Microsoft Copilot for Excel.
Here's what it actually does.
Microsoft Copilot for Excel lets you analyze data, write formulas, clean messy tables, build charts, and spot patterns in your spreadsheets using plain-English prompts. You type what you want to know (or do), and Copilot works in the grid with your actual data. As of June 2026, the Copilot panel sits inside Excel on the desktop and in Excel for the web, and the May 2026 redesign made it substantially easier to discover and use in daily work.
I've run Copilot in Excel sessions with organizations across just about every category (insurance, banking, healthcare, startups, retail, and more), from mid-sized to enormous. I've also delivered dozens of these sessions in English over Zoom with distributed teams. What follows is a practical breakdown of what Copilot actually does inside Excel: where it earns its keep, where it falls short, and how to get useful output fast. If you want the broader picture, see the complete Microsoft Copilot guide.
What Microsoft Copilot for Excel does
The short version: Copilot in Excel is a conversational layer on top of your spreadsheet. You describe what you want in plain English, and it acts on your data (adding columns, generating formulas, creating charts, filtering rows, summarizing ranges). It works inside your existing workbook. It doesn't create a separate document or require you to paste data somewhere else.
The five things it does most usefully are natural-language analysis, formula writing, data cleanup, chart creation, and (as of early 2026) Agent Mode for multi-step automation. Each of those is worth understanding separately, because they have genuinely different strengths and different failure modes.
01Natural-language data analysis
You can ask Copilot to summarize a dataset, find the top performers in a column, compare two groups, or flag anomalies. It reads your table, figures out the column structure, and returns an answer in the Copilot panel. For quick questions that would normally require a pivot table, this is fast. The output can be an explanation, a highlighted range, or a new formula, depending on what you ask. (The key word here is "quick." For complex multi-variable analysis, it still helps to know what you're asking for.)
02Formula writing with Copilot in Excel
Copilot in Excel is genuinely good at generating formulas. Describe the calculation you want ("give me the average sales for Q2 where the region is EMEA"), and it writes the formula with the right cell references for your actual data. It also explains formulas you already have: paste in a VLOOKUP or an IF-AND chain you inherited from someone else, and Copilot tells you what it does in plain English. For finance and ops teams that work with complex models, this saves real time. It doesn't replace knowing Excel, but it removes the "I know what I want but I can't remember the exact syntax" friction that slows everyone down.
03Data cleanup and formatting
Messy imported data is one of Excel's most common pain points. Copilot can help split columns (first name and last name in one cell is a classic), standardize inconsistent text values, remove duplicates, and reformat date strings. You describe the problem, it generates the fix. This isn't magic: it still needs reasonably structured data to work with. But for the tedious cleanup work that eats the first half-hour of any analysis, it's useful. Teams that import data from CRMs, ERPs, or third-party exports tend to notice this one immediately.
04Charts and visualizations
Copilot can generate charts from selected data based on a plain-English request ("show me monthly revenue as a bar chart comparing this year vs last year"). It picks an appropriate chart type, sets up the axes, and adds a title. You can then edit the chart normally in Excel. This is especially useful when you know what insight you want to show but you'd rather not spend ten minutes wrestling with chart settings. That said, for publication-quality or highly specific visuals, manual polish is still necessary.
05Agent Mode: multi-step tasks
The most significant addition in 2026 is Agent Mode in Copilot. Instead of handling one request at a time, Copilot in Agent Mode can execute a sequence of steps: clean the data, add calculated columns, build a summary pivot, and highlight exceptions, all from one prompt. It shows you the plan before it runs, and you can approve or adjust it. This is the version of Copilot in Excel that genuinely changes how people approach repetitive analytical workflows. It's also the version that requires the most care with prompting, because a vague instruction at the start produces a lot of steps going in the wrong direction.
Real tasks people use Copilot in Excel for
Knowing the feature list is one thing. Knowing what people actually use in day-to-day work is more useful. Here are the tasks that come up repeatedly in the Excel Copilot sessions I run with organizations, along with notes on where the results hold up and where they need a second look.
"Explain this formula to me"
This is consistently the most popular thing people try first. You click a cell with a complicated formula (inherited from a colleague, copied from the internet, built three years ago), ask Copilot to explain it, and get a plain-English breakdown. It works reliably. Finance teams with complex models find this particularly useful during audits or onboarding. The explanation isn't always perfect for highly unusual formulas, but it's correct more often than not.
"Flag the rows where [condition]"
Highlight anomalies, find outliers, mark rows where a value exceeds a threshold or where two columns don't match. Copilot handles these reasonably well. It typically applies conditional formatting or adds a helper column with TRUE/FALSE values. For exception reporting in operations, logistics, or finance, this cuts several steps out of the normal workflow.
