Unique visitors versus total scans and why the two numbers keep diverging

Somebody in marketing builds a slide for the sales team from the exhibitor performance export. It adds up the unique visitor count for every stand on the floor and reports that the show delivered 155,300 visitors to its exhibitors.

The show's audited attendance is 23,100.

Nothing in the export is wrong. Every one of the 468 per-exhibitor figures is a correct count of distinct badges scanned at that stand. The slide is wrong because unique visitors and total scans behave differently under addition, and almost nobody realises that until the number gets big enough to be obviously silly.

That is the mild version. The dangerous version is when the two counts are close enough that nobody notices which one they are looking at.

The audit standard already made this distinction

Exhibition organisers have a formal answer to this and it is older than most of the software involved.

UFI's Calculation Standards and Definitions, the 2007 document that sits behind the auditing rules for UFI approved events, sets out a layered set of counts. An entry, or admission, is each individual entering the site of the event during hours officially open to the public. A visit is the action of a person entering with an access document on an official open day, and a person may only be counted once per day. A visitor is a person attending the exhibition, counted once for the entire duration of the fair regardless of the number of visits. A repeat visit is each additional visit after the first, again capped at one per day. Total visits is visitors plus repeat visits.

The document then does the thing that matters most. It says figures for either visits or visitors are accepted, provided the definition of the applicable term is clearly understood in the context of its application.

Neither count is the right one. The rule is that you have to say which you mean.

Nobody applied any of this at stand level, because when the standards were written the stand level count was a stack of business cards. The structure transfers exactly, and it is worth transferring into exhibitor analytics, because it gives you names for three numbers that most exhibitor reports currently collapse into one.

• A capture event is the stand's equivalent of an entry. Every time somebody pressed the button.
• A stand visit is a distinct badge on a given show day, capped at one per badge per day.
• A stand visitor is a distinct badge across the whole event, counted once.

Three counts on one exhibitor's file

Take a stand at a three day show. The scan table holds 1,410 capture rows.

Distinct badges across the three days: 980. So the headline pair is 1,410 scans and 980 unique visitors, and the ratio is 1,410 divided by 980, which is 1.44.

Now compute the middle number, which almost nobody does. Distinct badges within each day: 470 on day one, 520 on day two, 310 on day three. Those sum to 1,300 stand visits.

Three numbers on one stand: 1,410 captures, 1,300 visits, 980 visitors. Top to bottom, a difference of 44 per cent, and all three are correct.

The middle number earns its place because it splits the gap into two pieces with different causes.

The gap between 1,410 and 1,300 is 110 rows. Those are second and third captures of the same badge on the same day: device retries, a passive reader, a visitor read again at a demo station, a second conversation before the doors closed. Whether any of those should survive is a windowed deduplication question with a real answer.

The gap between 1,300 and 980 is 320 badge-days. Those are people who came back on another day, which is a visitor deciding to walk back across the hall on Wednesday morning and has nothing to do with how the kit behaved. Reading that decision as an intent signal is worth doing separately.

Splitting at the day boundary is the whole trick, and it works because the day boundary is the one place where a repeat has an unambiguous meaning. UFI caps visits at one per person per day for exactly that reason. Coming back tomorrow required a decision. Being scanned twice in four minutes did not.

Unique counts do not add

Return to the slide. The sum of per-exhibitor unique counts across 468 stands is 155,300. Distinct badges scanned by any exhibitor anywhere on the floor is 18,940.

The ratio of those two numbers is 155,300 divided by 18,940, which is 8.20. That is the average number of stands that scanned each scanned attendee, and it is a genuinely useful figure that nobody computes because they are too busy being embarrassed about the 155,300.

The arithmetic is not subtle. A unique count is the size of a set, and set sizes only add when the sets do not overlap. Exhibitor lead files overlap enormously, by design, because a visitor who talks to eight stands is what a trade show is for.

Practical consequences, and each one is a real report somebody has published.

Never sum a unique column across exhibitors. If a roll-up needs a total, sum captures, which do add, and compute the floor-wide unique count separately from the scan table with a distinct count over the whole floor.

Never sum unique counts across days for one exhibitor either. That gives you the 1,300, which is visits, not visitors, and calling it visitors overstates by 320 on the stand above.

