Modelling the venue and hall dimension when a show changes buildings

The request sounds like an afternoon of work. Add venue to the model so we can compare performance across buildings.

Then you look at the source data and find that the venue and hall dimension has to describe two things moving at completely different speeds. The building changes once every few years and outlives the show team. The hall layout changes every single edition, sometimes twice, because a hall came free or a feature area was moved.

Putting both in one table gives you a dimension that either loses the building's history or duplicates it forty times.

Two speeds in one set of attributes

Sort the attributes by how often they change and the design writes itself.

The building's attributes are stable. Gross indoor exhibition space, total hall count, city, country, time zone, operator, transport links. UFI's World Map of Exhibition Venues, 2025 edition, published on 9 December 2025, is a census of exactly these attributes: 1,530 venues worldwide with at least 5,000 square metres of gross indoor exhibition space as of the end of 2024, holding 44.3 million square metres between them. Divide 44.3 million by 1,530 and the average venue in that census carries about 28,950 square metres of gross indoor space.

Those numbers move slowly. A venue adds a hall and the figure steps up once, then sits still for years.

The staging's attributes change every edition. Which halls were taken. How many. Gross square metres occupied. Net square metres sold. The floorplan version. Whether the feature area sat in hall 2 or hall 4. Whether the conference ran in the same building or across the road.

The split follows the speeds. The building goes in a venue dimension with its own key. The staging goes on the edition row, which already exists and already has one row per staging.

What belongs to the building and what belongs to the staging?

The test is whether the attribute would still be true if your show never ran again.

Gross indoor exhibition space is a property of the building. It would be the same if you cancelled. Venue dimension.

Halls occupied is a property of your contract with the building for one edition. It would not exist if you cancelled. Edition row.

Time zone is a property of the building and it is the attribute people forget. Every timestamp in the badge scan file is meaningful only against it, and a portfolio with editions in Frankfurt, Singapore and Chicago has three offsets that need to be applied consistently or half your on-site hourly analysis is an hour out.

Net square metres sold is the interesting boundary case. It is a commercial outcome of one staging, so it goes on the edition row, and it should be sourced from the space contract fact rather than typed in. The ratio between the gross space occupied and the net space sold is a measure organisers argue about properly, and the gross to net question has its own treatment.

Five venue rows and forty edition rows

Here is the shape on a realistic portfolio.

Eight show brands, five years of history, so 40 editions. Those 40 editions were staged across 5 buildings, because two shows share a venue and one moved in 2025.

The venue dimension has 5 rows. Each carries gross indoor exhibition space, hall count, city, country, time zone and operator. Forty edition rows point at those 5 rows through a venue key, and each edition row carries its own hall count and the gross square metres it occupied.

Take one edition occupying three halls of 12,000, 9,400 and 7,600 gross square metres. That is 29,000 gross square metres occupied. The following edition drops the smallest hall and occupies 21,400, a fall of 7,600, which is 26.2 per cent of the previous footprint.

That single number explains a lot of what a post-show report will otherwise blame on demand, and it is available from a dimension with 45 rows in it before anyone opens a fact table.

Averaging across the 5 venues, the portfolio stages 8 editions per building over the period. That is enough repetition for a venue comparison to be worth running and nowhere near enough for it to be statistically clean, which is worth saying out loud before somebody builds a league table of buildings.

Why does a venue move wreck the trend?

Because most models attach venue attributes to the show and then overwrite them.

The Kimball Group's technique page on slowly changing dimension type 1 describes what that does: "the old attribute value in the dimension row is overwritten with the new value; type 1 attributes always reflects the most recent assignment, and therefore this technique destroys history." Applied to a venue attribute on a show record, the 2023 and 2024 editions start reporting the 2025 building's square metres, and every space utilisation figure in your archive shifts.

Pointing each edition row at a venue key avoids this without any type 2 machinery, because the edition row is already the historical record. Edition 2023 points at venue 2. Edition 2025 points at venue 4. Nothing was overwritten, and a query asking for space sold by building groups correctly across the move.

The venue dimension itself still needs a history policy for its own attributes, which is where type 2 comes back. When a building opens a new hall and its gross space goes from 29,000 to 41,000 square metres, an overwrite makes your 2022 utilisation figures look worse than they were. The mechanics of effective and expiry timestamps are the same ones an exhibitor company dimension needs, and they are worth applying here on the two or three venue attributes that actually feed calculations.

For attributes that feed nothing, an overwrite is the right call. Nobody needs the history of a venue's postal address.

Is a separate hall dimension worth building?

Only if you report at hall level inside a show.

Where your post-show pack breaks attendance and stand performance down by hall, a hall dimension keyed on venue and hall identifier earns its place: one row per physical hall per building, carrying gross square metres, floor loading, ceiling height and whether it is ground level. Five venues with an average of six halls gives about 30 rows.

The edition then needs a bridge saying which halls it occupied, at one row per edition per hall. Forty editions occupying three halls on average is 120 rows, and that little table answers questions the edition row alone cannot: which halls have been used by more than one brand, whether a hall's stand performance differs across shows, how often hall 4 gets taken.

If your reporting stops at show level, skip all of it. A hall count integer and a gross square metres number on the edition row cover every question you are actually asked, and 120 rows of bridge that nobody queries is 120 rows that will be wrong within two years because nobody maintains what nobody reads.

I would start with the integer on the edition row and add the hall dimension the first time somebody asks a question it cannot answer. Building it speculatively is how the data platform acquires tables with a stale load date and no owner.

Naming, and the venue as a string

The failure that costs the most is the simplest. Venue held as free text on the edition row.

Messe Frankfurt, Messe Frankfurt GmbH, Frankfurt Messe, Frankfurt. Four spellings, four groups in every chart, and somebody spends an afternoon each year fixing them by hand in a spreadsheet that then diverges from the warehouse.

A venue key with a controlled dimension ends it permanently, and the dimension is small enough to curate by hand. Five rows. Ten if the portfolio grows. This is the cheapest conformity win available, and it is the same principle that makes shared dimensions across shows work at portfolio scale.

Keep the source system's own venue string on the dimension row as an alias column, so a load that meets an unfamiliar spelling can be traced to the extract that produced it.

Where this stops

A venue dimension describes buildings and tells you nothing about why a show performs differently in one.

Two editions of the same brand in two cities differ by building, by catchment, by travel cost, by season and by whatever the economy was doing. Attributing the difference to the venue requires an argument the model cannot make for you, and with 8 editions per building there is not enough repetition to separate those effects statistically.

Hall-level attribution has a sharper version of the same problem. Stands in hall 1 near the main entrance outperform stands in hall 3, and that is a location effect confounded with the fact that the largest exhibitors buy the best positions. The hall dimension will make the pattern easy to plot and it will not tell you which way the causation runs.

There is also a practical limit on venue attributes. Gross indoor exhibition space is published by the operator and by UFI's census, and the number an operator quotes in a sales brochure is often not the number you can actually sell into. Record the source of each figure on the venue row, and where your own measured occupancy disagrees with the published capacity, trust your contract.

List the venues used by every edition in your portfolio this week, deduplicate them by hand, and give each one an integer key. If that list comes back longer than the number of buildings you have actually been in, you have found the spelling problem, and fixing it takes an afternoon now and a fortnight in three years.