Running a year over year registration comparison that holds up

Slide two of the pre-show pack has one number on it. Registrations are down 8 per cent on last year. The show director asks what is causing it, and the honest answer is that nobody knows, because a single delta cannot have a cause.

A year over year registration comparison that stops at one delta cannot be acted on. An 8 per cent gap is a summary of thousands of individual differences that partly cancel. Some of them are people who came last time and have not registered yet. Some are people who have never heard of the show and now have. Some are exhibitor staff badges that were issued in a different week this year. The number on the slide is the sum, and the sum is the least useful form the information comes in.

The gap is four gaps

Take the show above, read at day minus 45 in both editions so the comparison is at the same distance from doors rather than the same calendar date. When the show itself has moved in the calendar, that alignment needs its own rebasing rule, which is A6's.

The prior edition held 8,000 registrations. This edition holds 7,360. The gap is 640 registrations, which is 8.0 per cent of 8,000.

Split the file four ways. Returning buyers went from 4,120 to 3,720, down 400. New buyers went from 2,180 to 2,330, up 150. Exhibitor staff went from 1,280 to 990, down 290. Press, students and everything else went from 420 to 320, down 100.

Check the arithmetic before you believe it. The prior edition's four parts sum to 8,000 and this edition's sum to 7,360. The four differences are minus 400, plus 150, minus 290 and minus 100, which sum to minus 640. A decomposition that does not add up is a decomposition with a definition problem in it, and finding that out here is much cheaper than finding out in the meeting.

Now read it. The buyer side, meaning returning plus new, went from 6,300 to 6,050. That is down 250, or 4.0 per cent. The headline said 8 per cent. More than half of the gap that started the conversation is exhibitor staff and press, and exhibitor staff registrations are largely a function of when your exhibitor services team opens the badge portal.

Two different reports come out of the same file. One says the show is down 8 per cent. The other says the buying audience is down 4 per cent, driven by 400 missing returning buyers against 150 additional new ones, with a separate operational question about why 290 fewer exhibitor staff badges have been issued. The second report is the one somebody can act on by Friday. Whether the mix of registration types itself has drifted underneath all four lines is A31's question.

Freeze the definition of returning before you count anything

The split above rests on a definition, and the definition does more work than the data.

Count someone as returning if they appear in any of the previous three editions and this file gives you 3,720 returning this year against 4,120 last year, a shortfall of 400. Count someone as returning only if they appear in the immediately preceding edition and the same file gives you 2,980 against 3,240, a shortfall of 260, with the new group correspondingly larger and essentially flat year on year at 3,070 against 3,060.

Same registrations, same two editions, two stories. One says you have lost 400 returning buyers and gained 150 new ones. The other says returning is down 260 and new is unchanged. Neither is wrong. They answer different questions, and if the definition changes between one pre-show pack and the next, the trend line in the pack is measuring your own definitions.

Pick one, write it in a sentence at the top of the report, and keep it for at least three editions. Longer lookbacks suit shows with a long purchase cycle where a buyer attends every other year by design. Shorter lookbacks suit annual shows whose exhibitors renew on the strength of who came last time. How to measure the returning cohort properly, including what to do about people who registered and did not attend, belongs to A37.

Freeman's end of year trends recap, released in January 2026 from its 2025 research, reported a blended industry retention rate barely above 30 per cent, with the most optimistic historical data it could find putting the figure in the low 40s. On that base rate, roughly seven in ten of the people at your last edition were never going to come back, whatever you did. A shortfall of 400 returning buyers is a large number against a cohort that small, which is the argument for reporting it separately rather than letting it disappear inside a total.

Why does a mid campaign split understate your new audience?

There is a timing problem sitting under every new versus returning comparison read before the show opens, and it runs in a consistent direction.

Maritz's Registration Insights Report 2024, drawn from more than 360,000 attendee registration records across 30 trade shows over three years, found that almost twice as many first time attendees register late as repeat attendees, 47 per cent against 25 per cent, with 75 per cent of veteran attendees registering early. New registrants are the people who have not decided yet, so they decide later.

Apply that to the numbers above. If 47 per cent of your eventual new cohort arrives in the final four weeks, then at day minus 45 you are looking at roughly 53 per cent of the new registrants you will end up with. If only 25 per cent of returning registrants arrive that late, you are already looking at about 75 per cent of them.

Gross both editions up on those shares. This edition's 2,330 new registrants imply an eventual 2,330 divided by 0.53, which is 4,396. The prior edition's 2,180 imply 4,113. New is running about 283 ahead rather than 150 ahead. This edition's 3,720 returning imply 3,720 divided by 0.75, which is 4,960, against the prior edition's 5,493. Returning is running about 533 behind rather than 400 behind.

The net barely moves, and both halves get about a third larger. The composition question sharpens while the headline stays put, which is a good description of what a decomposition is for.

Treat the grossed up figures as an illustration of direction rather than a forecast. The 47 and 25 per cent shares come from someone else's 30 shows, and yours will differ. Compute the two shares on your own last three editions before you use them for anything that matters.

Cut it by geography before you cut it by anything else

Of the four cuts available on a registration file, geography is the one that most often finds the cause, because the causes of a lost cohort tend to be geographic: a visa regime, an airline route, a currency move, a competitor's show in the same region.

Take the 400 missing returning buyers and split them. Domestic returning fell from 3,090 to 2,880, down 210. International returning fell from 1,030 to 840, down 190. Those two look similar until you divide. The domestic cohort is down 6.8 per cent. The international cohort is down 18.4 per cent, nearly three times the rate, on a base less than a third the size.

Half the missing returning buyers came from a group that supplies a quarter of them. That is a finding with a plausible list of causes and a short list of people to ring. An 8 per cent headline has neither. Reading international share properly across editions, including what to do when the travel cost base moves, is A35's subject, and the full geographic cut of a registration file is A34's.

What does the decomposition actually change?

The point of doing this before the show rather than after it is that the four numbers route to four different people.

Missing returning buyers are a retention problem and they belong to whoever owns the database and the reactivation campaign. Additional new buyers are a channel result and they belong to the acquisition team, who should be asked which channel produced them. Missing exhibitor staff badges are an operations question about the badge portal and the exhibitor manual. The press and student line is usually a policy change somebody made and forgot to circulate.

A single delta routes to nobody, which is why it generates a meeting instead of a task. Getting a number to the person who can move it is most of what attendee analytics is for.

Where this stops

Everything above assumes you can tell whether a registration in this edition is the same person as a registration in a prior one. That is a matching problem, and on a file where people change employers, use different email addresses and get registered by assistants, the returning cohort you measure is a lower bound on the returning cohort you have. A poor match rate manufactures exactly the finding people fear, which is a collapse in returning registrations, and it does it silently.

The second limit is that historic editions keep changing. Cancellations processed after the show, deduplication runs and late data merges all move a prior edition's counts months after doors closed. If last year's 8,000 is a cleaned number and this year's 7,360 is a live one with the duplicates still in it, part of your 8 per cent is your own housekeeping. Rebuild both sides under the same rule, which is A8's problem.

The third is that a decomposition explains where a gap sits and never why it exists. Four hundred returning buyers who did not come back have four hundred reasons, and none of them is in the registration file. The file tells you who to ask.

Start by adding three columns to the pre-show pack you already produce: returning, new, and everything that is not a buyer, each with the prior edition's figure at the same days out beside it. Make the three columns sum to the headline number, publish the definition of returning in one sentence above the table, and do not change it for three editions.