Testing Claude for Excel: Practical Lessons from a Lifetime Excel User

I gave Claude in Excel a workout using some sample files with dummy data and prompts it created for me. Reach out if you’d like the files. Happy to share.

After my testing, and a professional lifetime using Excel, here are a few tips:

📉 𝗜𝘁’𝘀 𝗻𝗼𝘁 𝗽𝗲𝗿𝗳𝗲𝗰𝘁. Claude produced two errors during my testing. In one case, it included Closed Won deals in a pipeline analysis. In another, it made up an assumption that I did not explicitly supply. Everything else checked out.

🛠️ 𝗛𝗼𝘄 𝘁𝗼 𝗰𝗵𝗲𝗰𝗸 𝗶𝘁𝘀 𝘄𝗼𝗿𝗸. One of the future skills in working with AI is how to check its work. Here are a couple of methods that I used. Ask Claude to:

1️⃣ Explicitly define metrics, so there is no ambiguity.
2️⃣ Supply formulas in tables and calculations, not hard coded numbers, so you can check the math. Claude usually does this, but I got pasted values a few times.
3️⃣ Provide proof that summarized data ties back to the source with visible check formulas.

📊 𝗙𝗼𝗿𝗺𝗮𝘁𝘁𝗶𝗻𝗴 𝗶𝘀 𝗮𝗻 𝗲𝗮𝘀𝘆 𝘄𝗶𝗻. Use Claude or ChatGPT to generate a PDF style guide from an existing presentation, then ask Claude for Excel to format tables/charts to match. This is especially appealing for me since my design skills are weak.

🧹𝗗𝗮𝘁𝗮 𝗰𝗹𝗲𝗮𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝗺𝗮𝗻𝗶𝗽𝘂𝗹𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮 𝗻𝗮𝘁𝘂𝗿𝗮𝗹 𝗳𝗶𝘁. There are several existing tools and techniques for cleaning messy data, but Claude in Excel makes it easy. Claude can handle fixing inconsistent labels, unpivoting data, joining tables, and much more.

I’ve already started thinking about what an Excel skill file would look like here. Formatting rules pulled from a style guide. Model checking logic. A budget versus actual templates for business partners.

I haven’t seen examples of this yet. I would be curious to know if anyone’s already heading this direction.