Measuring early bird cannibalization on your own registration file

The margin review happens in April, after the show and before the budget cycle. Registration revenue was up four per cent on a file that was up nine per cent, and the finance lead wants to know where the other five points went.

The answer is on a slide nobody made. The early bird window took a larger share of the file than it did the year before, so more of the audience paid the lower price, and the average yield per paid registration fell. That much is easy to put in a report. Early bird cannibalisation is the harder question sitting underneath it, which is whether the discount bought anything at all, and answering that needs a different cut of the same file.

What did the discount actually buy?

Every organiser can tell you how many registrations came in at the early bird price. Very few can tell you how many of those people would have registered in that window with no discount available.

Those are different numbers and only the second one matters. A discount that moves nobody is a price cut, and a price cut applied to the largest window of your campaign is the single biggest pricing decision most shows make without deliberating about it.

The measurement you want is a comparison between the people who took the early bird price this edition and what those same people did last edition. It runs on data you already have, it takes an afternoon, and it produces a number you can put next to the giveaway.

Split the takers by what they did last time

Take the registrations created on or before this edition's early bird cutoff. Call that 3,000.

Now match each of those registrations back to the prior edition's file, and sort them into three groups. The first group registered at the prior edition inside the equivalent window, measured in days out from that edition's opening date rather than by calendar date. The second group registered at the prior edition but outside that window. The third group has no prior edition registration at all.

A shape that comes up often: 2,100 in the first group, 540 in the second, 360 in the third.

The first group is the one the whole exercise is about. Those 2,100 people demonstrated at the previous edition that they register early. They did it again. Nothing in that pattern needs a discount to explain it, and PCMA made the point directly in April 2024, reporting Maritz research to the effect that time-based discounts risk giving away money to the very people who would attend the show regardless.

The second group of 540 changed behaviour in the direction you wanted. They registered late last time and early this time. The third group of 360 is new, and their timing tells you nothing, because they have no prior timing to compare against.

A word on what counts as a returning registrant. CEIR's 2016 Attendee Retention Insights series found the most common industry definition of retention to be attendance at two of the last four editions, and found that 77 per cent of trade show organisers were tracking attendee retention at all. For this measurement I would use the immediately prior edition only, because you are comparing timing behaviour and timing behaviour two editions ago was formed under a different campaign. The wider returning and new split across five editions is the baseline a target gets built from, which is A11's job and a different cut.

Price the giveaway

Put the discount on it. Say the early bird price is 595 and the standard price is 715, so the discount is 120 per registration.

The gross giveaway is 3,000 times 120, which is 360,000.

Split it the same way as the takers. The 2,100 prior early registrants account for 2,100 times 120, or 252,000. The 540 movers account for 64,800. The 360 new registrations account for 43,200.

That first line is the whole argument. Seventy per cent of the discount, 252,000 of 360,000, went to people whose registration timing was identical last year without it. The share is worth computing as its own figure, because it is stable enough between editions to become a number your commercial team recognises. Here it is 2,100 divided by 3,000, or 70 per cent.

What did each registration that moved cost?

Now express the giveaway as a unit cost, which is where it stops being an abstraction.

The whole 360,000 bought 900 registrations whose timing was not already established, since 3,000 minus 2,100 is 900. That is 400 per registration moved, on a badge whose full price is 715.

Take the harsher reading. The 360 brand new registrations had no prior timing, so crediting the early bird with moving them is a choice rather than an observation, and a new registrant arriving in January is at least as likely to be responding to the campaign launch as to the price. Charge the whole 360,000 against the 540 people who demonstrably shifted from late to early, and the cost per moved registration is 666.67.

You are then spending 666.67 to change the date on a 715 registration. The organiser who is comfortable with that is doing it for a reason other than revenue, and there are legitimate ones. Ten weeks of extra lead time on a hosted buyer programme is worth real money. A readable geography mix before the visa deadline is worth real money on an international show. Say which one it is, and put the 666.67 next to it.

A control group you already have

The split above still leaves one hole. Some of the 2,100 might have drifted out of the early window if there had been no price to beat, in which case the discount held them rather than merely rewarding them.

There is a control group sitting in your file for this, and it costs nothing to use. Take the population that faces no price at all: complimentary registrations, invited guests, whatever your equivalent is. That group receives the same campaign emails, sees the same deadline messaging, and has no financial reason to respond to any of it.

Suppose the show closes at 11,600 registrations, of which 1,400 are complimentary. Count how many of those 1,400 were created on or before the early bird cutoff. If it is 370, then the price-insensitive population put 26.4 per cent of its registrations inside the early window.

Compare that with the paid population. The 3,000 early bird registrations sit inside a paid file of, say, 10,200 minus the comps, so 3,000 of 10,200 is 29.4 per cent.

Three points of difference between a population that responds only to the calendar and a population that also faces a price. On those numbers, most of what the early window is doing is campaign scheduling rather than pricing, and the deadline email is doing the work the discount is being credited with. If the gap comes out at fifteen or twenty points instead, the discount is genuinely moving people and the argument changes.

The comparison is imperfect, and I will come back to that. It is still the cheapest counterfactual available to an organiser who will not run a real price test.

What I would change

I would keep an early deadline and shrink the discount until the cannibalisation share tells me to stop.

The mechanism worth protecting is the deadline, which schedules the file. What the cutoff itself moves is A12's measurement, and it is worth running before you touch the price, because the deadline and the discount are usually credited together. The discount is one way of making a deadline bite and it is the most expensive one. On a show where 70 per cent of takers are prior early registrants, halving the discount from 120 to 60 gives back 126,000 against the prior early group alone, and the question is only whether it loses more than 126,000 divided by 715, which is 176 registrations, from the groups that were actually moved.

Where the new price sits inside the wider ladder is A13's argument, and if you find yourself tempted to soften the change by pushing the cutoff back instead, A15 has the bill for that.

Run the split for two editions before you touch the price, so you know whether 70 per cent is your show's number or an artefact of one year. Then change the discount by a meaningful amount in one direction and measure the same three groups again. That is a slow experiment and it is the only one most organisers can actually run.

Where this stops

The control group is the weakest part of the method, because complimentary registrations are not a random sample of your audience. They are disproportionately VIPs, speakers and sponsor guests, and they are often loaded in bulk by an administrator on a date that has nothing to do with when the person decided. If a sponsor uploads 300 guest names in one afternoon in February, the comp population's early share is measuring an administrator's calendar. Before using the comparison, check the daily distribution of comp creation dates and drop any day that contains an obvious bulk load.

The window boundary is the second problem. A registration created two days after the cutoff is treated as a different behaviour from one created two days before, and for a person who nearly made the deadline that distinction is noise. If the three groups shift a lot when you widen the window by a week in each direction, the split is not stable enough to price a decision on.

The deeper limit is that none of this measures the counterfactual you actually care about, which is what the file would have looked like with no early bird at all. Nobody has that file. What you have is a defensible upper and lower bound on what the discount is buying, and a giveaway figure that is exact.

Pull the last two editions' registration files this week and add one column to your attendee analytics extract for the current edition: for every registration created on or before the early bird cutoff, whether the same person registered inside the equivalent window at the prior edition. The share that comes back true, multiplied by your discount, is the money the early bird handed to people who were already going to be early.