Reading the new versus returning attendee mix in your post show numbers

Attendance came in at 24,800 against 24,700 the year before. A hundred people up on a base of nearly twenty-five thousand, which is flat by any reasonable reading, and the report says so.

Underneath that flat line, 14,300 of the people in the hall had never been to the show before. Well over half the audience was replaced in twelve months, and the report does not mention it, because a total that does not move gives nobody a reason to look inside it.

The new versus returning attendee mix is the most important operating fact about most trade shows and it is the one least often written down. It sets your marketing budget, it explains why your exhibitors keep saying the quality has changed, and it decides whether a retention programme is worth building. It is also the piece of attendee analytics that a total attendance figure is structurally incapable of showing you.

Why is a two-way new versus returning split not enough?

The instinctive split is new against returning, and it is one category short.

Cohort every badge at this edition against the previous three editions rather than only the last one. That produces three groups. Returning attendees came to the immediately previous edition. Recovered attendees came to one of the two editions before that, skipped last year, and are back. First-time attendees appear nowhere in the prior three.

On the 24,800: returning 7,400, recovered 3,100, first-time 14,300. In shares, 29.8 per cent returning, 12.5 per cent recovered, 57.7 per cent first-time.

The recovered group is the one the two-way cut destroys, and it is worth 3,100 people here. A two-way split calls them new, which overstates acquisition by 12.5 points of the audience and hides the fact that a meaningful slice of your show operates on a two-year cycle. Buyers whose capital budget runs biennially, or whose category refreshes every other year, do exactly this and they are not churn.

Getting the recovered group wrong has a direct cost. Win-back campaigns get aimed at people who were always coming back, and acquisition budget gets credited with an audience it did not find. The exact arithmetic that reconciles those groups to a headline movement is the year over year attendance variance decomposition in D13, and this post takes the mix as given rather than deriving it again.

Why does cohort retention rise every year?

Chart retention by year of first attendance and something odd appears. The people who first came four years ago return at a much higher rate than the people who first came last year, and the gap is large.

Take a stylised version. A cohort of 10,000 first-timers. Of those, 2,600 come back the next year, a 26.0 per cent rate. Of those 2,600, 1,040 come back the year after, which is 40.0 per cent. Of those 1,040, 520 return, at 50.0 per cent. Of those 520, 302 return, at 58.0 per cent.

The retention rate on that cohort goes 26, 40, 50, 58 without any single person's behaviour changing. If the show acquires 10,000 first-timers every year and each cohort follows the same curve, the blended retention rate across the whole audience sits at 4,462 returners against a prior-year total of 14,160, which is 31.5 per cent, and it stays there year after year while every individual cohort's rate climbs.

The mechanism is a sorting effect, and it is well described. Fader and Hardie set it out formally in Marketing Science in 2010, in a paper on customer-base valuation that opens with a firm whose cohort-level retention rates run 63.3, 68.9, 74.7 and 79.8 per cent while its aggregate rate sits almost flat at 63.3, 65.5, 67.5 and 69.1. Their illustration is a cohort of 10,000 split into two invisible segments, one third with a constant 0.9 annual retention probability and two thirds with a constant 0.5. Nobody changes. The high-churn segment simply drains away faster, so the survivors are increasingly made of the loyal type, and the observed rate rises on its own.

The consequence they draw is that valuing a customer base on a single aggregate retention rate understates it, in their worked case by 38 per cent. The consequence for an organiser is more direct. A first-time attendee and a fourth-time attendee are not the same asset, and averaging them into one retention number throws away the only structure in the data that tells you where to spend.

What the curve tells you to do

If the second visit is the hard one, and on the numbers above it is by a distance, then the highest-value intervention available to an audience team is converting first-timers into second-timers.

The arithmetic is worth doing. Moving the first-year rate from 26.0 to 30.0 per cent on a 14,300 first-time cohort produces 572 extra second-year attendees. Those people then enter a segment retaining at 40 per cent, then 50, then 58, so the same 572 carry forward roughly 229, then 114, then 66 in the following editions. One year's improvement on the hardest step delivers about 981 attendee-visits across four editions, from a cohort that cost you nothing extra to acquire.

