let it argue the chart type · Excel pane · Allow editing ```text Which chart type best answers the question "[how is the mix of failure modes changing over time]" for this data, and why? Make that chart, then make the strongest alternative next to it on the same sheet so I can compare and choose. ``` 17 · The infographic, Friday's rough prompt tidied up · Copilot Chat ```text Create an infographic from the table above: one headline, the three most important numbers, and a simple visual of [cases by region]. Keep every number exactly as in the table. This is for my own thinking, not for publication, so clarity beats decoration. ``` Rung 6 ## Verify: the rung that makes the others safe. Oskar's third frustration on Friday was the convincing answer that is not quite right. His advice, and mine: ask at the end of every conversation what is actually true. Copilot will speak with confidence from the data it has, and with equal confidence past the edge of it. These four prompts find the edge. 18 · The number anchor · either door ```text Before I trust this: show me exactly which rows and columns you used to get [145 claims for the X-series]. Describe the filter you applied and the count. Then calculate the same number a different way and tell me if the two agree. ``` 19 · Data versus assumption · either door ```text Separate what you know from this data from what you assumed. List every assumption you made to produce the result above, and rate your confidence in each conclusion as high, medium or low with a one-line reason. ``` 20 · The evidence-gap sheet · Excel pane · Allow editing ```text Add a sheet called Evidence gaps. List every place where the data was missing, ambiguous or contradictory while you built [the report], how you handled each case, and what I would need to supply to close the gap. ``` 21 · Three checks by hand · either door ```text Give me three quick checks I can do by hand in two minutes to confirm that [the report] is right, with the exact cells to look at. Then tell me the most likely way this report could be wrong without either of us noticing. ``` Rung 7 ## Repeat: never explain the same thing twice. 22 · Pull the plug, keep the instructions · either door ```text This conversation has become long. Summarise all the instructions, preferences and corrections I have given you in this chat as a numbered list that I can paste into a new conversation. Include the sheet and column names. Then stop. ``` That is Oskar's second workaround from Friday, and the one I use most. Paste the list into a fresh chat and the misunderstandings usually disappear with the old context. 23 · Save the workflow as a skill · Excel pane ```text @skill-builder Save what we just did as a new skill called [monthly-claims-report]. Include the sheet names, the report columns, the chart, and the verification steps we ran. Make the description say when to use it. ``` 24 · Add a rule to the workbook · Excel pane · Allow editing ```text Add a rule to the .Rules sheet: "[Treat an absent value as missing, not zero, and never invent actuals or plans.]" Then tell me in one sentence how you will apply it the next time I ask for a report. ``` If a prompt does nothing Check the mode (Chat only cannot edit), the calculation setting (must be Automatic), and whether the skill you @mentioned exists in your account. Then narrow the context: name the sheet, name the columns, add the file. That was Oskar's first workaround and it solves most of it. [Put them to work in one hour →](https://copilotinexel.jardenberg.net/practice.html)[Save the corrections you keep making](https://copilotinexel.jardenberg.net/personalize.html) Prepared by AI + JJ. Post-session self-study material after Convoy #4 (16 October 2026), checked against Microsoft documentation on 17 October 2026. Prompts and rules are proposed wording to adapt.