🏦
Meet Arjun
From non-tech to data analyst in Hyderabad

Arjun switched careers last year and now works nights on side projects while his family sleeps. He just landed a data role at SouthBank India in Hyderabad. On his first Monday, his manager hands him a CSV file with 8 days of branch transactions and asks: "Which branch is our star performer?"

Arjun opens the CSV. He sees 48 rows of raw data—dates, amounts, branch names. To answer the question, he needs to go from this messy table to a clear insight. This journey—from raw data to answer—is what data analysis is.

Data analysis in one picture

Here's what Arjun's workflow looks like: messy rows on the left, insights on the right.

Before analysis
Raw CSV data
date branch amount
2026-04-01 Hyderabad ₹12,400
2026-04-01 Bangalore ₹8,900
2026-04-02 Chennai ₹15,200
2026-04-02 Hyderabad ₹18,600
2026-04-03 Mumbai ₹9,800
2026-04-03 Hyderabad ₹22,100
2026-04-04 Delhi ₹11,500
2026-04-04 Hyderabad ₹16,800
After analysis
3 key insights
Top branch
Hyderabad → 42% of volume
Busiest day
Friday → 2.3× higher volume
Avg transaction
₹12,400 per txn

This is data analysis: turning the left side into the right side. No coding yet—just the idea.

The 5-step analyst workflow

Every analysis follows this path. Let's walk through each step with Arjun's branch question.

1
Ask a question
What do you want to know? "Which branch is our top performer?"
2
Load data
Find the CSV, JSON, or database. Read it into memory.
3
Clean
Fix typos, remove duplicates, handle missing values.
4
Explore
Group, filter, summarize. Find patterns. Ask follow-up questions.
5
Communicate
Build a chart or summary card. Tell the story your data reveals.

Here's the reality: cleaning (Step 3) takes 80% of your time. You'll spend one day asking questions and four days fighting dirty data. This is normal.

Time breakdown for a typical analysis
Bar chart showing data analysis workflow breakdown: cleaning dominates at 80% of time, ask and load each 10%, explore and communicate each 5%
What this shows: Cleaning dominates every real project. You load data, find typos, handle blanks, rename columns, validate dates. The messy middle is where you spend your effort. Once clean, asking questions and building charts is quick.

Tools analysts use

You don't need many tools. Start with these five.

Tool What it does When to use it
pandas Load, clean, group, filter data Your main workhorse. 80% of your time.
matplotlib / seaborn Draw charts (line, bar, scatter, histogram) When you're ready to visualize.
SQL Query databases (PostgreSQL, MySQL, etc.) When data lives in a database, not a CSV.
Excel / Google Sheets Quick summaries, pivot tables, sharing For non-technical stakeholders. Fast but not scalable.
Jupyter Notebook Interactive coding environment (Python + markdown + charts) Your main editor. Code + notes + results in one place.

Your first mini-analysis: 3 steps

Let's solve Arjun's problem with real code. We'll load the branch data, calculate totals, and draw a chart. Three Python blocks, three outputs.

Step 1: Load and preview the data

First, read the CSV file and see what you're working with.

load_data.py Load CSV → DataFrame
Output
date branch amount 0 2026-04-01 Hyderabad 12400 1 2026-04-01 Bangalore 8900 2 2026-04-02 Chennai 15200 3 2026-04-02 Hyderabad 18600 4 2026-04-03 Mumbai 9800
What happened: We loaded the CSV with pd.read_csv() and peeked at the first 5 rows with .head(). Now we know the columns: date, branch, amount. All numeric—no typos, no blanks. Lucky us.

Step 2: Group by branch and sum

Group all transactions by branch, then sum the amounts. This answers: which branch has the highest total volume?

group_by_branch.py Groupby + sum → branch totals
Output
branch Hyderabad ₹3,42,00,000 Bangalore ₹1,92,00,000 Mumbai ₹1,68,00,000 Chennai ₹98,00,000 Delhi ₹82,00,000
What happened: We grouped the 48 transactions into 5 branch groups, summed amounts within each group, and sorted them descending. Hyderabad's ₹3.42 crore is 42% of the total. Boom—answered the question.

Step 3: Draw a horizontal bar chart

A chart tells the story better than a table. Let's visualize the branch volumes side-by-side.

visualize.py Create bar chart
Transaction volume by SouthBank branch (Apr 1-8, 2026)
Horizontal bar chart of transaction volumes by branch showing Hyderabad leads at 342 crores (39% of total), followed by Bangalore 192, Mumbai 168, Chennai 98, and Delhi 82
What happened: We created a horizontal bar chart with matplotlib. Hyderabad towers above the others at 342 crore (42% of volume). The chart communicates this instantly—no math required. This is Step 5 of the workflow: showing, not telling.

What this track will teach you

Lesson 01 is the foundation. The rest of the 29-lesson track builds on these five steps, with deeper tools and real-world scenarios.