Compare that with finding 572 more first-timers, which costs the full cost to attract an attendee that D19 works through, and 74 per cent of whom will not be seen again.

This is also why the first-time share is a leading indicator and not a vanity metric. A show whose first-time share climbs from 51 to 58 per cent over three editions is running harder each year to stand still, and the retention curve will tell you whether that is because the curve itself is deteriorating or because acquisition simply grew faster than the base.

Publish the cohort table, not the percentage

The output is a small triangle and it should go in the post-show report unchanged.

Down the side, year of first attendance. Across the top, each subsequent edition. In the cells, the count of that cohort still attending. Five rows is plenty. Anybody can read the diagonal and see whether this year's first-year retention is better or worse than the equivalent figure two editions ago, which is the comparison that actually matters and which no single blended percentage supports.

Two rules keep it honest. Cohorts are defined by first appearance in your data, so a show with four years of history cannot distinguish a genuine first-timer from somebody who last came five years ago, and the earliest cohort is always contaminated. Say so in a footnote rather than pretending the left column is clean.

And the cohort assignment is fixed forever at first appearance. If somebody's identity gets merged next year and their true first attendance moves back two editions, the whole triangle changes shape. Version the identity resolution run that produced the table and store the run identifier next to it, or you will spend next October explaining why last year's published table no longer reproduces.

The exhibitor conversation this changes

Exhibitors complain about audience quality in a specific pattern, and the cohort table usually explains it.

An exhibitor who has been at the show for six years has a stand team who remember the buyers. When 57.7 per cent of the hall is new, most of the faces are unfamiliar, and the honest experience on the stand is that the audience feels less senior and less prepared, whether or not it is. What has changed is the tenure of the audience, and the job title mix can hold steady while that happens.

Giving exhibitors the first-time share alongside their own lead counts reframes that conversation into something useful. A first-time attendee genuinely does behave differently: less certain about where to go, more likely to browse, less likely to arrive at a stand with a shortlist. That is an argument for pre-show meeting programmes and better wayfinding, and it is a much better answer than defending the quality of a badge file.

It is also worth being straight about the limits of comfort here. Freeman's End-of-Year Trends Recap on 2025, released in January 2026, reports that 20 per cent of attendees do not believe their objectives are being met at events, and that only 40 per cent say they experienced what it calls a peak moment, while 85 per cent of those who did are likely to return. A first-time cohort retaining at 26 per cent is partly a measurement artefact of a busy first visit and partly a real signal about what that visit delivered.

Where this stops

Everything above depends on deciding when two registration rows are the same person, and that decision is doing more work than the analysis is.

Match too tightly and returning attendees who changed jobs are recorded as first-timers, which inflates your first-time share and depresses every cohort retention rate. Since senior buyers change employer more often than junior ones, that error concentrates in exactly the population you most want to track. Match too loosely and you merge people, inflating retention. The cohort triangle is highly sensitive to this in a way that a total attendance figure is not, and any published version should carry the matching rule and its version next to it. Settling the unique attendee definition underneath all of it is D6's job.

The second limit is that a trade show has no cancellation event. Somebody who skips two editions and returns in the third was never lost, they were on a cycle, and the model has no way to distinguish a long gap from a departure until the person either comes back or does not. That makes any projection of the curve beyond your observed history a modelling assumption rather than a measurement, and it should be labelled as one.

The third is that the curve says nothing about who should be retained. A cohort retaining at 58 per cent might be entirely composed of people who no longer buy anything. Retention measured on badges is a measure of attendance, and connecting it to commercial value needs the exhibitor side of the data, which most organisers do not have and cannot get.

Build the triangle this week from four editions of registration data, using whatever identity key you already trust. If the first-year retention on your most recent cohort differs from the one two editions earlier by more than a couple of points, that difference is the number worth the rest of the quarter.