← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Brutal Meta DS interview focused entirely on search feature evaluation. Four interconnected parts, all building on each other, and I felt like I was barely keeping up by the third one.

Questions Asked (4)

Q1

You have two binary labels per query for a new search feature: relevancy and accuracy. Design a composite success metric using these labels. What's your exact scoring rule, how do you justify it under different error cost assumptions, and how would you calibrate the weights using actual business outcomes?

Product Analytics & MetricsA/B Testing & ExperimentationTechnical Trade-offs
Author's notes

This is where I spent the most mental energy.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining a weighted composite score that combines relevancy and accuracy, then justify the weights based on the relative cost of errors (false positives vs. false negatives). Finally, explain how you would calibrate the weights using business outcomes like user engagement or revenue, potentially through A/B testing or regression analysis.

Pro tip: Emphasize that the weights should be dynamic and context-dependent, and propose a method to continuously learn and adjust them based on real-time feedback, showing a product-minded and iterative approach.

1. Define the Composite Metric

Propose a weighted sum of relevancy and accuracy, e.g., Score = w1 * Relevancy + w2 * Accuracy, where w1 + w2 = 1. Discuss possible variations like harmonic mean if both are equally important.

2. Justify Weights Under Error Costs

Explain that weights reflect the relative cost of errors: if a false positive (irrelevant but accurate) is more costly, increase w1; if a false negative (relevant but inaccurate) is more costly, increase w2. Use examples to illustrate.

3. Calibrate Using Business Outcomes

Describe how to use historical data or experiments to find weights that maximize a business KPI (e.g., click-through rate, conversion). Suggest methods like logistic regression, grid search, or multi-armed bandits.

4. Validate and Iterate

Propose A/B testing to validate the chosen weights and monitor for shifts in user behavior or business goals. Emphasize the need for periodic recalibration.

Key Points to Mention

  • Weighted sum vs. other aggregation methods (e.g., harmonic mean, product)
  • Error cost asymmetry: false positives vs. false negatives
  • Business KPI alignment: choose weights that maximize engagement or revenue
  • Calibration techniques: regression, A/B testing, bandits
  • Dynamic weighting: adapt to context (query type, user segment)
  • Monitoring and iteration: set up feedback loops for continuous improvement

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

At what level do you aggregate this metric: query, session, user, or day? How do you handle users with many queries, missing labels, and the fact that outcomes within a session are probably correlated?

Product Analytics & MetricsData ModelingA/B Testing & Experimentation
Author's notes

Honestly tripped me up more than I expected for something that sounds procedural.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the metric's purpose and the decision it informs, then recommend a primary aggregation level (e.g., user-day) while acknowledging trade-offs. Explain how to handle users with many queries via weighting or capping, missing labels via imputation or exclusion, and session correlation via clustering or mixed-effects models.

Pro tip: Emphasize that the choice of aggregation should align with the experimental unit and the business decision; for A/B tests, aggregating at the randomization unit (often user) avoids inflated false positives due to correlation.

1. Clarify the metric and goal

Define what the metric measures and the decision it supports (e.g., product change, model evaluation). This determines the appropriate aggregation level.

2. Choose primary aggregation level

Recommend aggregating at the user-day level as a balance between granularity and stability, but note that query-level may be needed for debugging and session-level for behavioral analysis.

3. Handle users with many queries

Use weighting (e.g., inverse query count) or capping to prevent power users from dominating; alternatively, model at the user level with query count as a feature.

4. Address missing labels

Assess missingness mechanism; if random, exclude or impute; if systematic, incorporate missingness as a feature or use multiple imputation.

5. Account for session correlation

Use cluster-robust standard errors, mixed-effects models, or aggregate to user level to account for within-session correlation.

Key Points to Mention

  • Alignment with experimental unit (e.g., user-level randomization in A/B tests)
  • Trade-offs between granularity and statistical power
  • Weighting or capping to handle heavy users
  • Missing label mechanisms (MCAR, MAR, MNAR) and appropriate handling
  • Correlation within sessions and methods to adjust (clustering, mixed models)
  • Practical implications for metric stability and interpretability

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

Design the online experiment for this search feature. Cover your randomization unit, which metrics are primary vs. guardrail, what inputs you'd use for power and sample size, how you'd pre-register the analysis, and how you'd prevent p-hacking when you're looking at many different slices.

A/B Testing & ExperimentationProduct Analytics & MetricsTechnical Trade-offs
Author's notes

This part I felt most comfortable with.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the search feature and its goals, then propose a randomized controlled experiment with a well-justified randomization unit (e.g., user-level). Define primary success metrics and guardrail metrics, outline power analysis inputs, and describe pre-registration and p-hacking prevention strategies. Emphasize the importance of pre-registering the analysis plan and using techniques like Bonferroni correction or false discovery rate control for multiple comparisons.

