A few years ago, we started saying "creative is the new targeting." The idea was simple: Meta's algorithm had matured to the point where your ad creative — not your audience definition — was doing most of the targeting work. You feed it a great ad, it finds the people who respond to it. You feed it a mediocre ad with a precisely-defined audience, and you get mediocre results at scale.
That idea was right. But in 2026, it's incomplete.
Meta's Andromeda algorithm — officially announced in late 2024 and fully rolled out globally through 2025 — represents the most significant overhaul to Meta's advertising infrastructure in over a decade. It didn't just confirm that creative is the targeting. It raised the stakes: now, creative velocity is the competitive variable. The accounts winning on Meta in 2026 aren't the ones with the best individual ad. They're the ones producing, testing, and refreshing creative fast enough that the algorithm always has something new and compelling to work with.
This post explains what Andromeda actually is, how it changed ad delivery, why creative velocity matters so much now, and what a velocity-first creative system looks like in practice.
What is the Meta Andromeda algorithm?
Andromeda is Meta's AI-powered ad retrieval engine — the system that decides which ads even get considered in any given auction. To understand why it matters, you have to understand what the "retrieval" stage is.
Every time a user's feed loads, Meta runs an auction. That auction decides which ads appear and in what order. But before the auction even happens, there's a retrieval phase: Meta needs to narrow down the universe of possible ads to a shortlist that the auction can then evaluate. This retrieval step is where Andromeda operates — and it's where the fundamental change happened.
Under the old system, retrieval relied heavily on the targeting parameters the advertiser set. If you targeted women aged 25–40 interested in fitness, the system filtered for ads matching that definition. It evaluated a few thousand ads per auction. Under Andromeda, the algorithm uses deep machine learning to analyse the content of the ad itself — the visual elements, faces, emotional tone, text, and format — and cross-references that with billions of real-time user behaviour signals. It can now scan tens of millions of ads per auction and predict which creative is most likely to resonate with which specific user, in that specific moment.
How Andromeda Delivers Ads
Your Creative
Ad enters the pool with millions of others
Andromeda Scans
Analyses visual, emotional and format signals in the ad itself
Matched to Users
Delivered to users most likely to respond — based on the creative itself
Old system: thousands of ads evaluated per auction | Andromeda: tens of millions
The hardware behind this is substantial — Meta invested in custom AI accelerators and NVIDIA Superchips specifically to power Andromeda's retrieval capacity. This isn't a software tweak. It's a fundamental infrastructure upgrade that made a new kind of ad delivery possible.
The practical effect: the algorithm no longer primarily relies on who you told it to target. It uses your creative to find its own audience. And if your creative is weak, repetitive, or fatigued — there's nothing in the audience definition that can compensate.
Creative is the targeting — here's what that actually means
When we first started saying "creative is the new targeting," some people heard it as a creative department talking point. It isn't. It's a description of how the algorithm works mechanically.
When you run an ad featuring a busy mum using a product, Andromeda reads those visual and contextual signals and delivers the ad to users whose past behaviour suggests they'll respond to that kind of content. When you run a spec-focused ad for the same product — performance numbers, technical comparisons, data — Andromeda finds a different audience automatically. You didn't change your targeting. You changed your creative. The creative is the targeting instruction.
This flips the traditional agency logic on its head. Less time engineering audience segments. More time engineering creative variety. And not superficial variety — Andromeda is sophisticated enough to recognise lazy variants. Changing a button colour or swapping a headline doesn't register as a new creative signal. The algorithm needs genuinely different creative concepts: different emotional angles, different formats, different hooks, different visual styles.
"Creative is the new targeting. The audience is whoever responds to your ad — your job is to make something worth responding to."
How ad delivery changed: pre-Andromeda vs now
| Factor | Pre-Andromeda (2022–2024) | Andromeda Era (2025–2026) |
|---|---|---|
| Ads evaluated per auction | Thousands | Tens of millions |
| Primary targeting mechanism | Advertiser-defined audience segments | Creative signals + real-time behavioural data |
| Audience definition | Set by the advertiser (interests, demographics) | Discovered by the algorithm from creative response |
| Creatives needed per campaign | 3–5 variations | 8–15 distinct concepts |
| Campaign structure | Segmented, audience-specific ad sets | Broad, consolidated — Advantage+ preferred |
| Learning phase | 1–2 weeks, relatively stable | 3–7 days, more volatile before it clicks |
| Competitive edge | Audience architecture and targeting precision | Creative production velocity |
Why creative quality alone isn't enough: the velocity problem
Here's where the 2026 evolution kicks in. "Creative is the new targeting" tells you what to optimise. It doesn't tell you for how long.
Creative fatigue on Meta has always been a problem. But Andromeda has accelerated it significantly. Because the algorithm is now much better at finding the exact right users for a given creative, it exhausts those users faster. It reaches the most responsive people first, quickly. Once it's saturated that high-response cohort, performance drops — frequency rises, CPMs climb, and the click-through rate falls. What used to take six weeks now often happens in two.
The solution isn't better creative. You might already have great creative. The solution is a system that produces, tests, and refreshes creative fast enough to stay ahead of fatigue. That's creative velocity: not just what you make, but how fast you make more of it.
8–15
Distinct creative concepts now needed per campaign (up from 3–5)
2–3×
Faster creative fatigue under Andromeda vs the pre-2025 algorithm
3–7
Days for Andromeda's learning phase before performance stabilises
Andromeda also changed what counts as a "new" creative. The algorithm is sophisticated enough to recognise cosmetic changes as duplicates. Swapping a background colour, changing capitalisation, or reordering bullet points doesn't give the algorithm new signal. It needs genuinely different creative concepts — new emotional angles, new formats, new hooks — to find new pockets of high-response users.
