Product design question

Design a friend recommendation system for Facebook.

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Technical and product thinking on recommendation systems, balancing accuracy with privacy and trust concerns.

Users: people with sparse social graphs who benefit most, and people who find suggestions creepy or invasive.
Pain points: mutual friends alone miss real-world connections, and over-aggressive suggestions raise privacy discomfort.
Signals: mutual friends count, shared groups or events, contact list overlap, and location or workplace proximity.
Prioritize signals by strength and privacy sensitivity, weighting mutual friends and shared groups higher than more invasive signals like location.
Design the ranking: combine signals into a score, then apply a diversity and freshness filter so suggestions do not feel stale or repetitive.
Success metric: friend request acceptance rate from suggestions, with a guardrail on user reports of unwanted suggestions.

Explicitly weighs privacy sensitivity when choosing signals, not just predictive power, appropriate for a sensitive recommendation surface.
Names concrete signals, mutual friends, shared groups, and contact overlap, grounded in how real friend-suggestion systems work.
Sets both an effectiveness metric, acceptance rate, and a trust guardrail, unwanted suggestion reports.

Using only mutual friend count with no other signals, producing weak recommendations.
Ignoring privacy risk from signals like location tracking.
No guardrail metric for user discomfort or reports.

Chapters of the AI PM course that teach what this question tests.