Data analyst internships: answering the question someone asked

What the job is made of

• SQL well past SELECT — joins, window functions, and the patience to check what a join silently dropped
• Cleaning data that arrives wrong, which is most of the work and none of the coursework
• One dashboarding tool, well enough that someone else can read your chart without you narrating it
• Enough statistics to know when a difference is noise, and the nerve to say so
• Writing a short, clear summary — the finding is worthless if the person who asked cannot act on it

Analyst or scientist — the difference that decides your prep

A data scientist is usually asked to build something predictive. An analyst is usually asked to explain something that already happened, to someone who will make a decision on the answer this week. The scientist's output is a model; the analyst's output is a recommendation with a number behind it.

The practical consequence is that analyst interviews lean hard on SQL and on how you reason about a messy business question, not on machine learning theory. Students who spend six months on neural networks and then interview for an analyst role are routinely caught out by a question about a join producing duplicate rows.

You do not need a CS degree for this one

Analytics is the most accessible of the data roles for students outside computer science, and the mechanical, civil and electrical students who take it seriously often do well, because the job rewards asking what a number means before computing anything with it.

What replaces the degree is evidence. Not a certificate — a piece of work where you took a real dataset, asked a specific question of it, and wrote up what you found including what you could not conclude. That last part is disproportionately convincing, because it is the thing an inexperienced analyst almost never does.

The portfolio project that actually works

Pick a dataset about something you genuinely care about, ask one narrow question, and answer it in a page. One chart that makes the point, two paragraphs explaining the method, and an honest note about what would change your conclusion.

This beats a sprawling notebook with twenty visualisations, because hiring managers read the first and skim the second. The narrow, finished, honest piece of analysis is also a far better predictor of what you would be like to work with, which is the actual thing being assessed.

A useful test before you publish it: hand it to a friend outside your branch and watch where they stop. If they need you to explain what the chart is showing, the chart is not finished. Analysts are paid partly for making a finding land with someone who does not share their context, and a portfolio piece is the first evidence you can do it.

Where to get real data to work on

See every curated opportunity →