Advertisers, consumers, and search engines have a healthy tension within the search environment. Advertisers want to reach the maximum number of qualified consumers while spending the least amount on advertising. Consumers want to see content with the least amount of interference from advertisers. And, search engines balance this tension, with their own needs by giving consumers a great experience, while maximizing their ad revenue and driving results for advertisers.
The latest episode in the evolution of search plays out after Google’s confirmation that they will stop showing ads on the right rail for desktop results globally (except for PLAs and knowledge panels). Keyword behavior will change along most metrics (positions, CPCs, impression share, etc) and the effects will differ across your keyword set depending on whether they are in high positions. Search management rules of thumb relating to average position and impression share will need to be refreshed and relearned. Performance will be temporarily volatile and for those who run campaigns manually, the next few weeks might entail several late nights.
But what about Adobe Media Optimizer’s (AMO) algorithms? Don’t they require massive adjustments too? Do you or your customers need to be worried about these changes? No.
AMO’s algorithms were built with three key principles, making them robust to search engine changes over the years. AMO’s algorithms are:
Resilient to changes in the auction mechanisms:
Keywords in our algorithms are modeled by bids and their effect is on number of impressions, clicks and revenue, not position. We realized very early on that position was an artifact, a side effect of the auction process. The true effect of a bid in the auction was on the number of impressions and clicks a keyword received. Thus, the reduction in number of available positions has no effect on the way our models work.
Self-correcting to marketplace changes:
Our models self-correct with volatility. While we will see volatility in performance in the coming weeks, our models account for the change in
keyword performance and correct their estimates without human intervention. What’s more, they smooth out short-term fluctuations to make for stable and accurate estimates.
Stable by statistical aggregation:
Long tail keywords with a click or less per day are notoriously hard to model. Data sparsity of both auction side (clicks, impressions, and CPC) as well as conversion side means that one cannot naively estimate how well a keyword will do. We overcame this limitation a long time ago by using finite mixture models that statistically aggregate similar keywords to build reliable models, even for the long tail. These models become even more relevant now as the changes in Google’s ad serving would mean that fewer keywords would get impressions and clicks. This would worsen the data sparsity problem.
We think that the current changes in Google’s ad-serving will be a test for many marketers. AMO’s stable and self-correcting models should assure customers that changes to the auction and ad placement are automatically incorporated into bid decisions (even if these changes aren’t public to market). Armed with robust and scalable algorithms (along with the right strategy), sophisticated marketers will continue to rule the day.