The event registration s curve and what its shape tells you

A board pack came round in February with a registration chart on slide four. Someone had drawn a straight line from the launch point through the current position and extended it to show open, and written on track underneath. The line hit 10,800 at doors.

The show closed on 11,900. An event registration s curve is closer to a flattened S than a straight line, and that straight line had been wrong by nine per cent in the direction that would have made the operations team under-order, for a reason that had nothing to do with the market. Cumulative registrations do not grow at a constant rate, and any projection that assumes they do will be wrong twice: too optimistic in the quiet middle and too pessimistic in the last fortnight.

Why is the middle of the curve so flat?

Plot cumulative registrations against days to show open for any B2B exhibition with a long campaign and you get a flattened S. Registration opens and a block of people who were always coming register in the first fortnight, often with the previous edition's rebooking exhibitors pushing their guest lists through at the same time. The line rises fast, then it slows, and it stays slow for a long stretch in the middle where marketing is spending most of its money. Then it turns up again as the show approaches and finishes with a tail that keeps running through the doors on day one.

The opening block is bigger than most teams credit. On the show below, the first fourteen days of registration being open accounted for around nine per cent of the eventual file, and roughly two thirds of those records carried an exhibitor guest pass code. That is a pool of people who had already decided, arriving early because somebody handed them a free badge, and it tells you almost nothing about how the campaign will perform against people who have not decided.

The middle is where the trouble starts, because the middle is where the campaign lives. Weeks are passing, invoices are being paid, and the line is doing very little. A show director looking at that section without a reference curve has no way of knowing whether the flat is normal, which is a separate problem and one A4 takes on directly.

Maritz's Registration Insights Report 2024, built on more than 360,000 attendee registration records across 30 trade shows over three years, put 45 per cent of attendee registrations inside the final four weeks before the event. If close to half your file arrives in the last month, the middle of your campaign will look flat no matter how well it is going, and the shape of the curve is doing exactly what it should.

Normalise to share of final, not to counts

Comparing raw cumulative counts across editions only works if the editions are the same size, and they never are. A show that grew from 9,400 to 11,900 over five years produces five curves that never touch, and reading shape off them is guesswork.

Divide each edition's cumulative count at each day out by that edition's final registration number. Now every curve starts at zero and ends at one, and the y axis is share of final registrations. Editions of different sizes sit on top of each other and the only thing you are looking at is timing.

This has one obvious cost. You cannot compute the share until the edition has finished, so the normalised curve is a historic object. The current edition has no share of final, because its final is what you are trying to work out. That is the right way round. History gives you shape, and shape is what you apply to the live count.

Five editions of one show

Here is a packaging show, five consecutive editions, finals of 9,400, 10,100, 10,600, 11,200 and 11,900. The share of final registrations held at each day out:

Four of those columns are the same curve with noise on it. At day minus 60 the first four editions sit between 0.55 and 0.58, a spread of three points of share. The fifth sits at 0.47, eight points below the nearest of the others, which is several times the spread of the rest.

The 2025 curve did not fail. That edition finished on 11,900 registrations, 6.3 per cent above the 11,200 of 2024. What changed was when the registrations arrived.

What was the broken year actually telling you?

Work the day minus 60 comparison in counts. The 2024 edition held 0.55 of 11,200, which is 6,160 registrations. The 2025 edition held 0.47 of 11,900, which is 5,593. The pacing index at that point was 5,593 divided by 6,160, times 100, which is 91. Anyone reading that index in isolation had a show running nine per cent behind, and the show closed six per cent ahead.

The cause was in the operations diary. Registration for the 2025 edition opened 34 days later than it had for the previous four, because the new platform migration ran over. The whole front of the curve was compressed into a shorter window and the shape shifted right by about a month.

Once you know that, the eight-point deficit at day minus 60 becomes readable. The 2025 campaign had 34 fewer days in which to accumulate anything, and the four prior editions had banked between five and seven points of share in the equivalent window at the front. The gap at day minus 60 is roughly the size of the front you never had. That is a testable statement, and testing it is cheap: take the prior editions, cut off their first 34 days of registration, rescale, and see whether the truncated curves land near 0.47 at day minus 60. If they do, the 2025 shape is explained and the correct report to the show director is that pacing is meaningless this year until the curves reconverge.

The consequence landed on the wrong team. From day minus 30 to doors, the 2024 edition added 0.25 of its final, which is 2,800 registrations. The 2025 edition added 0.34 of its final, which is 4,046. That is 1,246 more people arriving in the last month than the operations plan had ever had to absorb, and nothing in the pacing report said so, because the pacing report was reading a total. What a heavy tail does to badge stock, catering guarantees and floor staffing is A3's subject and it is the part of this that costs money.

The shape is a property of your audience

There is no universal registration curve, and borrowing someone else's is worse than having none, because it gives you a false reference.

The same Maritz 2024 analysis found the sector spread directly. Medical and healthcare conferences had 29 per cent of registrations arriving in the final four weeks, against an overall average of 45 per cent. Food and restaurant shows had 54 per cent inside the same window. Those are different shapes, and a food show benchmarked against a medical conference's curve would spend the last two months of every campaign in a panic it had no reason to be in.

Within your own portfolio the drivers are visible if you look for them. International share pushes the curve earlier, because a visa or a long-haul flight is booked further out. Drive-in local audiences push it later. Price tier boundaries create steps, and each step is a decision your commercial team made rather than a fact about the market. First-time registrants land later than returning ones, which means a show that is successfully recruiting new audience will see its curve flatten in the middle and steepen at the end, and will look like it is failing right up until it is not. Every one of those is a field on the registration record, which is what makes curve shape a question attendee analytics can settle on your own file.

Freeman's May 2026 report on non-attendance is a useful caution against generic benchmarks here. It drew on roughly 20,000 survey responses, and conferences made up 72 per cent of them against 28 per cent for trade shows. Any pooled figure across event types is weighted towards whichever type dominates the sample, and your exhibition is not the average of that blend.

Where reading the shape stops working

The method needs finished editions, and it needs enough of them. Two editions give you a shape and no sense of how much a shape naturally varies. The eight-point gap that mattered above was only interpretable because four other editions sat within three points of each other.

It also assumes each historic edition's final number is a fixed, known quantity. It is not, quite. Cancellations processed after the show, deduplication runs, and late data merges all move a final number months after doors closed, and every one of those moves rescales the entire curve for that edition. Fix a rule, apply it identically to every edition, and write it on the chart.

The deeper limit is that shape is silent about composition. Two editions can have identical curves while one of them replaced 1,200 senior buyers with 1,200 students. The share of final registrations at day minus 60 will not flicker. Reading the curve tells you about timing and nothing else, and the moment you use it as a proxy for health you have started making a claim the data does not support.

There is also a genuine floor on precision. With five editions you have five observations of the share at any given day out, and a median of five numbers is a weak estimate with a wide interval around it. Building the band properly, with percentiles, is A7's job, and how far you can push a parametric fit on five points belongs with O11.

Start with one show and one afternoon. Pull the final registration count for each of the last five editions, then take every registration record with its created date, compute days to show open against that edition's opening morning, and build the cumulative share for each edition at day minus 90, 60, 30 and 14. Put the five columns side by side. If one of them is more than a few points of share away from the rest at the same day out, go and find out what your own team did that year, because the answer is almost always in the diary rather than in the market.