How the 'attention funnel' can help us understand cross media measurement
This week, we launched a new report in partnership with TVision and Ebiquity called ‘The Challenge of Attention’. You can download the full, interactive turtlised report here (and even read it on your mobile phone!). It is an interesting companion piece to the work we have been doing with Dentsu.
But in the meantime, here’s a taste of one of the chapters:
There is a big difference between what people can see and what they do, in fact, end up looking at. Attention specialists Lumen and TVision have created the concept of the ‘attention funnel’ to help us understand the differences between attention to advertising across different media.
In this model, attention can be thought to flow from top to bottom: from what people could see (whether or not it is technically ‘viewable’ or not) to what people do, in fact, look at, and how long they actually look at it for.
At the top of the funnel is what people could see: was the ad served on the screen? After all, you can’t look at something that isn’t there. We count the ad as being on the screen even if only one pixel is viewable for less than a second.
Next, it’s worth thinking about which ads technically ‘viewable’, according to the Media Ratings Council (MRC) standards. A digital display ad is viewable if at least 50% of the pixels of the ad are available to be seen for 1 second or more, or 2 seconds plus for digital video advertising. According to BARB, the audience rating organisation for TV in the UK, someone has to be in the room and ‘available to view the ad’, without a minimum time requirement.
However, it’s important that we appreciate that ‘technical viewability’ is a man-made standard. It defines a minimum threshold: if your ad doesn’t achieve this level of viewability, then it doesn’t count (and, under certain trading deals, you don’t have to pay for it). And because it’s man-made, it’s a bit arbitrary: why 50% of the pixels and not 37%? Why 2 seconds of video time and not 3.1 seconds? Or 10% of the run time? Or something else entirely? We have included in the following charts for reference purposes, but as we will see, ‘technical viewability’ has only a tangential connection with actual viewing.
Next, we come to actual attention itself: not just the opportunity to see an ad, but actual viewing. At Lumen, we define an ad as viewed if it receives a single eye fixation on the pixels of the ad. ‘Fixations’ can be variously defined and are in their own way almost as arbitrary as viewability standards. The way we define fixations assumes that they occur 3-4 times a second.
Crucially, this means that people can look at ads even if they are not technically viewable by MRC standards. For instance, only 49% of the pixels might be peaking up ‘over the fold’ – not enough to be technically viewable, but sometimes enough to get looked at. Or the whole ad might be on the screen, but not for the requisite 2 seconds: it might get looked at even if it doesn’t quite cut the mustard as a viewable ad, technically speaking.
And, equally crucially, it means that ads that are fully viewable according to the MRC don’t actually get viewed. Your ad was technically viewable, but people politely declined the ‘opportunity to see’ it.
But seeing the ad at all is only half the story. The next important piece of information is how long people look at the ads for: the eyes-on dwell time. This does not need to be continuous: you can look at the first 2 seconds of an ad, look away and then look at the last 2 seconds and we will record the attention as being 4 seconds in total.
The data from the panels is all collected on an individual basis, but reported as mean averages: ads of a certain type or format, or on a certain platform or domain, will have a greater or lesser chance of being viewed, and their dwell time will be averaged. There is considerable variance in how long ads are engaged with: some people will merely glance at an ad, while others will invest a lot of time in the same ad. But for simplicity’s sake, we take a mean average.
Finally, we can create a composite metric that combines both the average likelihood that someone will view a particular type of ad and the average time that they spend looking at the ad. We call it ‘attentive seconds per thousand impressions’.
Attentive seconds per thousand = % viewing x eyes on dwell time (sec) x 1000
Imagine 1000 impressions served to a screen (let’s not consider imponderables like fraudulent or non-human traffic ads in our calculations for now). Of these 1000 ads, how many get looked at – whether they are technically viewable or not? And what is their average eyes on dwell time? If you multiply one number by the other, you get the average aggregate attention produced by 1000 ad impressions served to a screen – a unit of analysis that is consistent across TV, mobile and desktop advertising. An attention currency.
And we can put this attention currency to use by comparing attention across media.
We can use the ‘attention funnel’ approach to compare the ability of each media to get people to look at advertising at all, and how good they are at holding people’s attention. This will allow us to compare apples with apples.
