Company interview guide

PayPal Machine Learning Interview Guide (2026)

SQL and Python screens, medium-difficulty algorithm coding, ML fundamentals and fraud-flavoured cases, with loops that vary by team.

PayPal's machine learning interview loop has 8 reported stages, from recruiter call to hiring manager, project deep dive and behavioral; typical length: 3-8 weeks.

Roles covered: Data Scientist, ML Engineer and Applied Scientist. The question bank holds 14 questions reported from PayPal interviews, 1 of them free to solve in your browser.

Try a free question

Interview process length
3-8 weeks
Stages
8
Reports linked to bank questions
15
Bank report dates
May 2021–Oct 2025
Questions in bank
14 (1 free)
Last reviewed
October 2026

How the PayPal interview loop works

A recruiter or talent call, sometimes followed by a hiring manager screen, then three to five interviews held virtually on Microsoft Teams. Data science loops usually run a SQL and Python round, one or two case or experimentation rounds and a hiring manager round; ML engineer loops usually run one or two coding rounds, ML breadth and depth, ML system design and a behavioral round with the manager. Some teams replace the first technical round with a take-home and a review of it. Ask your recruiter which rounds your team uses, since formats differ a lot between teams.

The PayPal interview process, stage by stage

01Recruiter call

30-45 min phone or video call; on some teams a hiring manager call follows straight away

A talent partner reviews your CV, your experience and why you want PayPal, and explains the role. On a few teams the next call is with the hiring manager, and it can include a coding question without warning.

What it tests

  • A clear, specific reason for wanting PayPal and this team
  • How your past work fits the role, especially any fraud, risk or payments experience
  • Logistics: level, location, timeline

How to prepare

  • Ask which rounds follow and whether any call with the hiring manager will include live coding; several candidates were surprised.
  • Ask whether the team uses a take-home, since some teams start with one.
  • Tell the recruiter if you have applied to more than one PayPal role.

A recruiter or HR screen opens the loop in 21 candidate reports (18 since 2024), usually 30-45 minutes; the company's interviewing page and a 2021 company article on DS hiring (20-45 min) also describe it.

02Take-home modelling task (some teams)

Tabular classification or prediction task, often timed at about 4-5 hours, then an interview reviewing it

Teams mostly in Israel and Ireland send a take-home after the screen: clean a customer or sales dataset, build a binary classifier and explain it. The next interview walks through what you did and why you made each choice, often together with a past project.

What it tests

  • Data cleaning and feature preparation on messy tabular data
  • Model choice and validation for a binary target, including class imbalance
  • Justifying each decision, for example why you resampled or which metric you used
  • Communicating results clearly under questioning

How to prepare

  • Write down the reason for every choice as you go; the review interview is mostly why this and not that.
  • Handle imbalance deliberately and be ready to compare resampling with reweighting or threshold tuning.
  • Budget the full time; timed versions of four to five hours are reported.

11 candidate reports, mostly Israel and Ireland data science and ML scientist roles (8 since 2024), describe a take-home; most were classification tasks followed by a review interview.

03SQL and Python screen

45-60 min live coding on HackerRank or a shared screen; usually 2-3 SQL plus 1-3 Python questions, easy to medium

The standard first technical round for data scientists and common for ML engineers too. SQL runs on transaction, customer and merchant tables; Python is either pandas cleaning or a small algorithm, and some interviewers ban pandas and numpy. One 2026 team let candidates use an AI assistant and watched how they used it.

What it tests

  • Joins, subqueries, grouping and window functions on transaction-style tables
  • Cleaning messy fields: dates stored as strings, null or padded ids
  • Bucketing customers by spend or percentile
  • Pandas cleaning and weighted aggregates, or plain Python without libraries
  • Explaining your logic as you write

How to prepare

  • Interviewers care more about correct logic than exact syntax; talk through edge cases before you run anything.
  • Practise window functions until rolling averages, ranks and running totals are automatic.
  • Check which SQL dialect the platform uses before you start.
  • Be ready to write a small Python function without pandas if asked.

Live SQL, often with Python, appears in 26 candidate reports (21 since 2024), mostly data science roles; several name HackerRank. A 2021 company article on DS hiring lists a programming and data-handling round.

04Algorithm coding

45-60 min live, usually one medium-difficulty algorithm problem, any language; ML engineer loops can have 1-3 such rounds

ML engineer loops nearly always include data-structures coding, and some data science screens add one algorithm question. Most problems are standard mediums on arrays, two pointers, trees and grids, but a few reports describe a hard problem in a short slot.

