Booth size unit conversion errors that quietly corrupt a portfolio roll up

The portfolio report says the group's average booth size is 45 square metres and rising. Somebody who has walked all ten shows knows that is wrong by a wide margin, because most of the floor is 9 and 18 square metre shell scheme, and says so in the meeting.

Booth size unit conversion errors are what produced that 45, and the interesting part is where they hid. Nobody made a mistake with the conversion factor. The four United States shows in the group were loaded in square feet, the six European shows in square metres, and the column they share has no unit in it.

Where does the error actually enter?

Rarely at conversion. A team that knows it is converting usually converts correctly, and the factor and rounding rules for a single figure are P16's subject.

The error enters where nobody thinks a conversion is happening. A registration or space management system exports a booth area as a number. The warehouse has a column called booth_area and the loader writes the number into it. Six months later somebody joins two shows' data and computes an average, and the join has silently mixed two measurement systems that both look like plain integers.

The two authoritative worlds it spans are both legitimate. UFI's Auditing Rules for the Statistics of UFI Approved Events, dated June 2021, certify net exhibition space and report exhibitor counts against their number of square metres of exhibition space. The CEIR Index, published through IAEE, builds its measure from four components starting with NSF, net square feet (IAEE, 2026). A portfolio with shows in both markets holds both, permanently, and no amount of standardisation at head office changes what the source systems produce.

Two stands that look identical and are not

Start with the two most common products in the industry.

A 3 by 3 metre stand is 9 square metres. Converted, it is 96.9 square feet. A 10 by 10 foot booth is 100 square feet. Converted, it is 9.29 square metres.

Put those side by side and the difference between them is 3.2 per cent. Small enough that an exhibitor moving from a United States show to a European one gets what they expected. Small enough that a product catalogue can call both of them a standard shell scheme unit. And small enough that when the two numbers, 9 and 100, land in the same column, nothing about either value looks wrong.

That is the property that makes booth areas dangerous compared with, say, dates or currencies. A 2026 date in a column of 2025 dates is visible. A euro amount in a column of yen is visible by magnitude. A 100 in an area column is a completely normal booth in one system and a good sized one in the other, so row level validation has nothing to grip.

What an unlabelled column does to a mean

Take a small mixed set and work it by hand.

Forty per cent of the stands in the group are 10 by 10 foot booths, stored as 100. Sixty per cent are 3 by 3 metre stands, stored as 9. Average the raw column and you get 0.4 times 100 plus 0.6 times 9, which is 40 plus 5.4, or 45.4.

Now do it properly. Convert the first group to 9.29 square metres each. The true mean is 0.4 times 9.29 plus 0.6 times 9, which is 3.72 plus 5.4, or 9.12 square metres.

The reported average is 45.4 and the real one is 9.12, a factor of 5.0. That is the 45 in the opening, and the reason nobody caught it is that a mean is a derived figure nobody reconciles. Totals get reconciled against finance. Counts get reconciled against registration. Averages get published.

The distortion also moves with the mix, which is the part that turns a static error into a fake trend. If the United States share of stands rises from 40 to 50 per cent next year with no change in what anyone bought, the raw mean rises from 45.4 to 54.5, an apparent 20 per cent increase in average booth size. The real mean barely moves, from 9.12 to 9.15. Whether average booth size is genuinely falling is F32's question and it cannot be answered from a column like this one.

The roll up that is off by a factor of five

The portfolio total behaves the same way and is easier to demonstrate.

Six European shows contribute 84,300 square metres of net space. Four United States shows contribute 612,000 net square feet. Added raw, the group reports 696,300 of nothing at all.

Converted, 612,000 square feet is 56,857 square metres, so the group's real total is 84,300 plus 56,857, which is 141,157 square metres. The raw sum overstates it by a factor of 4.93.

An error that large gets caught, eventually, by somebody who knows roughly how big the group is. What survives is the same error applied to a subset: one show loaded in the wrong unit inside a set of nine loaded correctly. A single 30,000 square foot show entering a 141,157 square metre portfolio as 30,000 adds 27,213 square metres of floor that does not exist, inflating the group by 19.3 per cent, and the total is still the right order of magnitude.

Why is a segmentation error worse than a total error?

Because a total has an owner and a segmentation does not.

Define exhibitor size bands the way most portfolios do: small under 20 square metres, medium 20 to 99, large 100 and above. Apply those thresholds to a column holding square feet and every 10 by 10 booth, at 100, lands in the large band, along with every 10 by 20 at 200. The United States shows appear to have no small exhibitors at all.

The consequences run straight into commercial decisions. Small exhibitor retention looks like a European problem because the American shows report no small exhibitors to lose. A pricing analysis of the entry level tier has no United States data in it and nobody notices, because an empty band looks like a market fact. A first time exhibitor programme gets scoped against a population that is missing four shows.

None of this trips a check. The row counts are right, the joins are complete, the totals are within a plausible range, and every individual number traces to a real source system. The only symptom is that a band is empty for four shows, which reads as information rather than as a defect.

The unit column, and four rules for it

The fix is dull and it holds.

• A unit code on every area row, in its own column. Populated at write time by the loader that knows where the file came from, never nullable, and with no default value. A default is what turns an unknown unit into a confidently wrong one.
• A constrained domain of exactly two values. A free text unit column will fill up with sqm, m2, SQM, sq m, SF, sqft and Sq Ft inside a year, and then somebody writes a case statement that misses one.
• Source value preserved, canonical value derived. Keep the number as it was measured and certified, because that is the figure with an audit trail behind it, and materialise a single canonical column for analysis using the full factor.
• A range check on the canonical column. Flag any stand below 4 or above 5,000 square metres for review. It will not catch a 100 that should be 9.29, and it will catch the 30,000 that should be 2,787.

Backfilling an existing warehouse is the harder half. The usable signal is the distribution: a show whose booth areas cluster on 100, 200, 400 and 800 is almost certainly in square feet, and one clustering on 9, 12, 18 and 36 is in square metres. Classify by show and edition rather than by row, check each classification against one known certificate, and write the result into the new column with a note saying it was inferred.

Where this stops

A unit column fixes the unit. It does nothing about the definition, and the definitional differences between two shows' booth area fields are usually larger than the unit gap.

One show may record contracted area, another the built area after a downsize, another the area on the original floorplan. All three are called booth size, all three will convert perfectly, and a portfolio series built across them measures three different things in consistent units. Establishing which of the three you are holding takes a conversation with each operations team and cannot be inferred from the data.

The other limit is that inference has a failure mode. A large United States show whose booths are mostly 400 and 800 square feet looks, by clustering alone, plausibly like a European show with 400 and 800 square metre halls of pavilions. Where the distribution is ambiguous, the certificate settles it, and the definitions behind that certificate are what tell you which figure you should have been storing.

This week, run a count of distinct booth area values per show and eyeball the top ten for each. Any show whose modal values are round hundreds is a square foot show, and if your warehouse does not know that already, the net space figure it feeds and the hall ratio built on top of it are both currently wrong.