How to read registration pacing in the weeks before a show

On the Tuesday campaign call somebody reads out the registration number. We are at 6,808. Last year on this date we were at 7,300. The room goes quiet for a second and then the show director asks the only question that matters, which is whether that is bad.

Nobody on the call can answer it, because registration pacing was never the thing that got measured. The comparison they just made is between two numbers taken at different distances from their own show opening. Last year the show opened on 14 March. This year it opens on 23 March. The 7,300 was measured 69 days before that edition opened. The 6,808 was measured 60 days before this one opens. Those are not the same measurement and the gap between them is nine days of a campaign that is running at its steepest.

Why is the calendar date the wrong axis?

Registrations do not arrive on dates. They arrive at distances from show open, because that is what the buyer is responding to: a diary decision, a travel budget, a manager's approval, an agenda that has finally been published. Nothing in that chain is anchored to the fifteenth of January.

So the axis is days to show open. Every registration record gets a single derived field, the number of days between its created timestamp and the opening morning of the edition it belongs to. Day minus 60 means sixty days before doors. Day zero is opening day. Everything after is the on site tail.

Once you have that field, pacing has a definition you can defend. Pacing is the cumulative count of registrations at a given days-out value, compared against the cumulative count of a prior edition of the same show at the same days-out value. Not the same date. Not the same week number. The same distance from doors.

Take the call above and redo it properly. At day minus 60 the prior edition held 7,400 registrations. This edition holds 6,808. The pacing index is 6,808 divided by 7,400, times 100, which is 92. The show is pacing at 92 per cent of the prior edition at the same point. The 7,300 that started the argument was the prior edition at day minus 69 and it was never the right comparison.

Working an index through to show open

An index of 92 on a Tuesday in January is a number, and the useful question is what it turned into.

That edition finished on 11,979 registrations. The prior edition finished on 12,100. The closing index is 11,979 divided by 12,100, times 100, which is 99. The show came in one per cent under a year it had been eight per cent under two months earlier.

The arithmetic of why is worth doing on the page. The prior edition held 7,400 of its eventual 12,100 at day minus 60, so it had already banked 61 per cent of its final number and it added 4,700 registrations in the last two months. This edition held 6,808 of its eventual 11,979, which is 57 per cent, and it added 5,171 in the same window. The tail was 471 registrations heavier and it arrived in a period where the earlier edition had already gone quiet.

Notice what this does to the naive projection. If you take the index at face value and assume it holds, you predict 0.92 times 12,100, which is 11,132. The actual number was 11,979. The projection was 847 short, seven per cent low, and it was low because it assumed the shape of the two curves was identical, which is A2's subject. Converting an index into a defensible number for show open needs the historic share at that day out, which is A9's method and not this post's.

Freeman's end of year trends recap, released in January 2026 from its 2025 research, reported that 50 per cent of the events it tracked ran behind their regular registration pace that year. An index of 92 in that context is not automatically an emergency. It is a reading that has to be judged against how this show has behaved before, and against how much of the year is still in front of you.

Pick the days out where you actually decide something

There is a temptation to plot the index daily and stare at it. The line will jump around, because a single exhibitor uploading 140 guest registrations on a Thursday moves an index built on a base of 3,000 by more than four points.

Read the index at the points where a decision gets made. For most B2B shows that is four readings.

Day minus 120 is the first honest read, and it is honest only if registration opened at a comparable distance in both editions. Day minus 90 is where a media budget reallocation still has time to work. Day minus 60 is where the index starts to be predictive, because enough of the file is in. Day minus 30 is where you stop trying to move the number and start telling operations what to plan for.

Day minus 60 earns its place for a second reason. Maritz's Registration Insights Report 2024, built on more than 360,000 attendee registration records across 30 trade shows over three years, found that registrants who book within 60 days of the show spend the most on ancillary items, and named 31 to 60 days out as the strongest window. If that holds on your file, day minus 60 is the last point at which you can influence the segment that carries the most revenue per head.

Between the four readings, look at the index once a week and resist commentary on the wobble.

What moves the index without any change in demand?

An index is a ratio of two counts, and either count can be corrupted by an operational decision nobody logged.

The registration open date is the biggest one. If this edition opened registration on day minus 210 and the prior edition opened on day minus 175, the early part of the curve is not comparable at all and the index at day minus 150 is measuring a policy change. Check the earliest created timestamp in each edition before you read anything.

Price tier boundaries move volume across days, which is the point of them, so an index read the week after an early bird deadline in one edition and the week before it in another is comparing a pull-forward against its own absence. Note where every tier boundary sits in days out for both editions and mark them on the chart.

Bulk uploads distort the base. Exhibitor guest pass allotments, association member imports, a delegate list loaded from a co-located event. These land as hundreds of records with the same created timestamp, and if the equivalent upload happened at a different distance last time, the two curves diverge for a reason that has no buyer in it.

Then there is the definition itself. Cancellations, duplicates removed in a later cleanup, and test records all change historic counts after the fact. If your prior edition's curve is being rebuilt from a cleaned final file while this edition's curve is a live count with the rubbish still in it, the index is biased against this edition by however much rubbish you normally remove. Rebuilding comparable curves from a flat export is A8's problem and it is more delicate than it sounds.

What I would do with an index of 92

Two months out, at 92, with the tier boundaries in the same place and no upload anomalies, I would not authorise unplanned spend on the strength of that number alone.

The reason is that the index at a single day out has an error bar you can estimate from your own history, and on most shows it is wide. Take five editions, compute each one's index against the edition before it at day minus 60, and compute each one's closing index. On the show above, the day minus 60 index was 92 and the closing index was 99, a movement of seven points. If the other four editions moved by four, six, nine and three points, then a reading of 92 is consistent with closing anywhere from roughly 95 to 101, and the honest thing to say on the call is that range.

That is a different posture from the one most teams take, which is to treat the current index as a forecast and start spending against it. It is also different from doing nothing. A 92 with an unusually thin new-registrant share is a real signal even when the headline index recovers, because the composition is telling you something the total is hiding.

The comparison against a single prior edition is the weakest part of everything above, and the fix is a band built from several editions, which A7 covers.

Where a pacing index stops

Early in the campaign the index is arithmetic on a base too small to mean anything. At day minus 150 an edition holding 380 registrations against a prior 300 has an index of 127, which looks like a strong year and is in fact a difference of 80 records, roughly one exhibitor uploading its guest list a fortnight earlier than last time. Below about a thousand registrations in the base, I would report the raw counts and refuse to compute an index at all.

The index also breaks when the prior edition was not normal. A year with a venue change, a strike, a competitor launching against you, or a date that landed inside a national holiday is a poor denominator, and an index against it will flatter or damn this edition for reasons that have nothing to do with this edition. When the prior year was odd, say so on the slide and index against the year before it as well, so the reader can see both.

The last limit is the one people forget. A pacing index measures registrations. It says nothing about who those registrations are, and a show can pace at 105 while losing the buyers its exhibitors renew for. Pace and mix are separate reads, and any attendee analytics worth running puts both in the same weekly pack.

Start this week by adding one field. Take the two most recent editions of one show, compute days to show open for every registration record from its created date, and build the two cumulative curves on that axis. Read the index at day minus 90, day minus 60 and day minus 30, and write down the registration open date and every price tier boundary for both editions next to those three numbers. If the tier boundaries sit at different days out, you have found the reason last year's comparison felt wrong before you have done any modelling at all.