If you have ever spent more time lining up the boxes on an Excel dashboard than thinking about what goes inside them, Copilot's new interactive canvas is going to get your attention. Give it some data and a description of what you want, and it can build a polished dashboard with filters, charts, and the kind of visual hierarchy that usually takes a fair amount of nudging things around.
The appeal comes with a change in what you are maintaining: the result behaves more like a small web application living in your workbook than a collection of ordinary Excel charts. That gives you some impressive presentation possibilities, along with less direct control over editing and a less familiar route to checking the calculations.
Let's put it to work with the Palmer Penguins dataset, then compare the experience with a traditional Excel dashboard and a Python in Excel dashboard. The penguins have already dressed for the presentation. We still need to check their numbers.
What is interactive canvas in Excel?
Microsoft's announcement of interactive canvas describes a way to turn workbook data into an interactive report through a prompt. It stays connected to the workbook, and Microsoft says it updates automatically as the source data changes. You refine the report by asking Copilot to change the layout, emphasize a different finding, or adapt it for another audience.
The important distinction is what Copilot creates. The generated artifact I inspected in the test workbook was an HTML-based interface, using web styling and JavaScript for the display and interactions. HTML describes the page, CSS controls its appearance, and JavaScript handles behavior such as filtering. You do not need to write those languages to try it, but they explain why this feels different from selecting an Excel chart and opening Format Data Series.
The source data still lives in cells. The dashboard occupies its own canvas tab. Its cards, menus, and charts belong to that generated interface, so you should not expect each visual to be a native Excel chart or each headline number to have a formula you can inspect in the formula bar.
Availability, checked November 12, 2026: Microsoft's October announcement described an Insider release. Its current built-in skills guide says interactive-canvas is enabled by default and rolling out to production, so it may not be available to every customer yet. You need access to Copilot in Excel and the skill in your account. A missing skill is an availability issue, not something you can fix by finding a more persuasive adjective for your prompt.
Start with a small dataset you can check
Our example uses Palmer Penguins, a dataset made available by Allison Horst, Alison Hill, and Kristen Gorman, with observations collected by Dr. Gorman and the Palmer Station Long Term Ecological Research Program. It includes three penguin species, islands, body measurements, sex, and year. The data are available under CC0.
The starting workbook contains one Excel table named Penguins. All 344 observations are retained. The source's NA markers have been converted to blank cells, while measurements remain numeric. This matters because two observations have no body-mass measurement: a count of penguins and a count of measured body masses should give different answers.
Download the workbook and prompts (ZIP)
Extract the ZIP, open penguins-canvas-start.xlsx, and save a working copy. The download includes that workbook, a copy-ready prompt file, and one short README. You can also browse the reader repository on GitHub.
Open Copilot in Excel with workbook editing available. Under Manage plugins & skills, look in Visualization for interactive-canvas. You can invoke it with @interactive-canvas; if Excel offers a skill picker when you type @, select it there. This is a Microsoft-provided skill, so this exercise does not require installing a custom skill folder.
Give Copilot a clear dashboard brief
Microsoft's general prompting guidance suggests four useful ingredients: goal, context, source, and expectations. It also encourages following up on the result. That is general Copilot guidance; the dashboard-specific choices below are my application of it.
For this example, the goal is to explore size differences between species. The audience is a beginner. The source is the Penguins table, and the expectations include the metrics, filters, missing-value rules, and visual style. Those decisions give Copilot a much better starting point than "make an amazing dashboard," which leaves a surprising number of analytical choices hiding inside the word "amazing."
Paste this prompt, also included in practice-prompts.txt:
@interactive-canvas Create an interactive dashboard called "Penguin Field Guide" from the Penguins Excel table for a beginner exploring how penguin species differ in size. Keep it connected to the table and leave the source data unchanged.
Show cards for observation count, average body mass in grams, and the number of nonblank body-mass measurements. Add a species-count bar chart, average body mass by species, and a scatterplot of flipper length (mm) versus body mass (g), distinguished by species.
Add species and island filters and a Reset filters button; all cards and charts must use the same filtered rows. Count every row as an observation. Exclude missing measurements only from calculations or plots that need them; never treat blanks as zero. Show the sample size and units, and a clear message for an empty result.
