How the Vantage Forecast Works, and What It Will Not Tell You
By Evan Taylor, Founder
Every event platform can tell you how many people registered. That number is a fact about the past. The number an organizer actually needs is how many will walk in, and they need it while the catering order is still changeable.
This is a plain description of how ours produces that number: what it reads, what it hands back, when it gets sharp, and the questions it will not pretend to answer. The last part is the part most vendors skip.
The problem the number is solving
Registration is not attendance, and the gap is not small. Across in-person events the no-show rate averages around 32%, and it splits hard on price: free events run 40% to 50%, paid events 3% to 10%.1 Conferences sit near 15%.2
An average is not a forecast. Your October bootcamp in Dallas and your February workshop in Chicago do not share a no-show rate, and neither matches a category benchmark pulled off a blog. What you need is a number built from your event, your audience, and your history.
What the forecast reads
Six things, all of them either about your event or about the world your event happens in.
Registration velocity. Not the total, the shape. Fifty registrations in the last 72 hours means something different from fifty spread across six weeks, and late-surge events behave differently at the door.
Your own attendance history. Every check-in you run writes back to the registrant's record, so the model learns how your audience behaves rather than how audiences behave.
Repeat attendees. Someone who came to your last three events is not the same bet as a first-time registrant from a cold ad.
Weather on the event date. Pulled for your venue's location on your event's date.
Day of week. A Tuesday evening and a Saturday morning are different commitments.
Seasonal patterns. The same seminar in July and in October is not the same seminar.
The model also weighs how close the event is and how many registrants have confirmed, which is why the number moves as you get closer.
What it gives back
One attendance number, with a confidence score. The forecast returns an expected attendance figure and a confidence rating from 0 to 100, labeled High, Medium or Low. Confidence rises with registrant count, confirmed ratio, how recent your data is, and how near the event is.
A confidence score is not a margin of error, and we do not print one. Plenty of tools show a tidy plus-or-minus band. Ours does not, because the model does not produce a statistical prediction interval and inventing one would make the output look more precise than it is. You get the number and an honest read on how much weight it carries.
Per-registrant risk, once there is enough history. After your account has 50 registrant records with known outcomes, the model starts scoring individual registrants: a predicted show probability and a risk band per person, plus high, medium and low counts for the event. That is the difference between "expect 30 no-shows" and a list of who they probably are, which is a list you can actually do something about. Manager and Admin roles see it.
Vantage Points. The engine scores 19 signals on every event and surfaces the highest-importance ones first: registration spikes and slowdowns, cap projections, confirmation-rate movement, and the rest. How many you see depends on your plan: 1 on Free, 6 on Growth, 15 on Pro, all of them on Business.
It gets better the longer you use it
The first event on a new account is the hardest one to forecast, because the model has nothing of yours to learn from. It leans on general patterns. By the second and third event it has seen which of your registrant segments show up, which drift, and what your reminders actually move.
Real events, hundreds of guests, forecast attendance within 3%a, improving with every event.
What it will not tell you
This list exists because a forecast you cannot trust is worse than no forecast, and the fastest way to lose that trust is to overstate the machinery.
It does not know how far your registrants are driving. Travel distance and drive time are not inputs. Other write-ups of forecasting treat them as standard inputs. Ours does not read them.
It does not read email opens. The engagement signals it uses are ticket views, confirmation-page views and RSVP-page views, which are things the product actually records.
It does not break the forecast down by ticket tier. You get one number for the event, not a split across your standard, premium and VIP packages.
It does not re-forecast live during check-in. During the event you get a live check-in feed that updates as guests arrive. The re-forecasting happens for your next event, from what this one taught it.
It starts learning on your first event
The full model is the one that learns your audience, and it starts learning on your first event. You can start a 30-day Pro trial and run that event on it.
Sources
- Nunify, "Event Attendance Rate," 2025 to 2026. nunify.com/blogs/event-attendance-rate
- Who's In, "Event Attendance Statistics 2026," 2026. whos-in.app/research/event-attendance-statistics-2026
- Observed average error was 7.3% across three consecutive real events, improving with each event. Forecast attendance within 3% is the expected accuracy as the model learns from every closed event.