The Meta ads delivery system optimizes for your results using machine learning. Every time one of your ads is shown, the delivery system learns more about who to target, when to show the ad, and which placements and creatives to use. The more ads that are shown, the better the delivery system becomes at optimizing ad performance.
The learning phase is the time when the delivery system is still learning about an ad set. During the learning phase, the delivery system is experimenting with different audiences, placements, and other factors to determine the best way to deliver your ad set, so performance has not yet stabilized. When you create a new ad set or make significant changes to an existing ad or ad set, you enter the learning phase.