How to use Claude for marketing data analysis

Open your data in Google Sheets or Excel.

Headers: clean, descriptive, no spaces or special chars. "customer_email" not "Customer Email (verified)."

No merged cells. Each row = one record, each column = one field.

Remove sensitive PII if possible: anonymize email IDs, mask phone numbers.

Export as CSV (preferred for clean parsing) or Excel.

File should be under 30MB for Claude to handle efficiently.

Open a new Claude chat. Click the paperclip icon to upload the file.

Claude processes the file and confirms: "I have analyzed the file. It has X rows, Y columns. Columns are: [list]. Here are 3 sample rows: [shows]."

Verify the parsing: column names correct? Data types right (numbers as numbers, dates as dates)?

If parsing is wrong: tell Claude. "Re-parse the date column as datetime. It is currently being read as string."

Example for ad spend data: "What is the average CPA by campaign? Identify the top 3 most efficient campaigns and the top 3 least efficient by CPA. Cite specific numbers."

Example for customer data: "What is the LTV distribution? What percent of customers drive 80% of revenue? Cite specific numbers and customer IDs."

Example for behavior data: "Group sessions by traffic source. Average session duration by source. Identify sources with abnormally short sessions or high bounce rates."

Claude reasons through the data and produces written analysis.

Verify the answer: do the numbers match your spot-checks of the source data?

After getting a text answer, prompt: "Generate an SVG bar chart of the top 10 campaigns by CPA, sorted ascending."

Claude generates an Artifact with the chart.

For time-series: "Generate a line chart of monthly revenue over the last 12 months. Highlight the highest and lowest months."

For distributions: "Generate a histogram of customer LTV. Show median + 90th percentile."

Export: download the SVG from the Artifact panel. Embed in reports or share via screenshot.

For cohort analysis: "Split customers into cohorts by signup month. For each cohort, calculate average revenue per customer at 30/60/90 days. Is retention improving or declining over cohorts?"

Claude analyzes and writes narrative + numbers.

For trend analysis: "Looking at the last 12 months, identify the top 3 trends in [metric]. For each: describe the trend, quantify it, suggest the most likely cause based on the data."

Narrative analysis is where Claude shines vs ChatGPT — output reads like an analyst wrote it.

For every finding, ask: "Does this match what I already know about the business?"

If Claude says "Campaign X has highest ROAS" but you know Campaign X had a tracking issue last month, the analysis is wrong.

Validation tactics: cross-check 1-2 numbers manually in source data. Compare against last period's known values. Run the same analysis in a different tool (Excel, Looker) and compare.

AI analysis is a starting point, not the final answer. Validate before acting on findings.

After analysis is complete: "Summarize the top 5 findings as a shareable report. Format: executive summary (3-5 sentences) → key findings (5 numbered, with numbers) → recommended actions (3-5). Markdown format."

Claude formats the output. Copy to Notion/Google Docs.

For team sharing: "Format as a Slack message. 3-5 bullet findings with key numbers. Under 500 chars."

Reuse the chat as audit trail — anyone questioning the analysis can see prompts and responses.