The digital media platforms have grown enormously in recent times and have also become more intelligent. Real-time bidding is a must for many DSPs and many platforms also come up with smart algorithms that can maximise performance. However, the whole concept of algorithms is often a black box, which makes it more difficult to estimate how to optimise your campaigns. To get the best out of the algorithms, one principle is very important: give the algorithm as much room as possible to optimise. This is also called liquidity. In this article, some best practises will follow to maximise liquidity and get the best out of these algorithms.
Let the algorithm compose the best creatives
Both Facebook and Google have features to assemble different creatives for campaigns. With Facebook's Dynamic Ads and Google's Responsive Display Ads (RDA), you have the option to enter multiple images and multiple copies. Facebook and Google then compile different combinations of these texts and photos in order to show different versions of advertisements to users. The algorithms' estimations concerning the user will allow the ad to be "personalised" to the user and his predicted behaviour in a relatively easy way.
With responsive search ads (RSA), you create ads for Google Search Engine Advertising by setting up multiple headlines and descriptions. Here too, Google serves the most ideal combination as an advertisement to the user. As of June 2022, this will be the only usable form of search ads. It will no longer be possible to create expanded text ads (ETA). In this way, Google forces advertisers to let go of control and put more power in the hands of the algorithm.
Experiment with broad targeting
The more room an algorithm has, the more it can learn and the greater the probability of optimal performance. Therefore, try experimenting with broad targeting. This means that you once again loosen the control by setting broad (or even no) audiences when you set up a campaign. By using a smart campaign objective such as maximising conversions, the algorithm will use its knowledge of the users and the type of content to target the right audiences. Modern media channels such as Meta, Google, Pinterest, TikTok... know from each user what websites they visit, what content they interact with and what products they buy. This makes an algorithm much smarter than any human media buyer, allowing the algorithm to select audiences where the media buyer's gut feeling or data falls short.
For search, this principle can be applied by experimenting with broad match types and introducing Dynamic Search Ads (DSA).
Aggregate campaigns and ad groups
The days when campaigns and ad groups had to be composed in a very detailed and granular way are fortunately behind us, thanks to machine learning. To give the algorithm as much room as possible to optimise, campaigns and/or ad groups should be aggregated where possible.
In the past, it was recommended to group campaigns according to content, but nowadays it is best practice to group by objective.
Provide sufficient feed to the algorithm
In order for an algorithm to learn optimally, it also needs sufficient signals. Therefore, always foresee sufficient budget for a campaign so that the campaign can collect the necessary volume of conversions to draw its conclusions and get out of the learning phase. Sometimes, it is necessary to set up a micro conversion as a campaign objective (e.g. Step 1 of the checkout process instead of a completed purchase). If you take a "too difficult" conversion as an objective, the algorithm does not have enough conversion data to draw its conclusions. Therefore, it can be rewarding in that case to lower the bar and optimise for a conversion that occurs more often. Finally, you can also increase the conversion volume by adjusting your attribution window. For example, a post-view window will count for more conversions than a post-click attribution window. Choose an attribution window that makes sense to you, but can still collect enough data to give the algorithm enough data.
By strengthening the algorithm, you will have to accept some uncertainty, a black box you cannot control 100%. Therefore, you must always consider what is most important: optimal performance or maximum control? When you rely on the power of the algorithms, you must keep at least two principles in mind: make sure the algorithm has sufficient (clean) data at its disposal and always maintain the correct (business) objective.