Data Analysis Process: 7 Steps With a Free Checklist

The data analysis process is the repeatable path from a business question to an answer someone can act on: ask, collect, clean, explore, analyse, interpret and communicate. Different books and courses list four, five, six, seven, eight or even ten steps. They are naming the same work with different cuts. In this guide I use seven steps, show them on an illustrative training-institute enquiry dataset, and give you a checklist you can download.

I'm Gulab Chand Tejwani. I teach data analysis, SQL and Power BI at TechNet Consultancy in Vallabh Vidyanagar, Anand, and online. Beginners usually ask me for "the" official steps. There isn't one official list. There is a workflow that stops you jumping straight to charts.

Why do sources disagree on the number of steps?

Search results for "data analysis steps" mix four-phase research models, six-phase analytics frameworks, seven-step teaching outlines and ten-step exam checklists. The labels change; the jobs do not: clarify the question, get data, fix it, examine it, draw a careful conclusion, and share it.

So this post does not invent a separate process for every count. It teaches one 7-step version and maps the common variants onto it.

What you see online	Where it sits in this 7-step flow
4 steps / 4 phases	Ask+Collect, Clean, Analyse, Communicate (Explore and Interpret folded in)
5 key steps	Ask, Collect, Clean, Analyse, Communicate
6 phases	Ask, Collect, Clean, Explore, Analyse, Communicate
7 steps (this post)	Ask, Collect, Clean, Explore, Analyse, Interpret, Communicate
8–10 stages	Same seven, with extra sub-steps for modelling, validation or deployment

The practice dataset (illustrative)

All numbers below come from a synthetic enquiry file built for teaching: 854 rows of training-institute leads with deliberate blanks, duplicate IDs, mixed date formats and currency text such as ₹25,000 or Rs. 12,000. It is labelled illustrative in the README. Download:

• CSV
• Data dictionary
• README

I regenerated and counted the file on 4 November 2026: 300 Enquired, 212 Follow-up, 162 Enrolled, 135 Lost, 45 Duplicate; 46 blank course values; 117 blank sources.

Step 1: Ask

Write one question a decision-maker cares about. Example for this file:

"Which enquiry sources are converting into enrolments, and where are we losing people after the first contact?"

Also write what a useful answer looks like: a short table of source → enquiry count → enrolment count → conversion rate, plus one recommended follow-up.

Step 2: Collect

List the fields you need and where they live. For this exercise the whole answer is in one CSV. In a real institute you might pull the CRM export, a WhatsApp follow-up sheet and a fees spreadsheet, then join them later with Excel, SQL or Power Query.

Collection techniques that show up in beginner questions (surveys, interviews, observation, system exports, web forms, third-party lists) are just ways to feed this step. Pick the method that can actually answer the question; more data is not always better.

Step 3: Clean

Open the dictionary first. Then fix, in order:

1. Trim names and text fields.
2. Parse dates from mixed formats into one calendar type.
3. Parse fees by stripping ₹, Rs., INR, commas and spaces before you sum. Treat markers such as NA, - or a blank as missing — not as zero.
4. Handle blanks in course and source (flag them; don't silently fill).
5. Deduplicate on enquiry_id (the practice file includes intentional duplicates).

On this file, 46 rows have a blank course and 117 have a blank source. Those are not "insights"; they are quality issues that will distort every later percentage if you ignore them.

Step 4: Explore

Before fancy analysis, profile the columns:

• status counts (see above)
• enrolments by course (illustrative counts from the same file on 4 November 2026): Python for Data Analysis 36, Power BI Mastery 34, SQL Fundamentals 33, Data Analysis 28, PMP Exam Prep 26, blank course 5
• how many enrolled rows have a parseable enrolled_amount (all 162 in this export)

Exploration is where you notice that enrolled_amount sometimes appears on non-enrolled rows — another deliberate mess to catch in cleaning rules.

Step 5: Analyse

Answer the question with comparisons. Example: count enquiries and enrolments by source, then compute conversion as enrolments ÷ enquiries for each source. Exclude or separately report blank sources so you don't pretend missing marketing channels are a channel called "(blank)".

You can do this in Excel PivotTables, SQL GROUP BY, Power BI measures or pandas groupby. The process is the same; only the tool changes. Later posts in this series reuse the same CSV for Excel and SQL walkthroughs.

Step 6: Interpret

Say what the numbers mean and what they do not prove. A high conversion rate on Walk-in might mean good counselling on site, or it might mean Walk-in volume is small and noisy. Blank sources make channel rankings incomplete. Duplicate IDs inflate enquiry counts if you don't remove them first.

Write one sentence of finding and one sentence of uncertainty. That habit matters more than any chart style.

Step 7: Communicate

Share a short narrative, one clear table or chart, and a next action ("call back Follow-up leads older than seven days", "fix the form so course and source are required"). If nobody can act on the output, the analysis isn't finished.

Free checklist download

Use this sheet on every project until the seven steps feel automatic.

How this fits with the rest of the series

• Definitions and a tiny worked example: what is data analysis
• Descriptive, diagnostic, predictive and prescriptive views of the same file: types of data analysis
• Tool skills: Excel, SQL, pandas

If you want guided practice with feedback, our data analysis training runs in Vallabh Vidyanagar, Anand, and online.

FAQ

What are the steps of data analysis?

A practical sequence is: ask a decision question, collect the data, clean it, explore it, analyse it, interpret carefully, and communicate a next action.

Why do some articles say 4 steps and others say 7 or 10?

They split the same work differently. Four-step versions fold explore/interpret into analyse; longer lists split modelling or deployment into extra stages.

What is the most important step?

Asking a clear question. Without it, cleaning and charts have nowhere to aim. On messy files, cleaning usually takes the most clock time.

How do I practise the process?

Use a labelled dataset with known errors, like the illustrative institute enquiry CSV linked above, and walk all seven steps end to end before switching tools.