How to Build a Data Analyst Portfolio With No Work Experience

With entry-level hiring down 15%, the portfolio is not a nice extra. In interviews it is usually the whole conversation.

With entry-level data analyst hiring down 15% year on year, the portfolio has stopped being a nice extra. It is the mechanism by which someone with no employment history becomes a credible candidate, and in interviews it is usually the whole conversation.

Most beginner portfolios fail for the same three reasons. Here is how to avoid them.

The three failure modes

No question. A dashboard exists, but nobody can say what it was for. Interviewers cannot assess judgement they cannot see.

Clean data. A tidy Kaggle CSV analysed without incident demonstrates that you can use a tool. It does not demonstrate that you can do the job, because the job is mostly the mess.

No recommendation. The analysis ends at a chart. Real analysis ends at "so we should do X".

Fix those three and you are ahead of the large majority of applicants at the same stage.

What to build: three projects, three different muscles

Project 1: an Excel or SQL analysis of genuinely messy data. Purpose: prove you can clean and reconcile. UK government open data is ideal, because it is real and untidy. Document what you fixed and why. Example: analysing three years of local authority spending to find category inconsistencies and identify where reported totals diverge from line-item sums.

Project 2: a Power BI dashboard with a proper data model. Purpose: prove you can model, not just visualise. Multiple related tables, a star schema, a date table, a handful of DAX measures. Answer one question clearly. Full walkthrough in our Power BI beginners guide.

Project 3: an end-to-end capstone with a recommendation. Purpose: prove you can carry a business problem from question to decision. This is the one you will talk about most in interviews, and it is where a structured programme helps, because it is the hardest to scope well alone.

Optionally a fourth: a small Python automation. Useful, not essential at junior level.

Choose data that matches where you want to work

This is the most underused tactic on the list. Applying to an NHS trust with an analysis of NHS Digital data, or to a housing association with an analysis of housing stock data, changes the conversation entirely. It shows you understand the domain and that you chose them deliberately.

Strong UK sources: data.gov.uk, NHS Digital, ONS, TfL open data, Police UK crime data, Land Registry price paid data, Companies House.

Structure every project the same way

Five sections, in this order:

The question. One sentence. Written before you touched the data.
The data. Source, size, time period, and its known limitations.
What you did to clean it. Specific decisions with reasoning. This section impresses more than any chart.
What you found. Two or three findings, stated plainly, with the visual that supports each.
What you would recommend. One or two concrete actions, plus what you would investigate next with more data.

That last line matters more than it looks. Saying what you do not know signals maturity; presenting a finding as definitive when the data cannot support it is the most common junior mistake.

Where to host it

GitHub for SQL and Python work. Use a real README with the five sections above. Recruiters rarely read code, but a clear README is scannable.

Power BI service for dashboards, or a recorded walkthrough if you cannot publish publicly.

One simple landing page tying it together. A basic site, a Notion page, or even a well-built LinkedIn Featured section. One link on your CV should reach everything.

Do not over-invest in building a portfolio website. Nobody has ever been hired for their CSS.

Using AI without undermining the exercise

Be pragmatic and be honest. Using AI to draft code, explain an error or tidy your write-up is normal professional practice, and nearly 40% of UK analyst adverts now mention AI.

The line is ownership. You must be able to explain every decision in your project, in an interview, without notes. If an AI made a modelling choice you cannot justify, it is not your project and the interview will expose that within two questions.

A genuinely strong move: include a short note on how you used AI and what you had to correct. It demonstrates exactly the validation instinct employers are trying to hire for.

Talking about it in an interview

You will be asked to walk through one project. Interviewers are listening for how you handled ambiguity, not for the chart.

Prepare answers to: what surprised you, what did you get wrong first, what would you do differently with more time, and how do you know your numbers are right. That last one is the question that separates candidates.

Rehearse the walkthrough out loud until it takes four minutes and does not need the screen.

Frequently asked questions

How many projects do I need? Three well-executed ones. Beyond that you get diminishing returns, and a fourth mediocre project weakens the set.

Do employers actually look at portfolios? Not always during initial screening, which is often CV-keyword driven. Almost always at interview, where it becomes the main evidence you are assessed on.

Can I use work data from my current job? No, not without explicit permission, and usually not even then. Recreate the problem shape with public data instead.

Should I include a project that failed? If the analysis was sound and the finding was inconclusive, yes. "The data could not support the hypothesis, here is why" is a genuinely strong answer.

The capstone in Uptrail's AI Data Analyst Career Programme is designed specifically as the third project above: a scoped, mentor-reviewed, end-to-end piece with a real recommendation, so you finish with the one thing that is hardest to produce alone.

Sources: DfE AI & Future of Work Unit with LinkedIn, April 2026; ITJobsWatch, Data Analyst skills co-occurrence data, 6 months to 1 September 2026.