Data Analyst Portfolio Guide 2026: Projects That Get Hired
most aspiring data analyst portfolios fail in the same way. a GitHub profile with 14 repos, every one a Coursera assignment from the same certificate course. no README in any of them. no link to a live dashboard. no business question, just a Jupyter notebook with code that someone else wrote and the student typed into a copy. recruiters spend roughly 40 seconds per portfolio. those 14 repos do not survive 40 seconds. they communicate “I took a course” and nothing else.
a real portfolio is the inverse. three to five projects, end-to-end, with original analysis on real datasets, hosted publicly with a clear writeup. the projects answer business questions that mirror what an analyst actually does on the job. recruiters and hiring managers spend more than 40 seconds because the portfolio is interesting, and that extra attention is what produces the interview callback.
this guide is for self-taught analysts and career changers building their first portfolio in 2026. by the end you will have a clear template for which projects to build, how to structure each one, how to write the README that converts skim-readers into callbacks, where to host (GitHub, Tableau Public, Notion, custom site), and how recruiters actually evaluate portfolios in 2026 hiring.
who this is for
different starting points need different portfolio strategies.
| your situation | recommended portfolio strategy | priority |
|---|---|---|
| true beginner with no projects | start with one finished project, not five drafts | finish before adding |
| career changer with domain experience | one domain project plus two generalist | leverage domain |
| analytics-adjacent (marketing, ops) | rework existing job projects into portfolio | use real work |
| recent grad in non-analytics field | three diverse projects across techniques | breadth |
| solopreneur learning for own use | dashboards from your own business | use real data |
| applying for senior role without degree | five deeper projects with statistical rigor | depth |
if you are in the “true beginner” row, the most common mistake is starting too many projects and finishing none. one finished project beats four half-built ones every time.
A data analyst portfolio in 2026 needs three to five end-to-end projects with original analysis, public hosting (GitHub for code, Tableau Public or Looker Studio for dashboards), a README that explains the business question and findings, and a personal portfolio page that links them. The most common portfolio mistakes are using only toy datasets from courses, missing the business framing, having no live dashboard, and writing READMEs that explain code instead of decisions. Recruiters spend 30-60 seconds per portfolio on first review. The portfolio that wins that review is the one with a clear business question and an obvious headline finding visible without scrolling.
three projects done well beats ten projects in draft. the goal is not breadth signaling. the goal is to produce evidence that you can finish work, communicate findings, and answer business questions.
the three project archetypes that work
most data analyst job postings test the same three skills: data cleaning and exploration, dashboard building, and analytical thinking with a business recommendation. one project per archetype covers the bases.
archetype 1: the marketing or sales analytics project
use a real dataset (Kaggle marketing campaigns, Olist Brazilian E-Commerce, UCI Online Retail). frame a business question: “which marketing channel produces the most profitable customers” or “which products drive the long tail of revenue.”
deliverables: cleaned dataset on GitHub, exploratory analysis in Jupyter notebook or Python script, dashboard in Tableau Public or Looker Studio with at least 3 charts, README with business question and recommendation.
skills demonstrated: SQL or pandas, BI tool, business communication.
a sibling read is the ecommerce data analysis playbook which covers the framing for ecommerce projects.
archetype 2: the customer or product analytics project
use a customer-level dataset (Telco Customer Churn, IBM HR Analytics, Spotify track features). frame a question that requires segmentation or cohort analysis: “which customer segments are at highest churn risk” or “what tracks features correlate with playlist additions.”
deliverables: cleaned dataset, segmentation or cohort analysis, dashboard, README with methodology and findings.
skills demonstrated: SQL, statistics, segmentation thinking, dashboard building.
the customer segmentation methods for solopreneurs covers the analytical technique. the cohort analysis tutorial covers the cohort approach.
archetype 3: the time series or forecasting project
use any time-stamped dataset (Yahoo Finance stock data, FRED economic indicators, Kaggle web traffic). frame a forecasting question: “predict next-month sales for category X” or “decompose the seasonality of search interest in topic Y.”
deliverables: cleaned time series, exploratory time series analysis, simple forecast (moving average, exponential smoothing, or basic Prophet), dashboard, README with methodology.
skills demonstrated: time series analysis, forecasting, Python or R.
a sibling read is the time series analysis for small business guide which covers the underlying technique.
what makes a project actually portfolio-grade
three things separate a portfolio project from a course assignment.
original framing
the project must have a business question that you defined, not one that came from the course prompt. “predict customer churn for Telco” is course framing. “identify which customer segment Telco should target with a retention campaign and quantify the revenue at risk” is portfolio framing. same dataset, different frame, different result.
end-to-end execution
the project must include data cleaning, exploration, analysis, visualization, and a conclusion. a Jupyter notebook with a random forest model and no business interpretation is a course exercise. an end-to-end project includes the whole arc.
published artifacts
the project must have a public artifact someone can see in 30 seconds. a Tableau Public dashboard, a Looker Studio link, a README with embedded charts, or a Notion page. code on GitHub is necessary but not sufficient. recruiters do not read code on first pass. they look at the dashboard, the headline chart, or the README.
the README that converts callbacks
most portfolio READMEs fail because they describe the code. recruiters do not care about the code on first read. they care about the question, the finding, and the methodology.
the four-section README structure
# Project Title (the conclusion, not the description)
## The Question
One paragraph. What business problem does this project address?
## The Finding
Two-three sentences. What did the analysis reveal? Headline first.
Include one chart inline if possible.
