Your Meta Ads Shifted and You Changed Nothing. The Engine Underneath Was Rewritten
- Meta rewrote the engine that decides who sees your ad. Its retrieval system, Andromeda, now matches ads to people by the content of the creative, not by the audience you set by hand.
- Many advertisers saw results worsen without changing a thing. Accounts still run narrow interest targeting and a couple of ads, while the system moved on.
- The money lever shifted from targeting to creative production. A budget that feeds three near-identical ads a month now starves the engine.
- Entering Czechia, this flips your instinct. The urge to narrowly target “Czech consumers interested in X” now hurts; localised creative volume is the real lever — and the harder one in a new market.
What actually changed
Meta swapped the brain that decides who sees your ad. That choice used to come mostly from the targeting you set. Now it comes from how well the content of your creative fits a specific person. You changed nothing; the rule of the game changed.
The engine has a name: Andromeda. Meta introduced it on its engineering blog on 2 December 2024 and rolled it out across most accounts through 2025, globally by year end. That is why the impact lands only now — it was not a launch day with a press release, it was a quiet transition that finished during 2025.
Andromeda is Meta’s proprietary machine learning system for retrieval in ad recommendation … enabling a meaningful increase of model capacity (10,000x) for enhanced personalization … +6% recall improvement to the retrieval system, delivering +8% ads quality improvement on selected segments.
Retrieval is the first step of delivery: before the auction, the system narrows tens of millions of possible ads down to a few thousand that fit a given user. When a machine makes that call from the content of the creative, the centre of gravity in your work moves. It stops being the manual definition of an audience and becomes what you give the machine to choose from. And what you give it is creative.
Why Meta did it
Because the system now guesses the right viewer better than you can define one by hand, and because Meta benefits when an account can be run by someone who is not a targeting specialist. Fewer manual levers means more advertisers and more spend. The whole Advantage+ suite points the same way: Meta builds the audiences, splits the budget and places the ad across its surfaces, and Andromeda is the motor that makes that possible.
The catch is that most brands, and plenty of agencies, did not react to the shift. They keep tuning narrow interest audiences and running one or two “proven” ads. In a system built for the opposite approach, that is not neutral — it is a brake.
Why this flips your instinct when entering Czechia
A foreign brand launching in a new, smaller market reaches instinctively for control. You do not know the Czech audience, so the urge is to target narrowly: “Czech users, this city, this interest, this age.” Under Andromeda that instinct now works against you. The engine has less room to find the right person, precisely where you most need it to explore on your behalf.
- Narrow targeting is a false comfort. It feels safer in an unfamiliar market, but it starves the one system that could map that market for you.
- Creative volume is the real entry cost. You need several genuinely different, localised concepts, not one global hero ad with translated text pasted on.
- The small market punishes thin creative faster. Fewer people means the engine exhausts a weak ad quickly, so a shortage of fresh concepts shows up sooner than at home.
What it costs, in real terms
The money lever moved from media buying to creative production. Do the arithmetic plainly. If a budget feeds three similar ads a month, you give the engine almost nothing to choose from; it falls back on the cheapest impressions and performance drops. The same budget reallocated to produce ten to fifteen genuinely different concepts gives the machine material it can work with. This is not “spend more” — it is “spend differently.”
Across 3,014 e-commerce advertisers and $834M of 2025 spend, ROAS fell about 7% overall after Andromeda rolled out — and by over 30% for advertisers who had been top performers under the old, targeting-led playbook.
The second, hidden cost looks like activity. Twenty variants of one ad with a new headline or colour cost you production work, but Meta reads them as one and the same signal. You pay for output that creates no new learning. New signal comes only from a creative that differs by a real idea: a different angle, format, opening second of video, or setting.
How we would approach it
This is a change of centre of gravity, not rocket science. Here is how we would rebuild a Meta account from 2020 logic to 2026 logic for a brand entering Czechia.
- Loosen targeting. Replace narrow interest audiences with a broad brief and let the engine map the Czech market for you.
- Build a creative line. A regular supply of distinct, locally fluent concepts — not one translated hero ad. This is now the core of the media buy, not a side task.
- Do not produce in vain. Spend on concepts Meta will read as genuinely new, not cosmetic variants it treats as one.
- Keep measurement clean and read early signals. Server-side events feed the engine honest outcomes; watch how fast a new concept catches, not just lagging CPA.
FAQ
Why did my Meta ads get worse when I changed nothing?
The change was on Meta’s side. Its retrieval engine, Andromeda, now decides who sees an ad from the content of the creative rather than the audience you set. If you run narrow interest targeting and a few near-identical ads, the system has little to choose from and performance slips, even though you changed nothing. The lever moved from audience settings to creative volume and variety.
What is Meta Andromeda?
Andromeda is Meta’s ad retrieval engine, introduced in December 2024 and rolled out globally through 2025. It is the first step of delivery: before the auction it narrows tens of millions of possible ads to a few thousand that fit a given user. Meta reports a 10,000x increase in model capacity. In practice, machine matching of creative content to a person now decides delivery, not your manual audience.
For a new market, is narrow targeting or broad audience better?
After Andromeda, broad reach plus many different creatives wins, and this matters even more on entry. Narrow targeting feels safer in an unfamiliar market, but it starves the one system that could map that market for you. Give the engine room to explore and feed it enough distinct, localised concepts to choose from.
How many creatives should I run on Meta?
Difference matters more than count. Meta clusters near-identical variants as one ad, so twenty versions of a single visual with a new headline give it no new signal. Aim for several genuinely different concepts — different angle, format, opening second, setting or story — and a steady supply over time rather than one perfect ad a month.
Should I still pay for precise targeting as a core service?
Precise targeting lost most of its value after Andromeda and in places hurts performance. If a partner mainly charges for tuning audiences while delivering a couple of creatives a month, you are paying for the wrong thing. What makes the difference now is the ability to produce enough distinct, localised concepts and read which one works.
Sources
- Engineering at Meta — “Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine” (2 December 2024; 10,000x capacity, +6% recall, +8% ads quality). engineering.fb.com
- Confect.io — full-year 2025 analysis of 3,014 e-commerce advertisers and $834M spend (independent data on ROAS decline after Andromeda). confect.io
We build the creative line, not just better targeting
We loosen dated targeting on your Meta account and set up a steady supply of distinct, locally fluent concepts the engine can learn from. Budget moves to where performance is actually decided — creative and clean measurement. From Prague, in native Czech.
Talk to our Prague team