Diagnose a single-day drop in Facebook birthday greetings

Quick Overview

A metric-diagnosis question for the product manager of Facebook Birthdays after greetings fall sharply on a single day. It tests validating the data, choosing a fair baseline, decomposing the greeting funnel, segmenting by platform and region, ordering hypotheses and deciding on a response.

Diagnose a single-day drop in Facebook birthday greetings

Company: Meta

Role: Product Manager

Category: Analytics & Experimentation

Difficulty: medium

Interview Round: Other

You are the product manager for Birthdays on Facebook: the reminders that tell people a friend has a birthday and the places where they can post or send a greeting. On one particular day, the number of birthday greetings drops. Walk through how you would find out why it happened and decide what to do. ```hint Use the shape of the drop A fall confined to a single day points to different causes than a slow slide. Pin down what that day would normally have looked like, then break the greeting count into pieces you can check one at a time. ``` ### Clarifying Questions - What counts as a greeting: a post on the person's profile, a comment, a message sent from a birthday reminder, or all of these? - How large is the drop, and against which baseline: the day before, the same weekday in recent weeks, or the same date last year? - Is it spread across every country, platform and app version, or concentrated in a few of them? - Did the number recover the next day, or is that day the start of a lower level? ### What a Strong Answer Covers - Checking that the drop is real: logging and pipeline health, metric definition changes, and a baseline that respects weekday and calendar effects. - A decomposition of greetings into birthdays that day, friends reminded, reminders seen, share who greet and greetings per greeter, used to locate where the loss happened. - Segmentation by platform, app version, region and entry point to separate a release or outage from a behavior change. - Internal and external hypotheses checked in a sensible order, each with the data that would confirm or rule it out. - A response tied to the cause (roll back, fix, accept, or redefine the metric) and a way to catch the next drop sooner. ### Follow-up Questions - The whole drop is on Android, in the newest app version. What do you do in the next hour? - Greetings posted on profiles fell, but birthday messages sent privately rose by about the same amount. Is that a problem? - How would you set up alerting so a drop like this is noticed the same day rather than the next week?

Overview: A metric-diagnosis question for the product manager of Facebook Birthdays after greetings fall sharply on a single day. It tests validating the data, choosing a fair baseline, decomposing the greeting funnel, segmenting by platform and region, ordering hypotheses and deciding on a response.

Meta logo
Meta
Mar 9, 2025
mediumProduct ManagerOtherAnalytics & Experimentation
0
0

You are the product manager for Birthdays on Facebook: the reminders that tell people a friend has a birthday and the places where they can post or send a greeting. On one particular day, the number of birthday greetings drops. Walk through how you would find out why it happened and decide what to do.

Clarifying Questions Guidance

  • What counts as a greeting: a post on the person's profile, a comment, a message sent from a birthday reminder, or all of these?
  • How large is the drop, and against which baseline: the day before, the same weekday in recent weeks, or the same date last year?
  • Is it spread across every country, platform and app version, or concentrated in a few of them?
  • Did the number recover the next day, or is that day the start of a lower level?

What a Strong Answer Covers Guidance

  • Checking that the drop is real: logging and pipeline health, metric definition changes, and a baseline that respects weekday and calendar effects.
  • A decomposition of greetings into birthdays that day, friends reminded, reminders seen, share who greet and greetings per greeter, used to locate where the loss happened.
  • Segmentation by platform, app version, region and entry point to separate a release or outage from a behavior change.
  • Internal and external hypotheses checked in a sensible order, each with the data that would confirm or rule it out.
  • A response tied to the cause (roll back, fix, accept, or redefine the metric) and a way to catch the next drop sooner.

Follow-up Questions Guidance

  • The whole drop is on Android, in the newest app version. What do you do in the next hour?
  • Greetings posted on profiles fell, but birthday messages sent privately rose by about the same amount. Is that a problem?
  • How would you set up alerting so a drop like this is noticed the same day rather than the next week?