What it tests

  • Arrays and two pointers, including sliding windows
  • Trees, graphs and grid search with DFS
  • Binary search, including on two sorted arrays
  • Dynamic programming such as coin change
  • Clean code and complexity analysis

How to prepare

  • Drill sliding window, DFS on grids and binary search; these cover most of the reported questions.
  • Expect a resume chat to eat into the slot; aim to code a working solution in 20-25 minutes.
  • State the complexity and test a small case before the interviewer asks.

Free reported questions for this stage

Algorithm coding appears in 19 candidate reports (15 since 2024), nearly all ML engineer and ML scientist loops; difficulty is mostly medium with two reports of a hard binary search.

05ML and statistics concepts

Mixed into the technical rounds or a dedicated breadth and depth round; occasionally an ML coding exercise

Rapid questions across classical ML, statistics and some deep learning, then follow-ups on why. Tree ensembles and boosting come up most often; newer loops add LLM fine-tuning and GenAI questions. A few ML engineer loops ask you to code logistic regression or reason through TF-IDF or a transformer block.

What it tests

  • Bagging vs boosting, XGBoost vs LightGBM, tree splits
  • Metrics and the confusion matrix for imbalanced classes
  • Hypothesis tests, p-values, multicollinearity and probability
  • RNN, LSTM, CNN basics and, more recently, LLM fine-tuning
  • Implementing a simple model such as logistic regression from scratch

How to prepare

  • Be able to explain gradient boosting step by step and compare its main libraries.
  • Prepare fraud-shaped answers: rare positives, delayed labels, cost of false positives.
  • Write logistic regression with its loss and gradient in numpy once before the loop.

ML or statistics questions appear in 25 candidate reports (16 since 2024); only 3 describe a hands-on ML coding exercise. A 2021 company article on DS hiring lists a data science fundamentals and algorithms round.

06ML system design

About 60 min open discussion, mainly in ML engineer loops

Design an end-to-end ML system, usually in payments risk: real-time fraud scoring, spending limits or a production pipeline that handles drift. One 2025 loop ran separate rounds on a reinforcement learning spending-limit policy, a classic fraud model and whether LLMs can help in fraud detection.

What it tests

  • Framing fraud as a real-time approve, review or decline decision
  • Features, labels and sampling when fraud labels are rare and arrive late
  • Serving latency, monitoring and handling drift or model decay
  • Trade-offs between classic models and newer approaches such as RL or LLMs

How to prepare

  • Prepare one full real-time fraud design, from features and labels to thresholds and monitoring.
  • Explain how you would pick negatives and positives and how label delay affects training and evaluation.
  • Have a balanced view on where LLMs help in risk and where classic models still win.

8 candidate reports since 2024 describe an ML system design round, 7 of them ML engineer loops; fraud or risk prompts are the norm where the prompt is named.

07Business, risk and experimentation case

Data science loops: 1-2 rounds, sometimes split into product case and A/B testing

Data scientists get a business problem tied to the hiring team: whether to launch payments in a new country, product metrics for a feature, an A/B test design or a credit and fraud scenario. Risk teams ask you to spot red flags in transactions or find the root cause of a policy problem.

What it tests

  • Structuring a vague business question and choosing success metrics
  • Designing an A/B test and reading its results
  • Risk reasoning: fraud red flags, credit policy, loss and approval trade-offs
  • Domain knowledge of payments and the hiring team's product

How to prepare

  • Learn the basics of payments risk: chargebacks, authorization rates, credit losses.
  • Ask clarifying questions early; several candidates found the prompts vague.
  • Tie every metric to a decision the team would actually make.

18 candidate reports (12 since 2024), almost all data science roles, describe a case, product or A/B round; risk teams favour fraud and credit cases. A 2021 company article on DS hiring also lists a business case round.

08Hiring manager, project deep dive and behavioral

1-2 conversations with the manager, a senior leader or director; sometimes a project presentation

A deep dive into your most relevant project plus behavioral questions on stakeholders and teamwork. Some loops end with a senior manager or director round, and one ML engineer loop asked for a project presentation. PayPal asks candidates to prepare against its values and Leadership Principles.

What it tests

  • Your own contribution, decisions and impact on a past project
  • Working with non-technical stakeholders
  • Motivation for PayPal and the specific team
  • Fit with the hiring manager's style and the team's domain

How to prepare

  • Pick one project and rehearse the reason for every decision; interviewers push on why.
  • Read PayPal's values and Leadership Principles and map one story to each.
  • Give specific examples from more than one job, as the company's own guidance suggests.
  • Follow up politely after the final round; slow replies are common.

