Skip to content
Join the community
Home/Resources/Projects/Data analytics projects
Project guide · Analytics

Data Analytics Project Ideas for Your Portfolio (2026): 10 Projects With Datasets and SQL

By Younus FardeenUpdated 31 Oct 202612 min read
Short answer

Strong data analytics portfolio projects start from a business question, use a real dataset, answer it with SQL and a dashboard, and end with a recommendation. Build three to five: sales performance, customer segmentation, churn, an A/B test and one AI-assisted analysis with a validation log.

Analyst portfolios are screened on SQL first and storytelling second.

Start here#1 e-commerce performance, #2 RFM and cohorts and #3 churn drivers cover SQL, segmentation and a dashboard.
Product or growth analyst?Add #6 A/B test, #7 funnel and the PM portfolio projects.
Want an AI angle?Finish with #10, an AI-assisted analysis that records every place the AI was wrong.
  1. 1Business questionA decision someone must make
  2. 2Load and checkSQL, row counts, nulls, duplicates
  3. 3AnalyseJoins, windows, cohorts, tests
  4. 4VisualisePower BI or Tableau, 3 charts max
  5. 5RecommendWhat to do, expected impact, limits
The pipeline every project below follows. Recruiters read the last step first.

The 10 projects

01

E-commerce sales and delivery performance

Beginner6 to 8 hours

The question: Where is revenue growing, and does late delivery hurt customer ratings?

PostgreSQL or DuckDBPower BI or Tableau Public
Start with
Olist Brazilian E-Commerce: about 100,000 orders from 2016 to 2018 in nine tables.
Steps
  1. Join orders, items, customers and reviews; check row counts after each join.
  2. Monthly revenue and growth (starter query A).
  3. Delivery lateness vs review score (starter query B).
  4. Build a one-page dashboard with revenue, AOV, on-time rate and rating by lateness.
Measure
Every dashboard number reconciles with a SQL query you can show.
Ship
Dashboard, SQL file and a half-page recommendation on delivery.
Why it counts
Classic business metrics plus a causal-looking question to discuss with care.
Resume bullet (fill in your numbers)

Analysed 100k marketplace orders in SQL; showed late deliveries cut average review score from [x] to [y] and recommended a delivery-promise change.

02

Customer segmentation (RFM) and cohort retention

Beginner to intermediate8 to 10 hours

The question: Who are our best customers, and how fast do new customers stop buying?

SQL window functionspandasTableau or Power BI
Start with
UCI Online Retail II: 1,067,371 transactions from 2009 to 2011, CC BY 4.0.
Steps
  1. Remove cancellations and rows without a customer ID.
  2. Score recency, frequency and monetary value with NTILE (starter query C).
  3. Build the monthly cohort retention table (starter query D).
  4. Name four segments and give each one action.
Measure
Segment sizes add up to total customers; cohort percentages are between 0 and 100.
Ship
Segment table, retention heatmap and an action plan.
Why it counts
RFM and cohorts are staples of marketing and growth analytics interviews.
Resume bullet (fill in your numbers)

Segmented [n] customers by RFM and built a cohort retention view; identified a segment worth [x]% of revenue with [y]% month-3 retention.

03

Telecom churn drivers dashboard

Beginner6 to 8 hours

The question: Which contract, tenure and service patterns separate customers who leave from those who stay?

SQLExcel or Tableau
Start with
Telco Customer Churn (Kaggle): about 7,000 customers.
Steps
  1. Compute churn rate overall and by contract type, tenure band and payment method.
  2. Test the biggest gaps with a simple chi-square or two-proportion test.
  3. Rank drivers by size and actionability.
  4. Dashboard with churn by segment and a retention proposal.
Measure
Each finding states the gap, the sample size and whether it is statistically supported.
Ship
Dashboard and a one-page memo.
Why it counts
Easy to explain and close to real telecom and SaaS analyst work.
Resume bullet (fill in your numbers)

Identified contract type and tenure as the largest churn drivers ([x] vs [y] percentage points) and proposed a targeted offer with estimated impact.

04

Marketing campaign conversion analysis

Intermediate6 to 8 hours

The question: Which contact approaches and customer groups convert best, and where is spend wasted?

