Tutorial projects give you a clean dataset and a list of charts to make. Real work does not. You get a vague problem from someone busy, data that is messier than promised, and a deadline. Live projects are practice for that reality. Here is a method that works on any of them.
Step 1: Read the brief twice
The business requirements document (BRD) or brief tells you who is asking, what decision they need to make, and what they already believe. Read it once quickly, then again slowly with a pen.
Write down:
- The stakeholder and their goal in one sentence.
- The questions they explicitly ask.
- Definitions: what exactly counts as a "failure", a "late delivery", a "churned customer"?
- Constraints: time period, regions, products in scope.
If a definition is unclear, write down the assumption you will use. In a real job you would ask; in a project, a stated assumption shows the same judgement.
Step 2: Turn the problem into specific questions
"Why are payment failures up?" is too broad to analyse directly. Break it into questions data can answer:
- Is the rise real, or did the way failures are recorded change?
- Is it across all banks and apps, or concentrated in a few?
- Did it start on a specific date? What changed around then?
- Which failure reasons grew the most?
- What does it cost the business?
Each question should point to a table, a calculation and a chart.
Step 3: Explore and check the data
Before analysing, understand what you have:
- Row counts, date ranges, columns and their types.
- Missing values, duplicates and impossible values (negative amounts, dates in the future).
- Whether keys actually join between tables.
- Whether totals roughly match any numbers given in the brief.
Keep a short data issues log: what you found, what you did about it, and how many rows it affected. This becomes part of your write-up and is one of the things interviewers most like to hear about.
Step 4: Analyse, one question at a time
Work through your questions in order, with SQL or pandas. For each one, record:
- The query or code.
- The result (a small table or a number).
- One sentence on what it means.
Resist making charts too early. Find the answer first, then choose the chart that shows it best.
Step 5: Look for the "why"
Descriptive numbers ("failures rose in the last month") are the start, not the finish. Dig one level deeper:
- Segment: by region, product, channel, customer type.
- Compare: before vs after, this group vs that group.
- Rates, not counts: a branch with more failures may simply have more transactions.
The best findings are specific: not "failures went up", but "the rise is concentrated in one partner bank's timeouts during evening peak hours, starting after a specific date".
Step 6: Recommend
A finding without a recommendation leaves the manager asking "so what?". For each key finding, suggest an action, who should take it, and how you would measure whether it worked. Be honest about uncertainty; "the data suggests" is fine when it is true.
Step 7: Present it
Structure your output for a busy reader:
- Summary first: three to five findings with recommendations, on one page.
- Supporting analysis: one section per question, each with a chart and a short explanation.
- Data notes: issues found, assumptions made, limitations.
Use titles that state the point ("Evening timeouts drive most of the rise"), not just the topic ("Failures by hour").
Step 8: Turn it into a resume line
When you finish, write the project up in the problem โ work โ tools โ finding format described in our resume projects guide. That one line is often what gets you an interview.
Where to start
Our live projects give you a fictional company, a full brief and realistic datasets in ten industries. Good first picks are UPI payment failures in banking or OPD appointment no-shows in healthcare. If you want more structure first, try the step-by-step guided projects.
Practise what you just read
Reading helps; writing the code yourself is what sticks. Start here: