Sourcing produces more dashboard-able numbers than any other recruiting activity, most of them useless. Five metrics carry the signal.
1. Qualified candidates per role per week. The basic throughput number — candidates meeting the hiring bar, as judged by the hiring manager, not the sourcer. If this is under ~10 for a mainstream role, the constraint is upstream (search or bar-calibration), not outreach.
2. Reply rate. The honest measure of outreach quality. 2026 benchmarks: 5–15% for cold email at volume, 15–30% for well-personalized or warm-channel outreach, higher for opt-in marketplaces like Wellfound where intent is declared. Personalization quality moves this number more than channel choice.
3. Sourced-to-interview pass-through. What fraction of sourced candidates the hiring manager agrees to interview. Under ~30% means the hiring bar isn't calibrated between sourcer (or agent) and manager — the most common and most fixable sourcing failure. Agentic tools like Noon attack exactly this with explicit calibration loops.
4. Recruiter minutes per delivered candidate. The number that justifies (or kills) tooling spend. A hands-on stack might cost 45–90 recruiter minutes per qualified, engaged candidate; done-for-you and agentic models compress this toward review-only time. Multiply by loaded cost per hour before comparing tool prices.
5. Cost per sourced hire. The executive metric: all sourcing software plus loaded sourcing labor, divided by hires from sourced pipelines. Comparing this against agency fees (typically 20–25% of first-year salary) tells you whether the function earns its keep.
The three to drop: profiles viewed (activity theater), InMails/emails sent (volume without reply rate is spam measurement), and database size in vendor decks (every aggregator claims 700M+ profiles; coverage of your specific talent pool is the only version that matters — test it in a pilot).