KPI tile labels that carry the filter and the window

The slide said Attendance, and underneath it 24,300, set large. It had been screenshotted out of a report by somebody in the commercial team and pasted into a deck for a prospective exhibitor.

Two weeks later a different deck, from a different team, said Attendance 41,600. Both had come from the same warehouse. The first was verified professional attendees across three show days. The second counted visits, so the same person walking in on Wednesday and again on Thursday contributed twice.

KPI tile labels are the last thing between a correct measure and a number that leaves the building meaning something else. They are also the cheapest fix in reporting, and almost every report I have looked at has the same defect: the label is one word.

What has to be on the label?

Three things, and a fourth when instruments differ.

The population. Who or what is being counted. Professional attendees, registrations, exhibiting companies, accounts with an outstanding document.

The filter. What narrows the population from everybody to this group. Verified, paid, excluding staff, excluding cancelled.

The window. What period the figure covers. Show days, to date, the prior edition, the trailing twelve months.

The instrument, when two of them could produce the same number. Entry scans against registration records is the classic event example, and the two disagree by whatever your scan coverage is.

The industry's own audit machinery already works this way, which is a useful argument to have available when somebody objects that the label is too long. UFI's requirements for a UFI Approved Event, read on ufi.org in August 2026, set the audited statistics as "The number of visitors (international and national/domestic) the net exhibition space and the number of exhibitors (international (foreign and multinational) and national/domestic)". On the visits question, UFI is explicit: "Visits (visitors + repeat visits) can also be counted but this should be clearly indicated on the standard audit certificate."

Read that last clause as a labelling rule written by an auditor. The figure may be reported. The basis must be stated alongside it. A tile is a certificate with a shorter print run.

Work the two numbers from the opening. 41,600 visits against 24,300 visitors is 1.71 visits per visitor across three days, which is a perfectly ordinary figure for a trade show and a completely unusable one if the reader thinks it is a headcount. The gap is 17,300 people who do not exist, and no arithmetic error was made by anybody.

What one extra line costs

The objection is always space, so it is worth pricing.

A label line set at 11 point occupies about 14 pixels of vertical height including its leading. On a page with eight tiles arranged four across in two rows, the extra line costs 14 pixels twice, so 28 pixels total. On a report canvas of roughly 900 usable vertical pixels that is a little over 3 per cent of the page.

Three per cent of the page height to remove the most common category of reconciliation argument is not a close decision. If the tile genuinely has no room, the tile is too small, and the fix is the size and position question rather than the label.

Set the label below the number rather than above it. Above, it reads as a heading and gets skipped by anyone scanning for figures. Below, it reads as a qualification attached to the value, which is what it is, and it stays attached when somebody crops a screenshot around the tile.

Eight tiles relabelled

Take a real weekly page and rewrite its labels. The whole exercise took an afternoon, most of which was arguing about two of the eight.

Registrations became Registrations created, all types, to date. The all types matters because this show has six registration types and two of them are exhibitor staff.

Attendance became Professional attendance, show days, entry scans. Three phrases, one line, and the audit team stopped asking which figure the page was showing.

Exhibitors became Exhibiting companies, contracted, current edition. Contracted rather than confirmed, because the show tracks both and they differ by around 40 accounts in any given month.

Space sold became Net square feet, contracted, current edition. Net rather than gross, and the label says so, which ended a recurring disagreement with the operations team who work in gross.

Rebooking became Rebooked on site, per cent of exhibiting companies, last close. The denominator is the part that had been invisible, and it is the part two teams had been reading differently.

Revenue became Booked revenue, exhibit space only, current edition. Excluding sponsorship, which had its own tile and had been double counted in a summary once.

No-shows became Registered non-attendees, professional only, last edition. Nobody liked the phrase and everybody agreed it was unambiguous, which is the right trade for a tile.

Website sessions was deleted, because when the label was written out in full nobody could say what decision it fed. That happens more often than you would think, and writing the label honestly is a surprisingly effective way to find tiles that have no purpose.

Six of the eight labels were agreed in about ten minutes. Attendance and rebooking took the rest of the afternoon, and both arguments were worth having, because in each case two people had been quoting different numbers to different audiences for a year.

The rebooking one is worth working through, because the denominator argument recurs on every show. Sales had been reporting 68 per cent, calculated as 280 rebooked accounts over the 412 exhibiting companies at the last edition. Finance had been reporting 74 per cent, calculated as the same 280 over the 380 accounts eligible to rebook, since 32 had already been told the show would not be inviting them back. Both figures are defensible. The gap is six points, which on a renewal target is the difference between a good year and a difficult conversation, and it existed entirely because the word Rebooking on a tile does not carry a denominator.

Writing the denominator into the label forces the choice to be made once, in daylight, by the person who owns the number. That is a better outcome than the alternative, which is two departments each quietly believing the other one is using their figure.

Why isn't the measure name enough?

Because the measure name lives in the model and the label lives on the page, and only one of them gets screenshotted.

A well-formed measure name carrying its own filter and window solves the problem for the report author choosing from a field list. It does nothing for the exhibitor who receives a PDF, or the board member reading a slide, or the colleague forwarding a picture of a tile. Naming and labelling are two controls at two different points, and a team that has done the naming work properly still needs the labels, because the audiences are different.

There is also a translation step between the two that is worth doing deliberately. A measure named Attendance Professional Show Days is precise and reads like a database object. The tile label Professional attendance, show days, entry scans says the same thing in the register a show director uses out loud. Keeping the two mapped one to one, so that every label corresponds to exactly one measure, is the discipline that makes the pair work.

Labels that leave the building

Anything shown to an exhibitor, a sponsor or a board needs more label than an internal page, for the simple reason that the reader cannot ask you what it means.

Add the edition explicitly. Internal readers know which show they are looking at. An exhibitor holding scorecards from four of your events does not, and Current edition means nothing on a page that will be opened again in eighteen months.

Add the date the figure was calculated, near the tile rather than in a footer. Whether the underlying feed is fresh is a separate question with its own controls, and the point here is narrower: a figure with a visible as-at date is a figure nobody can accidentally quote as current two years later.

Events Industry Council publishes an industry glossary that ran to roughly 1,350 terms when I checked it in August 2026, which tells you two things. The vocabulary genuinely is contested enough to need one. And nobody receiving your report is going to open it.

Where this stops

A label states what the tile claims to count. It cannot prove the claim, and a confidently labelled tile fed by a broken pipeline is more dangerous than an ambiguous one, because it invites the reader to stop checking.

The label also has a hard length limit that real definitions sometimes exceed. Professional attendance in some shows depends on a job function mapping with two dozen rules and a manual review queue, and no one-line label carries that. What the label can do is name the population precisely enough that the reader knows which definition to go and read, which means the full definition has to be reachable from the tile itself rather than from a document nobody opens.

The last limit is social. Labels get shortened, by people who find them ugly, in the week before a board meeting. If the labelling convention is not written down and owned by someone, the page will drift back to single words within two reporting cycles, and it will drift silently because a shorter label never looks like a defect.

Take a screenshot of your main report this week, crop it to a single tile, and send it to a colleague in a different team with no context. Ask them what the number means. Whatever they get wrong is the label you rewrite first, and it will take you an afternoon to do all eight, which is the best return available anywhere in an event business intelligence layer.