What Can Microsoft Copilot in Excel Actually Do (Formulas, Python, and Data Analysis)?

Detailed Explanation

Copilot in Excel started as a chat pane bolted onto the side of a workbook — useful for explaining a formula or drafting a simple one, but limited to conversation rather than action. Through 2026, Microsoft extended it into something closer to a genuine in-workbook assistant, with three capabilities worth understanding separately, since they solve different problems and aren't interchangeable.

Edit with Copilot switches Copilot from describing changes to making them. Instead of Copilot suggesting a formula for you to copy in yourself, an April 2026 update introduced step-by-step reasoning where Copilot outlines a plan for a multi-step change, applies it directly to the workbook's cells and formulas, and shows what changed so you can review or adjust it — a materially different interaction than the original chat-only mode, closer to a "Plan with Copilot" edit flow than a Q&A box.

Python in Excel, driven through Copilot, brings real code execution — pandas for data manipulation, matplotlib for charting, scikit-learn for statistical models — directly into a worksheet, invoked through a natural-language Copilot request rather than requiring you to write the code by hand. This is the feature that actually closes the gap between "what a formula can calculate" and "what a proper statistical or forecasting analysis requires": correlation analysis, clustering, regression-based forecasting, and similar work that spreadsheet formulas alone were never well suited to.

The =COPILOT() function puts an AI-generated result directly inside a cell, usable the same way as any other Excel function — reference other cells as its input, drag it down a column, combine it with regular formulas. A May 2026 update extended it further, letting the function ground its answer in a live web search, so a formula can pull in current information (a company's current stock ticker, a benchmark figure) rather than being limited to what's already in the workbook.

How This Differs From a General AI Assistant Analysing Your Data

Using an AI assistant like Claude or ChatGPT to analyze a spreadsheet means uploading or pasting your data into a separate tool, getting an analysis or a summary back, and manually bringing any useful output back into your actual workbook. Copilot in Excel works the other way around: it operates inside the live file, edits the real cells and formulas, and (with Python in Excel) runs code directly against your actual data range — there's no export-analyze-reimport cycle. The trade-off is scope: a general assistant can reason about anything you describe or paste in from anywhere, while Copilot in Excel is specifically bound to what's inside the open workbook.

Things to Consider

• Licensing and rollout stage matter more than the feature list. The =COPILOT() function specifically requires a Microsoft 365 Copilot licence on Business or Enterprise plans and was still rolling out toward general availability later in 2026 — verify current availability for your specific plan before building a process around it, since Copilot feature rollout timing has shifted before.
• Python in Excel's output still needs the same review discipline as any AI-generated analysis. A regression or clustering result Copilot's Python code produces can be confidently wrong the same way any AI output can — treat it as a draft analysis to sanity-check against what you already know about the data, not an automatically authoritative result.
• This is a prompted, in-session tool, not a scheduled automation. Copilot in Excel does what you ask while you're working in the workbook; it doesn't run unattended on a schedule the way Power Automate flows do — for a recurring, hands-off process, Power Automate (or Office Scripts triggered by it) is still the right tool.
• "Show Changes" and similar review features exist because Copilot editing your actual data is a real risk to manage. Because Edit with Copilot changes real cells rather than just describing a change, review what it actually did before trusting the result, especially on a workbook other people also rely on.

Common Mistakes

• Assuming =COPILOT() is available on every Microsoft 365 plan. It's a Copilot-licensed premium feature on Business and Enterprise plans, not a general Excel function available by default — check licensing before planning a workflow around it.
• Using Copilot in Excel for a task that needs to run unattended on a schedule. It's a prompted, in-session assistant; a nightly or triggered process still belongs in Power Automate or a scheduled Office Script, not in a Copilot conversation someone has to run manually.
• Trusting a Python-generated statistical result without spot-checking it. A confidently presented forecast or cluster analysis is still an AI-generated output with the same accuracy risks as any other — verify a sample of the underlying calculation before relying on it for a real decision.
• Confusing Copilot in Excel with a general AI assistant analysing an uploaded spreadsheet. They solve overlapping but different problems — Copilot in Excel edits your live workbook in place; a general assistant analyses a copy of the data you gave it, separate from the source file.