With our analysis of account data through Google Analytics, we identified and segmented the account. This wasn’t based on traditional segments such as branded or generic campaigns, but on performance.
We bucketed campaigns based on historical performance and set the strategy constraints based on these.
Historically viewing each campaign’s ROI, revenue, as well as upper and lower CPCs, we were able to group them into 8 segments. Each segment was given a unique target based on this historical information and then adjusted while the algorithm learned and adapted.
This highly structured approach gave us the ability to guide the machine learning as it optimised the campaign performance.