Revolut Data Scientist Interview Questions 2026

TL;DR

Revolut’s Data Scientist interviews in 2026 focus on behavioral alignment, technical execution, and product intuition—not just model accuracy. Candidates fail not because they lack coding skills, but because they misread Revolut’s risk-first culture. The process spans 3 to 4 weeks, includes 5 rounds, and tests real-time decision-making under ambiguity.

Who This Is For

You’re targeting a Data Scientist role at Revolut in 2026, likely at L4–L6 (Senior Data Scientist to Staff), and have already passed the resume screen. You’ve worked with transactional or financial data, but you’re unsure how Revolut’s regulatory constraints shape interview expectations. This isn’t for entry-level candidates or those seeking academic-style ML debates.

What are the most common Revolut data scientist interview questions in 2026?

The most common questions probe judgment, not syntax. In a Q3 2025 debrief, a candidate correctly built a churn model but lost the offer because they ignored KYC triggers in feature selection. The panel concluded: “You optimized for precision, not compliance.” Revolut doesn’t want data scientists who treat fraud or AML as edge cases. They’re core.

Common technical questions include:

  • “How would you detect suspicious transactions in real time with incomplete user history?”
  • “Build a model to flag high-risk transactions—now explain how you’d reduce false positives without increasing exposure.”
  • “Design an A/B test for a new credit limit algorithm when 60% of users have <3 months of activity.”

Behavioral questions are weaponized for cultural fit. “Tell me about a time you pushed back on a product team” isn’t about conflict—it’s about whether you prioritize stability over speed. One candidate lost points for saying they “moved fast” to deploy a segmentation model; the hiring manager noted, “We don’t move fast with money.”

Not “What models do you know?” but “Why didn’t you choose the other model?” is the real question. The difference isn’t technical depth—it’s risk calibration.

In a Q2 2025 interview, a candidate proposed a neural net for transaction categorization. The interviewer shut it down: “This isn’t a Kaggle competition. How will you explain this to compliance?” The follow-up wasn’t about layers or loss functions—it was, “Can you document this for an audit?” Revolut’s data scientists write for regulators as often as they write for engineers.

How is the Revolut data scientist interview structured in 2026?

The process has 5 rounds over 18 to 26 days—shorter than Google, tighter than fintech peers. Round 1 is a 30-minute recruiter screen focusing on tenure, domain exposure (payments, fraud, lending), and location fit (London, Vilnius, or remote).

Round 2 is a take-home: a 72-hour case involving transaction data, missing labels, and a compliance constraint. Candidates must submit code, a 1-page summary, and a risk assessment. In Q1 2026, 60% of submissions were rejected for omitting data lineage—a required section.

Round 3 is a 60-minute technical interview: live SQL and Python on HackerRank-style interface. Queries involve time-series joins, lag features, and handling duplicate events. The coding bar is moderate, but the data interpretation bar is high. A candidate once wrote perfect Pandas code but failed because they treated a negative balance as an outlier, not a potential fraud signal.

Round 4 is the behavioral and product sense round with two senior PMs or DS leads. They ask about trade-offs: “Would you rather reduce false negatives in fraud detection by 5% or improve customer approval rates by 10%?” There is no correct answer—but your justification must reflect Revolut’s risk posture.

Round 5 is the hiring committee (HC) calibration. No interview, but your packet is reviewed. In a November 2025 HC, a candidate with strong technical scores was rejected because their take-home showed “over-reliance on supervised learning without fallback logic.” The verdict: “Unsuitable for edge-case density in live financial systems.”

Not “Can you code?” but “Can you contain failure?” is the structure’s true test. The process rewards defensive thinking.

What technical skills do I need to pass the Revolut DS interview?

You need SQL, Python, and statistical reasoning—not deep learning or NLP. SQL questions involve rolling windows, recursive CTEs for sessionization, and deduplication logic. One 2025 question: “Calculate 7-day rolling transaction volume per user, but exclude test accounts added after fraud spikes in June.” The trap wasn’t the window function—it was identifying test accounts via pattern-matching in user IDs.

Python tests focus on pandas and datetime manipulation. You’ll clean transaction timestamps across time zones, detect sequence gaps, or impute missing merchant categories. In a 2026 mock, candidates were given a dataset where 15% of “completed” transactions had null settlement times. The expected response was to flag them as pending—not impute or drop.

Statistical questions avoid theory. Instead: “If our false positive rate in fraud detection doubles, what’s the impact on customer support volume?” You’re expected to estimate using historical ticket data and churn elasticity. One candidate failed by quoting p-values instead of business impact.

Machine learning expectations are minimal. You should know logistic regression, decision trees, and evaluation metrics—but the emphasis is on operationalization. “How would you monitor model drift in a transaction risk score?” is more common than “Derive the gradient for XGBoost.”

Not “Are you technically strong?” but “Can you operate under production constraints?” is the real benchmark. Revolut’s systems log every transaction decision—your model must be traceable, not just accurate.

How do Revolut’s behavioral questions differ from other tech companies?

