What Junior Data Analysts Get Wrong About Talking to Interviewers

SELECT
  c.customer_id,
  COALESCE(SUM(o.amount), 0) AS total_spend,
  CASE
    WHEN COUNT(o.order_id) = 0      THEN 'Never ordered'
    WHEN SUM(o.amount) >= 20000     THEN 'High value'
    WHEN SUM(o.amount) >= 5000      THEN 'Regular'
    ELSE 'Occasional'
  END AS segment
FROM customers c
LEFT JOIN orders o
  ON o.customer_id = c.customer_id
 AND o.status = 'delivered'
GROUP BY c.customer_id;

To a technical interviewer:
"I left-joined delivered orders to customers so that
customers with no orders stay in, then used CASE on total
spend to create four segments."

To an HR or business interviewer:
"I grouped every customer into four groups based on how
much they had spent, including people who signed up but
never bought anything. The 'never ordered' group was
useful because the marketing team could target it with a
first-order offer."

Same query. Different listener. Different words.

Interviewer: What did your review analysis find?

Candidate: So first I downloaded the dataset, it had around
ten columns. Then I loaded it in pandas and checked nulls.
There were some missing ratings, so I dropped them. Then I
converted the date column. Then I made a histogram of
ratings, and a bar chart by category, and then I merged
the delivery data, and then I saw that... late deliveries
had lower ratings.

(The finding arrives at the very end, almost as an aside.)

Interviewer: What did your review analysis find?

Candidate: Late deliveries were the main driver of one-star
reviews.

When I compared on-time and late orders, late ones had a
much lower average rating in every product category, so
it was not just one bad category.

The hardest part was joining the review data to delivery
dates, because the two files used different date formats.

Would you like me to go into the cleaning, or how I
compared the groups?

USEFUL CHECK-IN PHRASES

- "Should I go deeper into the method, or is the result enough?"
- "I can show the query if that helps."
- "Does that answer what you were asking?"
- "There are two parts to this. Shall I start with the data issue?"
- "In short, the key point is ..."

Use one per answer at most. More than that sounds unsure.

Interviewer (HR round): Tell me about your project.

Candidate: My project looked at why some customers stop
buying from an online store. I found that customers whose
first delivery was late were much less likely to order
again.

(Interviewer nods, does not write anything.)

Candidate: In short, it showed the business that fixing
delivery could matter more than offering discounts. I can
go into the technical side if that is useful, but I am
happy to move on.

Interviewer: That's fine, let's move on.

(The candidate kept it short for an HR round and saved
technical details for the technical interviewer.)

TRANSLATION DRILL

Write 10 tool phrases from your projects, e.g.:
- LEFT JOIN on customer_id
- groupby and agg
- DAX measure with CALCULATE
- removed outliers above 99th percentile

For each, write one plain sentence a sales manager
would understand. Then say both versions aloud.

Test: could your parents follow the plain version?

FIRST-LINE DRILL

Take 5 questions about your projects:
1. What did you find?
2. What was hardest?
3. Who would use this?
4. What would you change?
5. Why this dataset?

For each, write ONLY the first sentence.
It must contain the answer, not the method.
Read them aloud. Then build the rest of each answer.

INTERRUPT DRILL (with a partner)

Partner asks a project question.
You answer normally.
At a random point, partner interrupts with:
- "Can you summarise that?"
- "Skip the method. What was the result?"
- "Explain that to someone from HR."

You adjust immediately in one or two sentences.
Swap roles after five questions.