AI for Data Analytics (For People Who Aren't Analysts)
Co-founder at King’s Cross Labs · ex-LinkedIn PM & Forbes 30 Under 30
Send this guide to yourself
Get the link in your inbox so you can read it whenever you're ready.
Your email will also be saved for future updates.
AI for data analytics means giving an AI your numbers and asking questions in plain English instead of building formulas, pivot tables, or dashboards. The tool you pick matters less than one habit: make the AI run real code on your data, and then check its math.
I studied economics and data analysis in college, and I was a product lead at LinkedIn, where decisions started with a spreadsheet. I'm not saying that to flex. I'm saying it because I've seen AI get an analysis right and get one confidently wrong, and the difference is almost always how you ask. This guide shows a full worked example with Claude and the exact check I run before I trust any number.
What is AI for data analytics?
AI for data analytics is using an AI assistant to clean, calculate, chart, and explain data you give it, like a CSV export or a connected spreadsheet. You ask "why did refunds jump in June?" and it writes and runs the code that answers it. Most of the guides ranking for this topic are enterprise software explainers. This one is for the person with a spreadsheet and a deadline.
What is AI good at, and where does it fail?
AI is good at the slow parts of analysis and bad at the judgment parts. Here is how I split it.
Good at:
- Cleaning messy columns, fixing dates, and merging two exports
- Totals, averages, month-over-month changes, and ratios across thousands of rows
- Drawing a chart in seconds
- Explaining a result in plain English
- Suggesting questions you didn't think to ask
Fails at:
- Math it does in its head instead of by running code
- Knowing what your columns mean (is "revenue" before or after refunds?)
- Noticing a number is wrong when the code ran without errors
- Deciding which question matters to your business
Which AI tool should you use for data analysis?
Start with whatever assistant you already pay for, as long as it can run code on a file you upload. These are the common options, summarized from each vendor's own pages. Check current pricing and limits before you buy.
Claude. Upload a CSV or spreadsheet and Claude can write and run code to analyze it, as long as code execution and file creation is switched on in your settings. Claude is my main assistant for everything, including analytics, because I can connect my own tools to it, which matters later in this guide.
ChatGPT. Its data analysis feature runs Python on files you upload and draws charts. OpenAI's own help page recommends reviewing the generated code, outputs, and assumptions before relying on results, which is the habit this whole guide is built around.
Julius AI. A tool built only for analyzing uploaded data. It has a small free tier, and third-party sources disagree on paid prices, so check its plans page.
Microsoft Copilot (Excel and Power BI) and Tableau. These make sense if your company already runs on Microsoft or Tableau, since the AI works inside the reports and dashboards your team already uses.
How do you analyze a spreadsheet with Claude? (Walkthrough 1)
Upload the file, give Claude context about what each column means, and ask a specific question. The sample below is made up so you can follow along. It is not real data from my business.
Sample data: 12 months for a small online shop. Copy this into a file called shop-sample.csv. Orders is the number of orders. Refunds is the number of orders that were refunded.
| Month | Orders | Revenue ($) | Ad spend ($) | Refunds |
|---|---|---|---|---|
| Jan | 310 | 12,400 | 1,800 | 14 |
| Feb | 295 | 11,800 | 1,800 | 12 |
| Mar | 340 | 13,900 | 2,000 | 15 |
| Apr | 360 | 14,700 | 2,200 | 17 |
| May | 405 | 16,800 | 2,600 | 19 |
| Jun | 388 | 16,100 | 2,600 | 31 |
| Jul | 352 | 14,300 | 2,400 | 18 |
| Aug | 330 | 13,200 | 2,400 | 16 |
| Sep | 395 | 16,400 | 2,800 | 21 |
| Oct | 450 | 19,300 | 3,400 | 22 |
| Nov | 610 | 27,600 | 5,200 | 38 |
| Dec | 570 | 25,100 | 5,200 | 41 |
The prompt. Attach the file and paste this:
I've attached 12 months of sales data for a small online shop.
Columns: Month, Orders (count), Revenue ($, before refunds),
Ad spend ($), Refunds (number of refunded orders).
Questions:
1. Is the shop actually getting more efficient with ad spend,
or just spending more?
2. Anything unusual in the refund numbers?
3. What would you look into next?
Run code to calculate everything instead of estimating, and show
me the code. Tell me any assumptions you made. Keep the answer
under 200 words and lead with the single most important finding.What a good answer looks like. It should contain these facts, which I calculated separately so you can compare:
- Revenue per ad dollar dropped from about 6.9 in January to about 4.8 in December. Ad spend roughly tripled (from $1,800 to $5,200 a month) while revenue only doubled. The shop is buying growth, not getting more efficient.
- June's refund rate spiked to about 8% of orders (31 of 388), against roughly 4% to 5% in the surrounding months. A good answer flags it and suggests checking what changed that month, like a product batch, a promo, or a shipping problem.
- Refunds crept up again in November and December, to about 6% and 7%.
- Full-year totals: $201,600 revenue, 4,805 orders, $34,400 ad spend, 264 refunds.
If Claude's numbers don't match, that is the lesson. Don't ask it to try again yet. Run the checklist below first.
How do you check the AI's math?
Treat the AI's answer like work from a smart new hire: probably right, and worth ten minutes of checking before it goes in front of anyone. This is the checklist I use.
1. Make it run code, and ask to see the code. "Run code to calculate this and show me the code" is the most valuable sentence in any analytics prompt. A number the AI typed from memory can be invented. A number produced by code can be traced, even if you can't read Python.
2. Reconcile the totals. Compare one or two totals against a source you trust. Does the total revenue match your payment dashboard? Here it should be $201,600. If the totals don't match, nothing built on them can be trusted.
3. Spot-check a few rows. Pick two or three rows and calculate them by hand. June refund rate is 31 divided by 388. A calculator takes ten seconds, and it proves the AI is reading your columns correctly.
4. Recount the rows. Ask "how many rows did you analyze, and did you drop any?" The AI may silently skip rows with blank cells or odd formats. This sample has 12 rows, so anything else is a red flag.
5. Ask what it assumed. "List every assumption you made, including how you defined each metric." This is where you find out the AI treated revenue as after refunds when yours is before, or that it ignored a column.
Key insight: AI errors rarely look like errors. They look like clean, confident answers, so the checking has to be a habit, not a reaction.
How do you connect live data so you stop exporting files? (Walkthrough 2)
Connect your tools to Claude through connectors so it reads live data instead of a stale CSV. Connectors are built on MCP, a standard that lets an AI assistant read from and act in other apps. Once your data is connected, "how did last week go?" becomes a one-line question.
Which apps you can connect (spreadsheets, payments, web analytics) changes often, so check the connector directory in your Claude settings for what's available on your plan. Two things I'd tell anyone who is starting out:
- Start read-only. Connect for reading data first. Add anything that can change data later, once you trust the setup.
- Run the same checklist. Live data doesn't make the math safer. Step 2, reconciling a total against the source, matters even more when you can't see the file.
My example: reading social and ad performance with Adeli. I built Adeli, which exposes Instagram, TikTok, and ad accounts as one MCP, so Claude can read post analytics and ad performance across them. I can ask Claude which posts did best this month without exporting a report from each app. The same idea works for ads on its own: How to Connect Meta Ads to Claude shows the Meta route step by step. For a personal-data version of the same pattern, see How to Build an AI Running Coach with Garmin, where Claude reads training data from connected apps, and the Health Coach dashboard.
What should you keep private?
Don't upload anything to an AI tool that you wouldn't be allowed to email outside the company. Before any upload:
- Remove customer names, emails, phone numbers, and addresses. Analysis almost never needs them.
- Replace IDs with row numbers if you only need patterns.
- Skip payment card data, health records, and anything legally regulated.
- Check your plan's privacy and training settings, and your company's AI policy.
- If you work at a company, ask first. A personal subscription is not a company-approved tool.
For a CSV like the sample above, there's no personal data at all, which is why it's a good file to practice with.
What is a reusable weekly analytics prompt?
Save this once and run it every week. It works best with live data connected, or with a fresh export each Monday. If Claude should remember your definitions between sessions, see How to Give Claude Memory.
Analyze my [business/channel] data for last week and compare it to
the previous four weeks.
Definitions: [e.g., Revenue = before refunds. Conversion = orders
divided by sessions.]
Give me:
1. The three numbers that matter most and how each changed.
2. The biggest positive and negative surprise.
3. One thing to investigate and a specific first step.
Rules: Run code for all calculations and show it. List your
assumptions and any rows you excluded. If a number looks off,
say so instead of explaining it away. Under 200 words.New to Claude? How to Setup Claude gets you the basics in ten minutes.
Additional reading
Here are some related guides to check out:
Frequently asked questions
- Can Claude or ChatGPT analyze a CSV file for me?
- Yes. Both let you upload a CSV or spreadsheet and ask questions in plain English, and both can run code on the file to calculate totals, find trends, and draw charts. The key is to make the AI run the calculation and show you the code instead of estimating the answer in its head.
- Does AI make up numbers when it analyzes data?
- It can. The usual causes are guessing instead of calculating, misreading a messy column, or silently assuming something you never told it. You can catch almost all of it by asking the AI to run code, reconciling its totals against your own, spot-checking a few rows, and asking what it assumed.
- Is it safe to upload company data to an AI tool?
- It depends on the data and your plan. Strip out customer names, emails, and anything regulated before you upload, and check the privacy and training settings on your plan. For anything sensitive, ask your company what it allows first.
- Will AI replace data analysts?
- Not the judgment part. AI is fast at pulling numbers and drawing charts, but someone still has to decide which question matters, notice when a number looks wrong, and act on the answer. It makes non-analysts far more capable and makes analysts faster.
Want to build your first AI agent?
Build Your First Agent 101 is the next step! You'll get step by step guides, video tutorials and starter prompts to create your first agent in half a day, customized to your own workflow.
Learn more