You've launched an ad on Facebook or Instagram and you're wondering why it shows to some people and not others. The answer lies in a precise mechanism Meta runs in real time every time someone opens the app. In 2026, with the arrival of Andromeda, this system changed fundamentally. Understanding how it works means understanding why some advertisers spend half as much for twice the results.
An auction system, not just an ad network
As we explained in our first article on Meta Ads, Meta's advertising platform is much more than a tool to "boost posts". Every time a user opens Instagram or Facebook, thousands of advertisers compete in a fraction of a second to show them their ad. This mechanism is called the Meta auction, and contrary to what you might think, the biggest budget does not always win.
Meta gives every ad an overall score combining three factors: your bid, the estimated action rate (will this specific user click, buy, sign up?) and the quality of your ad. It is the combination of these three elements that determines whether your ad is shown, to whom and at what cost.
In 2026, Andromeda's computing power has been multiplied by 10,000, allowing Meta to analyse billions of signals in real time before an ad even enters the auction. This massive upstream filtering changes everything: if your creative doesn't pass this first filter, no budget can make up for it.
The Meta Ads algorithm in 2026: Andromeda changed everything
With the Andromeda update, you no longer force a message onto an ultra-targeted audience. You deliver relevant content and the algorithm surfaces it to the right profiles at the right time. It is a complete paradigm shift for advertisers.
Andromeda observes everything: behaviours, weak signals, implicit intent, the way users consume content. It no longer relies only on declared segments or fixed interests. It learns continuously and enriches its data all the time.
What this means in practice for your campaigns:
❌ BEFORE ANDROMEDA
Manual targeting by precise interests
15 ad sets with different audiences
Complex micro-segmentation
Structure decides results
Targeting does the work
✅ AFTER ANDROMEDA (2026)
Broad targeting, the AI chooses the audience
2 campaigns max (test + scale)
A simple, readable structure
Creative decides results
The message does the targeting
Creative has become the new targeting
Targeting is no longer the core of performance. As Henry Kelly, Head of e-Commerce at Meta, put it: "Creative is the new targeting". Andromeda doesn't just deliver an ad to a defined audience. It constantly tests a message's ability to trigger a reaction: scroll, stop, click, engagement, conversion.
In practice, your images, videos and copy tell the algorithm who to show your ad to. A creative that speaks to young parents will automatically be shown to young parents, without you having to specify it in your targeting. The algorithm reads the message and finds the audience.
Direct implication in 2026: Meta now recommends launching at least 20 new creatives per week. And by "creatives" it doesn't mean changing a button colour, but unique concepts that are clearly different from one another. Each concept is seen as a new signal for the algorithm.
The numbers that show Andromeda's impact
×10,000
Andromeda computing power vs the previous Meta algorithm
−30%
Average cost per acquisition with well-configured Advantage+ campaigns
50
Minimum conversions per week to exit the learning phase
Advantage+: when Meta takes control of targeting
Broad targeting means letting the algorithm work with as few constraints as possible. Everything else (interests, behaviours, refinement) is left to Meta's AI, which relies on your conversion signals (Pixel + CAPI) to find the right people.
Advantage+ Audience goes even further: the algorithm can automatically expand beyond the audiences you entered if that improves performance. In practice, Meta may decide to show your ad to profiles you would never have targeted manually, and get better results.
The over-restrictive targeting trap: Rigid strategies with audiences that are too narrow and excessive exclusions limit delivery and push up costs. In 2026, constraining the algorithm with overly precise targeting means deliberately capping your performance. Let the AI work, but give it good signals.
What the algorithm needs from you to work
Andromeda is powerful, but it isn't autonomous. It needs three things from you to optimise properly:
Reliable conversion data.
Meta's algorithm needs at least 50 conversions per week to exit the learning phase. Below that, your campaign stays in permanent learning and performance is unstable. That is why a properly configured Pixel + CAPI is essential.
Varied, high-quality creatives.
The more differentiated your concepts, the better Andromeda understands the different facets of your audience. The algorithm needs diversity to identify which messages resonate with which profiles.
A consistent campaign objective.
Mixing several objectives in one campaign blurs the algorithm's signals and hurts performance. One objective = one campaign = clear signals for the AI.
Patience during the learning phase.
The algorithm needs a learning phase to identify the most relevant profiles. During this period costs may rise, because the system is testing different combinations. Changing your campaigns too early resets that learning.
The 3 mistakesthat stop the algorithm from performing:
Changing your campaigns every 48 hours.
Every change restarts the learning phase. The algorithm needs time to optimise; constant interference prevents it from reaching its real performance.
Managing only by ROAS or CPC.
Focusing only on metrics such as CPC or CPM often gives a skewed view of real performance: a Meta Ads campaign should be judged on the quality of its conversions, not on isolated numbers.
Not excluding audiences who have already converted.
Showing your acquisition ads to existing customers wastes budget and blurs the signals sent to the algorithm. Always exclude your recent buyers.
TAKE THE NEXT STEP:
Is your Meta Ads account giving the algorithm the right signals?
A poorly fed Andromeda algorithm optimises in a vacuum, and your budget disappears with no measurable return. Most of the accounts we audit suffer from the same problems: an overly complex structure, insufficient tracking data and creatives that lack variety.
At VD Adsconsulting, we analyse your account structure, the quality of your conversion signals and the diversity of your creatives in detail. You leave with a clear action plan so the algorithm finally works for you, with no commitment.
Request my free audit