Design LinkedIn's Suggested Connections: Goals, Signals, Ranking and Metrics

You are a product manager at LinkedIn. Explain how you would build the feature that recommends people a member may know and might want to connect with: what it should achieve, which members it should serve, which signals it should use and how suggestions are ranked, how members see and act on suggestions, and how you would measure whether it works. ```hint Define a good suggestion for both people A LinkedIn connection needs one person to send an invitation and another to accept it. Decide what makes a suggestion valuable to both of them, and to LinkedIn, before choosing any signals. ``` ### Clarifying Questions - Are we building the feature from scratch or improving the existing one, and on which surfaces: the My Network page, onboarding after sign-up, profile pages or email? - Should the first focus be new members with almost no connections, or established members with large networks? - Is success measured by invitations sent, invitations accepted, or what happens after two members connect? - Which data may we use, for example imported address books, shared employers and schools, or profile views? ### What a Strong Answer Covers - A goal tied to LinkedIn's mission and business, with a definition of a good connection that works for both the sender and the recipient. - Member segments with different networking needs and a reasoned choice of where to focus first. - Candidate sources and ranking signals, prioritized, with a clear position on data that must not be used for privacy, fairness or legal reasons. - The member experience: where suggestions appear, how they are explained, how invitations are sent and dismissed, and how a member with no network gets useful suggestions. - Two-sided success metrics, guardrails against spam and unwanted invitations, and how to run experiments when a treated member's invitations reach people outside the test group. ### Follow-up Questions - An experiment shows that suggesting more distant contacts lowers the acceptance rate but leads to more job changes later. Which version do you ship, and why? - A member complains that LinkedIn suggested someone they never want to see, such as a former partner or their doctor. How would you prevent sensitive suggestions without making the feature useless? - How would you stop heavy users, such as recruiters, from turning the feature into a mass invitation tool?