13 Data Analyst Portfolio Projects That Get Interviews

JobRise Team9 min read

Stop Analysing the Titanic#

If your portfolio opens with Titanic survival, Iris flowers or a Superstore sales dashboard, a hiring manager has seen it forty times before lunch. That is not a survey result, just what most analysts who hire will tell you in private.

What gets a call is a question someone would actually pay to answer, a dataset that is a bit messy, and a finished thing a non-technical person can read in two minutes.

Below are 13 projects grouped by the kind of question they answer. Each one lists the data, what you ship, the resume bullet it earns, and rough hours. All the data is free. Pick three or four, not all thirteen.

One rule before you start: every project needs a written "so what". A chart without a recommendation is decoration.

Get the next guide by email

New guides and fresh jobs in one short email. Free.

Unsubscribe in one click.

Group 1: Questions About Money#

1. Where does a city's parking ticket revenue actually come from?

Data: NYC Open Data parking violations (or the equivalent from your city's open data portal). You ship: A Tableau Public dashboard plus a one-page memo naming the three streets, violation types or hours that drive most of the revenue. Bullet: "Analysed 5M+ parking violation records in SQL and Tableau, found that three violation types produced the majority of fines, and recommended a staffing change." Hours: 25. Check your own row count before writing a number like 5M.

2. Would a small cafe survive a price rise?

Data: The Kaggle "Coffee Shop Sales" style datasets, or your own made-up till export built from a real menu. You ship: A Google Sheets model with a sensitivity table: price up 5%, 10%, 15% against three guesses of lost customers. Bullet: "Built a price sensitivity model showing break-even volume loss at each price tier." Hours: 8. This is a good first project because it needs no code.

3. Do interest rate moves show up in mortgage searches?

Data: FRED (Federal Reserve Economic Data) mortgage rates plus Google Trends exports for "mortgage calculator". You ship: A Python notebook with a lagged correlation chart and an honest paragraph on why correlation here is weak evidence. Bullet: "Joined FRED rate series with search interest data in pandas and tested 0 to 12 week lags." Hours: 14.

4. Which suburbs are overpriced on short-term rental sites?

Data: Inside Airbnb city snapshots (London, Toronto, Sydney and New York are all published). You ship: A Power BI report comparing nightly price to review score and number of listings per host, by neighbourhood. Bullet: "Cleaned 70k listing rows, flagged neighbourhoods where price outran review quality, built an interactive Power BI report." Hours: 22.

Advertisement

Group 2: Questions About People#

5. Why did bike share trips drop in one specific month?

Data: Transport for London cycle hire data, Citi Bike or Divvy trip files. You ship: A SQL notebook that tests weather, holidays and station closures one at a time and rules causes in or out. Bullet: "Used SQL window functions to isolate a 20 percent month-on-month drop to two causes." Only write this if your number matches. Hours: 20.

6. What do patients complain about in public hospital reviews?

Data: Publicly released NHS England patient survey free text, or Care Quality Commission reports. You ship: A categorised theme count (do the first 200 rows by hand, then use code) and a short report. Bullet: "Coded 1,000 free text comments into 9 themes and ranked them by frequency." Hours: 18.

7. Who are the repeat customers in a retail dataset?

Data: The UCI Online Retail dataset (a real UK gift retailer's transactions). You ship: An RFM segmentation (recency, frequency, money) in Python or SQL, with one recommendation per segment. Bullet: "Segmented 4,000+ customers with RFM scoring and proposed a win-back offer for the lapsed group." Hours: 16.

Group 3: Questions About Quality and Operations#

8. How late is the train, really?

Data: UK National Rail performance data, or Statistics Canada and Australian state transport open data. You ship: A Looker Studio report showing delay by hour of day and line, with the one fix you would try first. Bullet: "Built a Looker Studio punctuality report used to rank lines by delay and peak hour." Hours: 15.

9. Clean a deliberately dirty dataset and document every decision

Data: Any messy government spreadsheet: council spending files are famously inconsistent. You ship: A GitHub repo with the raw file, the cleaned file and a changelog of every rule ("merged 14 spellings of the same supplier"). Bullet: "Standardised 40k rows of inconsistent supplier names and documented 11 cleaning rules in a reproducible script." Hours: 12. Hiring managers love this one because it shows the unglamorous cleanup work most analyst jobs involve.

