A new engineered feature for video buying has been released and is available for use with the AppNexus Programmable Bidder (APB). We are happy to announce that
predicted_iab_video_view_rate can now be utilized in Bonsai Trees, bringing viewable video buying to the programmable age. "Predicted video view-rate" is the likelihood that a video impression will be measured as viewable by the IAB's industry standard definition of "view-rate" for a video creative. The standard states that a video is considered viewable if 50% of the creative's pixels are on screen for a minimum of 2 seconds.
To predict whether an impression will meet the IAB's criteria, AppNexus begins by gathering metrics for creatives via our native viewability technology on pages where an AppNexus tag is embedded. Our Data Science team has worked hard to create updated viewability models that utilize those metrics for video impressions, augmenting the successful models that we have already productionalized for display viewability prediction. While the math behind this calculation is not in the scope of this article, you can read more about our technique in a previous Tech Blog post, "Ad Viewability and Feature Selection for Big Data". Be prepared to dive into the wonderful world of logistic regression and L1 Regularization!
Similar to how many users of APB have already found success using existing viewability features for display buying, we envision that our APB alpha partners will be able to extend Bonsai's functionality to their video buying.
The two short example Bonsai trees below illustrate how this can be used in practice.
In the code examples below, comment lines (beginning with
#) are used to explain tree logic.
predicted_iab_video_view_rateis a value between 0 and 1.
The following tree optimizes towards video placements with a predicted viewability rate of 60% or more, and does not bid in other cases:
# If the predicted rate is greater than 60%, bid 2.50 # Otherwise, do no bid on this impression if predicted_iab_video_view_rate > 0.6: 2.50 else: value: no_bid
Please note that in some cases our prediction may not be available for every bid request. To test for the presence of our prediction, you can use the
absent keywords and handle that case as you wish.
Engineered features can also be used in Bonsai "Smart Leaves" to calculate a final bid price from the predicted value. In this more elaborate example, we test for the presence of segments, and use the
predicted_iab_video_view_rate feature in a compute() function:
# If the user is in segment id: 123, or id: 456 # use the value of the predicted rate, multiplied by 3 if any segment, segment: value: compute(predicted_iab_video_view_rate, 3, _, _, _) else: value: no_bid
We are excited to get this feature into the hands of our APB Alpha testers. As always, we welcome feedback on your experience in using the product, and look forward to hearing about how video buying in APB works for you!