Company interview guide

Visa Machine Learning Interview Guide (2026)

SQL and Python screens, ML fundamentals, and consulting-style cases built on payments data.

Visa's machine learning interview loop has 9 reported stages, from recruiter screen to behavioral and hiring manager; typical length: 3-8 weeks.

Roles covered: Data Scientist, ML Engineer and Applied Scientist. The question bank holds 4 questions reported from Visa interviews, 1 of them free to solve in your browser.

Try a free question

Interview process length
3-8 weeks
Stages
9
Reports linked to bank questions
4
Bank report dates
Dec 2021–Apr 2026
Questions in bank
4 (1 free)
Last reviewed
October 2026

How the Visa interview loop works

Data scientist track: recruiter screen, a live technical interview with one or more SQL questions and a short Python task, then a superday of about four virtual interviews (business case, ML or statistics, behavioral, sometimes a second case) in one day. ML engineer and scientist track: recruiter screen, a CodeSignal assessment, then two or three rounds that pair ML concept questions and a project deep dive with a hiring manager conversation; some senior loops add algorithm coding and API or system design. Outside the US, some data scientist teams use a take-home case presented to the team. Several candidates describe long silences after the final round, so follow up with your recruiter.

The Visa interview process, stage by stage

01Recruiter screen

15-30 min phone or video call; some recruiters send written questions first

A screening call about your background, relocation, pay expectations and how many years of SQL and Python you have. Some recruiters read from a fixed list and, for ML scientist roles, may include a quick ML concept question.

What it tests

  • Fit with the role and why Visa
  • Years of hands-on SQL, Python and ML work
  • Logistics: location, relocation, salary range
  • Occasionally one ML definition, such as bagging versus boosting

How to prepare

  • Have a short summary of one ML project ready; some recruiters pick a project and ask you to walk through it.
  • Ask which rounds follow and whether a preparation document exists; several candidates received one before the superday.
  • Research Visa as a payments network, not a card issuer.

A recruiter or HR call opens the loop in 24 candidate reports across data science and ML roles, 11 of them since 2024; a few teams sent a coding test first.

02Online assessment

Timed CodeSignal or HackerRank test, about 3-4 tasks or 1 hour; ML variants mix multiple choice with ML coding

Standard for ML engineer and ML scientist roles and for many data scientist roles outside the US. Variants range from four algorithm questions of medium difficulty to an ML framework test with multiple-choice questions, a computation problem, two ML coding tasks and one algorithm task, or data science tasks in pandas and scikit-learn.

What it tests

  • Medium-level algorithm and data structure problems
  • Data preparation and model fitting with pandas and scikit-learn
  • ML coding plus short multiple-choice ML questions
  • Speed under a fixed time limit

How to prepare

  • Practise data cleaning, a train/test split and a baseline model in pandas and scikit-learn without looking up the API.
  • Expect the score to decide whether you move on; one candidate was told the result right away.
  • Practise in a browser IDE with a timer running.

An online coding test appears in 12 candidate reports, 10 since 2024, mostly ML engineer or scientist roles and data scientist roles in Europe and India; US data scientists usually get a live screen instead.

03Live technical screen: SQL and Python

About 45 min video call, coding in a shared editor such as CodeSignal while the interviewer watches

The core filter on the data scientist track. Expect one or more SQL questions, often with window functions or aggregates, and a basic Python or pandas task; some interviewers add ML concept questions or a short business case.

What it tests

  • SQL window functions, ranking, aggregates and joins
  • Computing medians or quartiles without built-in functions
  • Basic Python or pandas: group-by, simple algorithms such as Fibonacci
  • Explaining your logic as you write it
  • Short ML or business follow-ups

How to prepare

  • Drill multi-step SQL with window functions, such as consecutive-year growth or top-N per group.
  • Talk through the query before running it; several candidates were asked to explain their logic.
  • Keep Python solutions plain and readable; the bar is fluency, not tricks.

A live SQL and Python round after the recruiter call appears in 14 data scientist reports, 8 since 2024; contents are consistent, with SQL window functions the most named topic.

04Take-home data case and presentation (some teams)

Dataset given to work on over a few days or a weekend, then presented to the team

Some data scientist teams, mostly outside the US, replace or follow the live screen with an open-ended dataset. You build an analysis or model and then present it to a tech lead or a client-facing panel, who probe your choices.

What it tests

  • Turning an open dataset into a clear business finding
  • Sound model evaluation choices
  • Presenting results to technical and commercial listeners

How to prepare

  • Lead with the business question and the answer, then the method.
  • Prepare to defend the evaluation metric you chose for a classifier.

