Data Scientist Interview Questions in Lagos (2026)

Data scientist interviews in Lagos test four layers: SQL fluency (the daily reality), statistics and ML fundamentals (the foundation), a business case framed in Nigerian data conditions (the differentiator), and the communication round where your take-home write-up matters as much as your model. With Nigerian data scientists averaging around ₦8M/year (BusinessDay) and fintech driving demand, here are the 15 questions that decide Lagos loops — with answer frameworks.

TL;DR:

• SQL is tested mercilessly; notebooks impress nobody who can't query.

• Local cases (fraud, credit scoring, messy data) separate prepared candidates from imported answers.

• The write-up is the silent half of every take-home grade.

How Lagos data loops are structured

Typical sequence: recruiter screen → SQL/Python technical → take-home (48–72 hours, usually a messy real-ish dataset) → presentation and defence → stakeholder/culture round. Banks add statistics depth and governance questions; fintechs add fraud/credit scenarios; internationals add live coding. The bands you're interviewing for are mapped in our data scientist salary guide — know them before the final round asks your expectations.

SQL and data handling (questions 1–5)

1. "Write a query: top 5 customers by transaction value last month, excluding reversed transactions." Joins, filters, aggregation, ranking — the bread-and-butter screen. Practise window functions (ROW_NUMBER, LAG) until automatic; Lagos interviews love them.

2. "This table has duplicate transactions from a double-webhook bug. Find and remove them safely." Very Nigerian-fintech: dedupe logic with idempotency awareness. Discuss identifying true duplicates (same reference, amount, window) versus legitimate repeats.

3. "How would you handle 30% missing values in the income field?" No single right answer — they want reasoning: missingness mechanism, imputation options and their bias trade-offs, or modelling missingness as signal (in Nigerian credit data, missing income is often informative).

4. "Explain a time you found a data quality issue that changed a conclusion." Have a real story: the metric that doubled because of a logging bug, the duplicated rows that flattered retention. Data scepticism is a hireable trait here.

5. "Pandas or SQL for this task — and why?" Judgment question: volume, repeatability, where the data lives. "SQL for anything the warehouse can do; Python for what it can't" is a safe spine.

Statistics and ML fundamentals (questions 6–10)

6. "Explain precision vs recall to a non-technical manager — using fraud detection." The Lagos framing writes itself: precision = of the transactions we blocked, how many were truly fraud (false blocks anger customers); recall = of all fraud, how much we caught (misses cost money). Pick the trade-off based on which error is costlier.

7. "Your model shows 94% accuracy on fraud data. Why am I unimpressed?" Class imbalance — if 1% of transactions are fraud, predicting 'never fraud' scores 99%. Discuss precision-recall curves, F1, and cost-weighted evaluation.

8. "Explain overfitting and how you'd detect it with limited data." Train/validation gaps, cross-validation on small Nigerian datasets, regularisation, and the honesty to prefer simpler models when data is thin.

9. "When would you NOT use machine learning?" Senior signal: rules beat models when patterns are simple and explainability is regulatory (CBN cares), when data is too scarce, or when a SQL threshold solves it. Knowing when not to model is the difference between a scientist and a notebook operator.

10. "Explain how you'd validate a credit scoring model before deployment." Out-of-time validation, population stability, fairness across segments, and a backtest narrative — borrowed straight from what Lagos lenders actually do.

The business case (questions 11–13)

11. "Design a churn early-warning system for a Nigerian neobank." Define churn first (dormancy ≠ closure here — wallets sleep and wake), pick observable signals (transaction frequency decay, airtime patterns, login gaps), propose intervention windows, and name the metric the bank actually moves (reactivation rate, not AUC).

12. "Our agent network's float requests spike unpredictably. How would you forecast them?" Local texture matters: payday cycles, market days, fuel scarcity shocks, regional events. A candidate who mentions naira cash-cycle seasonality instantly outranks one reciting ARIMA hyperparameters.

13. "You have three months and one analyst. Which data project do you pick for maximum business value?" Prioritisation under constraint — pick something measurable, near revenue or risk, shippable in weeks (the data career-path guide explains why deployed beats elaborate).

Communication and behavioural (questions 14–15)

14. "Walk us through your take-home." Structure the defence: business question → data issues found (there are always planted ones) → approach and trade-offs → result → what you'd do with more time. Defending a simple, honest model beats overselling a complex one — interviewers probe the parts you understand least.

