What Separates a ₹4–7 LPA Candidate From the Rest

Interview Insights · Data Analyst Offers

What Separates a ₹4–7 LPA Candidate From the Rest

SECTION 01Framing the business problem before touching the data

The first difference shows up within seconds. When I ask an entry-level candidate 'why are customers leaving?', they usually start with a tool: 'I would make a churn dashboard in Power BI.' A stronger candidate starts with a definition and a decision: 'First I would agree what churn means here, for example no purchase in a set period, and what the team wants to do about it, such as a win-back campaign.'

That framing does two things. It shows you understand that analysis exists to support a decision. And it prevents the most common analytics mistake, which is answering a slightly different question from the one the business asked.

BEFORE ANY ANALYSIS, SAY:

1. Decision:   'What will the team do with the answer?'
2. Definition: 'What exactly counts as ___?'
3. Scope:      'Which period, segment, product?'
4. Baseline:   'Compared with what?'
5. Output:     'Do you want a number, a chart or a
                recommendation?'

Then: 'Here is how I would approach it...'

Interviewer: Why are customers leaving our app?

Candidate: Before pulling data, I want to define
churn. For a monthly subscription, I would say a user
churned if they did not renew. For a free app, maybe no
login for a set number of weeks. Which fits better?

Interviewer: Subscription.

Candidate: Then I would compare renewers and
non-renewers on usage in their last month, plan type
and support tickets, so the team knows where to focus
a retention offer.

The stronger candidate asked one question and then committed to a plan. Framing is not endless clarifying. It is one or two questions that make the rest of the answer precise.

SECTION 02Live SQL fluency: writing, testing and explaining

Almost every analyst candidate lists SQL. The difference is what happens when the interviewer shares a blank editor. Entry-level candidates often write a query in one go, run it and accept the result. Stronger candidates write in small steps, check the row count, look for duplicates from joins and explain their choices aloud as they go.

Fluency does not mean speed or memorising rare functions. It means you are comfortable with joins, aggregation, CTEs and window functions, and you know how to check that the output is correct. That checking habit is what makes an interviewer trust you with real data.

-- Monthly active users (MAU) from an events table
-- Active = at least one event in the month
SELECT DATE_TRUNC('month', event_time) AS month,
       COUNT(DISTINCT user_id)          AS mau
FROM events
WHERE event_time >= '2026-01-01'
GROUP BY 1
ORDER BY 1;

-- Spoken checks:
-- 1. COUNT(DISTINCT) so one user is counted once a month
-- 2. Are test or internal accounts in events? Filter them.
-- 3. Time zone of event_time matches the business?

-- Before trusting a join, compare counts
SELECT COUNT(*) FROM orders;               -- rows in orders

SELECT COUNT(*)
FROM orders o
JOIN payments p ON p.order_id = o.order_id; -- after join

-- If the second count is higher, some orders have
-- several payments (retries, partial payments).
-- Fix: aggregate payments first.
WITH pay AS (
  SELECT order_id, SUM(amount) AS paid
  FROM payments
  GROUP BY order_id
)
SELECT o.order_id, o.total, pay.paid
FROM orders o
LEFT JOIN pay ON pay.order_id = o.order_id;

The join check is one of the clearest separators I know. Candidates who compare counts before and after a join have usually been burned by duplicated revenue at least once, and interviewers know that experience is valuable.

SECTION 03Insight, action and the project that proves both

The last two behaviours are where many good candidates stop short. They produce a correct number and then wait. A stronger candidate adds one sentence of meaning and one sentence of action: 'Renewal is much lower for users who never used the export feature, so I would test an onboarding nudge toward export.' That is the difference between an output and an insight.

Projects should show this too. A dashboard project described as a list of steps sounds like coursework. The same project described as a business question, a finding and a recommendation sounds like work an employer would pay for.

OUTPUT:
'Renewal rate for users who used export is higher
than for users who did not.'

INSIGHT:
'Users who try export in their first weeks seem much
more likely to renew. It may be that export is where
they first get real value.'

ACTION:
'I would test an onboarding prompt that guides new
users to export, and compare renewal against a
control group.'

CAUTION (say it):
'This is correlation. Engaged users may simply do
both. The test would tell us if export causes it.'

TWO-MINUTE PROJECT PITCH

Question:  Which product categories should a store
           discount next season?
Data:      Sales and returns, one year, CSV exports
Problem:   Returns were recorded as negative sales
           rows; I separated them in Power Query
Finding:   Two categories looked strong on sales but
           weak once returns were removed
Action:    Review those before discounting; protect
           margin on the rest
Limit:     No cost data, so margin is estimated
Next:      Add cost data and test the discount plan

The caution line in the first tab matters. Stronger candidates do not overclaim. Saying 'this is correlation, and here is how I would test causation' is one of the clearest signs of analytical maturity an interviewer can hear.

SECTION 04Practice plan to build the four behaviours

These behaviours are habits, and habits grow through repetition. Use these tabs for focused practice in the weeks before your interviews. In Uncodemy mock interviews, these are also the four things trainers score most closely.

FRAMING DRILL (10 minutes a day)

Pick one business question. Write:
 - the decision it supports
 - one definition you must agree
 - the baseline you would compare with
 - the output the stakeholder wants

Questions to rotate:
 - Is our loyalty programme working?
 - Why did delivery complaints rise?
 - Which customers should get a discount?
 - Should we stop selling a product line?

SQL FLUENCY DRILL (20 minutes a day)

Use any sample sales or events database.

[ ] Write the query in CTE steps
[ ] Run each step and check row counts
[ ] Check joins for duplicate rows
[ ] Decide how NULLs are treated
[ ] Explain each step aloud, recording yourself

Task list: MAU by month, top products per category,
repeat customers, revenue by cohort month,
day-over-day change with LAG().

INSIGHT CARDS (5 minutes each)

For any chart or table you make, write three lines:

Output:  what the number or chart shows
Insight: what it probably means for the business
Action:  what you would recommend or test

Add a caution line if the result could be
correlation, seasonality or a data quality issue.

Say all three lines aloud in under 45 seconds.

SECTION 05Habits that lift your interview above entry level

Small changes in how you answer can shift how an interviewer sees your level. These habits make the difference visible.

• Ask about the decision first — Open analytical answers by asking what the team will do with the result, so your analysis aims at the right target.
• Normalise before comparing — Compare rates such as tickets per order, not raw counts, when the base number has also changed.
• Check every join aloud — Mention row counts before and after joins, because duplicated rows are a very common source of wrong totals.
• State one caution — Name a limit of your analysis, such as missing cost data or correlation, which shows judgement rather than weakness.
• Close with a recommendation — Finish answers with a concrete next step, even a small test, so the interviewer sees impact rather than just output.

SECTION 06Interview Q&A — ₹4–7 LPA candidates

SECTION 07Test yourself — ₹4–7 LPA candidates quiz

SECTION 08Frequently asked questions

SECTION 09Related reads