10. Does response time predict customer satisfaction?

Data: The public Customer Support on Twitter dataset on Kaggle, or a synthetic one you generate and say so. You ship: A regression or even a simple grouped bar chart, with a note on what you could not conclude. Bullet: "Tested whether first response time predicted sentiment using Python and reported effect size with caveats." Hours: 14.

Group 4: Projects That Show You Can Build, Not Just Look#

11. An automated weekly report that emails itself

Data: Open-Meteo (free weather API, no key needed) or the GitHub API. You ship: A scheduled script that pulls data, writes a summary chart and posts it to a Slack channel or inbox. Bullet: "Automated a weekly reporting pipeline with Python and GitHub Actions, removing a manual step." Hours: 18.

12. Rebuild one chart from a news article

Data: Any newspaper that publishes its data: The Guardian, ABC Australia and Our World in Data all do. You ship: Your own version of their chart, plus a note on what you would change and why. Bullet: "Reproduced and critiqued a published data visualisation, improving axis scaling and labelling." Hours: 6. Quick win for a portfolio that is still thin.

13. A metrics dictionary for a fake product

Data: None. You define ten metrics for a food delivery app: what counts as an active user, a completed order, a late delivery. You ship: A one-page document with definitions, formulas and edge cases. Bullet: "Wrote a metrics dictionary of 10 KPIs with SQL definitions and edge cases." Hours: 5. Few candidates do this and every analytics lead wishes they did.

Advertisement

What a Good "So What" Memo Looks Like#

Every project above should end with a half page memo. Use the same four lines each time:

  1. The question in one sentence, written the way a manager would ask it.
  2. The answer in one sentence, with the number that backs it up.
  3. How sure you are and why. Missing rows, a short date range or a proxy metric all belong here.
  4. What you would do next if you were on the team.

Here is the shape for project 5, with placeholders where your numbers go: "Why did trips fall in March? Mostly because of two weeks of station closures and one cold snap. I am fairly sure about the closures, less sure about the weather because I only had daily averages. Next I would ask for dock level availability data."

Notice what is not in there: a list of libraries. The tools go in the README. The memo is for the person who will never open your notebook.

Mistakes That Sink Otherwise Good Projects#

  • Starting with the tool. "I wanted to learn Power BI" is a fine private goal and a weak project brief. Start from a question and pick the tool that fits.
  • No row counts. Always state how many records you started with and how many you kept after cleaning. It shows you looked.
  • Pretty but unreadable charts. Three colours, one message per chart, and a title that says the finding rather than the metric name.
  • Ignoring the licence. Check each dataset's terms. Government open data portals are usually open licences, but Kaggle uploads vary, and some scraped datasets are not safe to republish.
  • Unrealistic claims. If your sample is a single city or a single year, say that. Analysts who overclaim lose trust faster than analysts who say "I do not know yet".

A Sensible Toolkit for These Projects#

You do not need all of these. Pick the ones that match the jobs you are applying for.

  • SQL: SQLite or DuckDB on your laptop is enough for every project here, and both are free.
  • Spreadsheets: Google Sheets or Excel for projects 2, 12 and 13.
  • Python: pandas and a plotting library for projects 3, 5, 7, 10 and 11.
  • BI tools: Power BI Desktop, Tableau Public or Looker Studio for projects 1, 4 and 8. Check the current licence terms before publishing publicly, because they change.
  • Version control: a plain GitHub repo with a clear README for everything.

How to Package Any of These#

  1. Lead with the finding. The first line of your README is the answer, not the tool list.
  2. One link per project. A live Tableau Public, Looker Studio or GitHub link beats a PDF.
  3. Show the query. For SQL roles, link the actual .sql file.
  4. Say what you would do with more data. It signals judgement.

Which Three to Pick#

  • Aiming at finance or consulting: projects 2, 3 and 7.
  • Aiming at product or tech: projects 5, 10 and 13.
  • Aiming at government, health or NGO work: projects 1, 6 and 9.
  • Short on time: projects 2, 12 and 13 total about 19 hours.

Then run your resume through the free ATS checker against a real job description. If the posting says Power BI and your bullets say "dashboard", fix that. For the other documents in your application, see the data analyst cover letter examples and, if you are an engineering student, the branch-wise project ideas guide.

Build one thing this week. Ship it ugly, then fix it.

Advertisement

Send this to whoever has the interview this week.