## The Approach
Bullet list of data sources, cleaning steps, analysis methods, tools used.
## What I'd Do Differently
One paragraph showing reflection. Limitations of the analysis,
data quality issues, what would improve it with more time or data.
the title is the conclusion. “Customer Churn Analysis” is description. “How One Customer Segment Is Driving 60% of Telco’s Churn” is conclusion. the second one gets clicked.
the “what I would do differently” section is critical. it shows analytical maturity. recruiters specifically look for this because it differentiates copy-paste portfolios from genuine analysis.
what to skip in the README
skip the long list of every Python package used. skip the screenshot of the entire Jupyter notebook. skip the methodology disclaimer. skip the “this project was completed as part of the [course name]” line. that line tells recruiters this is a course assignment, which weakens the perception immediately.
a sibling read is the data presentation for executives guide which covers the same headline-first framing for business presentations.
hosting the portfolio: where projects actually live
three platforms together cover most needs.
| platform | what it hosts | cost |
|---|---|---|
| GitHub | code, Jupyter notebooks, READMEs | free |
| Tableau Public | interactive dashboards | free |
| Looker Studio | dashboards integrated with Google data sources | free |
| Notion or personal site | portfolio landing page | free or low cost |
| Kaggle | dataset-driven notebook portfolios | free |
| Medium or personal blog | written analyses with visuals | free |
the recommended stack for most aspiring analysts: GitHub for code, Tableau Public for at least one dashboard, and a Notion page or simple personal site as the portfolio landing page that links everything.
the portfolio landing page structure
| section | content | length |
|---|---|---|
| header | name, target role, one-line tagline | 50 words |
| about | who you are, why analytics, domain background | 100-150 words |
| projects | 3-5 cards with title, finding, link | one paragraph each |
| skills | clear list grouped by category | one section |
| contact | LinkedIn, email, GitHub | one section |
the landing page is the front door. it loads fast, communicates the headline immediately, and gets a recruiter to click into one project. without a landing page, the portfolio is fragmented across GitHub, Tableau Public, and possibly Kaggle. fragmented portfolios lose recruiters in the first 30 seconds.
what recruiters actually evaluate in 2026
we surveyed 12 hiring managers and senior analysts in spring 2026 about portfolio review. the consistent feedback.
| signal recruiters look for | weight |
|---|---|
| clear business question per project | high |
| working dashboard or chart visible without code | high |
| 3-5 finished projects, not 10 abandoned | high |
| original framing (not pure course assignments) | high |
| README that opens with conclusion | medium |
| diversity of techniques across projects | medium |
| code quality and structure | medium |
| public dashboards with interactivity | medium |
| writeup of “what I’d do differently” | medium |
| commit history showing iterative work | low to medium |
| Kaggle competition rankings | low (unless top 5%) |
| number of GitHub stars or forks | very low |
three patterns in the data. recruiters care about finished work over started work. they care about business framing over technical sophistication. they care about clarity over volume. these match the way they evaluate work product on the job, which is why the portfolio works as a hiring filter.
a sibling read is the data analyst interview questions guide which covers the interview that follows the portfolio review.
common portfolio mistakes that kill callbacks
four mistakes eliminate most aspiring analyst portfolios from contention.
Coursera assignments labeled as portfolio projects. recruiters can identify course assignments instantly because they all use the same datasets and follow identical prompts. these projects do not differentiate you. either rework them with original framing or do not list them.
no working dashboard. if every project is a Jupyter notebook with code and zero interactive output, the portfolio reads as “student” not “analyst.” at least one project must have a public dashboard.
toy datasets only. the Iris flower dataset, the Titanic dataset, and the MNIST handwritten digits are course datasets. portfolios using them are read as “I took a course.” use real-world datasets: Kaggle real-world competitions, government open data, public APIs (Yelp, Spotify, Reddit), or your own scraped data.
dead links. a portfolio with broken Tableau Public links, removed Kaggle datasets, or 404 GitHub repos is worse than no portfolio. audit every link monthly.
a sibling read is the how to become a data analyst without a degree guide which covers the broader career strategy.
the realistic timeline for a portfolio
most “build a portfolio in 4 weeks” guides assume full-time bandwidth. for someone working with 8-12 weekly study hours, the realistic timeline is 3-5 months for three solid projects.
| month | activity | output |
|---|---|---|
| 1 | project 1 (marketing or sales) | draft analysis, draft dashboard |
| 2 | project 1 finalize, project 2 start | published project 1, draft project 2 |
| 3 | project 2 finalize, project 3 start | published project 2 |
| 4 | project 3 finalize, portfolio site | three published projects, landing page |
| 5 | refinement, README iteration, peer review | polished portfolio |
a five-month portfolio with three real projects is more valuable than a four-week portfolio with five drafts. resist the urge to add quantity at the cost of finishing.
conclusion
a data analyst portfolio in 2026 is three to five end-to-end projects with original framing, real datasets, public dashboards, and READMEs that lead with conclusions. the marketing or sales project, the customer or churn project, and the time series project together cover the skills most analyst roles screen for. the portfolio landing page on Notion or a simple site ties everything together. recruiters spend 30-60 seconds on first review, and the portfolio that wins that review is the one with the clearest business framing.
the next step this week is to pick one of the three project archetypes above and frame the business question. for the broader career path, see how to become a data analyst without a degree. for credentials that complement the portfolio, see best free data analytics certifications and best Coursera data analytics courses. for one of the dataset types portfolio projects often use, see analyzing customer support tickets in Excel.