Disclaimer: No part of this newsletter is scripted using ChatGPT. We’re 100% human.
As a Google Sheets lover and fan forever, I will find any and every excuse to not use Microsoft Excel. So when Excel came out with their COPILOT function formula, I was genuinely intrigued and gave it a shot.
Introducing the COPILOT function, which now instead of chatting with the sidebar, you can prompt directly into the function and make it do things. It’s kind of like vibe coding but with Excel functions… or is it?
I mean now I don’t need to dump my Excel file into my LLM of choice and ask it a bunch of questions, I can do it right in a cell itself… or can I?
I mean the greatest minds at Microsoft had to have created a magic formula that put their users needs top of mind… or did they?
Short answer: it’s subpar at best based on it’s use cases.
The Syntax
For those of that don’t know, the syntax is the stuff inside of the Excel Formula.
To breakdown the COPILOT syntax:
prompt_part (required): text instructions (your natural-language “ask”)
context (optional but highly recommended): cell/range references that contain the data to use
The output: a spill like an array but multi-row/column
Use Cases and Limitations
Microsoft recommended the following ways to use COPILOT to work more efficiently and effectively:
Spur ideas: Whether you’re planning a marketing campaign or designing new product features, the COPILOT function makes brainstorming directly in the Excel grid easy. Need a set of SEO keywords based on a product description? Want to rewrite messaging for clarity or change the tone? COPILOT can handle that, too. - I don’t remember marketing or product teams ever using Excel
Generate summaries: The COPILOT function can distill large data ranges or long passages into concise narratives, highlight trends, or produce plain language explanations for complex calculations. This is especially valuable for reporting when you need to turn lengthy background material into concise, audience-ready content. - What’s the difference between this and putting it into an LLM?
Classify data: You can use the COPILOT function to categorize text data, such as customer feedback, support tickets, or survey responses, right in your spreadsheet, instead of having to export the data to another tool for tagging or sentiment analysis.- What’s the difference between this and putting it into an LLM?
Create lists or tables: The COPILOT function can generate lists and tables of data that fit seamlessly into your models. Whether you’re building a quick dataset for testing formulas, assembling a list of industry examples, or drafting a project plan outline, the function can return multi-row, multi-column outputs that spill directly into the grid.- What if I want to adjust or nicely format the list or table quickly?
Summarizing coffee machine feedback was the big demo Microsoft put out on YouTube:
Then they proceeded to list all the limitations related to COPILOT, including:
No large array support: Rows can be omitted when returning arrays. To work around this, restructure your queries to return smaller array results.
Low guidance: We are investigating providing user guidance when the COPILOT function is used for tasks not suitable for LLMs. For example: =COPILOT(“Sum these values”,A1:A10).
No enhanced knowledge: The COPILOT function is model grounded, with no access to web and enterprise data. We are investigating adding support for expanding these capabilities.
No date support: Currently, the COPILOT function returns dates as text rather than Excel’s date serial format.
No financial sense? I mean you must be playing a practical joke.
My general takeaway: the product managers at Microsoft need help because ~70% of the world’s Excel users use it for financial purposes, yet the COPILOT function is best used for summaries not for financial purposes and bonus: it cannot access web or enterprise data.
Personally I wouldn’t even bother releasing something that doesn’t change the life of a majority of my users in the way they use it. Product Management 101.
So now what?
This is a very good start, I’m excited to see what Version 2 will look like but for now it’s quite limited in it’s use cases as you can see above.
I’ll always be open minded to many things, including how COPILOT can materially change the way I work, so if you have any suggestions, please pass it along. Until then, long live Google Sheets.
Remember: Stay intelligent, not artificial