Counting impressions on physical signage without pretending you measured attention

The sponsor's post-show report said 42,000 impressions against a registration hall banner. Somebody asked how that had been counted and the answer came back as attendance times three, which turned out to mean attendance times three because three felt about right.

Counting impressions on physical signage is a defensible exercise if you do it in the open. Three inputs, multiplied, with every input written down and none of them dressed up as a measurement it is not. The version that gets a show into trouble is the one where the three inputs collapse into a single integer before the report leaves the building.

The three inputs, and why there are only three

An impression count for a physical asset is traffic multiplied by pass frequency multiplied by a visibility factor. Everything else is a refinement of one of those three.

Traffic is how many distinct people came within range of the asset. Pass frequency is how many times each of them did so. The visibility factor is the share of those passes where the sign plausibly registered with the person walking under it.

The first is close to measurable at most shows. The second is estimable with effort. The third is a judgement, and the whole credibility of the number rests on treating it as one. Multiply the three and you get a figure whose units are people-times-passes-times-a-guess, which is worth reporting and never worth reporting alone.

Where the traffic estimate is allowed to come from

The traffic input has to be a count of people who could physically see the asset, which is usually smaller than total attendance and occasionally larger.

Smaller, when the asset sits inside one hall of three and only 62 per cent of attendees ever entered that hall. Larger, when the asset sits in a public concourse that also carries hotel guests and staff, in which case you either exclude them or say plainly that they are in there.

The best source is your own access control data, which already knows who entered which hall and when. Where the asset is in a space with no scanning, you need a counting method, and choosing between camera counting, beacons and manual sampling is a proper subject in its own right that belongs with the sensing cluster. What matters here is that you record which method produced the figure, because a gate scan and a sampled manual count carry very different error, and the person reading your report in eight months will have no way to tell them apart.

Working one banner end to end

The registration hall banner, with illustrative figures.

Access control records 7,400 unique badges entering the registration hall across three days. That is the traffic input, and it excludes exhibitor staff, who scan through a separate entrance.

Pass frequency comes from re-entry scans. The 7,400 badges generated 19,240 entry events in total, so the average is 2.6 entries per attendee. That is a genuinely derived figure, which is unusual and worth exploiting where your access control supports it.

Gross exposures are therefore 19,240. Note that this equals the entry event count directly, because each entry is one pass through the banner's viewing zone. The multiplication of 7,400 by 2.6 is a check on the arithmetic and not a second source.

Apply a visibility factor of 0.30 and you get 5,772 adjusted impressions. The factor is where all the argument lives, and how to construct and defend one is a separate piece of work. For the count itself, the important discipline is that the 0.30 appears in the report as its own line, so that a reader can substitute their own value and see what happens.

How does a measurement body actually build this?

Out of home has spent decades on the same problem, and the structure of its answer is worth borrowing even where the specifics do not transfer.

Geopath's best practices, standards and protocols document, published by the Geopath Futures Council in August 2019, sets out seven components that feed an out of home impression: vehicular traffic counts, persons per vehicle, pedestrian traffic, illuminated circulation, visibility adjustment, speed and dwell time, and home location. The document defines an impression as "the number of eye contacts that people have with an OOH unit(s) in a week", which is a much stronger claim than a pass count and explains why so much machinery sits underneath it.

Two details from that document transfer directly to a hall. The first is that visibility gets decomposed into named attributes: the components table asks "How large does the media APPEAR within the audience's field of view? WHERE is the media within the audience's field of view? How much TIME does the audience have to see the media?" and records that "Observed dwell time, degrees off-center (at optimal view), and apparent size (at optimal view) taken into account."

The second is the level of specificity applied to inputs that sound trivial. The same document notes a change of default walking speed, recording that the "New default walking speed is 3.1 MPH (vs. 3.4 MPH)." An organisation that publishes its walking speed assumption to one decimal place has understood something about where credibility comes from. It also warns its own market that a methodology change breaks comparability, saying "we can no longer compare yesterday's impressions to today's", which is a sentence most shows should be prepared to write in a footnote when they change how they count.

Digital did the same thing in its own way. The Media Rating Council's Viewable Ad Impression Measurement Guidelines of 30 June 2014, produced with the IAB Emerging Innovations Task Force, requires 50 per cent of an advertisement's pixels in the viewable space for at least one continuous second. Both industries wrote a rule and lived with it. Exhibitions have not, which leaves the rule-writing to the individual organiser.

Publish all three inputs beside the answer

The single most useful change to a sponsor report costs nothing. Show the working.

Instead of 5,772 impressions, print the four figures: traffic 7,400 unique badges from access control, pass frequency 2.6 derived from 19,240 entry events, visibility factor 0.30 by assessment, adjusted impressions 5,772. Four lines where there was one.

Three things happen. The sponsor can substitute their own visibility factor, which they will do anyway and would otherwise do in private. The weakest input becomes visible, which in almost every case is the factor. And when the number changes next year, you can say which input moved, so a 22 per cent fall can be traced to a hall layout change and explained in one sentence.

There is a cost. A sponsor who has been receiving a single confident number for four years will notice that it has become an estimate with a method attached, and somebody will have to explain why. That conversation is easier before a competitor forces it than after.

The band around the product is a related question with its own arithmetic, and how wide the interval should actually be deserves its own treatment, and a footnote will not carry it.

What should you do when one input is missing?

Say so, and report the count as a range or as unavailable.

The common failure is to substitute a default and carry on silently. A hall with no scanning gets total attendance as its traffic input, a corridor with no re-entry data gets a pass frequency of three, and the report reads exactly like the ones built on real figures. Six months later nobody can tell which numbers were derived and which were filled in.

A workable convention is a short tag against every figure in the sponsor report: measured, derived, or assumed. Traffic from access control is measured. Pass frequency from entry events is derived. A visibility factor is assumed. On a typical signage asset you will find one measured input and two others, which is the honest shape of the thing.

The tags earn their keep in aggregate. Run them across a full sponsor report and count how many of the headline figures rest on at least one assumed input. On most shows the answer is nearly all of the physical inventory and almost none of the digital, which is a useful thing to know before you decide where next year's measurement budget goes. It also tells you which assets you can defend line by line in a renewal meeting and which you would rather discuss in general terms.

Where this stops

The count measures passes, adjusted by a judgement. Attention is what the sponsor wants and gaze is what an impression is supposed to represent, and neither is being observed here.

That gap is not fatal as long as the language matches. Calling the output adjusted exposures is more accurate than calling it impressions, and slightly harder to sell, and the trade is worth making for any sponsor you expect to keep for five years.

The second limit is that access control data tells you about entries into spaces, and signage lives at points inside those spaces. A banner at the far end of a hall is seen by a smaller share of the hall's entrants than a banner over the entrance, and no entry count knows the difference. Where an asset sits deep in a space, the visibility factor is carrying that geometry too, which is more weight than one number should hold.

Take one signage asset from your last edition and rebuild its impression figure as four lines instead of one. If you cannot produce the traffic input from access control, that asset's number is an assumption end to end, which is worth knowing before it appears in the pricing conversation or in the report the sponsor reads next season.