Query Your Data in Claude Code, Plain English

From the archive. Originally published 2026-06-26. Tool details, examples and offers reflect that publication date. See our current guides for newer material.

What is querying data with Claude Code?

Querying data with Claude Code means pointing Claude at a file you already have - a CSV export, a spreadsheet, a copied table - and asking it a question in plain English. Claude reads the file, works out the calculation, and returns the answer. You never write a query language or a formula unless you want to.

This is not a new dashboard or another tool to learn. It works on the export as it is. You hand it the file, ask the question the way you would ask a colleague, and the number comes back while the question still matters.

For PMs and operators, this changes one specific thing: you stop filing requests and waiting in the data team queue. The answer you need is almost always sitting in a file you already have access to. Claude Code lets you ask it directly.

Why this matters: the analyst queue problem

The question came up mid-meeting. The analyst is booked until Thursday. By the time the number lands, the decision has already been made without it. This is the default state for most teams - not because the data is unavailable, but because asking it requires a skill most people do not have and a queue they do not control.

Claude Code removes the queue. You bring a file and a question. Claude does the part that used to need the data team. The answer arrives in minutes, while the context is still live.

Minutes instead of days. Zero SQL written. Any export you already hold. If you can open the file, you can ask it.

Step-by-step: get your first answer in 30 minutes

Step 1: Drop in the file (5 min)

Export a CSV from whatever tool you use - your CRM, your product database, your billing system. A spreadsheet works too. Open Claude Code and give it the file path or paste the content directly. Let Claude confirm it can see the columns before you ask anything.

> here is signups.csv  -  can you confirm the columns you can see?

Step 2: Ask in plain English (12 min)

Ask the question the way you think it. No syntax, no schema knowledge required. Claude maps your plain-language question onto the columns in the file and runs the calculation in front of you.

> how many signups in March churned within 30 days?
142 of 1,090 March signups churned within 30 days (13%)

Notice that Claude returns 142 of 1,090, not just 13%. That fraction is how you check the answer, not just trust it.

Step 3: Sanity check the answer (8 min)

A confident wrong answer is the real risk. Before you act on any number, run these four checks:

1. Check the row count. Ask Claude how many rows it looked at. If that number is off, the answer is too.
2. Spot check one case. Pick a customer you know and confirm Claude got their record right.
3. Watch the date range. Make sure "March" means the March you meant, not a financial quarter or the previous year.
4. Ask it to show the working. Have Claude explain how it got there. A wrong method shows up in the explanation.

Trust the answers you can check. Question the ones you cannot.

Step 4: Make it routine (5 min)

The pattern is the same every time: file, question, sanity check. Once you have run it once, the next question takes two minutes. Follow-up questions build on the previous answer without starting over.

> now split that by plan
> which plan had the highest churn rate in March?

Three questions you can ask tonight

1. Churn by cohort

Pull your signups export and ask which cohort churned fastest. This is the question most teams wait a week for. With Claude Code it takes one prompt.

Prompt used

"How many signups in March 2026 churned within 30 days, broken down by plan type?"

2. Conversion across two files

Claude handles joins between two files. Give it your trials CSV and your paid accounts CSV and ask it to match them by user ID. You do not need to do the join yourself.

Prompt used

"Here are trials.csv and paid.csv - how many trials converted to paid last quarter, and what was the average time to convert?"

3. The mid-meeting number

Someone asks a question in a meeting and no one has the number. Export the relevant file during the break and ask Claude directly. The answer is back before the meeting resumes.

Prompt used

"How many enterprise accounts renewed in Q1 2026 versus Q1 2025, and what is the year-on-year change as a percentage?"

The one mistake that gets you a wrong answer

Most bad numbers come from one thing: a vague question Claude has to guess the meaning of.

× Avoid: "how are signups doing?"
Doing what, over what period? Claude guesses, and the number is meaningless.

✓ Do this: "how many new signups in March 2026, by plan?"
The count, the month, the breakdown. Claude answers exactly what you asked.

Ask the precise question and you get the precise number. Add the time period, the metric, and the breakdown you want - every time.

4 things that make this work every time

1. Name the time period explicitly. "March" is ambiguous. "March 2026" is not. Always include the year.
2. Ask for the fraction, not just the percentage. "142 of 1,090" lets you verify. "13%" does not.
3. Follow up freely. Each question builds on the last. You do not have to re-explain the file every time.
4. Start with one file. Get comfortable asking a single export before combining two. The pattern is the same, but one file at a time removes variables when you are learning.

Want to go deeper with Aki?

This session is part of the AI Native Product Management track - covering data, evals, and AI features live with Aki in a small cohort. Browse upcoming sessions →