Revolut’s behavioral questions are risk-filtered. At Amazon, “Disagree and commit” rewards challenge. At Revolut, it’s a red flag. In a 2025 post-mortem, a candidate said they “overruled compliance to launch a faster onboarding flow.” That ended the process. The debrief note: “This person will break things that can’t be broken.”

Questions like “Tell me about a time you failed” are landmines. Admitting a model error is fine—admitting it caused financial exposure without escalation is fatal. One candidate shared they delayed reporting a data leak for 48 hours to “fix it quietly.” HC rejected them for “lack of procedural integrity.”

Revolut uses a 3-part scoring rubric:

  1. Ownership – Did you act?
  2. Process adherence – Did you follow protocol?
  3. Escalation judgment – Did you know when to raise it?

A strong answer shows you acted within guardrails. A weak one shows initiative at the cost of control.

In a Q4 2025 interview, a candidate described pausing a campaign after noticing abnormal withdrawal patterns—even though it was pre-launch. They documented, alerted fraud, and waited 24 hours. That story scored “exceeds” in all three dimensions.

Not “Did you take charge?” but “Did you stay within bounds?” is the cultural litmus. Revolut hires for caution, not charisma.

How should I prepare for the Revolut data scientist take-home challenge?

Treat the take-home as a production artifact, not an analysis. In 2026, the average completion time is 6.2 hours—but the top 20% spend 40% of that on documentation. Revolut evaluates: code quality (40%), business logic (30%), risk assessment (20%), and presentation (10%).

A common failure is over-modeling. One candidate used a BERT-based classifier to label transaction descriptions. The model worked—but the feedback was, “This cannot be audited.” Simpler, rule-based approaches with fallbacks score higher.

Expected deliverables:

  • Jupyter notebook or script, well-commented
  • 1-page PDF summary: problem, approach, limitations, risks
  • Risk section must include: data gaps, regulatory exposure, false positive cost

In a January 2026 case, data included transactions with missing user risk tiers. Top candidates imputed conservatively (defaulted to high-risk) and noted it as a compliance gap. Bottom candidates used mean imputation and called it “efficient.”

You must version your code. Revolut checks git history. In two 2025 cases, candidates were disqualified for rewriting history to hide brute-force attempts.

Not “How good is your model?” but “How defensible is your process?” is what they grade. The output is secondary to the audit trail.

Preparation Checklist

  • Solve 3 real transaction fraud cases using public datasets (e.g., IEEE-CIS Fraud Detection) with emphasis on false positive cost
  • Practice SQL queries involving time-series gaps, sessionization, and data quality flags
  • Rehearse behavioral answers using the Ownership-Process-Escalation framework
  • Build a simple risk scoring model with documentation for non-technical stakeholders
  • Work through a structured preparation system (the PM Interview Playbook covers financial risk case interviews with real debrief examples)
  • Time yourself on a 72-hour take-home simulation with strict submission rules
  • Review GDPR and PSD2 basics—interviewers drop compliance questions casually

Mistakes to Avoid

  • BAD: Submitting a take-home with no risk assessment section. In Q2 2025, 12 candidates were auto-rejected for this. Revolut assumes you don’t understand their operating model.
  • GOOD: Including a “Regulatory Constraints” subsection, even if not asked. One candidate added a table mapping model decisions to potential audit questions—and got fast-tracked.
  • BAD: Saying “I would A/B test everything” in the product round. In a 2026 interview, a candidate suggested testing a high-risk loan feature. The interviewer replied, “We don’t experiment with solvency.”
  • GOOD: Saying “I’d run a shadow mode test and monitor exposure caps.” Shows you respect financial boundaries.
  • BAD: Using advanced ML without fallback logic. A candidate used LSTM for transaction prediction and failed when data lagged. No fallback was defined.
  • GOOD: Adding rule-based defaults (e.g., “if no history, use country median”), which demonstrates system resilience.

FAQ

What’s the salary range for a Data Scientist at Revolut in 2026?

L4 earns £75K–£90K base in London, L5 £90K–£110K, with 5–10% cash bonus and restricted stock units vesting over 4 years. Compensation is below FAANG but includes liquidity events from secondary sales. The real premium is role scope—Revolut DSs often influence compliance architecture, which isn’t typical at big tech.

Do Revolut data scientists need finance experience?

Not formally, but candidates without exposure to payments, fraud, or credit risk consistently underperform. In 2025, 70% of hires had prior fintech or banking roles. One exception was a data scientist from a ride-sharing company who focused on real-time fraud—not growth analytics. Domain adjacency matters more than title.

Is the interview different for remote candidates?

No. Remote candidates go through the same 5-round process. Time zone flexibility is expected, but asynchronous steps (take-home, coding) are identical. Revolut does not adjust bar for location—L4 in Vilnius is scored against L4 in London. The only difference is offer packaging, not evaluation.

Ready to build a real interview prep system?

Get the full PM Interview Prep System →

The book is also available on Amazon Kindle.

Related Reading