Use a white background (#FFFFFF), red (#CF3338), charcoal (#1F2120), gray, and readable Arial text. Keep the layout uncluttered. Do not invent a logo, add decorative arrows, or make causal claims.
You do not need all these details for every dashboard. Here, the missing-value instructions earn their place because we know the data contains blanks. In a sales dashboard, I would spend those words defining revenue, returns, and the reporting period. The useful detail is the detail that prevents a plausible-looking wrong answer.
Allow time for generation. In this test, Copilot took several minutes to build the interface. Your layout may differ, even with the same prompt and data.

Enjoy the dashboard, then make it earn your trust
The appeal is easy to see. Copilot handles the spacing, cards, controls, and charts together, so the result feels like a designed interface. A reader can explore the data without knowing which worksheet cells to select or how to configure a PivotTable. That is useful for a demonstration, an exploratory report, or a presentation where you want the audience to engage with a few clear questions.
In the live test, the opening cards showed 344 observations, 342 nonblank body-mass measurements, and 4,201.8 g average body mass. Those agree with independent calculations from the source data. The species counts should be 152 Adelie, 68 Chinstrap, and 124 Gentoo. These are counts in our sample, not estimates of the wild population.
Select Gentoo in the species filter. You should get 124 observations, 123 measured body masses, and a mean of 5,076.0 g, rounded to one decimal. That is a particularly useful check: the dashboard has to change both the numerator and the denominator while continuing to handle a missing measurement correctly.


Use Reset filters and confirm that the full-data figures return. Try a combination with no observations, such as Gentoo on Torgersen. In this test, the canvas showed zero observations and "No measurements" for the mean, with a message explaining that no observations matched. That is much more helpful than implying that the average penguin weighs zero grams.
You can check the main cards yourself in unused worksheet cells:
=ROWS(Penguins[species])
=COUNT(Penguins[body_mass_g])
=AVERAGE(Penguins[body_mass_g])
Those formulas calculate the full table, regardless of the canvas's filter selection. For the Gentoo check, use =COUNTIF(Penguins[species],"Gentoo") and =AVERAGEIF(Penguins[species],"Gentoo",Penguins[body_mass_g]). This gives you a separate route to the answer, instead of asking the same generated interface whether it thinks it did a good job.
Also test the connection to the workbook. In your working copy, temporarily change the first body-mass value, cell F2 on Penguins, from 3,750 to 4,750. With the canvas filters reset, the mean should become 4,204.7 g; the counts should stay the same. Then restore 3,750 and confirm the original mean returns. This worked in the live test: the canvas updated automatically after the source edit, and restoring the original value returned the mean to 4,201.8 g.
The tradeoff: interacting is easier than editing
A dropdown that works beautifully does not mean you have the same editing control you would have over a normal Excel dashboard. With a native chart, you can select a series, inspect its range, change the axis bounds, or move a legend yourself. In the canvas workflow, the normal route to a design change is another request to Copilot.
That can be convenient when your request is "make the comparison clearer." It can be frustrating when you want one label changed and everything else preserved. When you want to revise the design, start with a small request such as this one:
In the existing Penguin Field Guide canvas, change the average-body-mass card heading to "Mean body mass (g)". Keep the calculations, filters, source data, and other layout elements unchanged. Report what changed.
Then check the result. A narrow prompt communicates your intent, but it does not guarantee that the generated code changed only in that one place. If every minor formatting adjustment needs a fresh round of generation and review, some of the time saved during the first build can come back as maintenance work.
Auditability is a related issue. By that I mean being able to trace an answer to its inputs and calculation steps. A traditional dashboard can expose formulas, PivotTable fields, and chart ranges. A Python in Excel workflow can expose the Python code that filters the data and calculates the result. A canvas adds another layer of generated application logic that the usual worksheet auditing tools do not walk through for you.
HTML and JavaScript are inspectable technologies, so it would be too strong to call the result inherently unauditable. The practical problem is that inspecting web code is a different skill from tracing spreadsheet precedents, and the canvas experience does not offer the same familiar trail through cells. Asking Copilot to explain its calculations helps you review its intent; separate calculations and input-change tests help you check its behavior.
