Microsoft’s AI assistant for Excel has crossed a functional threshold: it can now generate entire workbook structures – dashboards, complex formulas, and multi-sheet reports – rather than merely answering questions or suggesting formulas. But a new hands-on test by How-to Geek, whose author had canceled Copilot 18 months ago out of frustration, reveals that while the tool saves serious setup time, the output still demands a skeptical human reviewer before it reaches a manager, a payroll system, or a customer.

From Digital Assistant to Spreadsheet Architect

When Microsoft first rolled out Copilot in Excel, it behaved more like a glorified help pane, offering formula suggestions and data insights from a sidebar. That’s now changed substantially. The feature once called Agent Mode has been rebranded as Edit with Copilot, and it can directly modify a workbook – adding sheets, charts, PivotTables, conditional formatting, and formulas. According to Microsoft’s own support documentation, edits are made through native Excel objects, meaning users can inspect, modify, refresh, or delete every AI-created element just as they would any manually built component.

Even more telling is a capability quietly added in recent months: the .Rules worksheet. This is a visible sheet where you list workbook-specific instructions – one per cell in column A – that Copilot will heed in subsequent interactions. Rules stay with the file, so you don’t have to re-explain your formatting preferences, naming conventions, or function choices every time you summon the assistant. Microsoft’s documentation recommends short, specific rules, and notes that formulas can even make rules responsive to workbook state.

The Three-Test Verdict: Gains You Can Touch, Flaws You Can’t Ignore

How-To Geek’s reporter put Copilot through three practical, everyday scenarios. Each test showed clear progress over the 18-month gap since the writer last used the tool, but also exposed persistent quality gaps.

Dashboards: Architecture Right, Polish Wrong

For a sales management dashboard, Copilot built KPI cards, charts, slicers, and conditional formatting from a 10-column data table. Critically, it didn’t just paint a static picture. It created a hidden PivotData sheet with live PivotTables linked to the source table, meaning the dashboard uses real formulas and will refresh with new data. Tables got meaningful names, and charts received titles automatically – small time-savers that add up.

But the bot also resized charts inconsistently, failed to link all slicers as requested, and froze panes so awkwardly that a slicer got sliced in half. Worse, it reformatted the source data without being asked. For anyone who works with strict input/output conventions – and most spreadsheet pros do – those are not cosmetic nitpicks; they’re violations of established workflow rules.

Formulas: From Nested IF Hell to Auditable Logic

A deliberately vague bonus-calculation prompt produced a clear win. Instead of a sprawling nested IF statement, Copilot used the LET function to assign names to intermediate steps (sales, target attainment, bonus rate, cap). This makes the formula far easier to audit and maintain – a welcome outcome for any Excel user who’s inherited a cryptic, multi-line monster. When asked for a version compatible with older Excel, Copilot broke the logic into helper columns and matched results exactly.

However, the reporter rightly warns that formula correctness isn’t proven by a first-pass plausible result. Business logic like a bonus calculation requires testing against edge cases, blank records, rounding rules, and conflicting conditions. Copilot is best seen as a formula drafting and refactoring accelerator, not as an arbiter of business rules. If a workbook handles compensation or sensitive data, the same review controls applied to human-written formulas must cover AI-generated ones.

Reports: Automating the Repetitive Grunt Work

Generating individual employee performance reports – with charts, KPI cards, conditional formatting, and print-ready layouts – is normally a copy-paste-and-tweak marathon. Copilot built separate sheets, linked calculations, and set up print output so reports fit on a single page. It handled the small stuff: hiding gridlines, applying a color scheme, formatting currencies.

But at scale, small errors become big problems. Chart labels overlapped, spacing was off, and currency formatting needed manual correction. When the same flaw propagates across 200 reports, it turns a time-saver into a quality-control headache. The author concluded that the tool gets a workbook 70–85% of the way there, leaving the final polish to the human analyst. That’s still a huge productivity gain – provided the last 15% is actually done and not skipped.

Why the Details Trip Up AI

A thread running through all tests is that Copilot stumbles on layout choices, visual consistency, and adherence to undocumented conventions that seasoned Excel users take for granted. Dashboard design isn’t just about formulas; it’s about readability, executive polish, and consistency across an organization. The AI doesn’t know your team’s color palette, your standard font sizes, or your preference for structured references over A1 notation – unless you explicitly tell it every time.

