Question 1

You are the engineer who owns Airbnb's search results page. On Monday morning you see that booking conversion from search results dropped 3% compared to last week. Walk me through exactly what you do.

Accepted Answer

Strong answers start with checking if it is a measurement problem (tracking bug, instrumentation change) before assuming a product problem. They then segment the drop by device, geography, listing type, and date range to isolate it. They check whether any A/B test, backend deploy, or ranking change was released that week. They form two or three hypotheses with evidence, build a quick investigation plan, and know when to escalate vs. keep digging.

Question 2

Only 40% of users who start the Airbnb host onboarding flow complete it and publish a listing. You are given no additional budget and 3 engineers for 6 weeks. What do you do?

Accepted Answer

Strong answers start by diagnosing where in the funnel people drop off (step-level funnel analysis), then segment by device, experience level, and listing type to find the worst-performing cohort. They propose a focused experiment on the highest-drop step rather than redesigning everything, measure both completion rate and listing quality (hosts who onboard faster may list lower-quality properties), and have a clear shipping criteria before starting.

Question 3

During bad weather or a major sporting event, Dasher supply drops and order volume spikes. Walk me through how you would design a dynamic pay boost system that incentivizes more Dashers to come online, and how you would communicate it to Dashers in the app.

Accepted Answer

Strong answers model the supply-demand imbalance per zone in real time, set boost amounts using a curve fitted to historical data on Dasher responsiveness to pay increases (price elasticity), and avoid sudden large swings that confuse Dashers. They address the communication design: push notification with expected boost amount, app UI showing the zone and time window, and transparency about when the boost ends. They should also mention that a too-large boost attracts Dashers from neighboring zones, disrupting those zones.

Question 4

You are a PM for Google Search. How would you define and measure the quality of search results? What metrics would you look at, how would you set up a measurement framework, and how would you distinguish between short-term and long-term quality signals?

Accepted Answer

A strong answer separates relevance (did the top result answer the query?), freshness (is the information current?), diversity (does the result set cover multiple intents?), and user satisfaction proxies like long-click rate, reformulation rate, and session abandonment. The candidate should mention the distinction between direct metrics (NDCG from human raters), behavioral proxies (click data), and long-term signals (return visits, query volume growth). Google uses human quality raters with defined guidelines, so an aware candidate mentions this.

Question 5

LinkedIn is testing a new format for sponsored posts in the feed. You want to measure whether the new format increases both click-through rate and member satisfaction (proxied by post-click dwell time), without cannibalizing organic engagement. How would you design this experiment?

Accepted Answer

A strong answer covers unit of randomization (member-level to avoid contamination), sample size calculation based on expected effect size, holdout groups, and the guardrail metrics (organic engagement rate, session length, complaint rate). The candidate should also discuss novelty bias, the tradeoff between running the test long enough for novelty to wear off versus business urgency, and how to handle multiple testing corrections across the two primary metrics.

Question 6

LinkedIn's stated mission is to connect the world's professionals to economic opportunity, and its long-term vision is to build an Economic Graph mapping every job, skill, and professional. If you were a data scientist on this team, what metric would you propose as the north star for measuring progress toward this mission, and how would you defend it?

Accepted Answer

A strong answer names a specific, measurable metric (for example, the number of economic opportunities accessed per member per year, or the wage growth attributable to LinkedIn-facilitated connections) and defends it against Goodhart's Law: could it be gamed, and does gaming it still constitute real progress? The candidate should also acknowledge the tension between member engagement metrics and outcome metrics, and propose leading indicators that predict the lagging outcome.

Question 7

A major concert at SoFi Stadium in Los Angeles is happening Saturday night, and Lyft expects 50,000 ride requests in a 90-minute window after the event ends. How would you forecast how many drivers you need in the area, and what mechanisms would you use to get them there?

Accepted Answer

Strong answers describe using historical data from similar events and comparable venues, a time-series demand model tuned on event attendance and typical post-event request patterns, and supply mechanisms like targeted push notifications to nearby drivers, a driver bonus for being in the area, and preemptive surge pricing to incentivize supply. They should note that supply incentives need to be balanced to avoid over-paying and bleeding margin.

Question 8

The Instagram feed ranking team wants to improve time-spent as a metric. You are running an experiment with a new ranking signal. How do you design the experiment, choose metrics, and decide whether to ship?

Accepted Answer

A strong answer frames the guardrail metrics (likes, comments, shares, follows) alongside the primary metric (time-spent), explains why time-spent alone can be gamed by addictive bad content, and designs an A/B test with holdout sizing based on minimum detectable effect. The candidate should mention how to handle network effects in feed experiments (users on the same social graph can contaminate control and treatment), and what Novelty effect means for ramp duration.

