The Data Science Interview

What the Process Looks Like

Data science hiring processes vary, but most, whether at a Lagos fintech or a London scale-up, include some combination of these stages:

1. Recruiter or HR screen (20-30 minutes): background, motivation, salary expectations
2. Technical screen: a SQL and/or Python test, often timed and online (HackerRank, CoderPad or a shared notebook)
3. Take-home project (usually 2 to 5 days): analyse a dataset and present findings
4. Technical interview: statistics, ML concepts, and a walkthrough of your projects or take-home
5. Case study / business interview: an open-ended business problem
6. Behavioural / culture interview: teamwork, conflict, handling ambiguity

Knowing the stages lets you prepare deliberately for each one.

1. The SQL Test

SQL appears in almost every data interview, and it is where many candidates fail. Expect:

• Filtering, aggregation and GROUP BY with HAVING
• JOINs, including spotting when a LEFT JOIN is needed
• Window functions: ROW_NUMBER, RANK, LAG, and running totals
• Date handling: monthly cohorts, week-over-week changes
• CTEs (WITH clauses) for readable multi-step queries

A typical fintech question: "For each customer, find their first transaction date and the total they spent in their first 30 days."

WITH first_txn AS (
    SELECT customer_id, MIN(txn_date) AS first_date
    FROM transactions
    GROUP BY customer_id
)
SELECT f.customer_id,
       f.first_date,
       SUM(t.amount) AS spend_first_30d
FROM first_txn f
JOIN transactions t
  ON t.customer_id = f.customer_id
 AND t.txn_date < f.first_date + INTERVAL '30 days'
GROUP BY f.customer_id, f.first_date;

Another classic: "Find each customer's second-largest transaction." (Use ROW_NUMBER() OVER (PARTITION BY customer_id ORDER BY amount DESC).)

Prepare with: LeetCode (database problems), HackerRank SQL, StrataScratch and DataLemur. Aim to solve 50+ problems, with window functions until they feel natural.

2. Python and pandas

You may be asked to clean a messy CSV, compute metrics with groupby, merge tables, or implement a simple function without libraries (for example, computing a moving average). Talk through your approach as you code. Interviewers care about how you think, not only the final answer.

3. Statistics and ML Concepts

Common questions, and what a good answer covers:

4. The Take-Home Project

Take-homes are your chance to shine, and where many candidates lose points on communication rather than technique.

Do:

• Start with a short summary of findings and recommendations at the top
• Explain your assumptions and any data issues you found
• Establish a baseline before complex models
• Keep the notebook clean: headings, comments, no dead code
• Respect the suggested time; say what you would do with more time

Don't:

• Dump 40 charts with no commentary
• Use a complex model without justifying it
• Ignore data leakage or class imbalance
• Submit without rerunning the notebook from top to bottom

5. The Business Case Study

You get an open-ended prompt such as: "Transaction volume on our USSD channel fell 15% last month. How would you investigate?"

Use a structure:

1. Clarify: How is volume measured? Across all users? Is this seasonal? Did anything launch or change?
2. Break it down: by state, bank, customer segment, transaction type, day and time of day
3. Form hypotheses: a telco network issue, a pricing change, a competitor promotion, users migrating to the app, or a data pipeline bug
4. Say which data would test each hypothesis
5. Recommend next steps and how to measure them

Talking through a clear structure matters more than landing on the "right" answer.

What Nigerian Companies Commonly Ask

Based on the kinds of roles advertised by Nigerian fintechs, banks and telcos, expect a strong focus on:

• SQL, often as a timed live test, because so much of the work lives in transaction databases
• Fintech metrics: transaction success rates, customer acquisition cost, lifetime value, retention cohorts
• Fraud and credit risk scenarios: "How would you build a model to flag suspicious transfers?"
• Excel / Power BI for analyst roles, especially in banks
• Practical impact: "Tell me about a project where your analysis changed a decision"
• Working with imperfect data: missing records, manual processes and inconsistent sources are common in local data environments, so show that you can cope

Global remote roles add more emphasis on written communication, asynchronous take-homes and system-level thinking.

6. Behavioural Questions: Use STAR

Structure answers as Situation, Task, Action, Result:

"In my churn project (S), I needed to prioritise 1,000 subscribers a week (T). I engineered recharge-trend features and compared three models, focusing on precision in the top list (A). The final model found 2.6 times more churners than random targeting (R)."

Prepare five STAR stories: a project you are proud of, a mistake you learned from, a disagreement, a time you explained something technical simply, and a time you handled messy data.

Questions to Ask Them

• "What does the data stack look like, and who owns data quality?"
• "How are data science results used in decisions here?"
• "What would success look like in my first 90 days?"

Good questions show that you are evaluating them too.