Company interview guide

Affirm Machine Learning Interview Guide (2026)

Practical, job-shaped ML interviews: EDA, feature engineering, metrics for credit and fraud models, plus medium-level coding.

Affirm's machine learning interview loop has 6 reported stages (Recruiter call, Hiring-manager chat, Technical screen, Final rounds: practical coding, Final rounds: practical ML, Behavioral & hiring-manager wrap-up); typical length: Not well reported; 4-6 interviews.

Roles covered: ML Engineer. The question bank holds 6 questions reported from Affirm interviews.

Try a free question from this loop's topics

Interview process length
Not well reported; 4-6 interviews
Stages
6
Reports linked to bank questions
9
Bank report dates
Jan 2020–Jun 2024
Questions in bank
6 (0 free)
Last reviewed
October 2026

How the Affirm interview loop works

Most reports describe a recruiter call, an early chat with the hiring manager, a one-hour technical screen, then a final set of three to five rounds of about an hour each with team members: practical coding, practical ML, and a behavioral or hiring-manager conversation. A 2024 senior offer described six interviews in all: a hiring-manager chat, two ML-flavoured coding rounds, two ML rounds and one behavioral round. One 2025 candidate's screen was about exploratory data analysis rather than a puzzle, while a 2024 screen was a medium-level string problem. No take-home appears in our reports. Experiences of pace differ: one candidate found it quick, another found four or five stages slow. All of this comes from few reports, so ask your recruiter for the type of every round.

The Affirm interview process, stage by stage

01Recruiter call

Phone or video call with a recruiter; length not reported

A conversation about your background, your goals and how your experience matches the team. At least one candidate was turned down here for lacking matching experience, so this is a real filter, not just logistics.

What it tests

  • Whether your past work matches the team's domain (credit, fraud, tabular ML in production)
  • Your career goals and reasons for the move
  • Logistics: location, level, remote eligibility

How to prepare

  • Lead with production models on tabular data you have shipped, with the business metric they moved.
  • Ask which team you are interviewing for (for example underwriting or fraud) and the type of every later round.

A recruiter call opens the process in 3 of 7 ML-engineer candidate reports (2021-2025); one candidate was rejected at this call for experience fit.

02Hiring-manager chat

Early conversation with the hiring manager, before the technical rounds

Several candidates met the hiring manager before any technical round. It is mostly about your projects and fit, but it can turn technical: one candidate was asked how to explain an individual prediction from a logistic regression model.

What it tests

  • Depth on your past ML projects and the decisions behind them
  • Fit with the team's problems and way of working
  • Basic model understanding, such as explaining a single prediction

How to prepare

  • Prepare to explain one model end to end: data, features, metric, launch and what you would change.
  • Be ready to explain a linear model's prediction from its coefficients and feature values.

An early hiring-manager conversation appears in 3 of 7 ML-engineer reports (2021, 2024, 2025), each time before the technical rounds.

03Technical screen

About 60 min, live; either a medium-level coding problem or a practical data task

A single one-hour technical screen. Reports differ on its content: a 2024 candidate had a medium-level string compression problem, while a 2025 candidate was asked about exploratory data analysis. One candidate said screen answers are reviewed by other engineers afterwards.

What it tests

  • Clean, correct code for a medium-level array or string problem
  • How you explore an unfamiliar dataset before modelling
  • Clear explanation of your reasoning while you work

How to prepare

  • Practise in-place array and string problems where you track read and write positions.
  • Have an EDA routine ready: shape and types, missing values, target balance, leakage checks, distributions by segment.
  • Write readable code and explain it: other engineers may review it after the call.

A one-hour technical screen appears in 3 of 7 ML-engineer reports since 2024; one was a string problem, one was about EDA, one gave no details.

04Final rounds: practical coding

1-2 x about 60 min with team members; problems lean towards practical ML and core programming

Coding rounds in the final set are described as practical rather than abstract puzzles, at about medium difficulty. The 2024 senior offer had two coding rounds with a practical ML focus; an older report had a string problem about finding, for each word, its shortest substring that no other word contains.

What it tests

  • Core data structures: hash maps, sets, string handling
  • Turning a brute-force idea into something efficient enough
  • Writing code that manipulates data the way an ML pipeline would

How to prepare

  • Practise medium string problems where you enumerate substrings and count them across inputs with a hash map.
  • Expect a data-flavoured task too: grouping, aggregating or transforming records in plain Python.
  • State complexity and test edge cases (empty input, duplicates) before you are asked.

Coding after the screen in 3 of 7 ML-engineer reports (2020, two in 2024); each describes it as practical. A 2025 report lists two teammate rounds without details.

05Final rounds: practical ML

1-2 x about 60 min with team members; discussion or hands-on work with a dataset

The ML rounds stay close to the job: set up a feature engineering pipeline for a given dataset, choose performance metrics for a modelling problem, and reason about trade-offs. Given the company's credit and fraud work, expect imbalanced classification, costs of false approvals versus false declines, calibration and explaining decisions to customers.