Three data scientist reports, from Australia and New Zealand, describe a take-home dataset followed by a discussion or presentation; it is not part of most US loops.

05ML fundamentals and project deep dive

Usually 45 min per round, often two rounds back to back; rapid-fire questions plus a walk-through of one project

The main technical signal for ML engineer and scientist roles, and part of many data scientist screens. Interviewers move quickly across classical ML, deep learning and newer GenAI topics, then ask you to explain a past project end to end in detail.

What it tests

  • Bagging versus boosting, random forests, bias and variance
  • Classification metrics, including choices for fraud detection
  • CNNs, RNNs, transformers and autoencoder losses
  • RAG: how it works, its benefits and limitations, metadata filtering
  • Project lifecycle: data, modelling, deployment and drift monitoring

How to prepare

  • Answer with trade-offs, not one-line definitions; the questions come fast but follow-ups probe why.
  • Pick one project you can explain down to every detail, including what you would change.
  • Have a fraud-detection metric answer ready: why precision-recall over accuracy under heavy class imbalance.

Free reported questions for this stage

ML concept questions or a project deep dive in a technical round appear in 14 reports: 8 from ML engineer or scientist roles (2023 to 2026) and 6 from data scientist screens.

06Algorithm coding and system design (ML engineer track)

About 45 min rounds; live algorithm coding plus an ML system or API design round

Some ML engineer loops, especially senior ones, add classic algorithm rounds and a design round. Reported designs include a store recommender that receives only a user ID and an API design; one candidate got three data structure rounds.

What it tests

  • Easy to hard algorithm and data structure problems
  • Designing a recommender with little input signal
  • API design for an ML service
  • Defending design choices under challenging follow-ups

How to prepare

  • Practise medium algorithm problems in Python while explaining complexity.
  • For design, cover cold start, candidate generation, ranking and latency for a payments-scale system.

Four ML engineer or scientist reports from 2024 to 2026 mention live algorithm coding, and two of them a system or API design round; other ML loops had no such rounds.

07Superday: business and consulting case

Part of a superday of about four virtual interviews in one day; one or two case rounds, about 45 min where reported

The signature data scientist round. Interviewers, often from Visa's consulting and analytics side, pose a client problem that you solve with hypothetical Visa transaction data, such as advising a retailer where to open a store or inferring where a cardholder lives.

What it tests

  • Structuring an open business problem for a bank or merchant client
  • Choosing metrics and data from card transactions
  • Business acumen and the so-what of an analysis
  • Communicating clearly to non-technical stakeholders

How to prepare

  • Think about who the client is (an issuer, a merchant, a government) and what they need before proposing analysis.
  • Practise transaction-data cases: home location from spend patterns, store siting, customer segmentation.
  • Use the preparation material the recruiter sends; candidates found it close to the real questions.

Business or consulting case rounds appear in 9 data scientist reports, 4 since 2024; superday panels consistently include at least one case, and some include two.

08Superday: ML and statistics

One round within the superday, usually case-based

A technical round inside the superday that frames ML and statistics as a case: how you would process data and model a problem, plus hypothesis testing or experiment design. Conceptual curveballs, such as what happens to a regression fit when every row is duplicated, also show up.

What it tests

  • Data processing and model choice for a business case
  • Hypothesis testing and experiment design
  • Regression assumptions and how data changes affect estimates
  • Supervised and unsupervised methods you have used

How to prepare

  • Know what duplicating data does to coefficients, standard errors and p-values.
  • Practise designing a test for a payments product change, including the metric and sample size.

An ML, statistics or technical round inside the superday appears in 6 data scientist reports, 4 since 2024; two of them name hypothesis testing or experiment design.

09Behavioral and hiring manager

15-30 min where reported; a superday round for data scientists, or a short hiring manager slot after the technical rounds

A behavioral round or hiring manager conversation appears on every track. Expect classic questions on teamwork and challenges, but many hiring managers turn it into a technical review of your experience or of the project they want you to take on.

What it tests

  • Teamwork, conflict and overcoming a challenge
  • Depth of your own experience and past impact
  • How you would approach the team's planned project
  • Alignment with Visa's leadership principles

How to prepare

  • Prepare STAR stories mapped to Visa's published leadership principles; one recruiter sent them in advance.
  • Expect the hiring manager to go technical even when the round is billed as behavioral.
  • Ask about the team's roadmap; one hiring manager spent most of the round on a future project.

