Learning phase
Advertising mechanicsDefined term
Ad delivery systems predict outcomes per impression, and predictions need data. After a campaign launches or changes significantly, the system enters a documented learning phase: delivery is deliberately exploratory while the model accumulates enough conversion events to stabilize its estimates.
Performance during learning is noisier and typically worse, which is the engineering reason behind the practitioner rule to avoid frequent edits — each significant change resets exploration. This is the exploration–exploitation trade-off of a bandit algorithm, not platform superstition.
Browse the glossary
Advertising mechanics6 of 28 terms
- Ad rank⦿ liveThe auction score that orders ad positions — in the classic model, bid × quality — and sets what the slot above pays.Advertising mechanics
- Generalized second-price auction (GSP)⦿ liveThe multi-slot auction behind most search advertising: each winner pays just enough to keep its position, not its own bid.Advertising mechanics
- First-price auction⦿ liveAn auction where winners pay exactly what they bid — the rule most programmatic display moved to in 2019.Advertising mechanics
- Learning phaseThe exploration period after a campaign change when a delivery algorithm spends budget gathering data before optimizing on it.Advertising mechanics
- Budget pacingThe control loop that spreads a fixed ad budget across a period, throttling auction entries to avoid early exhaustion.Advertising mechanics
- Frequency cappingA delivery constraint limiting how often one person sees the same ad within a window — an identity problem as much as a rule.Advertising mechanics