"Summarize this dataset"
You have a table of 5,000 rows. You want a quick answer: what's the total, what's the trend, who or what is performing above average. Copilot can return a narrative summary or a summary table. It's not a substitute for a real dashboard, but for a fast "what am I looking at here" read, it's faster than building a pivot from scratch. (I've seen analysts describe this as "the thing I didn't know I needed.")
"Create a column that calculates [X]"
Rather than writing the formula yourself, describe the calculation: "add a column showing the percentage change from last month," or "create a column that categorizes each row as high, medium, or low based on the score in column D." Copilot generates the formula and fills the column. It usually gets the logic right; the most common issue is cell reference drift in very large or oddly structured sheets, so it's worth checking the first few results before accepting.
"Clean up this import"
Data that comes from an export (CRM, ERP, survey tools) is often inconsistent. Copilot can standardize capitalization, remove leading or trailing spaces, parse combined fields, and flag duplicates. For ops and sales teams that work with imported contact or transaction data, this is one of the more time-saving uses. It doesn't catch every edge case, so a quick manual spot-check is still worthwhile.
How to prompt Copilot in Excel well
The biggest gap I see in organizations that have Copilot licenses and aren't getting value from them: they're prompting it the way they'd type a Google search. That doesn't work. Copilot in Excel responds to specificity. The more context you give it, the better the result. This is the skill that transfers across every Copilot surface, but it's especially visible in Excel because the output is right there in the grid and you can see immediately whether it understood you.
Be specific about columns and conditions
Don't write "find the top customers." Write "find the top 10 customers by total revenue in column D, for orders placed in 2025 only (column B), and add a new column marking them as 'Top 10'." The first prompt gets a generic result. The second gets exactly what you need. Name the columns. Specify the conditions. State what you want done with the output.
Use your column headers in the prompt
Copilot reads your table headers. If your column is called "Net Revenue USD," use that phrase in your prompt. It helps Copilot connect your request to the right column without guessing. This is especially important in tables where column names could be ambiguous (multiple date fields, multiple amount fields).
Tell it what format you want the output in
Do you want a new column? A chart? A summary in the chat panel? A pivot table? Say so. If you leave the output format unspecified, Copilot picks one, and it's not always the format you wanted. "Add a new column called 'Flag' that shows TRUE if…" is clearer than "flag the rows where…"
Check the first result before accepting
Copilot shows you what it's about to do before it applies changes. Read that summary. For formula generation and column additions, check a few rows manually before accepting the full result. Copilot is right most of the time, but "most of the time" is not the same as "always," and spreadsheets where errors compound are exactly where you don't want to discover that.
Copilot in Excel requires your data to be in a table
This trips up a lot of people. Copilot works best when your data is formatted as a proper Excel Table (Insert – Table, or Ctrl+T). A range of cells that looks like a table but isn't formatted as one produces inconsistent or unavailable results. Format your data as a table first. It takes thirty seconds and it's the single most common thing I see blocking Copilot from working in organizations that have already paid for it.
What Copilot in Excel can't do yet
Every Copilot session I run includes an honest conversation about where the tool falls short. Not because it's broken, but because knowing the limits saves people from wasted time and misplaced frustration. The people who get the most value from Copilot in Excel are the ones who know what to hand it and what to keep handling themselves.
It doesn't work well on unstructured data
If your spreadsheet is a mix of free-text notes, merged cells, and loosely arranged rows, Copilot struggles. It needs tabular data with consistent columns. If the data isn't clean enough for a human to analyze reliably, it's not clean enough for Copilot to analyze either. Clean up first, then prompt.
Cross-workbook analysis isn't there yet
Copilot in Excel (as of June 2026) works on the open workbook. If your analysis requires pulling from multiple files, you still need to consolidate manually or use Power Query. This is a real limitation for organizations that keep data across many sheets or export files.
Complex statistical modeling is still a manual job
Regression models, time-series forecasting, Monte Carlo simulations: Copilot isn't there for these yet. It can help with the setup and formula scaffolding, but the heavy analytical logic still requires someone who knows what they're doing. The tool is excellent for the 80% of Excel work that's analysis and reporting; the remaining 20% that's serious modeling still needs hands-on expertise.
It can get confident about wrong things
This is the most important limitation to understand. Copilot in Excel is not looking up a known fact. It's generating a formula or interpretation based on patterns. Most of the time, it's correct. Sometimes it isn't, and it presents the wrong answer with the same calm confidence as the right one. For any output that feeds a decision or goes into a report, verify before you use it. The tool is an assistant, not an auditor.
Getting your team to actually use Copilot in Excel
Most organizations that pay for Microsoft 365 Copilot licenses see a pattern: a wave of curiosity when it launches, then adoption that drops back within a few weeks. The Excel module is one of the most promising parts of the Copilot suite, but it requires some guided introduction to stick. People who open the Copilot panel for the first time and type a vague question, get a vague result, and close it again. That's not a product problem. That's a training problem.