What it tests

  • Building features from raw data without leaking future information
  • Choosing metrics that match the business cost of errors
  • Imbalanced classification, thresholds and calibration
  • Explaining a model's individual decisions

How to prepare

  • Practise turning a cost matrix into a decision threshold and a metric choice.
  • Talk about time-based splits and point-in-time features; leakage is a core concern for lending data.
  • Know gradient-boosted trees well; the team's job postings name them as the main model family.

ML rounds on features or metrics in 2 of 7 ML-engineer reports (2020, 2024); a 2025 report had two unnamed teammate rounds. Credit questions come from compiled lists.

06Behavioral & hiring-manager wrap-up

About 60 min, behavioral, often with the hiring manager at the end of the loop

A behavioral round about past experiences, projects and cultural fit, often led by the hiring manager. One candidate was asked for their most challenging project that needed critical thinking and analysis.

What it tests

  • Ownership of a hard project and the reasoning behind your choices
  • Collaboration with product, risk and engineering partners
  • Fit with the team

How to prepare

  • Prepare two or three project stories with a measurable outcome and a clear account of your own part.
  • Include one story about working with non-technical stakeholders, such as risk or product.

A behavioral or final hiring-manager round in 3 of 7 ML-engineer reports since 2024; two say the hiring manager led it.

Questions and themes reported most often

  • Exploratory analysis or a feature engineering pipeline for a given dataset (2 reports, final rounds: practical ml)
  • Medium-level string problems (in-place compression, shortest unique substring) (2 reports, final rounds: practical coding)
  • Choose performance metrics for a modelling problem (1 reports, final rounds: practical ml)
  • Explain a single prediction from a logistic regression or black-box model (1 reports, hiring-manager chat)
  • Most challenging project and the critical thinking behind it (1 reports, behavioral & hiring-manager wrap-up)

What Affirm asks about

  • Practical ML: EDA, features, metrics40%
  • Coding (medium, practical)30%
  • Behavioral & hiring manager20%
  • Credit & fraud modelling concepts10%

Affirm interview prep checklist

  • Solved in-place run-length compression and the shortest-unique-substring problem without hints
  • Walked through EDA and a leak-free feature pipeline for a transactions or loans table
  • Chose a decision threshold from a fraud cost matrix and justified the metric
  • Explained one rejected application from a logistic regression model and from a tree ensemble
  • Can discuss calibration, class imbalance and time-based validation for credit models
  • Prepared a most-challenging-project story with numbers and your own decisions
  • Asked the recruiter which team and which round types to expect

Tips

  • Prepare for practical ML, not theory puzzles: candidates consistently describe job-shaped questions.
  • Rehearse an end-to-end EDA and feature engineering walk-through on a tabular dataset in under 45 minutes.
  • Be fluent in picking metrics from business costs: approval losses, false declines, review capacity.
  • Know how to explain individual predictions for linear models and for tree ensembles.
  • Drill medium array and string problems with careful index handling and hash maps.
  • Watch for leakage: point-in-time features, time-based splits and labels that exist only for approved loans.
  • Expect the hiring manager early and late in the process; keep your project story consistent across both.
  • Ask your recruiter for the format of each round, since reports differ on whether the screen is coding or data analysis.

Practice the Affirm loop

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

Affirm interview FAQ

How many interview rounds does Affirm have for ML roles?

Candidates report 6 stages: Recruiter call, Hiring-manager chat, Technical screen, Final rounds: practical coding, Final rounds: practical ML and Behavioral & hiring-manager wrap-up. Most reports describe a recruiter call, an early chat with the hiring manager, a one-hour technical screen, then a final set of three to five rounds of about an hour each with team members: practical coding, practical ML, and a behavioral or hiring-manager conversation. A 2024 senior offer described six interviews in all: a hiring-manager chat, two ML-flavoured coding rounds, two ML rounds and one behavioral round. One 2025 candidate's screen was about exploratory data analysis rather than a puzzle, while a 2024 screen was a medium-level string problem. No take-home appears in our reports. Experiences of pace differ: one candidate found it quick, another found four or five stages slow. All of this comes from few reports, so ask your recruiter for the type of every round.

How long does the Affirm ML interview process take?

Reported timelines run Not well reported; 4-6 interviews.

What does Affirm ask in ML interviews?

By reported question mix: Practical ML: EDA, features, metrics 40%; Coding (medium, practical) 30%; Behavioral & hiring manager 20%; Credit & fraud modelling concepts 10%. The most repeated themes: Exploratory analysis or a feature engineering pipeline for a given dataset; Medium-level string problems (in-place compression, shortest unique substring); Choose performance metrics for a modelling problem.

How should I prepare for the Affirm ML interview?

Prepare for practical ML, not theory puzzles: candidates consistently describe job-shaped questions. Rehearse an end-to-end EDA and feature engineering walk-through on a tabular dataset in under 45 minutes. Be fluent in picking metrics from business costs: approval losses, false declines, review capacity.

Are there free Affirm interview questions?

LeetML has 6 questions reported from Affirm interviews. The stage sections above link the free practice problems for the topics each round covers; the reported-question set itself is in the app.

Related interview guides