Everything about Andromeda and the new reality of advertising on Facebook

In December 2024, Meta* announced its new advertising engine, Andromeda.

In the summer of 2025, a large-scale rollout of the algorithm began, and at the same time, most media buyers started running into problems: the effectiveness of familiar approaches dropped, CPM prices “spiked,” and ad campaign training dragged on.

At that time, Andromeda was the talk of the town, and most of the problems were exaggerated. Almost a year has passed since the engine’s full launch, and now we can soberly assess how Andromeda has changed the rules of the game within Meta* services.

Andromeda is an algorithm responsible for ad impression distribution. At the core of this system lie artificial intelligence and machine learning algorithms.

The way Andromeda works is illustrated in the image above. To save our readers from having to figure it out on their own, here’s an explanation:

• the algorithm takes a pool of creatives;

• it breaks down each creative into its constituent parts: text, video, CTA, colors, and more;

• the creatives are divided into groups by user segments: each large group is split into smaller ones;

• the AI sets aside irrelevant creatives and sends the “winners” to the auction.

The media buyer’s role during the initial campaign setup is limited to entering basic information: GEO, age, and language. If the targeting is precise but the creative doesn’t align with the audience’s interests, at best the buyer will get poor results, and at worst the creative won’t even make it into the auction.

The diminished role of targeting and the rise of creatives to the forefront is just the root cause that changed the rules of media buying within Facebook* and Instagram*.

Here’s what the launch of Andromeda has led to.

Rising CPM costs. Andromeda targets users who are most likely to take the desired action. This change has led to a rise in CPM costs for three reasons.

1. The number of impressions served to non-target audience segments has dropped significantly. Before Andromeda was introduced, algorithms tended to show ads to random segments of the target audience. However, in exchange for low costs, buyers were purchasing traffic of questionable quality.

2. Auction density has increased: advertisers now compete within narrow segments;

3. Weak creatives are weeded out at the start: as a result, only high-trust creatives enter the auction.

According to journalists at the Vibemyad portal, buyers who failed to adapt their ad campaigns to the new rules faced a 20–40% increase in CPM.

Previously, buyers targeted narrow segments to avoid wasting their budget on non-target users. With the introduction of Andromeda, the situation has been turned upside down. Running ads targeting narrow audience segments in the current environment is a surefire way to waste your budget.

The fact is that through targeting, the buyer sets the parameters within which Andromeda will search for potential leads.

By narrowing down to very specific segments, the user pool shrinks, and the algorithms can’t find a suitable audience. The result is nearly zero impressions.

Before Andromeda was launched, training times were shortened by targeting very specific segments. To put it simply, the ad engine “followed orders” from the buyer and strictly adhered to instructions.

Andromeda is an independent and unpredictable algorithm that evaluates various aspects of creatives and compares them to audience behavior in real time. Logically, this process takes longer than blindly following instructions.

The second reason for the long training period is a limited pool of creatives. If a campaign has only 3–5 creatives, the algorithms simply don’t have enough information. As you might guess, launch budgets will increase along with training times.

To train Andromeda, you need more creatives, but the catch is that quantity alone won’t cut it. The new engine needs diverse ideas: different structures, hooks, characters, and CTAs.

Creating a batch of identical creatives by slightly altering the color scheme, font size, or character design is a clearly losing strategy.

The fact is that, for Andromeda, these are identical creatives: the algorithm recognizes the context, unlike its predecessor.

Breaking campaigns down into small ad sets of 3–5 creatives will result in the engine once again lacking sufficient data for training. The issue is that due to this fragmentation, the algorithm will only see “snippets” of information rather than the full picture.

Andromeda is not the end of media buying, but the next stage in the development of Meta’s advertising ecosystem. If you don’t resist the changes and adjust your strategy, you can reach a new level of profitability.

For example, a marketer going by the nickname “Chris Marrano” stated on his LinkedIn blog that CPM prices dropped by 20% and conversion rates by 35% after adapting ad campaigns to the new rules.

We’ll explain which arbitrage principles you should revisit to keep turning a profit.

Andromeda is an AI algorithm that pays virtually no attention to targeting and shows creatives to whoever it deems necessary.

The engine considers those audience segments most likely to perform the target action to be the right ones. Andromeda may take the path of least resistance and repeatedly show creatives to “proven” leads. As a result, the buyer will receive traffic that won’t even meet basic KPIs.

To avoid this situation, you need to periodically intervene in the algorithm’s operation and filter out unnecessary segments in a timely manner.

The role of targeting has diminished, and most of the optimization work now falls on Andromeda.

If the tracker, pixel, and API conflict with each other and send inconsistent information to the algorithm, the AI gets confused, reduces the number of impressions, and generates duplicates.

To avoid confusing the algorithm, you need to standardize event names and parameter values.

Andromeda takes longer to train than its predecessor and is designed for diverse creative content rather than hundreds of variations of the same thing.

