Attendee reentry tracking without inflating the daily attendance number

The year the venue moved the food court outside the perimeter, day two entries jumped 8 per cent while every other indicator stayed flat. Nobody had run a bigger campaign. What had changed was that 900 people now walked out for lunch and walked back in, and the door readers counted them again on the way back.

Attendee reentry tracking is the work that stops a catering decision showing up as audience growth. It is unglamorous, it depends on data most shows do not collect, and getting it wrong moves a headline figure by more than most marketing campaigns do.

What does a re-entry look like in the data?

Like a first arrival, which is the whole problem.

An entry reader records the same event both times: this badge, this reader, this timestamp, granted. Nothing in the row says whether the person was already inside. The only signal available is the gap since that badge was last seen, and gaps are ambiguous. Four seconds is a double fire. Four hours could be a return from a client meeting or could be somebody arriving late for the afternoon.

Three patterns cause almost all of the volume at a business exhibition. Lunch, which produces a tight cluster of exits between 12:15 and 13:15 and returns forty to ninety minutes later. Client meetings, which spread across the day and last longer. Smoking, which is short, frequent, and concentrated among a small number of badges.

Those three have different signatures and only the first two matter for attendance, because a four minute step outside is handled by the debounce window described in entry scan deduplication and never reaches this question.

Two ways to see an exit

If you can observe the exit, the problem mostly dissolves. Two approaches do it.

Directional readers. A separate reader on the outbound side of the barrier, logging a badge as it leaves. This costs hardware and floor space, and it fails whenever people leave through a door that has no reader, which at most venues is most doors.

Anti-passback logic in the panel. Configure the access control system to refuse a second inbound grant on the same credential without an intervening outbound read. The value here is diagnostic rather than enforcement: an anti-passback deny tells you a badge came in twice without going out, which is either a re-entry through an uncounted exit or a badge being handed back through the fence.

Both leave gaps, and the size of the gap is a property of the building. Where a venue's exits are the same doors as its entrances, exit reads cover most departures. Where the exit is a fire door onto the car park, they cover none, and which doors are instrumented at all is the gate coverage question.

Sizing a minimum gap rule from your own data

Without exit reads, the rule is a threshold: consecutive entry events from the same badge on the same day, separated by more than G minutes, count as a re-entry. Below G, treat the second event as an artefact.

Pick G badly and you either inflate attendance with door noise or you erase genuine second arrivals. Pick it from data and the choice becomes defensible.

Where you have exit reads on even one door, use them. Take every exit followed by an entry from the same badge on the same day, and plot the distribution of the interval between them. At a business show this is usually bimodal: a spike between three and fifteen minutes, which is smoking and phone calls, and a broad hump from thirty-five to ninety minutes, which is food. The valley between the two is where G belongs.

Where you have nothing, ninety minutes is a reasonable default and should be stated as an assumption rather than presented as a measurement. Test it. Run the rule at sixty, ninety and one hundred and twenty minutes and publish the three answers internally before choosing one, because the spread across them tells you how sensitive your headline is to a decision nobody outside the data team will ever see.

The arithmetic on 12,700 arrivals

Day two at one show, counted doors only, after the debounce has already removed reader artefacts.

Raw arrival events: 12,700. Applying the ninety minute gap rule classifies 3,410 of them as returns, which leaves 9,290 distinct people through the doors that day. The raw figure was 26.9 per cent higher than the deduplicated one.

Now the sensitivity. A sixty minute rule classifies 4,180 events as returns, leaving 8,520. A hundred and twenty minute rule classifies 2,890, leaving 9,810. The gap between those two answers is 1,290 people, which is 10.2 per cent of the raw arrival count and 13.9 per cent of the ninety minute answer.

That spread is the number to put in front of anybody who wants a single attendance figure with no method attached. Three defensible rules produce three answers separated by more than a thousand people, and the difference between them is larger than the year on year movement most shows report.

The exit data, where it existed on four instrumented doors, recorded 3,180 outbound reads on that day, of which 2,760 were followed by an inbound read from the same badge within four hours. That gives a directly observed return count of 2,760 against a rule-based estimate of 3,410 across all doors, which is consistent once you allow for the ten doors with no exit reads at all.

The same correction changes the shape of the day as well as its total. Of those 3,410 returns, 2,050 landed between 13:00 and 14:15, which is 60.1 per cent of the day's returns inside a seventy-five minute window. Leave them in and the early afternoon looks like a second arrival peak, and somebody plans door staffing for a surge that is really the food court. Strip them out and the afternoon flattens to roughly a third of the morning rate, which is what the stewards on the doors have been saying for years.

How should re-entries be reported?

Against a published definition, so the number means the same thing to an auditor as it does to your sales team.

UFI's Auditing Rules for the Statistics of UFI Approved Events, in the June 2021 version, carry a set of calculation standards that settle this precisely. A visit is a person entering the event on an official open day and hour with an access document, and a person may only be counted once per day. A visitor is counted once for the entire duration of the fair, whatever the number of visits. A repeat visit is each additional visit by a visitor after the first that can be controlled, and there too only one visit per day is counted. Total visits equal visitors plus repeat visits, and total attendance equals unique visitors plus exhibitor staff, speakers and media representatives.

Read that carefully and the same-day question is already answered. A return from lunch is not a repeat visit under those rules, because repeat visits are counted per day. The 9,290 figure maps to the standard and the 12,700 maps to nothing at all.

That leaves three numbers worth publishing for each day: distinct people through the doors, total entry events, and the ratio between them. The ratio is the useful operational series, because it moves when the building changes and stays flat when the audience changes. The key you count people distinct on, before any of this applies, is the unique attendee definition.

Re-entry data has a second use that has nothing to do with reporting. HSE guidance HSG154, Managing crowds safely, published in its second edition in 2000, describes counting systems as a way to estimate the number of people inside a venue, and for sites with few defined boundaries suggests sampling flow rates at entrances to estimate the numbers entering or leaving. Occupancy needs both directions. A show with entry reads only can report arrivals and cannot report how many people are in the hall at 15:00, which is the number a safety officer actually wants.

Where this stops

The gap rule is a guess dressed in arithmetic, and it fails in one direction consistently. Somebody who arrives at 09:10, leaves at 09:25 because they went to the wrong hall, and comes back at 09:40 is counted once by a ninety minute rule and twice by a fifteen minute one, and no threshold gets every case right.

A worse failure is the shared badge. Two colleagues using one credential produce exactly the pattern of a re-entry, and the rule will quietly merge two people into one. Shows with a lot of comp badges have more of this, and the effect runs opposite to the re-entry inflation, so the two errors partly cancel in a way you cannot measure or rely on.

The last limit is structural. Everything here depends on doors you instrument, and the returns through an uncounted door are invisible on both legs. A perfectly tuned gap rule on 94 per cent coverage is still a rule operating on 94 per cent of the movement, which is why coverage and re-entry belong in the same paragraph of any attendee analytics method note.

Start this week by running your last edition's entry events for one day at three gap lengths, sixty, ninety and one hundred and twenty minutes, and writing the three answers side by side. If they differ by more than five per cent, your attendance figure has a parameter in it that nobody has ever approved.