Pro tip: At Meta, where experimentation velocity is high, it's crucial to balance statistical rigor with practical constraints; consider using sequential testing or variance reduction techniques like CUPED to detect smaller effects faster without inflating false positives.

1. Clarify the feature and define hypotheses

Ask clarifying questions about the search feature (e.g., ranking algorithm change, UI update) and articulate the null and alternative hypotheses for the experiment.

2. Choose randomization unit and metrics

Select the randomization unit (e.g., user, session, query) based on the feature and network effects. Define primary metrics (e.g., CTR, success rate) and guardrail metrics (e.g., latency, user satisfaction).

3. Determine power and sample size

Specify inputs: baseline metric values, minimum detectable effect (MDE), significance level (α), power (1-β), and variance. Use these to compute required sample size per variant.

4. Pre-register analysis plan

Document the analysis plan before launching: primary and secondary metrics, statistical tests, handling of multiple comparisons, and any subgroup analyses. This prevents p-hacking and ensures transparency.

5. Address multiple comparisons and slices

For many slices, use corrections like Bonferroni or Benjamini-Hochberg, or pre-specify a limited set of slices. Consider hierarchical models or holdout validation to avoid false positives.

Key Points to Mention

  • Randomization unit: user-level for most search features to avoid contamination, but consider query-level if user experience is not affected.
  • Primary vs. guardrail metrics: primary measures success (e.g., search success rate), guardrails ensure no harm (e.g., latency, abandonment).
  • Power analysis inputs: baseline conversion rate, MDE (often 1-5% relative), α=0.05, power=0.8, and variance.
  • Pre-registration: document hypotheses, metrics, and analysis plan in advance to prevent p-hacking and ensure reproducibility.
  • Multiple comparisons: use corrections (Bonferroni, FDR) or pre-specify slices; avoid data dredging.
  • Consider sequential testing or CUPED for efficiency and to handle peeking.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

Build out an offline evaluation pipeline: how do you collect labels, measure inter-rater agreement, maintain golden sets, and then monitor for label drift and Simpson's paradox when you segment results by query intent or locale?

Product Analytics & MetricsRoot Cause AnalysisData Modeling
Author's notes

Simpson's paradox question at the end caught me a bit off guard because I wasn't expecting it to get that specific.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Structure your answer around the full lifecycle of an offline evaluation pipeline: label collection, agreement measurement, golden set curation, and continuous monitoring. Emphasize how you would detect and address label drift and Simpson's paradox when slicing by query intent or locale, using statistical and visualization techniques. Tie everything back to ensuring reliable model evaluation and business impact.

Pro tip: Proactively mention that Simpson's paradox often arises when segment sizes shift over time, so you should monitor both segment-level and overall metrics, and use weighted averages or hierarchical models to reconcile them. Also, highlight the importance of documenting label guidelines and versioning golden sets to prevent silent regressions.

1. Design label collection and quality control

Define clear labeling guidelines, choose annotators (internal or crowdsourced), and implement quality checks like gold questions and consensus rounds. Use stratified sampling to ensure coverage across query intents and locales.

2. Measure inter-rater agreement and resolve disagreements

Compute agreement metrics (e.g., Cohen's kappa, Fleiss' kappa, Krippendorff's alpha) and set thresholds for acceptable reliability. Establish a adjudication process for disagreements and iterate on guidelines.

3. Curate and maintain golden sets

Build a representative, version-controlled golden set with high-quality labels, ideally from expert annotators. Periodically refresh it to reflect current data distributions and re-evaluate model performance.

4. Monitor for label drift and Simpson's paradox

Track label distributions over time and across segments (query intent, locale) using statistical tests (e.g., chi-square, PSI). For Simpson's paradox, analyze metrics at both aggregate and segment levels, and use visualization (e.g., scatter plots, heatmaps) to detect reversals.

5. Act on insights and iterate

When drift or paradox is detected, investigate root causes (e.g., changes in user behavior, labeling guidelines, or segment mix). Adjust the pipeline, retrain models, or reweight metrics as needed, and communicate findings to stakeholders.

Key Points to Mention

  • Label collection strategies: expert vs. crowd, stratified sampling by intent/locale, and quality control mechanisms.
  • Inter-rater agreement metrics (Cohen's kappa, Fleiss' kappa, Krippendorff's alpha) and their interpretation.
  • Golden set maintenance: versioning, refresh cadence, and ensuring representativeness.
  • Label drift detection: monitoring label distributions over time, using statistical tests like PSI or KL divergence.
  • Simpson's paradox: definition, how it can arise in segmented evaluation, and techniques to detect and resolve it (e.g., stratified analysis, causal inference).
  • Segmentation by query intent and locale: importance of consistent definitions and handling of small sample sizes.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.