A data analyst portfolio project in the AI era: show how you checked, not just what you built

A polished dashboard shows presentation and tool skills. An AI assistant can help produce the dashboard, notebook and write-up, so the final artefact alone does not show which decisions you made. A polished project now shows that you can use tools; it doesn’t show that you can be trusted with a number.

What is still hard to fake is the trace of judgement: which question you chose, how you defined the metric, what you checked, what you found wrong, and what you recommended.

Short version: pick a question someone would actually decide on, write the definition before the query, show two or three checks you ran (including failures if you found any), end with a recommendation and its limits, and keep it to one page plus the code.

What makes a project convincing now

1. A decision, not a dataset. “Should the marketplace extend its autumn promo?” beats “Exploratory analysis of e-commerce data”.
2. A written definition. “Net customer payments = payments after platform discounts minus completed refunds for paid orders created 1–30 September in Europe/Amsterdam, with refunds recorded by the stated snapshot date.” It shows you know that the word has several meanings.
3. Visible checks. Row counts before and after joins. Totals compared with a known number. The incomplete last week handled. A line like “my first total was 2.2× too high — here is why” says more about you than another chart.
4. Honest limits. What the data can’t tell, what you assumed, what would change the conclusion.
5. AI used openly. Say where an assistant helped and how you verified its output. It shows the verification habit that AI-assisted work depends on.

A one-page template

Question. Who asked (real or plausible), and which decision depends on it.

Definition. The metric in one sentence: measure, population, condition, period, exclusions.

Answer. The number, with its confidence and a chart.

Checks. Three bullets: what you verified and how. If a check found an error, show how you fixed it. Do not invent a failure for the story.

Recommendation. What to do, and what would change your mind.

Appendix. The SQL or notebook, with comments.

Where to get a good question and data

• Your own work or volunteering, anonymised — the strongest option, because the stakeholder is real.
• A realistic synthetic business with messy data: cancelled orders, refunds, promo codes, incomplete periods. Clean toy datasets hide exactly the traps you want to show you can handle.
• A simulator where requests come from “colleagues” and answers are checked. In Meridian, episode 1 ends in your own memo and every episode in a decision backed by your queries: practice in leaving the trace this guide is about (part of Atterna Analyst, €59).

Common mistakes

• Ten charts and no conclusion.
• A number without a definition.
• No sign of a single check.
• Copying an AI’s narrative without verifying its numbers — run it through the AI Answer Checker and the 7-question checklist first.
• Hiding what went wrong. The fix is the most convincing part.

FAQ

What is a good data analyst portfolio project in 2026?

One that starts from a decision, writes the metric definition before the query, shows the checks you ran (and any errors they actually caught), and ends with a recommendation and its limits. The trace of judgement matters more than the polish, because AI can generate the polish.

How many portfolio projects do I need?

Two or three strong, one-page projects beat ten dashboards. Each should show a different skill: a metric definition, an experiment, a drop investigation.

Should I say I used AI in my portfolio?

Yes. Say where it helped and how you checked its work. Verifying AI output is now part of the job, and hiding it looks worse than using it well.

Are Kaggle datasets good for a portfolio?

They are fine for practice, but popular datasets are often already cleaned and widely used, so projects on them look alike. Data with realistic mess — cancellations, refunds, incomplete periods — lets you show the checks that set you apart.