First, a quick refresher: what is Andromeda?

When someone opens Facebook or Instagram, Meta doesn’t compare every ad in existence in fine detail. It works in two stages:

  1. Retrieval (Andromeda): narrows a huge pool of eligible ads down to a short list of strong candidates in milliseconds.

  2. Ranking: takes that short list and studies it more closely to choose which ad to show.

Think of it like hiring. Retrieval is the recruiter who screens thousands of résumés down to a shortlist. Ranking is the interview panel that picks the final hire. If the recruiter screens poorly, great candidates never make it to the interview, and the same is true of your ads.

What actually changed

Per Meta’s engineering team, the latest version of Andromeda makes three major changes. 

1. A “team of specialists” instead of one generalist. Previously, retrieval relied on a single model of each user. Now it uses an ensemble of specialized “experts,” each capturing a different dimension of a person’s interests, such as recent ad engagement, interest cohorts, and behavior across Meta’s surfaces. For each ad, the system routes to the experts most relevant to it. Meta says this brings retrieval close to the quality of the ranking stage while using roughly 300 times less computing work than the 2024 version.

2. Ads are judged in context. The ad is evaluated together with real-time context, so the match is closer to “this person, this ad, right now” than a generic fit.

3. Everything runs on one piece of hardware. Meta redesigned the entire retrieval pipeline, including the step that enforces advertisers’ targeting rules, to run on a single GPU server rather than passing data across a network. The result is more than one billion ads scored per second, about 100 times the 2024 system.

Why Meta is doing this

The volume of ads is exploding. Generative AI creative tools and Advantage+ automation have helped advertisers produce hundreds of millions of new ads, according to Meta. A system that cut corners on scoring that many ads would risk missing good matches. This upgrade is designed to evaluate the whole pool without slowing down the experience for users.

Meta also reports that Andromeda has delivered a 2.4% increase in conversions on Facebook since 2025.

What Meta has claimed, and what we don’t know yet

The update is only days old, so it’s important to separate confirmed information from speculation.

What Meta has stated:

  • The new retrieval architecture approaches ranking-level quality at a fraction of the computing cost.

  • The system now scores every eligible ad at a very high speed.

  • Advertisers should see better customer matching and more conversions automatically, without changing anything.

  • Andromeda has contributed a 2.4% conversion increase on Facebook since 2025.

What we don’t know yet:

  • How much of any future performance change is attributable to this specific update.

  • How results will vary by industry, budget size, or objective.

  • Whether the 2.4% average applies to Instagram or to individual accounts like yours.

  • Whether Meta will release new advertiser controls or reporting tied to this update.

Meta’s post is also not an advertiser guide. It does not give creative or campaign-structure recommendations, so the guidance below is our interpretation, informed by how the broader industry has been reading Andromeda.

What this could mean for your ads

These are informed inferences rather than guarantees. We’ll update this post as data comes in.

1. Creative is likely your biggest lever

If retrieval gets better at reading ads and matching them to people, the substance of your creative matters more. Industry analysts note that Andromeda evaluates the content of the creative itself, not just the targeting settings around it.

2. Diversify concepts, not just colors

The industry’s consistent advice is to build ads for different people and situations: different customer motivations, angles, formats, and emotional tones. Swapping a headline or a background color usually isn’t enough. Reporting on Meta’s own guidance indicates that near-identical creatives may be treated as the same ad. Meta has also described creative diversification as a longstanding best practice and has not given a recommended number of ads.

3. You may not need hundreds of ads

More isn’t automatically better. For many smaller advertisers, a handful of genuinely different creatives is a reasonable starting point. For larger accounts, volume matters more, but only if the variations are meaningfully distinct.

4. Keep campaign structure simple

The consensus among leading performance marketing strategists (ourselves included) recommend a consolidated account structure so the system has room to allocate budget to what’s working, rather than splitting spend across many narrow campaigns.

5. Expect creative fatigue to be a bigger factor

As more advertisers produce more AI-assisted creative, standing out gets harder. Plan on a regular refresh cadence and watch for fatigue signals such as rising cost per result and falling click-through rates.

A word of caution

Some experts warn that advertisers often conflate Andromeda, which is a retrieval system, with campaign strategy as a whole. Andromeda is one stage of a larger pipeline. It does not replace sound offers, clear messaging, good landing pages, or accurate measurement.

What we’d do right now: a practical checklist

  1. Set a baseline. Record your current cost per result, conversion rate, and creative performance before making changes, so you can tell what actually moves.

  2. Audit for near-duplicates. Identify ads that say the same thing in slightly different packaging.

  3. Map creative to customer motivations. List the distinct reasons different customers buy from you, and make sure each has its own concept.

  4. Mix formats. Test video, static, carousel, and user-generated-style content, designed for mobile-first viewing.

  5. Consolidate where it makes sense. Review whether your campaign structure is fragmenting your budget.

  6. Schedule creative reviews. Check for fatigue regularly rather than waiting for performance to drop.

  7. Avoid overreacting. Make changes in a controlled way and give them time to produce a meaningful signal.

Where this is likely heading

Meta says the new architecture has substantial headroom: it is already working on next-generation NVIDIA hardware and expects to add model complexity, richer signals, and additional retrieval pathways without proportional cost growth. Our read, which is a prediction rather than a fact:

  • More automation. Advantage+ and AI-driven delivery will likely keep taking on work that was once manual.

  • Creative as the competitive edge. When the platform handles more of the matching, the differentiator shifts to what you put in front of people.

  • Higher bar for originality. With generative tools lowering the cost of making ads, distinctive, well-researched creative will stand out.

Myths to avoid

  • “More ads always means better results.” Only meaningfully different ads add new information for the system.

  • “Andromeda replaces strategy.” It improves matching. It doesn’t fix a weak offer.

  • “We should rebuild everything now.” The update is days old. Measure first, then adapt.

The bottom line

Meta’s latest Andromeda upgrade is a significant technical step, and it points in a direction that was already clear: smarter automation on Meta’s side, and creative strategy as the place where advertisers can make the biggest difference. We’re monitoring performance across our client accounts and will update this post as we learn more.

Want to know how this affects your account? Graphic Forms Inc. offers a creative and account-structure audit to help ensure your Meta ad account is set up to succeed in the age of Andromeda 2.0. Fill out the form on our website if you are interested in learning more.

Sources

This article reflects our analysis as of November 2026 and will be updated as more information becomes available.