Why most events over 25 people are not designed to produce outcomes. Ten findings from the first at-scale measurement of gathering design: 23,624 public event agendas, 316,643 agenda blocks, scored with the published Gathering Effectiveness Score rubric. Not a survey. The agendas themselves.
Across 23,624 scored public agendas, the average Gathering Effectiveness Score is 21.8 out of 100. The average gathering publishes who speaks and when; almost nothing about what should change because people met.
Zero of 23,624 records score above 60 under the strict public-evidence rubric. Even the strongest public agendas, scoring in the 50s, stop short of visible outcome architecture: owners, follow-up, and proof.
21,781 of 23,624 agendas show no visible follow-up, tracking, or impact evidence of any kind. Whatever momentum a gathering creates, the public record almost never shows a mechanism to keep it.
Follow-through averages 5.8 out of 100 corpus-wide, and between 5.3 and 6.8 in every category we measured, from tech summits to nonprofit convenings. No field has solved the bridge from the room to what happens after it.
99.4 percent of the top five percent of agendas include visible participant work, versus 9.1 percent of the bottom half. The strongest gatherings put people to work; the weakest put them in rows.
2,854 agendas are stage-only: passive formats end to end, with no visible participant work or commitments. In the AI era, broadcast information is the one thing a room full of people is least needed for.
Network design averages 17.5 out of 100. Most agendas offer proximity (receptions, breaks) rather than engineered connection: matched introductions, peer circles, or ask-and-offer structures.
Nonprofit and social-impact events average 27.6, the highest of any category; trade shows and expos average 17.8, the lowest. The spread shows design culture, not budget, drives outcome architecture.
The 1,000 hand-verifiable gold-set events average 37.8, fifteen points above the corpus. Their edge comes from participation and learning transfer; even they barely register on follow-through.
Before scoring anyone else, United We Transform gathered 1,245 registrants and ran fifteen live experiments in one day, recording 7,545 networking interactions. We published every result, including the failures: only 48 percent of AI networking recommendations were rated positive. Measuring the field honestly starts with measuring yourself honestly.
Every finding comes from the public agendas of real events, captured from the open web and scored with a deterministic, published rubric (version ges_agenda_judgment_v2_0) across eight pillars: participation architecture, follow-through, problem specificity, personalization, network design, learning transfer, evidence maturity, and future-of-work fit. Scores are opinions from visible public evidence: a missing signal in a published agenda is not proof the event lacked it. Source-backed evidence is separated from model inference throughout, a human validation study is underway against a stratified 120-event gold set, and every scored record links to its source. Full methodology, limitations, and the corrections process: the trust boundary.
United We Transform (2026). The State of Gatherings 2026: ten findings from 23,624 public event agendas. unitedwetransform.com/reports/state-of-gatherings-2026/