To fully unlock the engine’s potential, you need to “feed” the algorithm generously with creatives—and ones that are radically different from one another.

Whether to use neural networks or hire a team of UGC creators and designers is a matter of personal preference and the size of each buyer’s budget.

Focusing on narrow segments in the Andromeda era means cutting off a large portion of your target audience from the start, since the engine will still rely on its internal metrics.

It’s better to hand over campaign management to the “Advantage+” tool from the start so as not to confuse the algorithm.

If your landing pages aren’t intuitive, the forms are long, and optimization is lacking, the user will likely leave empty-handed. Andromeda will see zero conversion and blacklist the creative with all the ensuing consequences, even though the problem was an overloaded funnel.

A sudden change in budget or a quick relaunch of the campaign set is a short-sighted strategy.

Andromeda needs more time to train. If the budget were suddenly increased tenfold, the algorithm would have to start learning from scratch again. Every sudden change means going back to square one.

Andromeda hasn’t introduced any major changes to the account farming and warming process. It has built on the old principles and tightened them slightly.

In 2026, an account is no longer a disposable asset, but rather a link in the arbitrage chain just like creative content.

Here are the main changes:

Greater importance is placed on the account’s overall trustworthiness. Andromeda pays more attention to the profile’s history and the presence or absence of past violations. The era of hastily farming accounts is over. It’s better to create 3–5 high-quality accounts than a hundred “dummy” ones.

You need to warm up an account longer and more thoroughly. This point follows from the previous one: the algorithm will certainly not appreciate a profile that was unnaturally warmed up in a day and a half.

Don’t go all out aggressively from the start. Spending $100,000 on the very first ad campaign is a red flag for Andromeda.

The challenges with account farming and warming up aren’t limited to Andromeda. In 2026, the local anti-fraud system threw in a surprise of its own. Now, special importance is placed on the stability and plausibility of the fingerprint.

We’re not implying anything, but Vision boasts 99.98% uptime, real browser fingerprints, and the best Hymantype implementation on the market. No anti-fraud system will be able to detect your true intentions.

Vision works with a huge number of teams that use Meta services. One of them is our partners at LeadGenerals, who have been successfully working with Facebook for many years. We asked the team’s founder, Andrew, what he thinks about Andromeda.

I’ll answer without any fairy tales or whining, as someone who has survived more than one “update that will kill everything". What actually happened with the arrival of Andromeda?

Meta has become much smarter and much less manually controllable. Andromeda isn’t just an update. It’s the moment when:

• the algorithm started deciding on its own who to show ads to

• interests and detailed targeting started losing their meaning

• creative and signal became more important than settings

The "wide launch" strategy is actually working now

I've always advocated using Advantage+, and if you watch my YouTube videos from 2–3 years ago, you'll see that I specifically predicted we'd eventually end up here.

This has definitely made scaling much easier.

Your find Audience Faster

If the ad campaign is active:

• The algorithm finds the right people on its own

• No need to guess interests

• The time from idea to profit has been reduced

Scaling Has Become Easier

Loss of control

This is definitely the biggest downside, because you can no longer:

• understand exactly why traffic is flowing

• manage campaigns through audiences

• manually "fine-tune" ad sets

It's harder to optimize

Everything has come down to the creative, the signal, and the offer.

The algorithm has become more demanding in terms of quality

If:

• weak creative

• poor CTR

• no signal

then Andromeda simply won't "try."

The most important change: the focus has shifted away from link settings

• Less time on setup. Fewer ad sets / fewer audiences / less "tinkering." Setup now accounts for 20% of the work.

• More time on creative. The main focus is now on hooks, angles, and hypothesis testing. Creative has become the main lever — we’ve even set up our own UGC studio.

• Systematic testing. Now, the winner isn’t the one who sets things up better, but the one who tests faster, cuts faster, and scales faster.

• Signal processing. An algorithm is useless without data. That’s why tracking, events, and volume are so important.

Andromeda didn't "kill" affiliate marketing. It simply changed the rules:

• less manual control

• fewer "settings hacks"

• a greater role for creatives

• a greater role for speed

And to be completely honest—back then, the techies won, but now it’s the product-focused affiliate marketing specialists who are winning.

And the main takeaway:

If you have a strong campaign, Andromeda is your best friend. If it’s weak, it simply won’t show you to the market. This is a proven fact — just watch my video from when the algorithm was just being rolled out, or check out my conference presentations.

The launch of the Andromeda engine does not spell the end of media buying within Meta* services; it is merely a new phase. The rules have changed drastically, but that is no reason to write off Facebook or rush to switch your traffic source.

The key is not to ignore these changes and to adapt in a timely manner, as well as to use high-quality software, such as Vision, to minimize the risk of bans.

*The trademarks listed are the property of their respective owners and are mentioned solely for informational purposes.