Copilot übernimmt das Python-Skript in Ihrer Arbeitsmappe

Some questions push past what a formula can comfortably answer: a forecast across several quarters, a significance test, a grouping of customer records, a type of chart Excel simply doesn’t offer. In cases like those, you can now just ask Copilot. Describe in plain language what you want to know, and Copilot writes the Python code behind the scenes, runs it, and puts the result into your workbook. You don’t need to know how to program. Microsoft documented the feature at the end of August for Windows, Mac and the web. Here’s how to use it, and what happens to your data when you do.

 

Info Benefit: Analysis without the expertise
UseCase UseCase: Making sense of data
Zeit Reading time:
3 minutes
Schwierigkeit Difficulty: Easy

Two pieces come together here, and both have been around for a while on their own: Python in Excel, and Copilot’s editing mode. What’s new is that Copilot now puts Python to work itself, rather than handing you code to paste in. This has been rolling out since the spring; since late August it has been in the release notes for all three platforms.

In practice, analyses that used to mean reaching for a second tool now stay in the workbook. Copilot draws on the same toolkit data analysts use the world over – libraries with names like pandas, matplotlib and scikit-learn. You don’t need to know any of them, and there’s nothing to install; Copilot picks whatever suits your question. What matters for you is what it makes possible: forecasts, grouping similar records together, statistical tests, and charts such as box plots or heatmaps that are awkward to build with Excel’s own tools.

It’s worth knowing where the code actually runs: not on your machine, but in a sealed-off container in Microsoft’s cloud. Microsoft says the Python code can reach neither your device nor the network, and that the data isn’t stored there permanently. Every workbook gets its own isolated container.

  1. Open the workbook holding the data you want to look at, either in the browser or in the desktop app.
  2. Start Edit with Copilot.
  3. Describe what you’re after in plain language – “summarise the trends over the past twelve months”, say, or “give me a box plot of sales by region”. If you specifically want Python, mention it.
  4. Check the result in your workbook. If the analysis has landed there as a Python cell, you’ll spot it by the =PY() function – though you never have to read the code inside it.

  • The code runs in the cloud: execution happens in an isolated container in the Microsoft cloud. Microsoft states that the Python code can reach neither your machine nor the network, and that the data isn’t stored there permanently.
  • If you can’t find the feature: the update channel is the usual culprit. Copilot features aren’t available in the desktop apps on the Semi-Annual Enterprise Channel. The browser works regardless, and it’s the quickest way to try this out.
  • Check what comes back: Copilot does the number-crunching, but the judgement call stays with you. Give it the same once-over you would give a formula: are the magnitudes, the time period and the underlying figures right?
  • Not for everything: what comes back is analysis and charts – not macros, and not ready-made Excel formulas.
  • Ask in your own words: technical vocabulary isn’t required. “Which regions are out of line?” gets you there just as well as a precisely worded instruction. If you have a particular chart in mind, name it; if you don’t, just describe what the picture should reveal.
  • Start small: a specific question about a clearly bounded range of data beats a sweeping request aimed at the whole workbook.
  • Turn the answer into a template: ask Copilot explicitly to leave the analysis behind as a Python cell, which you’ll recognise by =PY(). Excel then recalculates it whenever the numbers change, and you can copy that cell into another workbook, where the analysis and its chart are built fresh rather than pasted in as a picture. Without asking, you’ll sometimes get only the finished result and no cell to reuse.

The real advance here isn’t the technology, it’s how far the barrier has dropped: analyses that used to call for specialist knowledge or a second tool now come out of a single sentence in plain language. You can look behind the curtain if you want to, but nothing makes you. A good place to start is an analysis you already run every month – describe it once, check the result, and reuse it the next time around.