← Meta Interview Insights

Meta·Software Engineer·Technical Phone Screen·Senior

Senior
Jul 2026

Summary

Meta data engineering interview, looked like a metrics design question dressed up as a simple 'define success' prompt but it went pretty deep pretty fast. The whole thing was one dense question with a lot of moving parts.

Questions Asked (1)

Q1

How would you define success metrics for a social feed feature? Walk through the primary metric with an exact formula, guardrail metrics, the eligible population and unit of analysis, how you'd handle bots and outliers and time zones, how you'd compute each metric from an events table, and the trade-offs between engagement, quality, and reliability.

Product Analytics & MetricsData ModelingTechnical Trade-offs
Author's notes

This question is basically five questions stapled together and they want you to treat it like one coherent answer.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the feature's goal (e.g., meaningful social interactions) and define a primary success metric that captures that goal, such as Daily Active Users (DAU) or a more specific engagement metric like 'Daily Engaged Users' (users who like, comment, or share). Then walk through the metric's exact formula, guardrails, data handling, and trade-offs, ensuring you cover all requested aspects in a structured manner.

Pro tip: Emphasize that metrics must be actionable and aligned with long-term user value, not just short-term engagement; mention how you'd validate the metric with A/B tests and monitor for unintended consequences.

1. Define the primary success metric

Choose a metric that reflects the feature's core goal, such as 'Daily Engaged Users' (users who perform at least one meaningful interaction). Provide an exact formula, e.g., DEU = count(distinct user_id) where event_type in ('like','comment','share') and date = current_date.

2. Identify guardrail metrics

Select metrics to ensure the feature doesn't harm other aspects, such as user retention, time spent, or report rate. For example, 7-day retention rate and average session length.

3. Define eligible population and unit of analysis

Specify who is included (e.g., users who have seen the feed at least once) and the unit (e.g., user-day). Exclude new users during onboarding or users in experiments.

4. Handle data quality issues

Describe how to filter bots (e.g., using user agent or behavior patterns), treat outliers (e.g., winsorize or exclude extreme values), and normalize time zones (e.g., convert all timestamps to UTC or user's local time for daily aggregation).

5. Compute metrics from events table and discuss trade-offs

Write SQL-like pseudocode for each metric, e.g., SELECT date, count(distinct user_id) FROM events WHERE event_type IN (...) GROUP BY date. Then discuss trade-offs: optimizing for engagement might reduce quality (e.g., clickbait) or reliability (e.g., system latency), so balance with guardrails.

Key Points to Mention

  • Primary metric formula: DEU = count(distinct user_id) with event_type in ('like','comment','share') per day.
  • Guardrail metrics: retention, report rate, session length, and system performance (e.g., latency).
  • Eligible population: users who have seen the feed; unit of analysis: user-day.
  • Bot filtering: use heuristics like rapid actions, unusual patterns; outlier handling: winsorization or percentile caps.
  • Time zone handling: convert timestamps to UTC or user's local time for daily aggregation.
  • Trade-offs: engagement vs. quality (e.g., clickbait) vs. reliability (e.g., system load); use A/B tests to validate.

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