SQLPython or ExcelPower BI
Start with
UCI Bank Marketing: direct marketing calls with outcome labels.
Steps
  1. Define conversion and baseline rate.
  2. Break down by month, contact type, previous outcome and job.
  3. Calculate lift of each segment over baseline.
  4. Recommend who to call, and who to stop calling.
Measure
Lift figures include group sizes so small groups are not over-read.
Ship
Lift table, chart and a campaign recommendation.
Why it counts
Shows marketing-analytics thinking without needing ad-platform data.
Resume bullet (fill in your numbers)

Analysed a bank campaign dataset to find segments with [x]x baseline conversion; recommended a targeting change to cut low-yield calls by [y]%.

05

Air quality across Indian cities

Beginner to intermediate8 to 10 hours

The question: Which cities and seasons are worst, and how did air quality move over the years covered?

pandasTableau Public or Power BISQL
Start with
Air Quality Data in India (Kaggle), compiled from CPCB readings, and data.gov.in for newer public series.
Steps
  1. Clean missing pollutant readings and state the rule you used.
  2. Rank cities by average AQI and by days in the worst categories.
  3. Show seasonality with a calendar heatmap.
  4. Write three insights a city planner could act on.
Measure
Missing-data rules are documented and results are shown with and without imputation.
Ship
Public Tableau dashboard and a short article.
Why it counts
A local, topical dataset gets more attention than another global sales sheet.
Resume bullet (fill in your numbers)

Built a public dashboard on multi-year air quality for [n] Indian cities; documented missing-data handling and highlighted seasonal peaks.

06

A/B test analysis and recommendation

Intermediate6 to 8 hours

The question: Did the new version increase the metric, by how much, and is it safe to ship?

Python (scipy or statsmodels)Excelone chart
Start with
A public A/B-test dataset such as the Cookie Cats mobile-game test on Kaggle, or simulate your own and state the assumptions.
Steps
  1. State the hypothesis, metric and guardrail metrics.
  2. Check the split is balanced and the sample is large enough.
  3. Run a two-proportion test and report the confidence interval, not only the p-value.
  4. Write ship, do not ship or collect more data, and why.
Measure
Report effect size with a confidence interval and the power you had.
Ship
One-page readout.
Why it counts
Experiment literacy is a differentiator for product and growth analyst roles.
Resume bullet (fill in your numbers)

Analysed an A/B test with [n] users; found a [x] percentage point change (95% CI [a, b]) and recommended [decision].

07

Product funnel and drop-off analysis

Intermediate8 to 10 hours

The question: Where do users drop between viewing a product and buying it, and which step is the biggest lever?

SQLMixpanel-style event thinkingTableau
Start with
Retailrocket e-commerce events: view, add-to-cart and transaction events.
Steps
  1. Build a user-level event table.
  2. Count users reaching each step and the conversion between steps.
  3. Split by new vs returning users.
  4. Pick one step and propose an experiment.
Measure
Funnel counts are consistent (each step is a subset of the previous).
Ship
Funnel chart and an experiment proposal.
Why it counts
Bridges analytics and product management.
Resume bullet (fill in your numbers)

Built a view-to-purchase funnel from [n] events; found the largest drop at [step] and proposed a test with an expected [x]% lift.

08

HR attrition analysis (people analytics)

Beginner5 to 7 hours

The question: Which factors are most associated with employees leaving?

SQLExcel or Power BI
Start with
IBM HR Analytics Attrition (Kaggle): a fictional dataset of 1,470 employees.
Steps
  1. Compute attrition by overtime, tenure, role and job satisfaction.
  2. Use window functions to rank departments by attrition.
  3. Highlight small groups that need caution.
  4. Recommend two policies to test.
Measure
State clearly that the data is fictional and that associations are not causes.
Ship
Dashboard and memo.
Why it counts
Teaches careful language about correlation, which interviewers probe.
Resume bullet (fill in your numbers)

Analysed attrition drivers across [n] employee records, highlighting overtime and tenure patterns and proposing two testable interventions.

09

Credit-card default risk segments

Intermediate8 to 10 hours

The question: Which customer groups carry the highest default rates, and where should limits be reviewed?

SQLpandasPower BI
Start with
UCI Default of Credit Card Clients: 30,000 Taiwanese card holders.
Steps
  1. Compute default rate by limit band, age band and repayment history.
  2. Find the strongest single predictor with a simple table.
  3. Create three risk tiers and quantify each tier's default rate.
  4. Recommend a policy per tier.
Measure
Tier default rates are monotonic and each tier has enough customers.
Ship
Risk-tier table and policy memo.
Why it counts
Fintech-flavoured analytics with a clear business decision.
Resume bullet (fill in your numbers)

Segmented 30,000 card holders into three risk tiers with default rates of [x]%, [y]% and [z]%; proposed tier-specific limit policies.

