Meta Andromeda

AI plays an important role in Meta’s advertising system by leveraging the power of machine learning (ML) to predict which ads a person will find most interesting. This helps people learn about a business or product they are interested in while helping an advertiser meet their objectives such as increasing brand awareness, acquiring new customers, and driving sales.

Retrieval is the first step in our multi-stage ads recommendation system. This stage is tasked with selecting ads from tens of millions of ad candidates into a few thousand relevant ad candidates. In the following stage, larger and more sophisticated ranking models predict people and advertiser value to determine the final set of ads to be shown to the person.

Volume of ad candidates: Retrieval processes three orders of magnitude more ads than subsequent stages. Features like predictive targeting, which dramatically improve advertiser outcomes, are computationally expensive. The continued positive momentum of Meta’s Advantage+ suite further increases the number of eligible ads through automation of audience creation, optimal budget allocation, dynamic placement across Meta surfaces, and creative generation. Finally, with the adoption of powerful new tools based on generative AI for creating and optimizing ad creative content, the number of ads creatives in Meta’s recommendation systems is expected to grow significantly.

Tight latency constraints: Selecting ads rapidly is essential for delivering timely and relevant ads, as any delay can disrupt the viewers experience by not providing the most current content. As advertising becomes increasingly dynamic, frequent updates to both delivery and each person’s interests demand increased model complexity in near real-time.