Never sum unique counts across editions to get a portfolio figure. The same buyer at three of your shows is one person, and no arithmetic on badge identifiers will discover that, because it is an identity resolution problem.

Never divide a floor-wide figure by a summed unique column. A percentage with 155,300 underneath it is not a percentage of anything.

And 18,940 against an audited attendance of 23,100 is itself a number worth publishing: 82.0 per cent of the people who came through the door were scanned by at least one exhibitor. That figure tells you something about your floor that no exhibitor level report will, and the 18 per cent who were never scanned by anybody are a population somebody should be curious about.

Are unique badges the same as unique people?

The word visitor in the count above is doing work the data cannot support, and the honest term is distinct badge.

A badge and a person come apart in several ways and all of them are ordinary. A visitor loses a badge on day two and gets a replacement printed with a new identifier, appearing as two people. Two colleagues take turns on one badge because only one of them registered, appearing as one person. A group registration issued through a trade association arrives as five badges collected by one person who hands four of them out in the car park. A manual entry, typed because the badge was in a hotel room, has no badge identifier at all and drops out of every distinct count silently.

None of this is fixable from the scan table and I would not pretend otherwise in a report. What is fixable is the label. Write distinct badges scanned, not unique visitors, in the column header, and put one line in the methodology note saying that a badge is a proxy for a person and where that proxy breaks.

The manual entry case deserves a number of its own because it is the one that biases in a predictable direction. If 60 of the stand's 1,410 rows are manual entries with no badge identifier, the distinct badge count of 980 excludes every one of them, and the stands with the most manual entries are usually the stands that talked to the most people. Report the manual share beside the unique count, or accept a denominator that is quietly wrong for your best exhibitors.

Which number goes in which sentence?

The two counts answer different questions and the questions are easy to keep straight once you write them down.

Total captures is a measure of activity on the stand. It is the right numerator for anything about staff effort, kit utilisation or cost per capture, and it is the number that adds across exhibitors and across days.

Distinct badges is a measure of reach. It is the right denominator for anything per person: share of the audience reached, percentage of a target list touched, penetration of a category. It never adds.

So the exhibitor scorecard line reads: 1,410 captures from 980 distinct badges across 1,300 stand visits. Three labelled numbers in one sentence, no ratio, no interpretation. The ratio between the first and second has its own healthy range and its own diagnostic value, which is a thing to work out deliberately rather than slip into a headline.

The one place I would refuse to publish either number alone is the pre-renewal pack. CEIR's 2026 Marketing Spend Decision Report found that sales metrics dominate how management evaluates exhibition return on investment, with lead volume and post-show closed deals ranking highest, which means the lead figure you print is the one somebody defends internally against every other channel. An exhibitor deciding whether to rebook is going to divide something by something, and if the pack contains a single unlabelled lead figure they will divide their whole show cost by it. On the stand above, dividing by 1,410 instead of 980 puts the cost per lead 30 per cent lower, and dividing by 980 instead of 1,410 puts it 44 per cent higher. Handing somebody a number that flattering or that harsh, from your own document, is a bad way to start a negotiation.

Where this stops

The three counts describe scanning, and scanning is not the same as visiting.

A stand with 980 distinct badges had some larger number of people stop, look, take a brochure and leave without being scanned. Nothing in a lead retrieval system sees them. Measuring the aisle takes a different instrument with its own error characteristics, and that work belongs to the on-site sensing posts in cluster C.

The per-day split also depends on having a reliable capture timestamp with a timezone on every row, and on a show day boundary that somebody has defined. A stand that ran a private reception until 21:00 on day two, or an event crossing midnight, will produce badge-days that do not match anyone's intuition unless the day boundary is stated explicitly rather than taken from the date part of the timestamp.

And the floor-wide distinct count of 18,940 is only computable if badge identifiers are consistent across every exhibitor's feed. Where one lead retrieval vendor returns a padded identifier and another strips leading zeros, the same person becomes two people in the roll-up, the floor total inflates, and the 8.20 figure falls. That is a five minute check and it is worth doing before anybody sees the number.

Take last edition's scan table for your ten largest exhibitors and compute three columns: row count, distinct badges per show day summed, and distinct badges over the whole event. If your current exhibitor report shows only the first and the third, the middle column is the one that tells you whether the gap you have been explaining as duplicate scans is actually people coming back.