That’s precisely the gap the .Rules worksheet aims to fill. By baking instructions like “use US dollars with two decimals,” “name tables using CamelCase,” or “prefer XLOOKUP over VLOOKUP” into the workbook itself, you reduce the prompting overhead. But Microsoft notes that rule support is not yet consistent outside English, and the feature will evolve. Moreover, rules guide Copilot – they don’t enforce constraints like data validation or sheet protection. A user can still manually break conventions after Copilot finishes.

Another subtle but serious concern: chat history loss. The reporter found that Copilot conversations sometimes fail to save inside the workbook. The formulas, charts, and sheets persist, but the reasoning behind them can vanish. If an auditor asks, “Why did you use that particular lookup table?” the workbook may show the result but not the original request. In regulated environments, this is a traceability gap that forces teams to document their AI prompts externally – adding friction back into the process.

What This Means for You

The practical implications break down by user type:

  • Home and casual users: You’ll see the most immediate benefit in time savings. Setting up a budget dashboard or a holiday planner is faster, but you should still double-check every figure and formatting choice before sharing the file with someone else.
  • Business analysts and power users: Copilot can now act as a junior assistant that prepares first drafts of dashboards, complex formulas, and repetitive reports. It frees you to focus on higher-value validation, design refinement, and communicating insights. However, you must treat its output as you would a colleague’s work – reviewable and changeable, not gospel.
  • IT and governance teams: The .Rules worksheet is a welcome step toward organizational consistency, but it’s not a compliance tool. Continue to enforce workbook standards via templates, data validation, documented review cycles, and – where necessary – VBA/Office Scripts. Be aware of the chat history gap and decide whether your users need a separate record of AI interactions for audit trails.

How We Got Here

Microsoft’s Copilot journey in Excel began with a cautious, sidebar-based assistant that offered formula suggestions and data insights. Users quickly realized that for anything beyond basic tasks, the AI needed too much hand-holding and still got context wrong. The rebranding to Edit with Copilot marked a strategic shift toward making the AI act inside the grid, not just beside it. Support pages now explicitly guide users on creating PivotTables, charts, and formulas with natural language, while company blog posts have framed the tool as part of an evolving “app skills” paradigm.

This reflects a broader industry trend: generative AI in productivity apps is moving from “answer a question” to “do a task.” But Excel is a uniquely unforgiving environment for that transition. A wrong number in a sales report can cost a deal. A mis-specified bonus formula can create HR nightmares. So Microsoft’s own documentation includes the warning that users should review AI-generated results, and third-party reviews consistently underscore the need for human judgment.

Your Move: Adopting Copilot Without Losing Control

If you’re thinking of giving Copilot another look – or trying it for the first time – here are concrete steps to get the most out of it while managing risk:

  1. Upgrade to Microsoft 365 Copilot. The feature is not available in classic plans. Check your subscription before spending time testing.
  2. Start with a .Rules worksheet. Create a sheet named .Rules and list your conventions – one short instruction per cell. Use it to enforce consistent currency formatting, table naming, function preferences, and color schemes.
  3. Test on a copy of your data, never the production workbook. Allow Copilot to build dashboards or formulas in a sandbox file. Inspect every formula, PivotTable connection, and chart setting before transferring work into a shared or official workbook.
  4. Build a review checklist. For formulas: test edge cases, blank values, and typical rounding scenarios. For reports: check chart labels, print layouts, and across-page consistency. For dashboards: verify that slicers and PivotCharts connect to all intended data.
  5. Document AI-generated logic separately. Since chat history may not save, maintain a simple log (in a cell comments, a separate worksheet, or a project management tool) of what Copilot was asked to do and why. This is especially critical in financial, HR, or compliance-heavy contexts.
  6. Don’t skip the final polish. Even if Copilot gets a report 90% right, the remaining 10% – visual alignment, font consistency, proper date formatting – is what separates a trusted deliverable from something that looks hastily assembled.

Looking Ahead

Copilot in Excel has graduated from an experiment to a tool that can genuinely shorten the path from raw data to structured output. Its ability to generate live, auditable Excel objects rather than static answers is a significant architectural win. The upcoming evolution of .Rules and deeper integration with the Microsoft 365 ecosystem may further reduce the manual corrections now required.

But the tool’s limitations are a reminder that spreadsheets are not just containers for numbers – they’re vessels for business logic and human judgment. As long as that logic must be verified, the responsible Excel user will remain the final editor, no matter how clever the assistant becomes. For Windows users who live in grids, the question isn’t whether to use Copilot, but how to build the discipline around it that turns a time-saver into a trustworthy partner.