10

AI-assisted analysis with a validation log

Intermediate6 to 10 hours

The question: Where does an AI assistant speed up analysis, and where does it produce wrong code or claims?

any LLM assistantSQLPython
Start with
Any dataset above. Re-do one project with an AI assistant and compare with your original.
Steps
  1. Ask the assistant to write SQL and an EDA summary.
  2. Verify every query on a small sample you can check by hand.
  3. Log each error: wrong join, wrong filter, invented column, unsupported claim.
  4. Report time saved, errors found and your rules for trusting output.
Measure
The validation log: number of checks, errors caught and the fix for each.
Ship
Before and after comparison and the validation log.
Why it counts
Shows you use AI with judgement, which employers now expect.
Resume bullet (fill in your numbers)

Compared AI-assisted and manual analysis on [dataset]; caught [n] errors through verification and cut analysis time by [x]%.

14 more project ideas

Shorter briefs for when you want a different angle. Apply the same rules: a real source, a baseline, a number and a README.

Hospital wait-time analysis

Beginner

SQL, Tableau Public

DataPublic hospital or OPD datasets on data.gov.in

Stand-outA staffing recommendation by hour and day

Food-delivery pricing and ratings

Beginner

pandas, Power BI

DataKaggle restaurant or delivery datasets

Stand-outPrice-per-rating analysis with a recommendation for one city

Cricket analytics dashboard

Beginner

SQL, Python, Looker Studio

DataPublic IPL or international datasets

Stand-outA player-selection recommendation backed by numbers

Electricity demand and tariff analysis

Intermediate

SQL, Power BI

DataState electricity board open data

Stand-outPeak-hour recommendation with estimated savings

Retail inventory stock-out analysis

Intermediate

SQL window functions

DataKaggle inventory or Olist

Stand-outReorder points per SKU and the cost of stock-outs

Customer support ticket analytics

Intermediate

SQL, text clustering

DataSynthetic or public ticket data

Stand-outTop 5 root causes and the SLA impact

Website analytics audit

Beginner

GA4 export, Looker Studio

DataYour own site or a sample GA4 dataset in BigQuery

Stand-outThree concrete fixes ranked by traffic impact

Survey analysis with statistical tests

Intermediate

Python, statsmodels

DataA survey you run with 100+ responses

Stand-outConfidence intervals, not just averages

Loan portfolio risk dashboard

Intermediate

SQL, Power BI

DataUCI credit data or public lending data

Stand-outRisk tiers and an expected-loss estimate

Education outcomes by district

Intermediate

pandas, maps

DataUDISE or state education data

Stand-outA district ranking with documented data gaps

Pricing elasticity estimate

Advanced

Python, regression

DataOlist or retail price and quantity data

Stand-outElasticity by category and a pricing recommendation

Dashboard quality review

Beginner

Any BI tool

DataThree public dashboards you critique

Stand-outA before-and-after redesign of one dashboard

dbt models with tests on a public dataset

Intermediate

dbt Core, DuckDB or Postgres

DataOlist

Stand-outTested, documented models an analyst can trust

Starter SQL (PostgreSQL)

Table and column names follow the Olist and Online Retail II files after loading them as shown. Adjust names to match your load.

A. Monthly revenue and month-on-month growth (Olist)

WITH monthly AS (
  SELECT date_trunc('month', o.order_purchase_timestamp) AS month,
         SUM(i.price) AS revenue
  FROM olist_orders_dataset o
  JOIN olist_order_items_dataset i USING (order_id)
  WHERE o.order_status = 'delivered'
  GROUP BY 1
)
SELECT month, revenue,
       ROUND(100.0 * (revenue - LAG(revenue) OVER (ORDER BY month))
             / NULLIF(LAG(revenue) OVER (ORDER BY month), 0), 1) AS mom_growth_pct
FROM monthly
ORDER BY month;

