From 155 Proposals to 45 Sessions — Without Losing a Gem

ESTO is the U.S. Travel Association's flagship education summit — the annual gathering where destination marketing leaders come to learn what's working, what's next, and what they should stop doing.

Every year, a volunteer committee reviews proposals, scores them, and builds a shortlist. It's the kind of work that depends on deep domain knowledge — and the committee is good at it. But reviewing 155 proposals on a 1–4 scale, in a spreadsheet, with limited time, means some decisions come down to pattern recognition rather than close reading. A strong proposal with a weak title gets skipped. A familiar topic gets the benefit of the doubt. An agency-submitted session gets cut on instinct, even when the content is solid.

The team also had context the committee didn't: agency restrictions, speaker limits per organization, sessions they'd already decided to cut for non-content reasons. That institutional knowledge lived in the team's heads, not in the spreadsheet — meaning the committee scored proposals without it, and the team couldn't easily backcheck every decision the committee made.

The team didn't need someone to replace the committee. They needed a way to pressure-test the committee's work — validate the cuts, challenge the assumptions, and surface anything worth a second look.

The EventEdIQ team received the full dataset — all 155 proposals, the committee's scores and comments, speaker organizations, session formats, and ESTO's content framework — and ran a six-part analysis:

The committee's instincts were largely confirmed. The AI's average score for their shortlisted proposals was 4.08 out of 5 — strong alignment.

But the analysis surfaced 30 proposals the committee had cut that scored 4.0 or higher. Eleven of them were in Innovation & Tech — a signal that the committee may have been more conservative on emerging topics than the content warranted. The team flagged several of those surfaced sessions as potential additions to the final program — or candidates to fill gaps later.

It also flagged 10 proposals in the shortlist that didn't meet ESTO's own 301-level standard — sessions that lacked actionable takeaways or didn't go beyond fundamentals. Having those flagged early gave the team a chance to work with speakers on strengthening the content before the event, rather than discovering the gap after the fact.

The combination analysis caught something else: after submitting their data, the team had independently decided to combine several of the same proposals the AI recommended — without coordination. When the recommendations align on their own, it builds confidence in both.

And the near-duplicate detection led to a concrete outcome. Two proposals in different tracks covered similar ground — "Using Major Moments to Lift the Whole State" and "Beyond the Host City." Rather than choosing one, the team decided to curate an entirely new session using speakers from both — something they wouldn't have seen without the cross-track view.

The team blocked time to review the analysis together — and finished quickly. The compliance flags caught a known sales pitch they were already managing, the oversaturated topic view helped them see where the program was heavy, and the 301-level analysis surfaced a session the team hadn't known how to handle — one they ended up recommending for the CEO's main stage instead.

But the real value wasn't any single finding. It was what three layers of review — internal team, volunteer committee, and AI — gave them when presenting to leadership. Content selections backed by three independent sources: internal review, an industry committee, and AI analysis. That's a stronger story for leadership — and a layer the team wants to keep using.

Most conference programming teams don't need to be told how to evaluate proposals. They've been doing it for years, and they're good at it. ESTO's volunteer committee is a coveted industry assignment — people want to be on it — and U.S. Travel has no intention of replacing it.

But when you're reviewing 155 submissions in a spreadsheet, the constraint isn't expertise — it's attention. You can't give every proposal the same close read. You make quick calls on the ones that don't grab you, and you spend your time on the ones that do. That's rational. But it also means the proposal with a bland title and excellent content gets skipped. The session that's solid but submitted by an unfamiliar speaker gets cut. The two proposals in different tracks that are basically the same talk never get compared.

AI scoring doesn't replace the committee's judgment. It extends it. Every proposal gets the same level of scrutiny — the same rubric applied the same way, without fatigue, familiarity bias, or time pressure. The committee still makes the decisions. They just make them with better information.

And as the team sees it, AI might change when the committee gets involved — not whether. In the future, AI could help streamline before the committee even sees the proposals. It's easier for them to react to a curated set than to review all 155 from scratch.