Microsoft Tests COPILOT Formula in Excel: Native AI That Recalculates Automatically

Microsoft has started testing a highly anticipated feature in Excel: a native AI function called COPILOT that embeds generative AI directly into the spreadsheet's calculation engine. Announced for Insider Beta Channel subscribers, the new function lets users write natural language prompts inside cells, optionally referencing data ranges, and receive AI-generated outputs that update automatically as source data changes—all without leaving the familiar formula bar.

The move, revealed on August 18, 2025, on the Microsoft 365 Insider blog, brings Excel much closer to Google Sheets, which launched its own AI function powered by Gemini back in June 2025. But Microsoft goes beyond a simple catch-up: the COPILOT function is designed to behave like any other Excel formula, fully integrated with dynamic arrays, logical operators, and custom LAMBDA functions. That means it can recalculate on the fly, spill results across rows and columns, and nest inside IF, SWITCH, and other formulas—features that power users and enterprises have been demanding.

The Google Sheets Precedent

Google’s AI function in Sheets, first announced on June 25, 2025, allowed users to type prompts like =AI("Write a formal ad copy for the product. Cater copy to the objective and target audience", A2:C2) to generate text, summarize data, categorize feedback, or perform sentiment analysis. It was available to Workspace Business Standard, Enterprise, and Gemini subscribers, and marked a significant step in bringing LLMs directly into the spreadsheet grid. However, Google’s implementation initially required a manual refresh when input data changed, and had a 200-cell processing limit per batch.

Microsoft’s COPILOT function is clearly a response, but one that refines the concept for Excel’s mature calculation architecture. Instead of a separate refresh step, COPILOT outputs recalculate whenever any dependent cell changes, just like a VLOOKUP or SUMIF. Microsoft’s blog post emphasizes that COPILOT “is part of the calculation engine” and “returns results that can spill and be chained with other functions.”

How COPILOT Works

The syntax is straightforward: =COPILOT(prompt_part1, [context_range1], prompt_part2, …). For example, =COPILOT("Classify this feedback", D4:D18) would analyze the text in that range and return a classification. The function can handle multiple prompt fragments and reference ranges, allowing complex, programmatic compositions.

Because COPILOT is a true Excel function, its outputs become part of the dependency graph. When you change a value in the referenced range, Excel automatically re-evaluates the function, just as it would for a VLOOKUP. This is a departure from Google’s initial approach, where the AI result became static until you manually refreshed. For business users who build entire reporting workflows on automatic recalculation, this is a critical advantage.

COPILOT also supports spilling—if the AI returns a table or list of values, it fills adjacent cells automatically, similar to the WRAPROWS or FILTER functions. You can then wrap the output in an IF statement, feed it into a LAMBDA, or use it inside a SWITCH. This composability opens up possibilities for semi-automated decision chains where AI-generated insights trigger conditional formatting, further calculations, or data manipulation.

Microsoft has however set initial usage quotas to manage server load: users are limited to 100 COPILOT calls every 10 minutes, and up to 300 calls per hour. The company recommends batching multiple cells into a single call by passing a range as an array, which can process many data points with one invocation. For example, instead of calling COPILOT once for each of 50 cells, you could pass the entire range and get all results in one spill. This is both quota-friendly and architecturally cleaner.

Comparing COPILOT and Google’s AI Function

At a glance, both functions serve the same purpose: they let users invoke a large language model from a cell. But the differences matter for enterprise adoption.

Google’s AI function also integrates with its side panel “Ask Gemini,” allowing users to generate content without blocking the grid. Microsoft has a similar Copilot side panel in Excel, but the COPILOT function is a direct in-cell formula, which may feel more natural for heavy formula users.

Neowin’s coverage of the feature noted that Microsoft is “finally testing an Excel feature that was introduced in Google Sheets months ago,” highlighting the perceived delay. But Microsoft insiders argue that the deep integration with Excel’s calculation engine sets it apart, turning AI from a one-shot tool into a persistent part of the spreadsheet’s logic.

Practical Use Cases

Microsoft provided several examples in its announcement, including classifying customer reviews from coffee machine feedback and extracting structured tables from text ranges. Here are some real-world scenarios where COPILOT can immediately add value:

• Sentiment Analysis: Tag thousands of open-text survey responses as positive, negative, or neutral with a single formula like =COPILOT("Determine sentiment of this feedback", B2:B500).
• Summary Generation: Create executive summaries from lengthy meeting notes stored in a cell, e.g., =COPILOT("Summarize these notes in three bullet points", A1).
• Data Enrichment: For an e-commerce dataset, generate SEO-friendly product descriptions based on specifications: =COPILOT("Write a 50-word product description for a " &B2& " with these features: " &C2, D2).
• Formula Generation and Explanation: While Excel already has a “formula by example” feature, COPILOT can write complex formulas from plain language, and even explain existing ones, reducing the learning curve.

Because COPILOT’s output is a formula, it can be audited like any other part of the workbook. You can trace dependents, evaluate portions of the formula, and include it in version control—a significant improvement over black-box AI tools that generate static values.