A behavioral or hiring manager round appears in 14 reports across data science and ML roles, 9 since 2024; several say the hiring manager round was more technical than billed.

Questions and themes reported most often

  • Walk through one past project end to end in detail (9 reports, ml fundamentals and project deep dive)
  • SQL with window functions, aggregates or top-N, written live (5 reports, live technical screen: sql and python)
  • Basic Python or pandas task: Fibonacci, group-by, middle element (5 reports, live technical screen: sql and python)
  • Bagging versus boosting and random forest theory or metrics (4 reports, ml fundamentals and project deep dive)
  • How to evaluate a classification model, including for fraud (4 reports, ml fundamentals and project deep dive)
  • Deep learning basics: transformers, CNNs, RNNs, autoencoder loss (4 reports, ml fundamentals and project deep dive)
  • Consulting case on Visa transaction data: store siting or where a cardholder lives (3 reports, superday: business and consulting case)
  • Hypothesis testing and experiment design (3 reports, superday: ml and statistics)

What Visa asks about

  • SQL & Python coding25%
  • ML & DL concepts24%
  • Business & consulting case18%
  • Behavioral & project deep dive16%
  • Stats & experimentation8%
  • Algorithms & system design9%

Visa interview prep checklist

  • Solved five multi-step SQL problems with window functions in under 15 minutes each
  • Completed a timed pandas and scikit-learn task from raw data to a scored baseline
  • Can explain bagging versus boosting, bias and variance, and fraud metrics with trade-offs
  • Practised two consulting cases built on card transaction data
  • Can explain what duplicating every row does to a regression
  • Rehearsed an end-to-end walk-through of one project, including deployment and drift
  • Wrote STAR stories that map to Visa's leadership principles

Tips

  • Find out your track early: data scientist loops are SQL and case heavy, ML engineer loops start with a CodeSignal test.
  • Make SQL window functions automatic; they are the most named topic in data scientist screens.
  • Prepare two or three payments cases out loud: merchant siting, home location from spend, fraud metrics.
  • Review classical ML theory in depth (ensembles, bias and variance, metrics), since rapid-fire concept rounds are common.
  • Know one project cold, from data sourcing to monitoring, because nearly every track includes a deep dive.
  • For ML roles, add GenAI basics: RAG trade-offs, retrieval filtering and transformer internals.
  • Map behavioral stories to Visa's leadership principles and expect hiring managers to probe technical detail.
  • Plan for slow feedback after the superday and keep other processes moving.

See the rest of the Visa loop

The later stages, the questions reported for them and how to prepare are in Pro.

Unlock with Pro

Practice the Visa loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Visa interview FAQ

How many interview rounds does Visa have for ML roles?

Candidates report 9 stages: Recruiter screen, Online assessment, Live technical screen: SQL and Python, Take-home data case and presentation (some teams), ML fundamentals and project deep dive, Algorithm coding and system design (ML engineer track), Superday: business and consulting case, Superday: ML and statistics and Behavioral and hiring manager. Data scientist track: recruiter screen, a live technical interview with one or more SQL questions and a short Python task, then a superday of about four virtual interviews (business case, ML or statistics, behavioral, sometimes a second case) in one day. ML engineer and scientist track: recruiter screen, a CodeSignal assessment, then two or three rounds that pair ML concept questions and a project deep dive with a hiring manager conversation; some senior loops add algorithm coding and API or system design. Outside the US, some data scientist teams use a take-home case presented to the team. Several candidates describe long silences after the final round, so follow up with your recruiter.

How long does the Visa ML interview process take?

Reported timelines run 3-8 weeks.

What does Visa ask in ML interviews?

By reported question mix: SQL & Python coding 25%; ML & DL concepts 24%; Business & consulting case 18%; Behavioral & project deep dive 16%; Stats & experimentation 8%; Algorithms & system design 9%. The most repeated themes: Walk through one past project end to end in detail; SQL with window functions, aggregates or top-N, written live; Basic Python or pandas task: Fibonacci, group-by, middle element.

How should I prepare for the Visa ML interview?

Find out your track early: data scientist loops are SQL and case heavy, ML engineer loops start with a CodeSignal test. Make SQL window functions automatic; they are the most named topic in data scientist screens. Prepare two or three payments cases out loud: merchant siting, home location from spend, fraud metrics.

Are there free Visa interview questions?

Yes. LeetML has 4 questions reported from Visa interviews, and 1 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources