Why this round exists
The PM role at data-intensive companies like logistics, marketplaces, fintech, delivery, has evolved closer to a product analytics hybrid. PMs are expected to pull their own data, diagnose metric anomalies without waiting on a data scientist, and communicate quantitative findings to engineers and executives with equal clarity.
As noted in a 2025 analysis on Medium (datainterview.com), the PM and data science interview format shifted decisively after 2022. Earlier interviews were largely standardized: a basic SQL or Python screen, a take-home, and a panel. By 2025, every round is tailored to the job you'll do and the way you'll do the work. Interviews now simulate real workplace conditions, asking you to debug a broken pipeline live, scope an ambiguous product metric, or pair with an interviewer on a system design task.
For PMs specifically, Uber formalized this with their "Jam Session" format: in the on-site, you interview with product managers, data scientists, marketers, and engineers in one room. You explore measurements and how you operate with data scientists to make data-driven decisions.
What the Data & Tech Screen tests
These screens are typically 45 to 60 minutes and break into three components: verbal analysis of data presented in the room, live SQL (usually on a shared doc or whiteboard), and problem-solving narration. The CloudKitchens Data & Tech Screen is a clear example where candidates are given data, asked to analyze it verbally, write some SQL, and walk through their problem-solving approach live with the interviewer.
The goal is to assess whether you can think like an analyst when the stakes are real, not to produce perfect code.
1. Verbal data analysis
This is the component most PMs underestimate in an interview. You're handed a table, dashboard, or raw output and asked to describe what you see, what's interesting, and what you'd want to investigate further in the data.
In 2025, interview questions tend to be ambiguous and more high-level to connect with a larger business problem. For a question like "Who are our most valuable users over time?", you get to define what "valuable" means, make trade-offs, explain why your logic makes sense, and then write the SQL that will correctly output the users according to your own logic.
The framing matters. Companies want to hear you move from observation to hypothesis to diagnostic question. A structure that works:
1) Anchor the metric: "This looks like weekly active orders per city, segmented by new vs. returning users."
2) Name the anomaly: "There's a dip in returning user orders in week 14 that doesn't appear in the new user line."
3) Form hypotheses: "That could be seasonality, a product change, a delivery quality issue, or a cohort maturity effect, and I'd want to look at each."
4) Propose a query: "To isolate the cohort effect, I'd pull retention curves by signup week and compare week-14 order rates across cohorts."
As documented in Interview Query's data analytics case study guide, product metrics cases tend to focus on a hypothetical situation: you'll propose a solution to a problem and supply the metrics you'll use to investigate or solve it, with an SQL query as an optional component.
What interviewers are scoring: Do you anchor the business model? Do you prioritize hypotheses with intelligence? Do you separate signal from noise without a prompt?
2. Live SQL
The SQL bar for PMs at data-heavy companies sits meaningfully above most candidates' expectations. Basic SELECT, WHERE, and GROUP BY queries will not get you past this round. You should be comfortable with the following:
1) Joins: Know when to use LEFT vs. INNER, and be able to explain why dropping unmatched rows would skew a result. This comes up constantly in marketplace data where order tables and user tables don't always share complete overlap.
2) Window functions: ROW_NUMBER(), RANK(), LAG(), and LEAD() show up in nearly every modern data screen. Per Prachub's Meta Data Scientist interview guide (updated 2026), you should be fluent in SQL fundamentals including joins, aggregations, window functions, CTEs, NULL behavior, and the difference between WHERE and HAVING, along with performance-aware query design.
3) CTEs: Complex queries should be built in stages using WITH clauses. Collapsing everything into nested subqueries signals you haven't worked with production-scale data.
4) Funnel construction: Given an events table, can you compute conversion rates across steps? A funnel query typically involves conditional aggregation with CASE WHEN logic using grouping by user_id, marking whether each user hit each funnel stage, then aggregating across the cohort to compute step-to-step conversion rates.
At Uber, the explicit advice is to verbalize why you choose each join or filter. This demonstrates analytical transparency and shows you can defend your work when results are questioned by stakeholders.
An important note on format: whiteboard or verbal walk-throughs, explaining how you'd structure a query without writing code, are used specifically for testing conceptual clarity under pressure. Hiring managers continue to test SQL directly because clean, optimized SQL is seen as a proxy for analytical rigor, data literacy, and attention to detail (Interview Query, 2025).
3. Problem-solving narration
The third component is the least structured and arguably the most telling of the three. You'll get something like: "Orders dropped 18% in one region last Tuesday. Walk me through how you'd investigate that." No data. No schema. Just your reasoning process.
One candidate at a delivery startup shared on Medium: "They asked why repeat orders dropped in LA. No data given. Just walk them through how you'd debug it." These cases are about structured thinking, metric design, and business impact.
The framework that works best for PMs in this format:
1) Scope the question first: Is this a data collection issue, a product issue, or an external issue? Check data pipeline health before diagnosing product behavior.
2) Segment the drop: Is it across all platforms, or isolated to iOS/Android/web? All customers or a specific cohort? All merchants or a subset?
3) Correlate with known changes: Did a release go out that day? Were there pricing changes, promotional changes, or driver/courier incentive changes?
4) Propose the query: "I'd start with a breakdown of order volume by platform, merchant category, and user tenure cohort for that Tuesday vs. the prior four Tuesdays."
The product case study format tests whether you can break an open-ended question into smaller, solvable chunks and narrow scope in the right, relevant product contexts. Interviewers want to see a few relevant metrics that tell a meaningful story without rambling, according to Bobby Deng's 2024 data science interview prep guide on Medium.
What companies look for by format
Understanding how each company operates is a critical skill. How many-sided is each marketplace? Below we’ll take a look behind the curtain of data-driven tech companies that prioritize the data interview.
CloudKitchens: A structured Data & Tech Screen with verbal data analysis, SQL, and a problem-solving walk-through. The emphasis is on operational data like order volume, kitchen utilization, delivery time variance.
DoorDash: Per InterviewKickstart's 2026 guide, the analytical assignment can include numerical and verbal reasoning through to Excel worksheets with hypothetical data. You'll need to use the data to analyze and then make recommendations from the point of view of a product manager. The focus is on marketplace thinking: Dasher supply, merchant health, consumer retention.
Uber: PMs are expected to explore measurements and demonstrate how they operate with data scientists to make data-driven decisions. The Jam Session includes live data discussion with data science peers who will probe metric definitions, trade-offs, and analytical approach. Per Exponent's Uber PM interview guide, you need competency in defining metrics as well as understanding what to do when metrics change, which shows a methodical approach to data-driven decision-making.
Stripe: Stripe's "Write Everything Down" culture means written communication skills are as important as verbal ones. Per Interview Query's Stripe PM guide, prepare concrete examples of how you've used data to identify problems or drive impact. Link every metric to either user trust, revenue growth, or operational efficiency - Stripe's three north stars.
Meta (Analytics PMs): Per Prachub's Meta Data Scientist interview guide (2026), the technical interview heavily focuses on SQL and product analytics reasoning. The format follows a standard pattern: SQL question first, using real product data context, followed by a product case where you use your query results to discuss product metrics or experiment design. At senior levels, add measurement platforms, data pipelines, causal inference principles, and communicating trade-offs to product and engineering partners.
How to prepare
1) SQL fluency: Work through 10 to 15 problems on DataLemur or Interview Query daily for two to three weeks. Focus on window functions, funnel construction, and cohort retention queries. These three patterns cover the majority of what you'll encounter in a PM data screen.
2) Verbal narration practice: Take any public dataset such as Airbnb listings data, NYC taxi data, Kaggle, and open it in a notebook or spreadsheet, and practice describing what you see out loud before you write a single line of code. Recording yourself is an uncomfortable but useful habit.
3) Business model grounding: Before any interview, map the company's core funnel. For a marketplace: acquisition, first order, repeat order, retention. Know which metrics live at each stage. If DoorDash's repeat order rate drops, you should already have an opinion on which cohort deserves the first look.
4) Practice the "why" before the "how": The strongest candidates start every analysis by stating the business question that drives the work. "I want to understand whether this drop is supply-side or demand-side" is more impressive than jumping to a GROUP BY query with no stated intent.
The real filter
Companies running Data & Tech screens are filtering for PMs who won't be bottlenecked by their analytics team, not trying to hire data scientists into PM roles. They are looking for people who can form a hypothesis, pull a sanity-check query, and come to a 1:1 with their data counterpart having already done 60% of the diagnostic work.
If you've shipped product but have never personally written a SQL query against production data, this round will expose that weakness. The good news is that the bar is learnable and well-defined with SQL practice tools like sql-practice.com to understand SQL application. Practice real-world cases and learn the metrics that matter for your target industry. When coding, strive for clean, optimized, and maintainable SQL.
The candidates who clear this round consistently do one thing the others don't: they narrate their thinking with the same confidence they bring to a product roadmap discussion. The data fluency is table stakes. The structured, confident communication is what gets you to the loop.