A hiring manager, project or behavioral round appears in 31 candidate reports (22 since 2024); the company's interviewing page describes competency-based team interviews and a company article a business team fit screen.

Questions and themes reported most often

  • Live SQL on transaction and customer tables: joins, windows, aggregates (26 reports, sql and python screen)
  • Walk through your most relevant project and defend each decision (18 reports, hiring manager, project deep dive and behavioral)
  • Tabular classification take-home, then a review of your choices (8 reports, take-home modelling task (some teams))
  • Boosting and tree ensembles: bagging vs boosting, XGBoost vs LightGBM (5 reports, ml and statistics concepts)
  • Product or market case, such as launching payments in a new country (5 reports, business, risk and experimentation case)
  • Medium arrays and grids: sliding window, product except self, word search (4 reports, algorithm coding)
  • Fraud and credit scenario: red flags, credit policy, risk KPIs (4 reports, business, risk and experimentation case)
  • A/B test design and hypothesis testing (4 reports, business, risk and experimentation case)

What PayPal asks about

  • SQL & Python data coding22%
  • ML & statistics concepts20%
  • Algorithm coding18%
  • Business, risk & experimentation cases17%
  • Behavioral & project deep dive15%
  • ML system design8%

PayPal interview prep checklist

  • Solved three SQL questions with joins and window functions on transaction tables in under an hour
  • Can solve a medium array or grid problem in about 25 minutes
  • Can explain gradient boosting and compare XGBoost with LightGBM
  • Sketched a real-time fraud model with labels, features, thresholds and drift monitoring
  • Practised one business or risk case and one A/B test design out loud
  • Prepared a project deep dive with the reasoning behind each decision
  • Mapped behavioral stories to PayPal's values and Leadership Principles

Tips

  • Ask the recruiter for the exact rounds; teams differ on take-homes, coding and case rounds.
  • For data science roles, make live SQL with window functions your first priority; it is the most reported round.
  • For ML engineer roles, split prep between medium-difficulty algorithm coding and one strong fraud-detection system design.
  • Learn the payments risk vocabulary: chargebacks, delayed labels, approval rate versus fraud loss.
  • Know tree ensembles in depth; boosting questions recur across every track.
  • Treat any call with a hiring manager as possibly technical.
  • Prepare one project story in depth and expect repeated why questions.
  • Keep other processes moving; many candidates waited weeks for a decision.

See the rest of the PayPal loop

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

Unlock with Pro

Practice the PayPal loop

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

PayPal interview FAQ

How many interview rounds does PayPal have for ML roles?

Candidates report 8 stages: Recruiter call, Take-home modelling task (some teams), SQL and Python screen, Algorithm coding, ML and statistics concepts, ML system design, Business, risk and experimentation case and Hiring manager, project deep dive and behavioral. A recruiter or talent call, sometimes followed by a hiring manager screen, then three to five interviews held virtually on Microsoft Teams. Data science loops usually run a SQL and Python round, one or two case or experimentation rounds and a hiring manager round; ML engineer loops usually run one or two coding rounds, ML breadth and depth, ML system design and a behavioral round with the manager. Some teams replace the first technical round with a take-home and a review of it. Ask your recruiter which rounds your team uses, since formats differ a lot between teams.

How long does the PayPal ML interview process take?

Reported timelines run 3-8 weeks (3-8 weeks, sometimes longer).

What does PayPal ask in ML interviews?

By reported question mix: SQL & Python data coding 22%; ML & statistics concepts 20%; Algorithm coding 18%; Business, risk & experimentation cases 17%; Behavioral & project deep dive 15%; ML system design 8%. The most repeated themes: Live SQL on transaction and customer tables: joins, windows, aggregates; Walk through your most relevant project and defend each decision; Tabular classification take-home, then a review of your choices.

How should I prepare for the PayPal ML interview?

Ask the recruiter for the exact rounds; teams differ on take-homes, coding and case rounds. For data science roles, make live SQL with window functions your first priority; it is the most reported round. For ML engineer roles, split prep between medium-difficulty algorithm coding and one strong fraud-detection system design.

Are there free PayPal interview questions?

Yes. LeetML has 14 questions reported from PayPal interviews, and 1 of them are free to solve in the browser; they are listed by stage above.

Related interview guides

Official sources