Should view through conversions count towards your registration numbers

The display line on the plan spent 34,000 and the platform reports 612 registrations from it. Nobody clicked anything. All 612 are view through conversions, credited because the person was served an advertisement at some point before they registered.

The question in the room is whether those 612 go in the registration report. My answer is that they go in a separate line with their own heading and never inside the headline count, and the reasoning is worth working through because the argument recurs every edition.

What is a view through claim actually asserting?

The claim is that an advertisement was displayed to a person, the person did not interact with it, and a registration that happened afterwards is partly the result.

That can be true. Brand advertising works without clicks, and an exhibition brand seen repeatedly through the spring does change whether somebody registers in the autumn. The problem is what the claim depends on.

It depends on the impression having been rendered. It depends on the impression having been in view, which is not the same thing. It depends on the person having noticed it. And it depends on the registration not having been caused by something else entirely, which for a show that also sends email, buys search and works with exhibitor lists is the usual case.

The Media Rating Council's viewable ad impression guidelines, first published in June 2014, set the desktop threshold at 50 per cent of the pixels in view for at least one continuous second, with two continuous seconds for video. Version 2.0 followed in August 2015 and kept that baseline for desktop. Half the pixels for one second is the floor for calling an impression viewable at all, and it is a low bar in absolute terms: an advertisement can meet it, sitting half on screen for one second while somebody scrolls past, and be counted as viewable.

So a view through conversion rests on an event whose weakest acceptable form is half an advertisement visible for one second. Write that sentence out before deciding whether to include the 612 in a board number.

Two settings decide how big the number gets

The window and the viewability requirement do almost all of the work, and both are usually left at whatever the account was created with.

Google Ads help documentation sets the view through conversion window to a default of one day, adjustable up to 30 days, against a click through conversion window that defaults to 30 days. A one day view through window is defensible: the person saw an advertisement and registered within 24 hours. A 30 day view through window on a show with a nine month cycle will claim a large share of every registration you get, because over 30 days a broad display or social campaign will have served an impression to most of your addressable audience at least once.

The arithmetic is easy to sketch on your own numbers. If a campaign reaches 180,000 people in a month and your show registers 12,000 people in that month, and reach is even mildly correlated with being in your target audience, the campaign will be able to claim thousands of registrations on impressions alone. It did not produce them. It was present.

Set the view through window to one day unless you have a specific reason, write the reason down when you do not, and require the viewable impression standard in the insertion order so that impressions which never met the floor cannot generate claims.

How do you de duplicate when two platforms both claim it?

They will both claim it. Two platforms measuring independently, each with its own tag, each with its own window, will each count the same registration, and neither has any way of knowing about the other.

Take one week of a live campaign. The registration file holds 2,400 rows for the period. Platform A reports 1,180 conversions and platform B reports 640, which sums to 1,820. Match both against the file on click identifiers and hashed email and you find platform A can point to 1,020 real rows and platform B to 540, with 260 rows claimed by both. The unique count of claimed rows is 1,020 plus 540 minus 260, which is 1,300.

The two platforms between them reported 1,820 registrations against 1,300 real rows, which is 520 of overcounting, or 40 per cent on top of the truth.

Now resolve the 260 contested rows with a rule stated in advance. Any click beats any view. Of the contested rows, 180 have a click from platform A and only a view from platform B, so they go to A. 40 have a click from B and a view from A, so they go to B. The remaining 40 have views from both and no click at all, so they go to neither and sit in a bucket named unresolved.

Final allocation: platform A 940, platform B 320, unresolved 40, and the total is 1,300. Nobody's number matches what they reported, and both are now describing the same file.

The ordering that makes this work has three levels. A click beats a view. Where two clicks compete, the later click wins if you are running last touch, and the rule must be written down either way. Where only views compete, credit nothing, because assigning a registration to one of two advertisers who both merely appeared is a guess dressed as a decision. The general case of one registration claimed by several channels, including partners and codes rather than platforms, needs a wider set of precedence rules than this.

What the evidence says about crediting mere exposure

Lewis, Rao and Reiley ran three controlled experiments at Yahoo and published the results in the proceedings of the twentieth international World Wide Web conference in 2011. Their subject was what they called activity bias: people who were served advertisements on a given day were doing more of everything online that day, so any comparison between exposed and unexposed people attributes ordinary activity to the advertisement.

Their sharpest demonstration involved sign ups at a competitor's website, where far more people signed up on days they saw an advertisement, though the true competitive effect was minimal. The apparent effect was a property of the measurement rather than of the advertising.

View through counting is that comparison, industrialised. It takes the population that was served an impression, observes which of them converted, and credits the impression. On a registration file where the same people are also receiving your emails, seeing your search ads and being chased by exhibitors, the overlap between served and converting is guaranteed to be large whatever the advertising did.

That is the strongest argument for the separate line. Not that the effect is zero, but that the measurement cannot distinguish an effect from a coincidence, and the headline registration count should only contain numbers that can.

What I would actually report

Three figures per campaign, every edition.

Registrations with a click path, matched to rows in the file. This is the number that goes into the acquisition report and the cost per registration calculation.

Registrations claimed on views alone, matched to rows in the file, reported underneath with the window and the viewability standard named. On the display example above that is the 612, with a note that they rest on a one day window and a viewable impression requirement.

Reach and viewable impression counts, reported as media delivery rather than as registrations, which is what they are.

Keep the view through line stable across editions so the series is readable. If the window changes from one day to seven between March and October, the two numbers are not comparable and somebody will read the increase as performance. Record the window and the viewability standard alongside the count in the same table, the way a currency is recorded next to a price, and the comparison stays honest even when the settings change.

Then, once a year, the only thing that will settle the underlying question: a holdout. Withhold the display campaign from a defined slice of your audience, compare registration rates, and you have a measured answer rather than an allocation rule. That test costs reach and is worth running on the campaigns where the view through claim is largest.

Where this stops

A separate line does not make the view through figure meaningless, and I want to be careful not to argue that it does. Brand exposure at the top of a nine month funnel is real, it is hard to measure with clicks, and a show that refuses to count anything without a click will systematically underinvest in being known.

The honest position is that view through counting is evidence of exposure and not evidence of causation, and that the two belong in different columns. The 612 registrations in the opening example may include 200 that the display genuinely produced. They also include some number that email produced, some that exhibitors produced, and some that would have happened with no marketing at all.

Everything here also depends on the match between a platform claim and a row in your file, which fails in the usual asymmetric ways, and the wider walk from a platform reported figure down to what the file supports is its own reconciliation. The economics of the social inventory generating many of these impressions sits with the placement level view, and the acquisition and attribution pillar holds the rest.

Check two settings this week. Find the view through conversion window on every conversion action in your ad accounts, and find out whether your display insertion orders specify a viewable impression standard. If the window is 30 days and the insertion order is silent, you already know what the next report is going to claim.