What Has to Be True at 8:15 on Day One

Your event gets judged in the thirty-five minutes between 8:15 and 8:50 on day one, by people standing in a line, and almost nothing you do afterward changes the verdict.

8:15 on day one

Doors opened at 8:00. The keynote is at 9:00, which means every person who matters wants to be seated by 8:50, which means your entire arrival curve lands in about thirty-five minutes no matter what the pre-registration report said. There are six check-in stations, two of them staffed by people who were trained yesterday afternoon.

At 8:15 the following is true at once. A VIP is at station three and is not in the list, because she registered under her assistant's email. The printer at station five has jammed on badge 240 and the queue behind it has stopped moving, and the person working it does not know that pulling the ribbon the wrong way will cost four more minutes. Lead retrieval licenses for a sponsor on the far aisle stopped syncing overnight and the first exhibitor has not noticed yet, but will, loudly, at 10:30. And the line is now long enough to be visible from the escalator, which is the moment attendees start forming an opinion about the whole event.

Every one of those errors is public, compounding and unfixable after the fact. You cannot send a correction email for a forty-minute line. The person who waited in it has already decided what they think, and your post-event survey will tell you about it in eleven days.

That hour is the real specification for onsite software. Not the feature list, and not the demo.

What has to be true

The standard is ninety seconds. Onsite software is good enough when a problem can be noticed and acted on inside the ninety seconds an attendee will tolerate before they start looking around for a staff member. That is the window where an intervention still counts as service rather than apology.

Three things have to be true to work inside it.

1. It has to see the floor, not the database. Throughput per station, queue depth, printer state, scan rate per device, license sync status. Most registration platforms do not instrument any of this in real time, because historically nobody asked them to, and you cannot act on what you cannot see.
2. The action has to be available where the problem is. Reroute the queue, reassign a station, re-provision a dead lead retrieval license, send a walk-up to a station with a working printer. If any of that requires a laptop in the back office or a support ticket, it does not happen before 8:50.
3. It has to fail into something a person can run. Degrade to a state the twenty-two-year-old you trained yesterday can take over in four seconds. Onsite software that fails into a frozen screen is worse than the clipboard it replaced.

That is the whole brief. It is also a useful filter for everything the category is currently announcing.

The pre-event half is the forgiving half

Event platforms have spent this year shipping agentic features, and they rhyme. An agent that drafts and edits your event website. One that runs a pre-launch health check on your setup. One that builds a session schedule from a prompt. One that finds the past attendees who have not registered yet and re-engages them. One that reads registration pacing against your own history and tells you what to do about it.

Taken on their own terms these are good products, and the pre-launch health check in particular is a real job that real people do badly under deadline. Notice where all of them live: the weeks and months before the event, where mistakes are cheap, slow and reversible. If the scheduling agent writes a bad session description, you fix it Tuesday. If the audience agent emails the wrong list, you send a correction and absorb a small amount of embarrassment. You have days of runway, the error surface is a screen that one person is looking at, and nobody standing in front of you is affected.

That is the right place to start, and it is why the demos look so clean. It also means none of it has been tested against the specification above. Pre-event software gets to be wrong quietly.

At 8:15 your staff is the loop

Every one of these features ships with the same assurance: nothing goes live until a person approves it. Before the event that assurance is real, and it is worth being precise about what it is. A human who genuinely reviews every change is a human doing the work of reviewing every change. The software drafted faster and the bottleneck moved rather than left. That is a real gain, and it is not autonomy.

At 8:15 the assurance stops meaning anything, for a dull and sufficient reason: nobody has attention to spare. Your ops lead is on her knees behind station five with a ribbon in her hand. Your registration manager is walking the VIP to the front of the line. The person whose approval is required is the person already elbow-deep in the problem, and the approval queue is a list nobody opens until 11:00.

So onsite, your staff is not supervising the loop. They are the loop. That sets the job for the software plainly: make the four people at that desk faster and better informed, inside ninety seconds, and fail into something they can run without you. It is a smaller claim than an autonomous event agent, and it is the one that decides what the line looks like at 8:50.

The useful question for any vendor, ours included, is not what the AI can do. It is this: when it goes wrong at 8:15, who is the human in the loop, and what else is that person doing at the time.

Where we stand

We build against that hour, which is why the Expo Pass onsite stack is deliberately boring hardware. Off-the-shelf iPads, a router, a badge printer, a kiosk a temp can be taught in an afternoon, and failure modes that hand a person back control instead of showing them a spinner. The standard above is the one we hold ourselves to, and the one we would rather be measured on than a feature grid.

Real-time floor telemetry is what comes next, and it comes before anything agentic, because software that acts on its own in a hall with printers, badge stock, convention center wifi and a union labor schedule has to be right the first time in front of an audience. Instrumentation first, autonomy second, in that order.

So when the next event agent gets announced, and one will before Christmas, ask where it works. If the answer is the planning window, it is a useful product. Whether any of this works in events gets answered at 8:15 on a Tuesday morning in a hall in Orlando, with a line out the door and a printer down, and not in anybody's booth demo.

Sources

• AI Agents, the Joy Factor: Our 7 Takeaways From IMEX America 2026, Event Marketer, October 16, 2026

This article was written with the help of AI and edited by humans. Graphics were generated with the assistance of AI.