Security, Privacy, and Governance

As with any AI feature in enterprise software, data handling is a top concern. Microsoft states that content processed by COPILOT is not used to train their models, and that enterprise tenant protections apply. However, details on data residency—where the prompt and data are processed—depend on the organization’s tenant type and licensing plan.

Administrators can use sensitivity labels to block COPILOT from working on confidential documents, providing a layer of control. But IT teams must still verify that the feature complies with industry regulations. Because COPILOT sends data to the cloud for inference, it may not be suitable for air-gapped environments or strict data localization requirements until Microsoft provides more granular controls.

Microsoft also reminds users that AI outputs can be inaccurate or “hallucinate.” The company recommends validating critical results and not relying solely on AI for high-stakes decisions—a caveat that should be emphasized in any AI adoption policy.

Licensing and Rollout

COPILOT is currently available only to Microsoft 365 Insider Beta Channel users who hold a Microsoft 365 Copilot license. The required minimum builds are:

• Windows: Version 2509 (Build 19212.20000)
• Mac: Version 16.101 (Build 25081334)

Web availability is planned through the Frontier program, with broader rollouts to follow. Microsoft has not yet committed to a General Availability date. Organizations interested in piloting the feature should ensure they have the correct licenses and run tests in controlled environments.

The 100-calls-per-10-minutes limit may sound restrictive, but for many ad-hoc analytical tasks it will suffice. Enterprises that plan to embed COPILOT in high-frequency reporting tools will need to architect around these quotas, perhaps by using helper columns to batch requests or by scheduling refreshes during off-peak times.

Critical Analysis: Strengths and Risks

Strengths:
- Seamless Integration: Making AI a native function that recalculates automatically reduces friction dramatically. Users don’t need to leave the formula environment to get AI help.
- Composability: The ability to nest COPILOT inside other functions means you can build complex logic that blends deterministic and probabilistic outcomes. For instance, =IF(COPILOT("Is this a valid email address?", A1)="Yes", "Send", "Skip").
- Lowered Learning Curve: Business users who already think in terms of formulas will adopt AI faster because it looks like just another function.

Risks:
- Hallucination Propagation: If an AI output is wrong, it can cascade through dependent formulas. Without robust validation, this could corrupt entire models.
- Quotas and Cost: The limits may force users to rethink spreadsheet design, and long-term, Microsoft is likely to charge for COPILOT usage beyond a certain threshold, similar to Azure AI services.
- Data Residency and Compliance: Sending spreadsheet data to an external AI service may violate some data protection policies, even with Microsoft’s tenant protections. Organizations must scrutinize the end-to-end data flow.
- Auditability Challenges: AI-generated outputs are not deterministic, so reproducing a workbook’s state requires logging the exact prompt, model version, and input data at the time of calculation. This adds complexity to spreadsheet auditing.

Architectural Implications

Embedding generative AI into the calculation engine marks a paradigm shift for spreadsheet applications. Until now, spreadsheets were purely deterministic: a given formula always produced the same output for the same inputs. With COPILOT, the same input might yield slightly different results over time as the underlying AI model is updated. This is a fundamental change that demands new best practices:

• Prompt Versioning: Treat prompts like code. Store them clearly in the workbook and document the expected behavior.
• Output Validation: Include sanity checks, e.g., verify that a sentiment classification is one of the allowed categories, or that a generated text falls within a length range.
• Model Lock-in: Over time, workbooks may become dependent on the behavior of a specific AI model version. Microsoft might offer version pinning in the future, but for now, users should anticipate potential drift.

Recommendations for Teams

For IT Administrators:
- Enable COPILOT only for pilot groups, using sensitivity labels to restrict it on sensitive data.
- Confirm data processing terms with your Microsoft account team, especially regarding residency.
- Set up usage monitoring to avoid unexpected quota exhaustion.

For Power Users:
- Start with low-risk tasks like classifying internal feedback, not financial calculations.
- Build validation columns that cross-check COPILOT outputs against known-good examples.
- Keep manual fallback formulas for critical paths.

For Developers and Automation Engineers:
- Wrap COPILOT calls in LAMBDA functions to create reusable, documented AI logic blocks.
- Use Excel’s built-in auditing tools to trace dependents and precedents of AI-augmented formulas.
- Version-control your spreadsheets more rigorously, perhaps by exporting to a text-based format like Power Query M or external scripts.

The Bigger Picture

Microsoft’s COPILOT function is more than a feature catch-up; it’s a statement that spreadsheets are evolving into conversational interfaces. Google took the first step with its AI function, proving that users want to talk to their data directly. Microsoft is now leveraging Excel’s mature calculation engine to deliver a more integrated experience.

This convergence of AI and spreadsheets will accelerate automation in finance, marketing, operations, and beyond. But it also raises the stakes for governance. Organizations that embrace AI-native formulas without updating their audit and validation processes risk generating beautiful but flawed reports.

The coming months will reveal how the quota system scales, what pricing models emerge, and how aggressively Microsoft integrates web and enterprise data access into COPILOT. For now, Windows and Microsoft 365 enthusiasts can get a head start by joining the Insider program and exploring how COPILOT can transform their spreadsheets—responsibly.