15. "Tell me about yourself" + salary expectations. The PPF framework answer in 90 seconds, and a researched band (₦6M–₦10M mid-level; see the salary guide) delivered with the negotiation scripts' calm.

"Lagos data interviews have matured fast — five years ago a Kaggle medal impressed; today hiring managers ask what you shipped and what it changed. The candidates who win talk about decisions and money, with the model as supporting cast," says Abraham Iyiola, Founder of CareerBuddy.

The take-home: where most candidates lose

1. Read the brief twice; answer the question asked. The single biggest failure is solving an adjacent, more interesting problem.

2. Spend 30% of your time on the write-up: one page, recommendation first, limitations stated honestly. Reviewers read prose before notebooks.

3. Surface the planted data issues explicitly — duplicates, leakage, impossible dates. Finding them is half the assessment.

4. Ship something that runs. A clean, reproducible notebook with a requirements file beats brilliance that errors on the reviewer's machine.

5. Timebox to the stated hours — a 40-hour submission to a 6-hour brief signals you can't scope (the PM loops grade this identically).

A model case answer, dissected

Question 11 — the neobank churn system — is where Lagos candidates most visibly separate. Here's the top-decile shape, compressed:

"First, definition: in a Nigerian neobank, 'churn' can't mean account closure — almost nobody closes accounts; they just go quiet. I'd define it behaviourally: no debit transaction in 45 days for a previously active user, with 'active' itself defined from the data. Second, signals: transaction-frequency decay rate, salary-inflow stoppage (job change or switched primary bank — very common here), airtime purchase gaps, and login-without-transaction patterns which often precede exit. Third, the model: start with a simple gradient-boosted classifier on those features — but honestly, the first version could be three SQL rules, shipped in a week, while the model trains. Fourth, intervention: scores mean nothing without action — route high-risk users to win-back nudges and measure reactivation rate, not model accuracy. I'd A/B the nudges where volume allows, and where it doesn't, use staged rollouts. Biggest risk: confusing dormancy with multi-banking — Nigerians keep four apps; sleeping here may mean thriving elsewhere, so I'd want device and inflow signals to separate the two."

Count the moves: a locally-correct definition, observable signals with Nigerian texture, a ship-fast bias, a business metric, an experiment plan sized to reality, and a named confound. No exotic architecture, no buzzwords — just evidence of having thought about Nigerian users with data. That's the entire bar, and it's reachable because you live inside the dataset these companies are modelling. The imported answer talks about churn like a Netflix subscription; yours shouldn't.

FAQ

Do Lagos data science interviews include live coding?

International employers and some fintechs, yes — usually SQL or Pandas exercises rather than LeetCode. Local companies lean on take-homes. Ask the recruiter "what does the technical assessment look like?" — it's a normal question.

How much depth in deep learning do I need?

For most Lagos roles: conceptual fluency, not implementation depth — tabular ML (gradient boosting, regressions) dominates banking and fintech work. Exceptions: computer-vision roles (KYC document processing) and the growing LLM-application niche.

What portfolio projects impress Lagos hiring managers?

Anything with Nigerian texture and a business number: a credit-default model on public data, a fraud-pattern analysis, an NGN-inflation basket tracker. One deployed dashboard with real users beats five Kaggle notebooks — the entry-level section of our salary guide expands on this.

Are statistics questions really that common?

At banks and serious fintechs, yes — expect hypothesis testing, distributions and experiment-design basics. The bar is explaining clearly, not deriving proofs.

How long do these processes take?

Three to six weeks typically; take-home turnarounds are the variable. Keep parallel applications running via the platforms guide rather than serialising your hopes.

What salary should I quote if asked first?

Deflect once, then quote the band from our data scientist salary guide for your level — and remember the remote market (also in that guide) is your true reservation price.

Should I mention AI tools like ChatGPT in my workflow?

Honestly and specifically, yes — "I use LLMs for boilerplate and documentation, never for analysis I can't verify" reads as current; pretending you don't use them reads as either dishonest or behind. Some take-homes now explicitly permit them; follow the brief's rules exactly, because process honesty is part of the grade.

Loop scheduled? Drill the fundamentals here, benchmark with the data salary guide, and find data roles with details upfront on jobs.thecareerbuddy.com.

Written by the CareerBuddy recruiting team — we run these loops with Lagos hiring managers weekly. Reviewed June 2026.

Featured image: Photo by Walls.io on Unsplash.