For important recurring reports, I would keep the agreed calculations visible in the workbook and ask the presentation to use those results where appropriate. I would also test adding rows, refreshing imported data, reopening the file, and viewing it with the intended recipient's Excel setup before relying on the report. A successful edit to one existing cell does not establish all of those behaviors.
This connects to a point I made on X about Claude dashboards: automating the build does not settle disagreements about what the numbers mean. People still have to agree on the business rules.
Canvas, traditional Excel, or Python in Excel?
I would choose based on what the next person needs to do with the report, including the person who has to maintain it next month.
| Approach | Where it helps | What you take on |
|---|---|---|
| Traditional Excel dashboard | Recurring reporting with familiar charts, formulas, PivotTables, and slicers. Direct control over individual components. | More manual design and wiring. Auditability still depends on clear formulas, ranges, and definitions; a spreadsheet is not automatically well documented. |
| Python in Excel dashboard | Statistical analysis and plots that go beyond standard Excel chart types. Transformation and plotting code can remain in Python cells. | Someone must understand the code. Typical Matplotlib and seaborn outputs are images; worksheet controls can trigger recalculation, but that differs from interacting directly with a web chart. |
| Interactive canvas | A polished exploratory interface, filters, summary cards, and presentation layouts built together from a prompt. | Prompt-driven revisions, less familiar calculation tracing, and a generated interface whose behavior needs testing. |
When I put together my earlier Python dashboard walkthrough, I ended by wishing for faster refreshes, tooltips, and more direct interaction with the visuals. Canvas moves toward that kind of interface.
Meanwhile, the checkbox-driven Python approach gives you a different bargain: more work to connect controls and plots, but Python logic you can open and revise. Here is how I wired those worksheet controls to the Python plot:
If you are returning to the older Python article, also see my update on Python in Excel returning to Copilot as a skill. The entry point has changed since that original walkthrough. These are different ways to use AI in the workbook, not interchangeable names for the same dashboard feature.
This is part of a bigger shift toward generated interfaces
If the experience reminds you of Gemini or Claude, that is a useful connection. Google's Sheets canvas announcement describes turning spreadsheet data into interactive mini-apps with Gemini. Its current help page explains how to create a canvas inside Google Sheets and refine it through prompts. That is a closer comparison than treating every product with "Canvas" in its name as the same thing.
I also shared Google's announcement about publishing a Sheets canvas as a standalone link. It is a useful reminder to check what each tool lets the recipient view, edit, and share.
Anthropic's Claude artifacts documentation likewise includes dashboards, websites, and small interactive tools created alongside a conversation. The shared idea is that AI can produce an interface you use, as well as an answer you read. The data connections, editing options, and sharing arrangements differ across these products, so the resemblance does not mean they have identical capabilities or limitations.
I expect this category to develop further. More direct editing, clearer calculation lineage, and dependable ways to move a report between viewing and sharing contexts would all make it more useful. That is my wish list, not an announced Microsoft roadmap or a promise that today's limitations will disappear.
For now, I would try canvas when the presentation and exploration experience matter, and keep a clear route back to the underlying analysis. Start with the penguins, change one filter, and verify one number. That will tell you more about the tool than admiring a dozen attractive cards from across the room.
Build your Copilot judgment along with your dashboards
My LinkedIn Learning course, Excel with Copilot: AI-Driven Data Analysis, is fully unlocked: the full course is free, and no account is required. It is a useful next step for practicing how to ask for help and evaluate the results.
For a finance, accounting, or operations team, I also offer hands-on Excel and Copilot training. We can work through the reporting tasks your team actually faces, from preparing the data to deciding which kind of dashboard fits the job. Tell me about the training you are planning.
Keep learning Python in Excel
If you want to build on the visualization ideas here, explore my book, Python in Excel for Data Analytics. It connects data preparation with Seaborn visualizations, exploratory analysis, and Excel reporting, using practical business examples and downloadable workbooks. That makes it a useful next step for learning how to choose a chart, understand the data behind it, and put the result to work in a spreadsheet.