B. Late delivery vs review score (Olist)

SELECT CASE WHEN o.order_delivered_customer_date > o.order_estimated_delivery_date
            THEN 'late' ELSE 'on_time' END AS delivery,
       COUNT(*) AS orders,
       ROUND(AVG(r.review_score), 2) AS avg_review
FROM olist_orders_dataset o
JOIN olist_order_reviews_dataset r USING (order_id)
WHERE o.order_status = 'delivered'
  AND o.order_delivered_customer_date IS NOT NULL
GROUP BY 1;

C. RFM scores (Online Retail II)

WITH base AS (
  SELECT customer_id,
         MAX(invoice_date)::date AS last_order,
         COUNT(DISTINCT invoice_no) AS frequency,
         SUM(quantity * unit_price) AS monetary
  FROM online_retail_ii
  WHERE customer_id IS NOT NULL AND invoice_no NOT LIKE 'C%' AND quantity > 0
  GROUP BY customer_id
)
SELECT customer_id,
       NTILE(5) OVER (ORDER BY last_order) AS r_score,   -- 5 = most recent
       NTILE(5) OVER (ORDER BY frequency)  AS f_score,
       NTILE(5) OVER (ORDER BY monetary)   AS m_score
FROM base;

D. Monthly cohort retention (Online Retail II)

WITH first_order AS (
  SELECT customer_id, date_trunc('month', MIN(invoice_date)) AS cohort
  FROM online_retail_ii WHERE customer_id IS NOT NULL GROUP BY 1
), activity AS (
  SELECT DISTINCT customer_id, date_trunc('month', invoice_date) AS active_month
  FROM online_retail_ii WHERE customer_id IS NOT NULL
)
SELECT f.cohort,
       (EXTRACT(YEAR FROM age(a.active_month, f.cohort)) * 12
        + EXTRACT(MONTH FROM age(a.active_month, f.cohort)))::int AS months_since,
       COUNT(DISTINCT a.customer_id) AS active_customers
FROM first_order f
JOIN activity a USING (customer_id)
GROUP BY 1, 2
ORDER BY 1, 2;

Using AI in analysis responsibly

  • Never paste personal or confidential data into a public tool.
  • Ask for the query and the reasoning, then run and check it yourself on a sample.
  • Re-compute one headline number by hand or in a second tool.
  • Keep a validation log. It is evidence of judgement and a strong interview story.

Portfolio checklist

  • One question per project, stated in the first line.
  • Data source and licence linked.
  • SQL file and dashboard both published (GitHub and Tableau Public or a Power BI screenshot).
  • Three charts at most, each with a title that states the finding.
  • A recommendation with expected impact and the limits of the data.

Frequently asked questions

How many data analyst projects should I include in my portfolio?

Three to five well-documented projects are enough. Include SQL, a dashboard and at least one project with a clear recommendation.

What is the best first data analytics project?

An e-commerce sales and delivery analysis in SQL with a one-page dashboard, because it uses real business metrics and can be finished in a day.

Do I need Python for a data analyst portfolio?

SQL is essential. Python helps for cleaning, statistics and automation, and many roles list it, but a strong SQL and BI portfolio can get interviews.

Where can I find datasets for analytics projects?

Kaggle, the UCI Machine Learning Repository, data.gov.in, GroupLens and Google Dataset Search. Check each licence.

Should I put analytics projects on GitHub?

Yes. Publish SQL, a README with the question and findings, and a link to your dashboard.

How long does it take to build a data analytics portfolio?

With 8 to 10 hours a week, three solid projects take roughly four to six weeks.

Can I use AI tools for my projects?

Yes, if you verify the output and document what the AI got wrong. A validation log turns AI use into evidence of judgement.

What projects should I avoid?

Projects with no business question, copies of tutorial notebooks, and dashboards with ten unlabeled charts.

Learn this with mentors

Data and business analytics programmes with admissions open: live classes, mentors and project work.

Business Analytics with AI (Pay After Placement)MICA
6 months→
Business Analytics with Gen & Agentic AIBITS School of Management
6 months→
Analytics with Agentic AIVishlesan i-Hub, IIT Patna
6 months→

Continue learning

Learning path

Data Analytics learning path

Spreadsheets, SQL, statistics and one BI tool, in the order employers test.

Data analytics projectsFree data analytics tools
Project guide · Product

PM portfolio projects

Teardowns, funnel audits, AI feature specs and experiments.

Open guide →18 ideas
Project guide · AI & ML

AI & ML projects

Churn, fraud, recommenders and RAG, each with data and a metric.

Open guide →24 ideas