Question 9

Data shows that 30% of Netflix subscribers open the app, scroll for a few minutes, cannot find something to watch, and close the app without playing anything. What product changes would you explore to fix this, how would you prioritize them, and how would you measure success?

Accepted Answer

A strong answer first diagnoses the root cause (poor personalization signal, overwhelming choice, low-quality thumbnails, wrong surface for mood-based browsing) before jumping to solutions. Solutions might include a mood-based filter ('I want something funny and short'), a continue-watching nudge, a social proof layer showing what friends watched, or a direct 'Play Something' button for decision-fatigued users. Measurement should cover play rate from browse sessions, not just clicks on a feature, and the candidate should distinguish between engagement metrics and the long-term retention metric Netflix actually cares about.

Question 10

Assume you are a data engineer who uses Palantir Foundry daily to build pipelines. What is one significant pain point you would focus on improving, and how would you approach the improvement? Be specific about the problem, the users affected, and the solution.

Accepted Answer

A strong answer shows familiarity with Foundry's actual capabilities (pipeline transforms, Code Workbook, dataset versioning, branch-based development) and identifies a real friction point such as debugging failed transforms, managing dependency graphs at scale, or slow feedback loops when iterating on large datasets. The candidate should frame the improvement as a user need first, propose a concrete solution, and reason about the trade-off between simplicity for new users and power for advanced ones.

Question 11

Your squad wants to ship a new personalization feature that changes how the Home tab orders content for users. Walk me through how you would design an A/B test, what metric you would use as the primary success metric, what guardrail metrics you would watch, and how long you would run the experiment before making a call.

Accepted Answer

A strong answer starts by picking a north-star metric tied to Spotify's business, such as weekly active listening minutes or track completion rate, and explains why that is better than CTR alone. It describes how to randomly split users, what sample size is needed to detect a meaningful lift (power calculation), and what guardrail metrics (like subscription cancellations or support tickets) would block a rollout even if the primary metric is green. It addresses novelty effects and how long to run to avoid them.

Question 12

X launched a paid subscription tier called X Premium. If you were the PM or lead engineer for this product, what metrics would you use to determine whether it is succeeding, how would you instrument the product to capture those metrics, and what would a healthy versus unhealthy metric trend look like 90 days after launch?

Accepted Answer

A strong answer separates acquisition metrics (conversion rate from free to paid, CAC), retention metrics (monthly churn rate, day-30 retention), engagement metrics (do paid users engage more or differently), and revenue metrics (MRR, LTV, MRR growth rate). The candidate should also identify leading indicators (feature adoption of premium-only features predicts retention) and define what healthy looks like (churn under 5% per month, LTV/CAC above 3). They should note that purely subscription metrics miss the ecosystem effect on advertiser confidence.

Question 13

You run an A/B test on a new tweet ranking feature. The treatment group shows a 4% lift in daily engagement (likes, replies, retweets) but also a 12% increase in content reported as harmful or abusive. The test is statistically significant on both metrics. What do you do?

Accepted Answer

A strong answer refuses to ship based on engagement lift alone, frames the decision around long-term user trust and platform health, and proposes decomposing the 12% harm increase to understand what kind of content is driving it (is it the ranking surfacing existing harmful content or generating new user behavior?). The candidate should recommend not shipping until a harm-neutral variant is found, escalate to a safety review if needed, and recognize that pure engagement metrics incentivize the system to surface outrage. X has historically struggled with this trade-off, so the candidate should show nuanced judgment rather than a simplistic 'do not ship' answer.

Question 14

You are the engineer who owns Airbnb's search results page. On Monday morning you see that booking conversion from search results dropped 3% compared to last week. Walk me through exactly what you do.

Accepted Answer

Strong answers start with checking if it is a measurement problem (tracking bug, instrumentation change) before assuming a product problem. They then segment the drop by device, geography, listing type, and date range to isolate it. They check whether any A/B test, backend deploy, or ranking change was released that week. They form two or three hypotheses with evidence, build a quick investigation plan, and know when to escalate vs. keep digging.

Question 15

Only 40% of users who start the Airbnb host onboarding flow complete it and publish a listing. You are given no additional budget and 3 engineers for 6 weeks. What do you do?

Accepted Answer

Strong answers start by diagnosing where in the funnel people drop off (step-level funnel analysis), then segment by device, experience level, and listing type to find the worst-performing cohort. They propose a focused experiment on the highest-drop step rather than redesigning everything, measure both completion rate and listing quality (hosts who onboard faster may list lower-quality properties), and have a clear shipping criteria before starting.

