Jobra
🐞Step 1 of Jobra:AI Debugging Rounds
DSA+AIPreparing you for both.
JOBRA · AI DEBUG ROUND (PREVIEW)sum.cpp
// buggy snippet
int total(vector<int>& a, int n) {
    int sum = 0;
    for (int i = 0; i <= n; i++) { // 🐞        sum += a[i];
    }
    return sum;
}
A real codebase. A real bug. You + AI.

Jobra drops you into a broken repo. Find it, fix it, ship it. AI helps you debug like on the job.

Prompt:Correct the loop condition
Tap 'Let Jobra fix it' below to simulate AI debugging...

Meta Data Scientist Interview Guide: SQL, Product Sense, Analytical Execution and Behavioral

5 min read

Prepare for Meta data scientist interviews with reported assessment patterns, interview questions, and preparation advice.

In this article
On this page

Last reviewed November 15, 2026.

For Data Scientist (Product Analytics) candidates, including PhD interns.

Built by the Codinzhub team from Meta data science interview experiences that candidates shared. Each experience is one person's interview, and Meta changes its cases over time, so use this as a map, not a promise.

The 30-second version. A recruiter call, then a 45-minute technical screen: two SQL questions and a product case on the same tables. Then a virtual onsite of four rounds: SQL, Analytical Execution (AE) with probability, Analytical Reasoning (AR) with product sense and experiments, and behavioral. The same well-known cases repeat (fake accounts, active ads, ads ranking), but interviewers change the numbers and follow-ups. Speed and getting to the key point quickly decide the result.

Round-by-round breakdown

Stage 1: Application, where candidates applied

How it works. Most candidates were contacted by a recruiter or a university recruiter. One senior candidate was reached by HR for an L5 Product Analytics role.

Stage 2: Shortlisting, how candidates moved forward

How it works. A short recruiter call, then the screen scheduled two to four weeks later. Candidates who passed heard within a day or a few days.

Stage 3: Technical screen

At a glance

  • Length: 45 minutes (5 intro, 20 SQL, 20 product, 5 questions)
  • Questions per round: 2 SQL questions and 3 to 6 product questions
  • What was asked: SQL on interaction or call tables, then a product case on the same data
  • Most common: a feature launch case with an A/B test

How it works. You may choose whether SQL or product comes first. The product questions build on the SQL tables.

Questions candidates reported:

Question (our summary)What it tests
Users who messaged 5 or more people today; users who got a reply within a minuteSQL with self-joins
Products with more than 3 buyers and 10 interactions; share of "validate" interactions for US products last weekSQL aggregation, joins
Share of today's active users with over 50 calls in the last 7 daysSQL with dates as strings
Should Instagram get a Facebook marketing feature? Metric, model, test and launch decisionProduct sense, A/B testing
Parents monitoring teen accounts: negative effectsProduct sense

More questions: Browse Meta questions.

Stage 4: The virtual onsite

How it works. Four rounds, each opening with a short introduction. Several candidates said the onsite is closer to known cases than the screen, but the details change.

SQL round

At a glance

  • Length: about 45 minutes
  • Questions per round: 3 or 4 SQL questions, then a hypothesis question
  • What was asked: ad revenue by source, shop visibility, Oculus engagement
  • Most common: active ads revenue by creation source

Questions candidates reported:

Question (our summary)What it tests
Daily revenue by ad creation source; countries with the fewest active advertisersSQL aggregation
Share of users per source who spent 1,000 more this year than lastYear-over-year SQL
Are new shops more active? Define your own metricSQL and metric design
Do social-app users engage more than game-app users?Hypothesis with SQL

More questions: Browse Meta questions.

Analytical Execution (AE)

At a glance

  • Length: about 45 minutes
  • Questions per round: a case with 4 to 5 probability questions
  • What was asked: Bayes' theorem, binomial probability, distributions, root cause of a metric drop
  • Most common: the fake accounts case

Questions candidates reported:

  • Fake accounts: how to detect them, then probabilities from the fake-to-real ratio, a binomial question and Bayes' theorem.
  • A chatbot response is good if helpful (0.8) and honest (0.9): the chance of a good response, of two in a row, and of the fourth given three.
  • The distribution of comments per daily active user, and of the mean when you resample 10,000 users many times.
  • A KPI suddenly dropped: find the cause.

Analytical Reasoning (AR)

At a glance

  • Length: about 45 minutes
  • Questions per round: one product case with 4 to 6 parts
  • What was asked: goals, metrics, experiment design, trade-offs, leadership communication
  • Most common: a new ads ranking algorithm

Questions candidates reported:

  • A new ads ranking algorithm: evaluate it, run the experiment, read the results, balance revenue, advertisers and users.
  • A feature recommendation: convince your boss, choose data, define a north star and counter metrics, design the test.
  • A Venmo-like transfer feature in Messenger: risks, monetisation, testing with network effects.
  • An AI-assisted ad creation tool for advertisers.

More questions: Browse Meta questions.

Behavioral round

At a glance

  • Length: about 45 minutes
  • Questions per round: 4 to 6 questions
  • What was asked: challenging projects, disagreement, making others feel welcome, learning fast
  • Most common: a challenging project; a coworker who did not feel welcome

Questions candidates reported: a breakthrough project; disagreeing with someone; reporting to someone new to the team; making others feel welcome; something you learned quickly; how you convinced others. Prepare six or more stories.

More questions: Browse Meta questions.

Stage 5: Decision and offer

How it works. Results came in a few days to two weeks. One candidate got an offer five days after the onsite. One senior case ended with presenting results on a single slide to executives.

The Codinzhub Next-Round Playbook

What data scientists who moved on did. Put together by the Codinzhub team from the DS experiences in this guide.

To get past the screen

  • Practise SQL until it is automatic; one candidate who uses SQL daily nearly ran out of time.
  • Read what the question asks: a count of products, not a list.
  • Go straight to the key point in the product case. Needing follow-ups to get there cost one candidate the screen.

To get past the onsite

  • Practise the known cases, then expect changed numbers and follow-ups.
  • Revise Bayes' theorem and binomial probability for AE.
  • In AR, state your assumptions, pick a north star and counter metric, then design the test.

Before you start

  • Prepare a two-to-three-minute introduction. Every session opens with one.

About this guide

This guide is researched and written by the Codinzhub team. We read every Meta data science interview experience we could find, kept only what candidates actually described, and rewrote it in our own words. Questions are our own short summaries. Found something out of date? Share your own experience on Codinzhub and we will update the guide.

Ready for the next topic?

Keep track of what you’ve learned.