Most accounts can't produce creative fast enough to keep up with this. Not because they don't understand the importance of creative, but because they don't have the production system to match the algorithm's appetite.
The Creative Velocity Cycle
Creative velocity isn't about producing more ads for the sake of it. It's a structured system: brief → produce → test → kill or scale → refresh → repeat. The goal is a continuous loop that keeps the algorithm fed without burning your production budget on ideas that aren't working.
The Creative Velocity Cycle
Brief
Identify new hooks, emotional angles, and formats to test. Based on what the last round learned — not guesswork.
Produce
UGC, founder-led video, statics, carousels. Genuinely different concepts — not colour swaps. Lean production, fast turnaround.
Launch
Enter the testing rotation alongside proven performers. Broad targeting. Let Andromeda find the right audience from the creative itself.
Measure
Hook rate, 3-second view rate, thumb-stop ratio, conversion rate. Read the creative signal quickly — not after six weeks of wasted spend.
Kill or Scale
Underperformers off fast. Winners get budget. No sentiment, no attachment — just signal.
Refresh
Proactively replace fatiguing winners before performance drops. Start the next brief before you need to — not after the decay starts.
This cycle runs continuously — not as a quarterly project
The key discipline in this cycle is the brief. A great brief doesn't just describe what to make — it encodes what the previous test taught you. If UGC with a pain-point hook outperformed a polished brand video, the next brief explores variations of that pattern while also testing an entirely new angle. The system gets smarter with each iteration because the brief carries the learning forward.
The other discipline is killing underperformers quickly. Letting a weak creative run "a little longer to gather more data" is how accounts waste budget and teach the algorithm bad habits. If the hook rate is poor after a few days of meaningful impressions, you have signal. Act on it.
What velocity-first accounts look like in practice
Accounts that are winning under Andromeda in 2026 share a set of common characteristics. They're not necessarily the ones with the biggest budgets or the most sophisticated attribution models. They tend to look like this:
Simplified campaign structure
One or two campaigns. Broad targeting or Advantage+. Budget consolidated, not fragmented across dozens of small ad sets. Less audience engineering, more creative engineering.
A testing rotation — not a hero ad
8–15 active creative concepts running at any time. No single ad carries the account. When one fatigues, there's already a replacement in the wings — not scrambled together at the last minute.
UGC and lo-fi creative in the mix
Polished brand video has a place, but UGC and founder-led content are often more effective because they provide distinctly different creative signals. Andromeda reads authenticity as a separate creative pattern — and finds a different audience for it.
First-party signal through CAPI
Andromeda's audience-matching is only as good as the conversion signal it receives. Accounts with CAPI implemented correctly give the algorithm better data to work with — and in a velocity-first system, the algorithm needs quality feedback to find the next right audience for each new creative.
Patience in the learning phase, aggression everywhere else
Andromeda's learning phase can be volatile for 3–7 days before performance stabilises. The worst thing you can do is kill a new creative during this window because the early numbers look uncertain. But once you have signal — act on it fast.
What to do if your account is still running on the old playbook
Most accounts we audit are still structured for 2022. Complex audience segmentation, 3–5 creatives per campaign, a single "winning" ad that's been running for months. They're not broken — they're just optimised for an algorithm that no longer exists.
If that sounds like your account, here's where to start:
1. Audit your creative library
How many genuinely different creative concepts are active right now? If it's fewer than five, the algorithm is working with a thin signal pool. That's the first constraint to fix — before anything else.
2. Consolidate your campaign structure
Merge fragmented ad sets. Combine budgets. Give Andromeda the room and the spend volume it needs to find audience patterns from your creative. Narrow audience constraints are now more likely to hurt you than help you.
3. Check your CAPI implementation
If you're not sending server-side conversion events, you're feeding the algorithm incomplete signal. It can't match creative to converting users accurately if it's only seeing a fraction of your conversions. Fix this before scaling anything.
4. Build a brief cadence
Decide how often you'll brief new creative. For most accounts, a fortnightly brief cycle works — enough time to produce, test, and read signal before the next round. Make it a recurring operation, not a crisis response.
5. Measure the right things
Hook rate (how many people watch past the first 3 seconds), thumb-stop ratio, and 25% video completion rate tell you whether a creative has a chance before it accumulates enough spend to read conversion data. Use these as early signals to kill underperformers fast and protect budget.
The bottom line
"Creative is the new targeting" is still true. Andromeda confirmed it at an infrastructure level. But in 2026, that insight alone isn't a competitive advantage — it's table stakes. Every serious Meta advertiser now knows creative matters.
The advantage now belongs to accounts that have built a production system to match the algorithm's appetite — one that briefs, tests, kills, and refreshes fast enough that the algorithm always has fresh signal to learn from. Creative velocity isn't a creative department problem. It's a business operations problem. And solving it is how you separate your Meta performance from everyone else who's stuck running the same three ads and wondering why CPMs keep climbing.
The accounts that build this system in 2026 will have a structural advantage that's genuinely hard to replicate quickly. The ones that don't will keep burning budget on an algorithm that's already moved on.
Andrea Atzori
Co-Founder, Ambire. 20+ years running paid media campaigns across Google and Meta. HeroConf speaker. Google Partner.
About Ambire →