Skip to content

Reading exhibitor category performance before you blame individual stands

Exhibitor analyticsUpdated 2026-08-187 min read

In short

Exhibitor category performance is the median unique leads captured by the stands inside one product category, compared against the median of all category medians on the floor. Flag any category more than one median absolute deviation below that reference, because the median absolute deviation is not inflated by the outlier it is meant to detect.

Nine complaints came in from the software and services aisle in the fortnight after close. Nine different companies, nine variations on the same message, and the sales team's working theory was that we had sold that hall badly and the stands had been under-staffed.

Then somebody sorted the complaints by floorplan position. All nine were in the same two aisles. Every stand in those aisles had complained, including two that had done well the previous edition with the same team and the same space.

Exhibitor category performance is where that kind of complaint gets diagnosed, because when an entire product category underperforms the cause is almost never nine independent staffing failures. It is audience mix, placement, or a category that was sold to exhibitors on a promise the registration file did not deliver. Diagnosing it takes one aggregation and one outlier-resistant threshold, and both are worth doing before anybody picks up the phone to an individual stand.

Aggregate to the category first

The instinct after a show is to rank exhibitors and look at the bottom. That ranking will always have a bottom, and it will always contain stands with plausible-looking explanations, so it generates work without generating findings.

Aggregating first inverts the question. Compute the median unique leads for every product category, then compare categories to each other. If eight categories cluster and one sits well below them, you have a structural finding, and the individual stands inside it are symptoms. This runs on the same category field that the peer groups behind every stand-level percentile are built from, which is E24's ground, so getting the categories right pays twice.

Use the median for each category, not the mean. Lead volume within a category is skewed in the same way it is across the floor, so one large stand in a small category will drag its mean up and hide exactly the problem you are looking for.

Take a floor of nine product categories and their median unique leads per exhibiting company:

Automation 212. Test and measurement 205. Fluid power 198. Materials handling 176. Safety and personal protective equipment 164. Bearings and drives 158. Welding 149. Coatings and finishing 121. Software and services 96.

Sorted, the middle value is 164, so the floor median of category medians is 164. That is the reference point.

Why not use the standard deviation to flag a category?

Now you need a threshold for what counts as far below. The standard deviation is the obvious candidate and the wrong one, for a reason that is easy to see with nine values.

Software and services at 96 is the outlier you are trying to detect. It is also one of the nine numbers going into the standard deviation, and it inflates it. The more extreme the outlier, the wider the band it creates, and the harder it becomes for that outlier to be flagged. The measure is contaminated by the thing it is supposed to find.

Leys and colleagues made this argument directly in the Journal of Experimental Social Psychology in 2013, in a paper whose title is the recommendation: do not use standard deviation around the mean, use absolute deviation around the median. Their point is that both the mean and the standard deviation are sensitive to the observations you most want to detect, while the median and the median absolute deviation are not.

Work it on the nine categories. Take the absolute difference between each category median and the floor median of 164.

Automation 48. Test and measurement 41. Fluid power 34. Materials handling 12. Safety 0. Bearings and drives 6. Welding 15. Coatings and finishing 43. Software and services 68.

Sort those deviations: 0, 6, 12, 15, 34, 41, 43, 48, 68. The middle value is 34. The median absolute deviation is 34.

Flag any category more than one median absolute deviation below the floor median. The threshold is 164 minus 34, which is 130. Two categories fall below it: software and services at 96, and coatings and finishing at 121.

Should you scale the median absolute deviation?

Leys and colleagues describe multiplying the median absolute deviation by a consistency constant of 1.4826, which makes it comparable to a standard deviation for data that is normally distributed. On these nine values that gives 50.4, and the threshold becomes 164 minus 50.4, or 113.6. Only software and services flags.

Which to use is a judgement and I would use the unscaled version at the screening stage. The constant exists to make the statistic comparable to a normal distribution, and category medians on a show floor are not drawn from one. More practically, the screening stage should be generous, because the cost of investigating a category that turns out to be fine is one afternoon and the cost of missing one is a category that quietly halves over two editions. Save the scaled threshold for anything you are going to put in a report or act on commercially.

Say which one you used. A threshold that changes between editions because somebody preferred a different constant is not a threshold.

