LinkedIn - Job Recommendation Redesign
Improving relevance & explainability for job seekers
The Situation
A product teardown of LinkedIn's job recommendation experience with a proposed redesign. The starting point was a platform I was using myself and finding frustrating.
What I Found
LinkedIn has more professional data than any platform in this space - skills graph, work history, learning history, network connections. But competitors like Indeed and Glassdoor outperform it on one thing: users know what they're getting and why. The gap wasn't relevance - it was explainability. A recommendation could be accurate and still feel wrong because nothing in the UI communicated why it appeared. Layered on top: promoted listings dominated the feed, the same company appeared multiple times, and a curated recommendation looked identical to a paid placement. That combination - opacity plus repetition - killed trust even when the underlying matching was reasonable.
What I Proposed
Two P0 features over the more technically ambitious ones: a "Why this job?" panel surfacing skills match, experience fit, and missing skills; and a layered company view that groups multiple roles from the same employer and surfaces only the most relevant. The tunable AI engine and LLM intent mapping went to P1/P2 deliberately - trust has to exist before users will engage with controls. The north star I settled on was Relevant Apply Rate, not raw applications. That framing forces optimization for quality over volume.
Reflection
A teardown taught me more about the product than using it every day had.