I run Microsoft Copilot training for companies across just about every category (insurance, banking, healthcare, startups, retail, and more), with teams from mid-sized to enormous organizations. Dozens of those sessions have been in English, over Zoom. For the full picture of what training looks like across the Copilot suite, see the Microsoft Copilot training for companies page. For the broader range of use cases across Word, Outlook, Teams, and PowerPoint, the Microsoft Copilot use cases page covers those in detail.
Start with the tasks your team already does in Excel
The fastest way to get adoption is to connect Copilot to a task that already happens every week. Not a demo dataset, not a fictional example: the actual report your team builds on Monday, or the actual import they clean up every Friday. When someone sees Copilot handle their actual data, the "I'll try this later" response changes to "I'm using this next time."
Make sure the data is actually in Excel Tables
Before any session I run on Copilot in Excel, I ask teams to format their main working spreadsheets as proper Tables if they haven't already. It takes a few minutes and removes the single most common technical blocker. This is the kind of thing that gets skipped in generic Copilot rollouts and then surfaces as "it doesn't work" frustration three weeks later.
Teach prompting alongside the feature
Copilot in Excel is only as good as the prompts people use. A fifteen-minute segment on how to write a specific, well-scoped prompt produces noticeably better outcomes than showing all the features and letting people figure out the prompting themselves. In every training session I run, this is the part people say they didn't expect to find useful but ended up using the most.
Questions about Microsoft Copilot for Excel
What does Microsoft Copilot for Excel actually do?
Microsoft Copilot for Excel lets you work with spreadsheet data using plain-English prompts inside Excel. It can analyze data and answer questions about it, write formulas based on what you describe, clean up messy imported data, create charts, highlight exceptions, and (as of 2026) execute multi-step workflows in Agent Mode. It works within your existing workbook, on your actual data. You type what you want, it does the work in the grid, and you review the result before accepting it. It works best when your data is formatted as a proper Excel Table.
Do you need a Microsoft 365 Copilot license to use Copilot in Excel?
Yes. Copilot in Excel is part of Microsoft 365 Copilot, which requires a separate paid add-on to your Microsoft 365 subscription. As of June 2026, the license is available for commercial and enterprise plans. It's not included in the standard Microsoft 365 plans. If your organization has already purchased Copilot licenses, the feature is available inside Excel on desktop and in Excel for the web, and you access it via the Copilot button in the Home tab ribbon.
Can Copilot in Excel write formulas?
Yes, and this is one of the things it does most reliably. You describe the calculation you need in plain English, and Copilot generates the formula with the correct references for your specific table. It can write IFs, VLOOKUPs, SUMIFS, XLOOKUP, array formulas, and most standard Excel functions. It also explains existing formulas in plain English if you ask it to. The output is usually correct; the main thing to watch is cell reference accuracy in unusually structured sheets, so it's worth reviewing the formula before applying it across a large range.
What is Agent Mode in Copilot for Excel?
Agent Mode is a 2026 addition to Copilot in Excel that lets it handle multi-step tasks from a single prompt. Instead of one action at a time, you describe a workflow (clean the data, add calculated columns, summarize the results, highlight exceptions), and Copilot creates a step-by-step plan, shows it to you for approval, and then executes the steps in sequence. It's the most capable version of Copilot in Excel and the one most useful for repetitive analytical workflows. It rewards clear, specific prompts; a vague starting instruction can lead to a chain of steps in the wrong direction.
Can you train our team on Copilot in Excel specifically?
Yes. I run Microsoft Copilot training sessions for organizations, including focused sessions on Copilot in Excel for teams that spend most of their time in spreadsheets (finance, ops, data, sales). Sessions are typically hands-on, with real data from your team, and include prompting practice alongside the feature walkthrough. I've delivered these sessions in English over Zoom to distributed teams across just about every category: insurance, banking, healthcare, startups, retail, and more. The Microsoft Copilot training for companies page has the full details on formats and how to get started.
What changed in the May 2026 Copilot redesign for Excel?
The May 2026 Microsoft 365 Copilot redesign brought a more integrated panel experience to Excel, making Copilot easier to access and more consistent across the Microsoft 365 suite. The panel is more persistent, the suggestions are more context-aware, and the feature discovery (knowing what to ask) improved noticeably. Agent Mode, which had been available in preview for some plans, became more broadly accessible around this period. The practical effect for most users: the barrier to getting started with Copilot in Excel dropped, and the results for common tasks like formula help and data cleanup became more reliable.
Want your team to actually use it?
An intro call takes 30-45 minutes. By the end you'll know exactly what I'd recommend and what it looks like for your organization.