What a flagged category usually turns out to be

Three explanations account for most of them, and the data to separate them is already in your systems.

The first is audience mix. Pull the registration file and count how many attendees declared an interest in that product area, or hold a job function the category sells to. If software and services has 41 exhibitors and 900 registrants who ticked software interest, the category is oversubscribed against its own audience and the stands are competing for a pool too small to support them. That is a sales and recruitment finding, and it is fixable in the next cycle by capping the category or by buying audience against it.

The second is placement. Join the flagged exhibitors to their floorplan coordinates and look at where they sat relative to entrances, catering and the main stage. A category housed in the far corner of hall three behind a build-up of large stands does not get walked past, and no amount of scanning discipline fixes a footfall problem. Check the staffing story before you settle on placement, because how many hours the stands were actually scanning is E31's measure and it explains a surprising share of what looks like position.

The third is category definition drift, which is the least visible. Software and services is a category that accumulates exhibitors who do not fit anywhere else, and a bucket full of unrelated companies will show a low median because its members have no coherent audience to draw on. The tell is high variance within the category alongside a low median.

Registration interest and floorplan geometry give you the first two cheaply. The third one needs somebody from sales to read the exhibitor list and say whether those companies belong together.

Where lead volume is the wrong measure of a category

The CEIR 2015 study on exhibitor ROI and performance metric practices found lead generation ranking as the top stated objective for exhibiting, ahead of brand awareness and reinforcement. It ranks first, and it does not rank first for everybody.

Categories with long sales cycles and small buyer populations exhibit for reasons that lead counts describe badly. A capital equipment category selling machines at a million each may have eleven serious buyers in the building, and a stand that spoke to six of them had an excellent show and will sit near the bottom of your ranking.

So before you flag a category, check whether its exhibitors are the kind who should be measured on volume at all. If your exhibitor analytics reporting rolls several measures into one composite, remember that the weights doing the rolling are E22's subject and they will move a category ranking as readily as a stand ranking. The practical screen is the ratio of graded leads to total leads. A category whose stands grade a high share of what they capture is qualifying carefully against a small pool, and a low median there means something different from a low median in a category that scans everything that walks past.

The limit

Nine categories is a small population and the median absolute deviation computed over it is itself unstable. Add one category next edition, or merge two, and the threshold moves. This method screens, and it does not test. Treat a flag as a reason to look, never as a finding in its own right.

The second limit is that categories are not independent of each other. Two adjacent categories in the same hall share footfall, so a placement problem in one produces a low median in both and the aggregation cannot tell you which is the cause. You will need the floorplan for that, and sometimes you will need to walk it.

The third is the one that catches people. This whole method compares categories within a single edition, so a category that has been quietly declining for three years alongside everything else on the floor will never flag, because it is not below the current floor median by enough. Run the same calculation on the previous two editions and look at whether any category's distance from the floor median is widening. That trend is the finding the single-edition version cannot produce.

Compute the median unique leads for each of your product categories from last edition, write the nine or fourteen numbers down in a column, and take the median of that column. Then work out the median absolute deviation by hand, which takes about five minutes, and see which categories fall more than one deviation below. Start with those before anybody rings an individual stand.

Questions people ask about exhibitor category performance

Why use the median absolute deviation instead of the standard deviation?
Because the category you are trying to find is one of the values going into the calculation, and it inflates the standard deviation. The more extreme the outlier, the wider the band it creates, and the harder it becomes to flag. The median and the median absolute deviation are not moved by the observations you most want to detect.
What causes a whole product category to underperform at a show?
Three things account for most of it. Audience mix, where the category is oversubscribed against the number of registrants who declared an interest in it. Placement, where the category sits away from entrances, catering and the main stage. And category definition drift, where a bucket has accumulated unrelated companies with no coherent audience to draw on.
Should you multiply the median absolute deviation by 1.4826?
Not at the screening stage. The constant makes the statistic comparable to a standard deviation for normally distributed data, and category medians on a show floor are not drawn from a normal distribution. Screening should be generous, because investigating a healthy category costs an afternoon. Save the scaled threshold for anything you will report or act on commercially.

Related reading

All exhibitor performance articles