Metrics
400 bites tagged Metrics — interview questions with model answers, and 60-second explainers.
Describe tracking a user event from frontend to BI tool
Tests your grasp of the modern data stack. A great answer outlines the five stages: frontend emission, ingestion, loading into a warehouse, transformation, and BI visualization. A red flag is describing only the frontend code and ignoring the data pipeline.
Correlate API Slowness with User Engagement
This tests your ability to design a controlled experiment and join disparate data. A good answer outlines an A/B test with an artificial delay, logging with shared IDs, and statistical analysis. A red flag is proposing a purely observational study.
How would you measure P95 latency by geographic region?
Tests your ability to translate a business need into a concrete observability implementation. A good answer involves instrumenting the API with a histogram metric, adding a region label via GeoIP, and querying with `histogram_quantile`.
How would you measure a sales forecast model's accuracy?
Tests if you can link statistical metrics to business outcomes. Define MAE (average error) and RMSE (penalizes large errors). Choose RMSE when large misses are costly (e.g., stock-outs), MAE otherwise. A red flag is reciting formulas without business context.
How would you analyze a time series of user sign-ups?
This tests your structured approach to decomposing time series data. A strong answer identifies trend (long-term growth), seasonality (weekly/yearly patterns), and irregular components like spikes or dips.
How CUPED increases statistical power in experiments
Tests your grasp of variance reduction in A/B testing. Explain how CUPED uses correlated pre-experiment data to reduce outcome variance, increasing statistical power. A red flag is confusing it with simpler difference scores, which can actually increase noise.
Primary vs. Guardrail Metrics in Experiments
Tests your grasp of risk management in A/B testing. A great answer defines a primary metric as the goal and a guardrail as a 'do no harm' check. A feature ships only if the primary improves without hurting guardrails.
What is the novelty effect in experimentation?
This tests your grasp of temporary user behavior changes that can invalidate A/B tests. A strong answer defines the effect, explains how it inflates metrics, and suggests running tests longer or segmenting by user tenure. A red flag is ignoring mitigation.
Why is stopping an A/B test early problematic?
Tests understanding of the 'peeking problem' in A/B testing. A good answer defines peeking, explains how it inflates false positive rates, and contrasts it with waiting for a pre-determined sample size. A red flag is not explaining the statistical mechanism.
How do you determine sample size and duration for an A/B test?
This tests your grasp of statistical power and business trade-offs. A good answer defines the four inputs (baseline, MDE, significance, power) to calculate sample size, then uses traffic to find duration.
How would you A/B test a 'Buy Now' button color change?
Tests your grasp of the A/B testing lifecycle. A strong answer defines a clear hypothesis (e.g., 'a green button will increase clicks'), selects a primary metric (CTR), and considers guardrail metrics. A red flag is skipping the hypothesis and metrics.
Why can't we t-test p99 latency in an A/B test?
This tests if you know why t-tests fail for percentiles. A t-test requires a normally distributed statistic (like the mean), but a sample p99's distribution isn't normal.
Explain Simpson's Paradox with a user engagement example
Tests if you see beyond aggregate data. Define the paradox, give a numerical example where a feature fails overall but wins in segments (e.g., new vs. returning users), and name the confounding variable. A vague definition without numbers is a red flag.
DAU dropped 10% overnight. Is this a significant change?
Tests your use of statistical hypothesis testing on business metrics. Outline the process: state a null hypothesis (no change), choose a Z-test, calculate the p-value, and compare to an alpha of 0.05. A red flag is guessing causes before proving significance.
Explain correlation vs. causation with a software example
This tests your critical thinking about data and ability to avoid logical fallacies. A good answer defines both terms, then gives a software example where a third, confounding variable (like traffic) is the true cause of two correlated metrics.
A/B test p-value is 0.08, PM wants to ship. What now?
Tests if you can translate statistical risk into business terms for a PM. A good answer defines the 8% false positive risk, weighs it against the cost of shipping, and suggests next steps like running the test longer instead of just saying no.
What does a p-value of 0.03 mean in an A/B test?
This tests your grasp of statistical significance and ability to make a data-driven decision. A great answer defines p-value, compares it to alpha (0.03 < 0.05), and concludes to reject the null hypothesis. A red flag is misstating the p-value's definition.
Mean vs. Median for API Response Times?
Tests your understanding of non-normal distributions like latency. Choose median as it's robust to outliers that skew the mean. Then, state that even median is insufficient; percentiles (p99, max) are crucial for capturing the full user experience.
A/B Test Results with Skewed Traffic: What's Next?
This tests your ability to spot confounding variables. A good answer invalidates the results due to sampling bias, proposes segmenting the data by device to find the true effect, and suggests re-running the test with correct randomization.
How to query a monthly cohort retention table in SQL?
Tests your ability to translate a core business metric into a multi-step SQL query. A good answer finds each user's acquisition month, joins that back to their activity, and pivots the data into a cohort grid. A red flag is calculating aggregate retention.
Average latency is up 50ms, but p99 is flat. How do you diagnose this?
Tests your grasp of latency metrics. A rising average with flat p99 means the *bulk* of requests (p50-p90) slowed, not the tail. Hypothesize a common bottleneck and segment data by endpoint/user to find it. Red flag: blaming new, slow outliers.
DAU dropped 10%. How do you investigate?
Tests structured problem diagnosis. First, verify the data isn't corrupt. Then, segment the drop by user type (new vs. returning), platform (iOS/Android/Web), and geography to isolate the 'what' before hypothesizing the 'why'.
How would you visually represent statistical uncertainty in a chart?
This tests your ability to accurately communicate statistical significance. A great answer discusses error bars (with 95% CIs), then more advanced options like gradient or violin plots, and frames the choice by audience.
Stakeholder claims correlation implies causation. How do you investigate?
This tests your scientific rigor beyond the simple "correlation isn't causation" mantra. Acknowledge the finding, probe for confounding variables, suggest cohort analysis, and propose an A/B test. A red flag is reciting the mantra without a concrete plan.
Get Metrics bites daily.
Five a day, five minutes, offline. With quizzes so it sticks.
Open testing — you’ll join as an early tester.