AI Chat Prompting: Using Copilot to Level Up in Excel

Generative AI chats (like Copilot, ChatGPT) make Excel more powerful than ever by removing the time spent searching how-to documentation.

Excel is arguably the top marketing technology tool (a remarkable feat given it’s not a marketing technology tool), but using it incorrectly wastes valuable time.

Excel is used so heavily by marketers to the point where it’s often used out of familiarity despite having purpose-built platforms for project management, analytics, etc.

And there are many of us who use Excel sporadically enough that it’s never made sense to deep-dive learn it. That is (was!) a problem.

Doing it the Hard Way Used to be Faster in the Short Term

I’m not special. I know there are many people like me who have never needed to use Excel frequently enough to spend the time learning its complex capabilities.

This has meant occasionally doing manual, time consuming work because it was often faster to do it the hard way than to poke around Google for a couple hours trying to find out how to craft a formula and troubleshooting when it doesn’t quite work for my use case.

Copilot has completely solved this problem for me.

But it doesn’t always solve it the first time it’s asked, and this is where prompt engineering comes into play.

Learn Prompt Engineering for More Effective AI Results

Prompt engineering is a fancy way to say “ask better questions.” If you’ve tried AI chats and haven’t had much luck, consider the below example.

I had a list of form submissions from a website that had thousands of nonsensical bot submissions. The names were random characters like xsdSKHdlkdOhhLL.

They needed to be filtered out to create a clean list but also to analyze the effectiveness of a recently implemented spam solution.

I put the list directly into Copilot and asked it to identify the bot submissions, which were exceedingly obvious to a human eye. It said it couldn’t do it!

But I knew it could, so I asked again in a different way – telling it to look for rows that had random sequences of characters.

It still couldn’t do it.

I tried again, asking it to identify any cells in the first name column that contained non-standard names. It said there were too many names to analyze and couldn’t complete the task.

I thought more about it and about Excel’s strengths with formulas and math. The bot submissions followed a particular pattern of alternating lowercase and uppercase letters with no spaces in between, something that’s uncommon in most first names.

Finally, I asked it to create a formula to identify any cells that contained a string of letters where a lowercase letter was immediately followed by an uppercase letter with no space in between. Then I asked it to create an extra column flagging each one so that I could quickly confirm if any rogue human names were included.

Copilot immediately returned the corrected Excel sheet with the new column and 100% of the bot submissions flagged that I could easily filter. It also explained the formula it used.

I was able to quickly identify the % of bot submissions and provide a clean list to my colleagues.

The Future of AI and Using the Right Tool for the Job

A great place to start today is working with AI to understand Excel to save yourself time and boost your ability to do more complex work. Once you start working with AI any time you find yourself doing something repetitive in Excel, you’ll start leveling up your work.

But as AI agents and memory become more available/usable I envision a future where AI will recommend another platform for the problem you’re solving.

For example, how many teams are using Excel to track project status despite having an enterprise project management platform? Or, how many teams are creating manual analytics reports in Excel despite having access to powerful BI platforms?

In the future, when you ask an AI agent/chat how to update your project management tracker for X function in Excel, AI might recommend implementing a certain feature in another platform that could work out of the box.

Until then, keep trying to use AI to problem solve and your prompt engineering will naturally get better and better.