The learning phase is a period at the start of a new Meta ad set during which the algorithm explores the audience and optimizes delivery based on performance signals. Performance during this period is typically unstable and cost per result is higher than post-learning. Meta requires approximately 50 optimization events per ad set within a 7-day period to exit the learning phase. Making significant changes to a running ad set such as adjusting budget by more than 20%, changing the audience, or swapping creatives, resets the learning phase entirely.