Pitch competitions
What the written round costs today#
| Figure | What it is |
|---|---|
| 300+ | Volunteer judges one flagship university competition publicly recruits, every year |
| 4–6 h | The real homework in one screening assignment: three written entries with feedback write-ups |
| ~5% | Share of the pool the average judge actually sees at a large event, so scores barely compare |
| 0 | Teams who see written-round feedback at most major competitions today |
The six steps#
1 · Rubric and tracks, locked. Criteria, weights and tracks are configured per competition. Publish a methodology statement in the competition rules — where fairness claims legally live — stating what the panel assists with and what the judges decide.
2 · Entries land in one place. A link or QR, open or invite-only, with a window and live statuses. Completeness is checked automatically, so staff chase exceptions rather than the pile. Existing intake keeps working if you are not replacing it this season. See Collect submissions.
3 · The panel reads every entry. Six independent reads score the whole pool on your rubric. One panel for the whole pool means no judge-assignment lottery in the written round: entry #1 and entry #300 are read under identical rules.
4 · Screeners confirm with packs, not piles. Prelim decisions stay human. Remote judges and alumni keep their role, but screening becomes a confirm-and-comment pass over evidence-linked reports instead of a reading weekend — the same touchpoint at a fraction of the ask, which is why the same judges say yes again next year.
5 · Live rounds run as designed. Semis and finals: same stage, same judges. Judges score as usual; the leaderboard is built from human scores times your weights; disagreements surface for deliberation rather than being averaged away.
6 · Feedback for every team. Structured feedback drafted from the evidence and reviewed by your staff before it is sent. For a student competition, feedback for 400 teams is a teaching outcome, not just an event.
The judge-hour arithmetic#
400 entries × 3 reads × 1.5 hours = 1,800 judge-hours consumed by a written round. Those hours are what moves to semis, finals and mentoring — while every entry still gets every page read.
Nothing here reduces judge count. Judge count is a program KPI and a sponsor perk; what changes is the ask. Nobody sponsors an event to do six hours of reading homework — sponsors come to see finalists and be seen.
What to say, and where#
- On stage: "Every entry received a full read under identical rules, and humans made every ranking decision."
- In the rules document: a methodology paragraph — what AI assists with, what humans decide, how a team can ask about its own record.
- On the application form: plain-language disclosure, so nobody discovers AI involvement after results.
- On conflicts of interest: your policy stays your policy. The record simply logs who scored what, which is what makes a recusal verifiable later.
Two product defaults worth knowing#
- A team can request its own record.
- No entry is eliminated without a named human sign-off.
Whether you run open appeals is your policy. The per-entry evidence, scores and decisions make either policy a five-minute conversation.
Validating before it counts#
Run a shadow pilot: your screeners work one written round as usual while the panel reads the same pool in parallel. Compare coverage, hours, and where the pre-ranking agreed with your screeners — validated against your own judges before it influences anything.
Institutional requirements#
Student submissions are never used to train models — contractual. The institution owns every report, score and decision log; retention and deletion follow your policy, student-data handling is structured to support FERPA obligations, and a DPA is available. PO and invoice are accepted, vendor registration forms and security questionnaires supported, and the sub-processor list is public.
Next steps#
- Brief your jury — the pack that raises agreement before scoring.
- Criteria and weights — the default rubric and the lock.
- Hackathons — the execution-weighted variant of the same round.