Shipping Your Analysis: Choosing the Right Export Format

Every analysis eventually has to leave its home: sent to a stakeholder, loaded into another system, attached to an email. The export format you choose silently decides how much of your analysis survives the journey. Match it to the consumer, not to habit.

The Four Formats and Their True Costs

CSV is the lingua franca of machines. Every tool reads it. But CSV is lossy in specific, predictable ways: types degrade to text (dates and numbers need re-parsing on the other end), no structure survives (a nested metric becomes a flat row), and there is no place for caveats or context. Export CSV when a machine will consume it and the receiving tool re-validates types. Export CSV with a header row that documents units and formats — the header is your only chance to speak to the next system.

JSON preserves structure and types, which makes it the right vehicle for feeding another application — a dashboard, an API payload, a script. Its weakness is human consumption: a 10 MB JSON file is unreadable in an email. JSON is for machines that already know the schema.

HTML is the most underrated export. A self-contained HTML report — one file, inline styles, inline charts as embedded images — is readable by every human and every browser, requires no special software, and can carry the full context: methodology, caveats, quality grades, footnotes. This is the format to choose when the consumer is a person who needs to understand the analysis, not re-run it. It is also the only format that travels as a single artifact with its charts intact.

PDF is HTML with a fixed layout and no interactivity. Choose it when the output is a formal deliverable — a document that will be filed, printed, or versioned as an immutable record. The cost: charts are static, long tables paginate awkwardly, and a PDF cannot be edited by its recipient.

A Decision Rule

Ask one question: what will the recipient do with this file?

• Load it into a system ? CSV or JSON (JSON when the schema matters, CSV when the target tool is older).
• Make a decision from it ? HTML report, with the numbers and the caveats in the same document.
• File it as a record ? PDF.
• Uncertain ? offer more than one. The right export for a CFO and the right export for a data engineer are different files. A tool that ships CSV, JSON and an HTML report — the set KPI Master offers — covers all three consumers without making anyone compromise.

Fidelity Checklist

Whatever you export, verify before sending:

1. Numbers round-trip. Re-import your CSV and check totals match the source to the cent (see floating-point hygiene).
2. Types survive. Dates should parse as dates, not text, on the receiving end.
3. Context travels. The recipient should be able to tell what the numbers mean, how they were computed and what data they came from — from the file itself, not from the email thread.
4. No data left behind. A report that omits the low-quality columns silently is worse than one that shows them with a grade; the omission is invisible, the grade is not.

The format is a communication decision, like the chart or the headline. Choose it for the audience, verify it end-to-end, and the analysis arrives as intended — not as a pale, flattened version of itself.