Measuring trade show no show rate on a denominator you can defend

Three people in the same post-show review quote a no show rate. Marketing says 22 per cent. The registration manager says 28. The finance lead, working from the file the agency sent, says 34. Nobody is lying and nobody has made an arithmetic error.

They are dividing the same number of bodies by three different registration counts, and none of the three has ever been written down as the definition. That is the whole argument, and it repeats every year because the definition lives in a spreadsheet formula rather than in a document.

The numerator is the easy half

Start with the part people argue about least. The numerator is the count of registered people who turned up, which for most B2B shows means a badge collected or a scan at an entrance reader.

It has its own problems. Coverage at the door is never perfect, a person who collects a badge on Tuesday and never returns counts the same as one who worked all three days, and group check-in at a hosted buyer desk can produce a single scan for six people. Those are worth fixing, and they are not what makes three people quote three different numbers.

The spread comes from the denominator, because a registration file is not one population. It is a stack of records in different states, and the rate changes by twelve points depending on which states you keep.

What states can a registration record be in?

Open the file the day before doors and sort it. You will find at least four kinds of row.

Live registrations. A real person, one record, still holding a valid registration at show open. This is the population that could have attended and might not.

Cancellations. Somebody registered and then told you they were not coming, or asked for a refund, or let a paid registration lapse. They did tell you. Counting them as no shows punishes the show for the one group that behaved well.

Duplicate records. The same person registered twice, or an exhibitor uploaded a guest who had already registered themselves, or a record was created by a data load and again by the person. One body, two rows, and only one of them can ever be marked attended.

Test and internal rows. Records created during configuration, staff accounts, a row named after the registration platform's support engineer. Small in number and always present.

There is a fifth population that never appears in the pre-show file at all, and it is the one that quietly breaks the arithmetic.

The people who were never in the denominator

Some of the bodies in your hall registered after the show opened. They walked up to a desk, filled in a form, and were issued a badge.

Maritz's Registration Insights Report 2024, built from more than 360,000 attendee registration records across 30 trade shows over three years, put 22 per cent of registrations in the final week before the show and nine per cent on site. Nine per cent is not a rounding error. On a show that verifies thirteen thousand people, it is more than a thousand attendances that were created after the pre-show denominator was frozen.

If you divide total verified attendance by registrations live at show open, those on site registrants inflate the numerator against a denominator they were never in, and the no show rate comes out flattering and wrong. The fix is not complicated. Count pre-registered attendances in the numerator and report on site registrations as their own line.

The same show, worked three ways

Take one edition and follow the numbers through.

The registration system has created 18,900 rows over the campaign. Of those, 150 are test and internal records, 1,450 are duplicate rows that resolve to a person already in the file, and 1,300 are cancellations processed before show open. That leaves 16,000 live registrations at the moment doors open.

Verified attendance across the three days is 13,630 people. Of those, 1,150 registered on site during the show. Pre-registered attendances are therefore 13,630 minus 1,150, which is 12,480.

Now the three rates.

Divide pre-registered attendances by live registrations at show open. That is 12,480 over 16,000, which is 78.0 per cent turnout and a no show rate of 22.0 per cent.

Add cancellations back into the denominator, on the argument that they were registrations once. That is 12,480 over 17,300, which is 72.1 per cent turnout and a no show rate of 27.9 per cent.

Use the gross count the marketing report has been quoting all campaign, every row ever created, duplicates and test records included. That is 12,480 over 18,900, which is 66.0 per cent turnout and a no show rate of 34.0 per cent.

Twelve points of spread on one file, and every one of those three numbers can be described as the no show rate without anybody misrepresenting anything.

There is a fourth version, and it is the one that sneaks into board packs. Divide total verified attendance, all 13,630, by live registrations of 16,000, and you get 85.2 per cent turnout and a no show rate of 14.8 per cent. It looks like the best year the show has ever had. It is 1,150 people counted in a numerator whose denominator excluded them, and the size of the error is exactly the size of your on site registration desk.

Which no show rate should you publish?

Registrations live at show open, after cancellations and duplicate merges, with test rows stripped. The 22.0 per cent.

The reasoning is about what the number is for. A no show rate is used to plan catering, badge stock, session room sizing and shuttle capacity. It also feeds the attendance forecast published before doors, which needs turnout by registration type and not one blended figure. All of those decisions are made against people who hold a live registration and might walk in. A cancelled registration cannot walk in. A duplicate row cannot walk in twice. Including them measures how messy your file is, which is a real thing worth measuring, and it belongs in a data quality report rather than in the operations forecast.

The gross count has one legitimate use, which is as the denominator for acquisition cost, because you paid to create all of those rows including the bad ones. Two different denominators for two different questions is fine as long as both are labelled.

The number that is hardest to defend is the cancellation-inclusive middle version, because it treats a registrant who told you they were not coming the same as one who ghosted. Those are different behaviours and the fix for them is different. Cancellations are a pricing and calendar problem. Silent no shows are a commitment problem.

Writing the definition down so it survives a year

The reason this argument recurs is that the definition is reconstructed from memory every time somebody builds the slide.

Write four lines and put them at the top of the post-show report template. Name the numerator, including whether a single scan on any day counts as an attendance. Name the denominator, including the exact cut point in time at which live registrations are counted. State how cancellations, duplicates and test rows are handled. State where on site registrations appear, which should be a separate line and never inside the rate.

Then freeze the cut point. Registrations live at show open means the state of the file at a specific timestamp, and if you rebuild it three weeks later from a cleaned export you will get a different number, because cleanup happened in between. Snapshot the file at the moment doors open and keep the snapshot. That single file is what makes next year's comparison possible.

CEIR's Q4 2024 Index results, released in March 2025, are a useful reminder of why the definition matters when you benchmark. The total index sat 4.4 per cent below the same quarter of 2019, but the four components were nowhere near each other: exhibitors were 0.1 per cent below, real revenues 1.1 per cent below, net square feet 3.0 per cent below, and attendees 12.9 per cent below. The attendee side is the slowest of the four to recover, which means it is the number under the most pressure, which means it is the number most likely to be quietly redefined. A rate whose denominator moves is not a series.

Where a defensible denominator stops helping

Getting the denominator right makes the number comparable across editions of your own show. It does not make it comparable across shows, and it does not make it a quality signal.

Two organisers can both compute the rate correctly and be uncomparable, because one runs a free expo where a registration costs the registrant nothing and the other runs a paid conference where it costs several hundred. Price is the strongest single driver of turnout there is, and a free show with a 40 per cent no show rate may be healthier than a paid one at 15 per cent, since the free show has 40 per cent more registrations to be absent from.

The rate also says nothing about who was absent. Losing 3,500 people who were mostly students and mostly local is a different show from losing 3,500 senior buyers who booked flights. The headline rate treats them identically, and it will be stable across a year in which the composition of the absence changed completely. Report the rate split by registration type or do not report it at all.

The last limit is that no show rate is a lagging measure with one observation per year. You get one number per edition, which means a five year history gives you five points, which is not enough to detect anything subtle. Anyone promising you a trend from it is reading noise. A rate this coarse also cannot say which registrations are at risk, since that needs a probability scored per registrant, and any attendee analytics worth running keeps the two reports apart.

This week, take last edition's registration export and add one column recording the state of every row at show open: live, cancelled, duplicate, test, or created on site. Compute the rate four ways from that single column and put all four numbers on one slide with their denominators named. The argument about which one to publish is worth having once, in a room, with the numbers visible, and then never again.