  • Arc 1: Foundations — Learn pandas, Jupyter, the workflow
  • Arc 2: Load — CSV, Excel, JSON, APIs, SQL, Parquet
  • Arc 3: Clean — Nulls, duplicates, types, regex, outliers
  • Arc 4: Explore — Descriptive stats, groupby, pivot, joins, time-series
  • Arc 5: Analyse — Correlation, regression, distributions, cohort, funnel, RFM, A/B tests
  • Arc 6: Visualise & Ship — Matplotlib, Seaborn, Streamlit dashboards, reports

Where you see this

Industries & use cases

🏦
Fintech & Banking
Branch performance, fraud detection, customer spending patterns, transaction volume trends. What Arjun does at SouthBank.
🛍️
E-commerce
Order value by region, top SKUs, customer lifetime value, churn prediction. Jumia Kenya uses this daily.
🏥
Healthcare
Patient outcomes by clinic, vaccination rates, ante-natal care coverage, donor impact reports. NGOs depend on this.

How this connects to AI

Every machine learning model starts as a data analysis project. Before you train a fraud detector or predict churn, you must understand your data—group it, visualize it, find patterns. This lesson is where AI people start. Clean data + clear questions = AI success.

data exploration feature engineering pattern discovery

Mini Challenge

Task: You have the same SouthBank transactions CSV. Calculate how many transactions each branch handled. Which branch processed the most txns (not the highest volume)? Use .groupby().size() to count rows per branch.

Hint: df.groupby('branch').size().sort_values(ascending=False) will count txns per branch and show the highest first.

Interview questions

Data analysis is the process of turning raw data into actionable insights. You load data, clean it, explore patterns, and communicate findings. The journey from messy CSV to clear answer.
Real data is messy. Columns have typos, dates are in 5 different formats, amounts have currency symbols, entire columns are blank. You can't analyze garbage. Cleaning takes 80% of time because it's the hardest part.
Cleaning fixes the data (remove blanks, fix typos, validate dates). Exploring asks questions of clean data (which branch is top? do Fridays have higher volume?). Clean = fixing, Explore = asking.
Use a chart, not numbers. A single horizontal bar showing "Hyderabad = 42% of volume" beats a table with five rows. Visuals communicate faster. Always pair the chart with a 1–2 sentence summary: "Hyderabad is our star branch, handling 42% of all transaction volume."
Use SQL when data is huge (millions of rows) and lives in a database—SQL is optimized for databases. Use pandas when you've already extracted the data to your laptop, or when you need complex calculations Python excels at. Often you'll query with SQL, then explore with pandas.

Test your knowledge

Question 1 of 5
Which step of the analyst workflow typically takes the most time?
Question 2 of 5
What does df.groupby('branch')['amount'].sum() do?
Question 3 of 5
Which is the best way to show "Hyderabad has 42% of volume"?
Question 4 of 5
Data analysis turns ___ into ___.
Question 5 of 5
Which tool is your main workhorse for data analysis?
0
/5
Excellent! You understand the data analysis workflow.

Cheatsheet: Data analysis foundations

Concept What it is Example
5-step workflow Ask → Load → Clean → Explore → Communicate Question: which branch is top? Answer: a bar chart.
Data cleaning Fixing typos, blanks, dates, duplicates Replace "Hyd" with "Hyderabad", drop empty rows
Groupby Group rows by category, then aggregate df.groupby('branch')['amount'].sum()
Data visualization Turn numbers into charts Horizontal bar, line, scatter, histogram
EDA Exploratory Data Analysis—asking questions of data How many txns per branch? Which day is busiest?
Pandas Python library for data manipulation pd.read_csv(), .groupby(), .describe()
Matplotlib/Seaborn Python libraries for charting .plot.barh(), .plot.line(), sns.heatmap()
SQL Query language for databases SELECT branch, SUM(amount) FROM txns GROUP BY branch
Jupyter Notebook Interactive Python editor (code + notes + charts) Write code, see output, add markdown notes
CSV file Plain text table (comma-separated values) transactions.csv with date, branch, amount columns

Common pitfalls

Skipping data cleaning
Garbage in = garbage out. If you don't clean, your insights are wrong. Always inspect the data first: df.info(), df.describe(), df.isnull().sum().
Making charts without understanding the data
Before you visualize, ask: What story am I telling? Is this bar chart or line chart? Who's the audience? Explore first, visualize second.
Forgetting to ask follow-up questions
"Hyderabad is 42% of volume" is one insight. Ask: Why? Is it growing? Do larger cities drive more? One insight leads to three questions.
Using a table when a chart works better
Non-technical people see numbers and glaze over. A single bar showing "Hyderabad 42%" beats a 5-row table. Communicate visually.
Assuming correlation = causation
"Fridays have 2.3× volume" doesn't mean Fridays cause more volume—it's payday. Always question why the pattern exists.
Africa & India's Best AI Academy
🤖 AI-first 🎓 Placement Assistance

Ready to dive in?

Join the Best AI Program: The AI-Era Analyst — 5 months, fully online, with placement assistance. Week 1 is free.

Explore The AI-Era Analyst →