howtocodeai.com
Tutorials / Beginner+

Analyze Any Spreadsheet With Claude (From CSV to Charts to Answers)

Sales exports, survey results, bank statements, sports stats — the workflow for turning a messy CSV into cleaned data, real charts, and answers you can defend.

DifficultyBeginner+
Time2 hours for the skill, 20 minutes per analysis after
You'll needClaude · A CSV or Excel export · Optional: Python 3 for big files
You'll buildNot an app — a repeatable analysis workflow: clean the data, ask questions in English, generate charts, and produce a shareable findings report. Plus the upgrade path when files get big.

Every business, hobby, and household generates spreadsheets nobody analyzes — the export sits there because 'analysis' sounds like a pivot-table certification. The new workflow: hand the data to Claude, interrogate it in English, and get charts plus caveats back. The skill worth learning isn't the clicking; it's how to keep the analysis honest.

Step 1 — Small file? Just paste it

Under a few hundred rows, attach or paste the CSV directly into a chat. First message sets the frame:

Here's a CSV of [12 months of Etsy sales]. Before any analysis: (1) describe the structure — columns, types, row count; (2) list data-quality problems — missing values, duplicates, inconsistent formats, suspicious outliers; (3) tell me what questions this data CAN'T answer. Don't analyze anything yet.

That third item is the honesty anchor. Most bad analysis comes from asking data questions it can't answer — knowing the limits first inoculates the whole session.

Step 2 — Clean before you conclude

Fix the problems it found: 'standardize the dates, merge the duplicate customer names you flagged, and tell me exactly what you changed.' Demand the change log every time — silent cleaning is how an AI quietly deletes the rows that mattered. Then save the cleaned CSV as your new working file.

Step 3 — Interrogate, then cross-examine

Ask your real questions: best-selling items by season, revenue trend excluding the viral spike, repeat-customer rate, which discount actually moved units. Then apply the two-question habit that separates analysis from vibes: 'How confident should I be in that, and what would change the conclusion?' and 'Show me the rows behind that claim.' If a finding can't survive those, it wasn't a finding.

Worth knowingWatch for the classic trap: AI happily narrates causation from correlation ('sales rose because you lowered prices'). Make it label every causal claim as hypothesis vs. supported, and you'll catch most overreach automatically.

Step 4 — Charts and the shareable report

Create a single HTML report file of this analysis using Chart.js from CDN: a monthly revenue line chart, top-10 products bar chart, repeat-vs-new customer split, and a findings section — each finding stated in one sentence with its supporting numbers and an honest caveat where one exists. Embed the cleaned data in the file so the report is self-contained. Clean, printable design.

One file, email-able to a boss or co-founder, no dashboard subscription involved. This is the same single-file pattern as everything else on this site, pointed at analysis.

Step 5 — When the file is too big to paste: graduate to scripts

Past a few thousand rows, stop pasting and start scripting: 'Write a Python script using pandas that loads sales.csv and answers these questions: [list]. Print results clearly and save charts as PNGs.' You run it locally (pip install pandas matplotlib), and here's the deeper win: the script IS the analysis, written down. Re-run it on next month's export and the report updates in seconds — which is the moment this stops being a chore and becomes a system. (Sound familiar? Add a cron and it's the self-updating pattern.)

What to analyze first

Your own exports are the perfect training data because you can smell wrong answers: bank/credit-card statements (where does it all go), your site's analytics export, fantasy league history, the family budget. Run the full loop once on data you know — then you'll trust yourself to run it on data you don't.

Keep going

Need somewhere to put it live? See where to host AI-built sites. Compare tool costs on the pricing tracker (or stick to the free options), then pick your next build.