Question 16

During bad weather or a major sporting event, Dasher supply drops and order volume spikes. Walk me through how you would design a dynamic pay boost system that incentivizes more Dashers to come online, and how you would communicate it to Dashers in the app.

Accepted Answer

Strong answers model the supply-demand imbalance per zone in real time, set boost amounts using a curve fitted to historical data on Dasher responsiveness to pay increases (price elasticity), and avoid sudden large swings that confuse Dashers. They address the communication design: push notification with expected boost amount, app UI showing the zone and time window, and transparency about when the boost ends. They should also mention that a too-large boost attracts Dashers from neighboring zones, disrupting those zones.

Question 17

You are a PM for Google Search. How would you define and measure the quality of search results? What metrics would you look at, how would you set up a measurement framework, and how would you distinguish between short-term and long-term quality signals?

Accepted Answer

A strong answer separates relevance (did the top result answer the query?), freshness (is the information current?), diversity (does the result set cover multiple intents?), and user satisfaction proxies like long-click rate, reformulation rate, and session abandonment. The candidate should mention the distinction between direct metrics (NDCG from human raters), behavioral proxies (click data), and long-term signals (return visits, query volume growth). Google uses human quality raters with defined guidelines, so an aware candidate mentions this.

Question 18

LinkedIn is testing a new format for sponsored posts in the feed. You want to measure whether the new format increases both click-through rate and member satisfaction (proxied by post-click dwell time), without cannibalizing organic engagement. How would you design this experiment?

Accepted Answer

A strong answer covers unit of randomization (member-level to avoid contamination), sample size calculation based on expected effect size, holdout groups, and the guardrail metrics (organic engagement rate, session length, complaint rate). The candidate should also discuss novelty bias, the tradeoff between running the test long enough for novelty to wear off versus business urgency, and how to handle multiple testing corrections across the two primary metrics.

Question 19

LinkedIn's stated mission is to connect the world's professionals to economic opportunity, and its long-term vision is to build an Economic Graph mapping every job, skill, and professional. If you were a data scientist on this team, what metric would you propose as the north star for measuring progress toward this mission, and how would you defend it?

Accepted Answer

A strong answer names a specific, measurable metric (for example, the number of economic opportunities accessed per member per year, or the wage growth attributable to LinkedIn-facilitated connections) and defends it against Goodhart's Law: could it be gamed, and does gaming it still constitute real progress? The candidate should also acknowledge the tension between member engagement metrics and outcome metrics, and propose leading indicators that predict the lagging outcome.

Question 20

A major concert at SoFi Stadium in Los Angeles is happening Saturday night, and Lyft expects 50,000 ride requests in a 90-minute window after the event ends. How would you forecast how many drivers you need in the area, and what mechanisms would you use to get them there?

Accepted Answer

Strong answers describe using historical data from similar events and comparable venues, a time-series demand model tuned on event attendance and typical post-event request patterns, and supply mechanisms like targeted push notifications to nearby drivers, a driver bonus for being in the area, and preemptive surge pricing to incentivize supply. They should note that supply incentives need to be balanced to avoid over-paying and bleeding margin.

Question 21

The Instagram feed ranking team wants to improve time-spent as a metric. You are running an experiment with a new ranking signal. How do you design the experiment, choose metrics, and decide whether to ship?

Accepted Answer

A strong answer frames the guardrail metrics (likes, comments, shares, follows) alongside the primary metric (time-spent), explains why time-spent alone can be gamed by addictive bad content, and designs an A/B test with holdout sizing based on minimum detectable effect. The candidate should mention how to handle network effects in feed experiments (users on the same social graph can contaminate control and treatment), and what Novelty effect means for ramp duration.

Question 22

Data shows that 30% of Netflix subscribers open the app, scroll for a few minutes, cannot find something to watch, and close the app without playing anything. What product changes would you explore to fix this, how would you prioritize them, and how would you measure success?

Accepted Answer

A strong answer first diagnoses the root cause (poor personalization signal, overwhelming choice, low-quality thumbnails, wrong surface for mood-based browsing) before jumping to solutions. Solutions might include a mood-based filter ('I want something funny and short'), a continue-watching nudge, a social proof layer showing what friends watched, or a direct 'Play Something' button for decision-fatigued users. Measurement should cover play rate from browse sessions, not just clicks on a feature, and the candidate should distinguish between engagement metrics and the long-term retention metric Netflix actually cares about.

Question 23

Assume you are a data engineer who uses Palantir Foundry daily to build pipelines. What is one significant pain point you would focus on improving, and how would you approach the improvement? Be specific about the problem, the users affected, and the solution.

