As a result, advertisers, because third parties are unable to measure attribution and viewability together, have started insisting on high overall viewability rates on their campaigns so that they can adjust the CPA provided by their attribution vendor (often DART or Atlas); we call this the “High Viewability Strategy”.

At Programmatic I/O recently, comScore Executive VP Anthony Psacharopoulos shared results about viewability that suggested that higher viewability leads to higher lift of in-store sales. In particular, he pointed out that increasing viewability from 50%to 70% increased in-store sales lift by 63%. But this doesn’t take cost or scale into account; in our experience, to produce that lift in viewability it may require spending at least 100% more on the media, killing all the gains and then some.

As an alternative, the industry should consider counting only viewable impressions for purposes of attribution.

The High Viewability Strategy suffers from a few problems:

  • Placements that are consistently highly viewable are easy for many buyers to find and hence are very expensive.
  • Most valuable “premium” sites have many placements that are not usually viewable, but when they are, they can be a great place to advertise.
  • Reach is constrained to a small percentage of the total inventory: To reach 70% viewability, display ads supply shrinks to only 2.79% of the total, as seen in the graphs below, which are based on data from the Rocket Fuel Avails Tool and include viewability rates based on third party measurement. 
  • Once a campaign “falls behind” the campaign goal for viewability, it can be impossible to catch up. If the target is 70% but it’s only running at 60% in the first half, an average of 80% would have to be met in the second half of the campaign to hit the goal, and that supply is really tiny. 
  • Some viewability measurement “games the system” e.g. by delivering only in mobile apps where viewability is assumed to be 100%, but which may not be the best context for the advertiser’s message.

Instead, we recommend showing ads based on real-time predictions of both viewability and performance of that impression. 

Which would you rather buy for $3?

  1. An ad with 25% predicted viewability but 1% predicted conversion rate.
  2. An ad with 75% predicted viewability but only 0.1% predicted conversion rate?

The answer is (1), which gives us 0.0025 percent predicted conversions, while (2) gives us 0.00075 percent predicted conversions. But the High Viewability Strategy never buys (1).

If we had a way to ignore non-viewed impressions when computing attribution and CPA, we’d be golden. We need to ask attribution vendors to incorporate viewability into their reporting and attribution (which is not yet possible), and then we need to ask DSP partners to optimise toward last viewable touch, to drive the right behaviour. Of course, that would mean the DSP has to measure and jointly optimise toward viewability and performance of every single impression.

To understand the costs, you could negotiate a vCPM price (pay only for viewable impressions), or just compute the effective CPM on the viewable impressions after the fact by ignoring the non-viewed impressions.  

But the most important metric is the CPA based on last viewable touch: that is, attribute each conversion back to the last ad shown that was measured as viewable (or clicked on). It’s simple, it aligns with real business value, and it should become the new currency of programmatic display advertising.

What to do for now

This kind of attribution is not yet available, though DART is close, which makes it possible to report on viewable impressions, but you cannot yet attribute credit only to viewable impressions. Instead, for now we can only separately measure viewability and “performance”, where performance is a CPA that includes attribution to some non-viewed impressions. But that doesn’t completely prohibit us from optimising to conversions based on viewed impressions.

Today we recommend that customers include viewability goals in each campaign, but work through A/B testing to find the optimal level of viewability that maximises performance of viewable impressions, assuming that impressions which got credit for last-touch attribution had viewability that was distributed the same as the viewability of the overall campaign. Also, if you are working with multiple partners, hold them all to the same overall viewability standard, and relax that from 70% down to 50% or lower if it can achieve a better “effective” eCPA (after adjusting for the lower viewability).

In the above chart, 50% viewability is best because it has the lowest eCPA, even though the “waste” spent on non-viewable impressions is lowest at 70% viewability. This would not be uncommon as the high cost of delivering highly-viewable impressions drives the CPA up more than the benefit of avoiding the waste on non-viewed impressions decreases the “adjustment factor”. The best way to figure this out for your campaign is to conduct a split test at various levels of viewability, splitting both the budget and the audience.