What Linear Learned About Preserving Demand Signal in Duplicates
TRIGGER
Feature requests arrived as individual issues from multiple customers through different channels, making it difficult to recognize when many users were asking for the same thing—demand signal was fragmented across dozens of separate tickets rather than aggregated into a clear prioritization input.
APPROACH
Linear's Triage Intelligence detects duplicates across incoming issues, which is 'particularly useful for aggregating feature requests' according to the team. When a new feature request arrives, the system identifies similar existing requests. Combined with Customer Requests, this links user feedback, company information, and original discussion records to Linear issues or projects. The product goalie reviews aggregated feedback and adds notes or context. Input: new feature request issue. Output: request linked to existing feature issue/project with customer and company metadata attached, surfacing total demand signal.
PATTERN
“Merged duplicates lose the customer metadata that makes prioritization defensible. Each linked request preserves a vote count with company context attached—the evidence that turns "we should build X" from opinion into data.”
✓ WORKS WHEN
- Feature requests come from multiple customers through multiple channels (support, social, community)
- Prioritization decisions benefit from knowing how many customers want something and who they are
- Requests use varied language for the same underlying need (semantic matching helps)
- Product team reviews aggregated requests periodically for roadmap planning
- Customer metadata (company, tier, use case) is captured and preserved with requests
✗ FAILS WHEN
- Feature requests are rare or come from a small, known set of customers (manual tracking sufficient)
- Requests are highly specific and rarely overlap (custom enterprise requirements)
- Prioritization is driven by strategic vision rather than customer demand aggregation
- Semantic similarity misses the point—superficially similar requests have different underlying needs
- No process exists to review aggregated requests, so linked duplicates never influence decisions