For years, advertising on Meta platforms felt like a direct negotiation. You told the system who you wanted to reach through detailed targeting, and Meta would hold an auction, ranking all the ads eligible for that user to decide a winner. Think of it as a single, massive audition where every potential ad gets a chance to perform for every user, every single time. This was computationally intensive and relied heavily on the advertiser's ability to define the perfect audience.

This legacy model put the burden of discovery on you. If your interest targeting was off, you missed your audience. If it was too narrow, you limited your scale. The system was powerful, but it was only as smart as the instructions you provided.

The old way worked, but it had a ceiling. Meta's new system, known as Andromeda, demolishes that ceiling by fundamentally changing how the process begins. It introduces a new first step: retrieval.

Instead of ranking every possible ad at once, Andromeda operates a two-stage system. The first stage is retrieval, and the second is the familiar ranking auction.

Think of the retrieval stage as a nightclub bouncer. Before, everyone got in line for the main event (the auction). Now, the bouncer quickly checks IDs and only lets a pre-approved list of guests through the door. This bouncer is a high-speed filter that instantly selects a small, relevant shortlist of ad candidates before the auction even begins.

This makes the system vastly more efficient. Instead of ranking millions of potential ads for a single impression, it only needs to rank a few hundred. This speed allows it to consider a much wider pool of ads at the retrieval stage than was ever possible with the old model.

How does the bouncer decide who gets on the shortlist? This is the most important shift for advertisers: the decision is now driven primarily by the ad creative itself, not your manual targeting settings.

Andromeda uses a technology called semantic embeddings to understand the meaning and context of your ad. It analyzes the text in your headline and body copy, the objects and concepts in your images or video, and even the sentiment of the language. It then converts this complex understanding into a numerical representation, or vector.

The system does the same for users, creating an embedding based on their recent behaviors and interests. The retrieval stage is essentially a massive, high-speed matching game. It looks for ads whose creative embeddings are the closest match to a user's embedding. Your detailed interest targets are still a signal, but they are now secondary to the signal provided by your creative.

This is why broad targeting is now often more effective. By giving the system a wide audience, you allow the retrieval algorithm to do its job: using the strong signal from your creative to find pockets of customers you might never have thought to target manually. Overly narrow targeting acts as a straitjacket, preventing the bouncer from considering a huge pool of potentially high-performing ads.

Your role has shifted from being an 'audience selector' to a 'signal provider.' Your primary job is to create ads with clear, potent signals that tell the algorithm exactly who the ad is for.

When all these signals align, Andromeda can build a powerful, accurate profile of your ideal customer and efficiently retrieve your ad for the auction when that person is online.

Ready to check your understanding?

This new approach requires a mental shift, moving trust from manual settings to the power of the algorithm, fed by high-quality creative.