Decrease of wasted impressions

Goal

Decrease wasted impressions for campaigns with viewable impression goals.

Case

Optimizing yield for ad inventory becomes harder and harder. Advertisers become more demanding and it is hard for publishers to keep up and proof the value of their inventory. An important metric is viewability, but it is hard for publishers to target on viewability.

Together with a publisher we added tracking to all ad slots. The tracking included:

  • How many times could we show an ad in the ad slot?
  • How many times was the ad slot unfilled?
  • How many times was the ad slot filled?
  • How many times was the ad slot viewed?

An ad slot was viewed if it was filled an at least 1 second for 50% in the viewport of the user.

With these metrics we could calculate the viewability for each adslot on each given time slot. Since the publisher collected the data themselves, they had the following benefits:

  • They were not reliant on their ad server for guessing whether an impression would be viewable.
  • They could use viewability targeting in both direct sales as well as their private marketplace.

All data was collected by Harvest Collect and stored in the Harvest Store. For each ad slot we calculated the viewability percentage, which was displayed in ranges of 10%. The data was available through an API of Harvest Connect.

This way, for each ad request to the ad server and header bidding, we were able to provide a key-value pair containing the viewability ranges. This allowed the publisher to specifically target certain campaigns to only serve for certain viewability ranges

The control group did not use the viewability ranges and therefore relied on the guesses the ad server made. For each ad request the ad server guessed whether it would become viewable or not.

Each wrong guess meant that an impression was served that did not count towards the viewable impression goal. So, each wrong guess was waste

The experiment group did use the viewability ranges. This way the publisher did not have to rely on the guesses of the ad server.

The question now was whether the viewability ranges we calculated were more accurate than the guesses of the ad server.

Outcome

The experiment group had 24% less wasted impressions. This meant that for each 1.000 impressions that were wasted by the control group, we were able to fill 240 in the experiment group.