Problem
Learners on Coursera run millions of searches a day. Usually they find what they need, but sometimes the right course does not exist, or it exists and is not good enough. The content strategy team already used market trend reports, external demand signals and partner feedback. None showed what learners search for on the platform and where they come up empty, because search logs had never been systematically used in content acquisition decisions.
Example from the article: a learner searches for "Data Visualization" and the catalog has good courses, but they sit on page 10 of the results, so the learner clicks nothing. This is a ranking problem, not a missing-content problem (the article marks its examples as hypothetical).
Solution in brief
Coursera built COIS (Content Opportunity Index via Search). It gives every query a saturation score from 0 to 1, assigns a gap type that points to an owning team, and multiplies the gap size by a business opportunity score to get a priority. LLM-based intent and skill labels are computed from the query text itself, not from the results.
Questions this case answers
- How to detect unmet demand in a catalog from user search queries and prioritize what to add or fix by expected impact
- How to classify the reason a query is poorly served so each cause goes to the team that can fix it
- How to find which search queries our catalog fails to serve, before users give up?
- How to tell whether a failed search needs new content, better ranking, a different difficulty level, fresher material or a partnership?
Tools and methods
- Saturation score — composite 0–1 estimate of how well the catalog serves a query
- Gap-type classification — content, quality, depth, freshness and partner gaps, each routed to an owner
- Priority score — gap size times business opportunity
- LLM intent and skill classification — difficulty, preferred format and skill inferred from the query
Google or a one-time email code. No password.