Accepted Answer

A strong answer shows familiarity with Foundry's actual capabilities (pipeline transforms, Code Workbook, dataset versioning, branch-based development) and identifies a real friction point such as debugging failed transforms, managing dependency graphs at scale, or slow feedback loops when iterating on large datasets. The candidate should frame the improvement as a user need first, propose a concrete solution, and reason about the trade-off between simplicity for new users and power for advanced ones.

Question 24

Your squad wants to ship a new personalization feature that changes how the Home tab orders content for users. Walk me through how you would design an A/B test, what metric you would use as the primary success metric, what guardrail metrics you would watch, and how long you would run the experiment before making a call.

Accepted Answer

A strong answer starts by picking a north-star metric tied to Spotify's business, such as weekly active listening minutes or track completion rate, and explains why that is better than CTR alone. It describes how to randomly split users, what sample size is needed to detect a meaningful lift (power calculation), and what guardrail metrics (like subscription cancellations or support tickets) would block a rollout even if the primary metric is green. It addresses novelty effects and how long to run to avoid them.

Question 25

X launched a paid subscription tier called X Premium. If you were the PM or lead engineer for this product, what metrics would you use to determine whether it is succeeding, how would you instrument the product to capture those metrics, and what would a healthy versus unhealthy metric trend look like 90 days after launch?

Accepted Answer

A strong answer separates acquisition metrics (conversion rate from free to paid, CAC), retention metrics (monthly churn rate, day-30 retention), engagement metrics (do paid users engage more or differently), and revenue metrics (MRR, LTV, MRR growth rate). The candidate should also identify leading indicators (feature adoption of premium-only features predicts retention) and define what healthy looks like (churn under 5% per month, LTV/CAC above 3). They should note that purely subscription metrics miss the ecosystem effect on advertiser confidence.

Question 26

You run an A/B test on a new tweet ranking feature. The treatment group shows a 4% lift in daily engagement (likes, replies, retweets) but also a 12% increase in content reported as harmful or abusive. The test is statistically significant on both metrics. What do you do?

Accepted Answer

A strong answer refuses to ship based on engagement lift alone, frames the decision around long-term user trust and platform health, and proposes decomposing the 12% harm increase to understand what kind of content is driving it (is it the ranking surfacing existing harmful content or generating new user behavior?). The candidate should recommend not shipping until a harm-neutral variant is found, escalate to a safety review if needed, and recognize that pure engagement metrics incentivize the system to surface outrage. X has historically struggled with this trade-off, so the candidate should show nuanced judgment rather than a simplistic 'do not ship' answer.

Questions

You own search. Conversion rate dropped 3%. What do you do?Product sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Improve the host onboarding completion rateProduct sensehardOccasional

As asked

Sample answer outline

Expect these follow-ups

How would you implement and explain Dasher pay boosts during demand surges?Product sensemediumOccasional

As asked

Sample answer outline

Expect these follow-ups

How would you measure the quality of Google Search results?Product sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Design an A/B test for a new LinkedIn sponsored content formatProduct sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Define the north-star metric for LinkedIn's Economic Graph missionProduct sensehardOccasional

As asked

Sample answer outline

Expect these follow-ups

Forecast driver supply for a major city eventProduct sensehardOccasional

As asked

Sample answer outline

Expect these follow-ups

Improve the ranking model for Instagram's main feedProduct sensemediumCommon

As asked

Sample answer outline

Expect these follow-ups

Improve content discovery for subscribers who stopped browsingProduct sensemediumCommon

As asked

Sample answer outline

Expect these follow-ups

How would you improve Palantir Foundry for data engineers?Product sensemediumCommon

As asked

Sample answer outline

Expect these follow-ups

Design an A/B test for a new recommendation featureProduct sensemediumCommon

As asked

Sample answer outline

Expect these follow-ups

How would you measure the success of X Premium (paid subscriptions)?Product sensemediumOccasional

As asked

Sample answer outline

Expect these follow-ups

An A/B test shows higher engagement but also higher harm reportsProduct sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

You own search. Conversion rate dropped 3%. What do you do?Product sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Improve the host onboarding completion rateProduct sensehardOccasional

As asked

Sample answer outline

Expect these follow-ups

How would you implement and explain Dasher pay boosts during demand surges?Product sensemediumOccasional

As asked

Sample answer outline

Expect these follow-ups

How would you measure the quality of Google Search results?Product sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Design an A/B test for a new LinkedIn sponsored content formatProduct sensehardCommon

As asked

Sample answer outline

Expect these follow-ups

Define the north-star metric for LinkedIn's Economic Graph missionProduct sensehardOccasional

As asked

Sample answer outline

Expect these follow-ups

Forecast driver supply for a major city eventProduct sensehardOccasional

As asked

Sample answer outline