Ask your own data anything in plain English

Why it's worth describing your data once

The first time you ask, Claude doesn't know what's in your spreadsheets or your systems. So the first step is just telling it, in a sentence or two: what you've got, and roughly what each thing means. After that, every question gets faster and more accurate, because it isn't starting from nothing each time.

This isn't a special feature or a technical setup step. It's the same thing you'd tell a new starter on their first day, before they could actually help you with anything.

The steps

This is the reel, written down.

1. Tell it roughly what you've got

   One sentence: what the spreadsheet or system is, what each column means. You don't need to get it perfect.

2. Then ask a real question

   Not a technical request. The way you'd ask a person who actually knew the numbers.

   The question from the reelCopy

   Has my Instagram growth actually turned into more revenue
   this year? Pull the follower count and the revenue side by
   side, day by day, and tell me if they move together.

3. Let it work out the rest

   It figures out which numbers actually answer the question, pulls them together, and does the maths. No spreadsheet formulas, no analyst, and you never have to see how it did it unless you want to.

4. Turn the answer into a chart

   The same two words from Day 10 do the work here too.

   Make it visualCopy

   Show me that as an artifact.

Five more questions worth asking

1. A trend over time: "How has [metric] moved week by week this year?"
2. Month-on-month: "Compare this month's numbers to last month's, and tell me what changed most."
3. Two metrics compared: "Does [metric A] actually move with [metric B], or are they unrelated?"
4. A filtered breakdown: "Break this down by [category] and rank them."
5. A ranked top-N: "What are my top five [things] by [measure], and why?"

Snapshot vs live

A daily-collected warehouse like this one is built from one snapshot a day, so it's the right tool for "how's this trending" and "compared to last month." It is not the right tool for "what's happening right now" — for that, a live query straight to the source (Stripe, your CRM, whatever it is) is the honest answer, because the warehouse might be up to a day old.

If you're building your own version of this, decide upfront which questions need "as of this morning" and which need "right now", and build both paths. Trying to make one do both jobs is where these systems usually go wrong.

Where this goes

That's Layer 2 done: your tools are connected, the numbers come to you every morning, you can see them on a dashboard, and now you can ask them anything without touching a spreadsheet. Layer 3 starts Day 13: Claude that remembers every meeting you've ever had.