Buying display advertising typically involves satisfying a set of goals for the campaign's key performance indicators (KPIs) — such as delivery, viewability, and performance. When we try to improve one KPI, we often find that other KPIs are affected as well; for example, an improvement in viewability may require an increase in CPA, as we raise bids to ensure that we're buying the right kind of inventory. Furthermore, as market conditions change over time, our bidding decisions need to change accordingly.
We recently published a paper on our method of addressing this without human intervention: a feedback control system that automatically adjusts bidding parameters over time, moving a campaign toward its goals and then making needed adjustments as market conditions change. This approach is capable of working toward multiple goals simultaneously, even when an improvement on one KPI affects the others. Because some goals are usually more important than others, the system needs to automatically choose which to focus on at any given time. Therefore, we also introduce Sequential Control, which considers the importance of each KPI to the campaign — along with how far each one is from its goal — in order to select which to optimize at any point in time.
This approach has been highly effective in simultaneously optimizing multiple KPIs, without the need for human involvement.