As previously noted, in general, there’s a big difference between what people could see and what they do, in fact, end up looking at. And there are big differences across different media too.
In the first place, we see that not all TV ads are ‘viewable’. Yes, they appear on the screen, but sometimes there’s no one in the room to watch them (or that person has fallen asleep!). TVision estimate that 74% of 30-second TV ads play out to someone in the room – meaning that 26% play out to empty rooms.
But just because someone is in the room, it doesn’t mean that they will definitely look at the ad. In fact, only 43% of 30-sec TV ads get looked at. People may be in the room, but they are checking out their phone, or reading the paper, or talking to their loved ones, or getting the kids ready for school.
Almost all YouTube ads are ‘viewable’, and the vast majority of them get some attention.
Interestingly, when it comes to social media, many ads fail to meet the stringent MRC-viewability standards, but do get some attention, leading to an interesting anomaly where viewing rates are higher than technical viewability rates.
Bringing up the rear of the chart, the desktop and mobile web data shows us the reverse: if ads are viewable to MRC standards, that is no guarantee that thy will get viewed.
Next, we can look at how long people look at ads for. Here we can see that if people look at TV ads, they tend to look at them for a long time, relatively speaking: a 30 second TV ad will generate around 13.8 seconds of eyes on dwell time, on average. Within this average there will be some people who watch the whole 30 seconds of the ad, and others who only glance at a couple of seconds, and a wide distribution in between. But for simplicity’s sake, we can use the mean average as a benchmark.
A 15-second YouTube ad will not get watched for the 15 seconds: people know the Skip button and are not afraid to use it. On average, dwell time with 15 second YouTube ads is 4.9 seconds.
Dwell time with social media ads is much lower, which is largely a result of the scroll velocity. If the ads are on screen, then they are extremely likely to be viewed. But they are frequently not on screen for very long, and so not available to be looked at for a long time. That said, there are plenty of examples of ‘thumb-stopping’ ads that people do stop and stare at: as with TV and YouTube, there is a wide distribution in the data.
Finally, there is the dwell time with desktop and mobile display, which is in line with the dwell time norms for social media.
Mean averages can obscure as much as they reveal. To get a true picture of the reality of attention we should also look at the distribution of attention. Sure, if someone looks up an TV ad, they will look at it for around 14 seconds on average – but how is that average constructed?
Below, we have plotted the distribution of average aggregate dwell time with ads in different media. This chart shows the percentage of people who look for one second, or two seconds and so on – in total. They may not be watching from the start of the ad, and they may not be watching consecutive seconds: they may look at the screen, look away, and then look back.
Looked at this way, we see that the distribution of attention varies greatly. Most digital and social media formats have a ‘fat head’ and a very ‘long tail’, suggesting that most people merely glance at ads, but, occasionally, if they find the ads useful or engaging, they can spend a very long time with them.
The YouTube data suggests that as soon as the skip button appears, the audience disappears.
And from the TV data, we can see that far more people get to the end of a 30-sec TV ad.
All these different views on viewing are interesting, but it would be helpful to have a single number we can use to compare between media: how many YouTube ads ad up to the same amount of attention as a typical TV ad? How many mobile web ads would I have to buy to create the same amount of attention as an ad on Facebook?
We can create this number by combining the viewing percentage (how many people actually look at the ad) with the mean average eyes-on dwell time (the time they actually spend looking at the ad) and multiplying it by a thousand (as media are always traded in thousands). We call this the aggregate ‘attentive seconds per thousand’ or ‘aPM’.
For instance, if you were to buy 1000 30-second TV ad impressions, we would predict that 43% or 430 of them would be viewed, but they would be viewed for around 14 seconds each, generating around 6000 attentive seconds. 920 of your 1000 YouTube impressions might get looked at, but for 4.9 second on average, generating 4500 attentive seconds. And so on.
By following the logic of the attention funnel consistently across media, we have been able to create a common currency of attention across media. And this very rough and ready calculation shows us that the average 30-sec TV ad generates the same amount of attention as 1 ½ YouTube ads, or 4 ½ Facebook in feed ads, or about 40 desktop display ads. Suddenly, we can start comparing apples with apples.