Build a lead generation account for Andromeda the way you would build it for a small budget: one campaign per offer, as few ad sets as your weekly budget can feed at about 50 leads a week each, and three to six distinct creative concepts per ad set. Andromeda is a retrieval system inside Meta’s delivery, not a setting, and it will not reward a particular account layout. The rules you have read about come from Meta’s creative diversification guidance and from practitioners filling the gaps, so treat them as decisions rather than switches.
What Andromeda is, and what it is not
Meta described the system in an engineering post on 2 December 2024. It called Andromeda its machine learning design for retrieval, the first step of a multi-stage delivery system, where tens of millions of candidate ads are cut down to a few thousand that larger ranking models then sort for one person.
The figures in that post are Meta’s own: a 6% gain in retrieval recall once it ran on Facebook and Instagram, an 8% gain in ads quality on selected segments, and a 10,000x increase in model capacity for more personalised retrieval. The reason given is volume, since Meta expects the number of creatives in its systems to grow as advertisers use generative AI tools, and a hierarchical index is how it plans to cope.
Nothing in that post tells an advertiser what to do. There is no advice, no count of ads per ad set, no campaign structure and no rollout date. A Meta for Business post from March 2025 places Andromeda among four AI systems in the ads stack, alongside GEM, Lattice and sequence learning, and says it helps Meta pick the right creative for each person as more creatives arrive. That is a statement about Meta’s internal work, not about your account.
What Meta did say to advertisers
The advertiser-facing guidance is almost all about creative diversification, and it arrived in pieces. Taken in order:
- November 2024: a post asks advertisers to launch new ads weekly and recommends at least twenty diversified ads with 20 to 30% of budget for testing. The context is Advantage+ shopping campaigns, e-commerce, holiday season.
- April 2025: a second post ties diversification to Andromeda by name and says the focus has shifted from niche targeting to creative diversification as the best lever for finding relevant audiences.
- Undated, still live: creative differentiation guidance warns that near-identical assets may not be recognised as different, and can end in a Creative limited or Creative fatigue status.
- Help Center: the article on ad volume pulls the other way, warning that too many ads at once means each delivers less often, fewer ads leave learning, and performance can get worse. Its advice is fewer ads per ad set with varied assets inside each one.
Those last two halves collide in a lead generation account. Variety needs ads, and each ad needs leads before anyone, Meta included, can learn from it.
What practitioners added, and what holds up
Agency posts turned Andromeda into rules. Some rest on Meta’s own words, others do not:
| Claim | Status | What to do |
|---|---|---|
| One campaign, one broad ad set | reasonable, though its support is Meta’s learning guidance on combining similar ad sets, which predates Andromeda | combine what you can feed, keep ad sets apart only for a real difference |
| Creative is the new targeting | Meta said a version of this in April 2025 | keep the constraints you need, such as area served, and let creative variety do the finding |
| You need twenty, fifty or a hundred ads | no Meta source says this for lead campaigns, and the Help Center warns against high volumes | size the count to your budget, not to a rule of thumb |
| Similar ads are grouped under one entity ID | Meta confirms the general effect but not the mechanics, and its spokesperson declined to describe them | vary the concept, the hook, the format and the person on camera |
| Winners fatigue faster since Andromeda | anecdotal, and consistent with the warning about similar creative, but nobody outside Meta can isolate the cause | watch frequency and CPL per ad, and change the concept when they move |
The structure a lead budget can feed
Meta’s October 2026 peak season guidance gives the formula: total weekly budget divided by fifty times your target cost per result gives the most ad sets your budget can support. In a leads campaign the result is a lead.
| Weekly budget | Target CPL €20 | Target CPL €40 | Target CPL €80 |
|---|---|---|---|
| €1,000 | 1 | 1 | 1 |
| €3,000 | 3 | 1 | 1 |
| €6,000 | 6 | 3 | 1 |
| €12,000 | 12 | 6 | 3 |
Round down, and never below one. A layout that stays inside that ceiling looks like this:
- One campaign on the Leads objective, at the conversion location you actually use.
- One or two ad sets on a broad audience, limited to the area you serve, or on an Advantage+ audience when nothing forces you to narrow it.
- Four to six ads per ad set, each a different concept, each pointing to a landing page that repeats its promise.
- New concepts added in one batch, because adding an ad is a significant edit and restarts learning.
- More ad sets only where something real differs: another country with its own landing page, another conversion location, retargeting, or an offer whose cost per lead sits in a different range.
How an account is built level by level covers naming and the campaign, ad set and ad decisions in more detail.
Why fuller ad sets beat one ad set per audience
The limit is arithmetic. An ad set needs about fifty results in the week after its last significant edit to leave learning, so at a €40 cost per lead one ad set needs €2,000 a week to learn. Five ad sets on a €3,000 weekly budget ask five groups of delivery to learn from a fraction of what one needs, and the extra targeting usually shows up as unstable results rather than better ones.
That is also why changing a live account is expensive: a significant edit restarts learning, so you pay for the same learning twice. Add creatives, audiences and budget in one pass, then hold the structure still for a week.
With one or two ad sets the budget type stops being a strategic question. Campaign budget and ad set budgets behave the same with a single ad set, and CBO vs ABO only matters once you run several and need to protect one from the others.
How many creatives one ad set can rank
Meta’s developer documentation recommends several ads per ad set so delivery can choose between images, video, links and text, and stops there. The real limit is your ability to read the result: a creative needs roughly ten times your target cost per lead in spend before its own CPL means anything.
Weekly ad set budget divided by ten times the target cost per lead gives the most creatives you can rank in a week. An ad set on €2,000 a week at a €40 target can carry five. Give it fifteen and each one averages about €133 a week, roughly three leads, which cannot separate a winner from a lucky draw.
Two habits keep the count honest. Make each ad a different concept rather than a new colour of the same one, since Meta’s warning about similar creative covers near copies. And give each concept its own ad rather than piling assets into one, so every lead arrives with an ad ID you can group by concept later. If you hand Meta the audience and the budget, Advantage+ leads campaigns explains what stays in your hands.
How to test when Meta chooses the winner
More creatives on the same budget means fewer leads per creative. A budget of €3,000 a month at a €30 cost per lead buys about 100 leads; across 20 live creatives that is five each, while five concepts get 20 each and enough data to rank.
Three rules make the test readable:
- Low spend is not a verdict. Meta’s delivery decides which ads get budget, and its Help Center says an ad with few results next to better performers in the same ad set is expected behaviour from a system predicting performance.
- Fund a first screen at three times your target cost per lead, and compare finalists only at ten times. Below the first threshold nothing has been tested yet.
- Read every creative together with the landing page behind it. One concept can be cheap on one landing page and expensive on another, and a campaign average hides that.
The last point is where Andromeda changes the job rather than the settings. Meta decides which creative a person sees; the landing page is still your choice, and it decides what the click is worth. DropCPL measures the real cost per lead of each creative and landing page pair on your own pages, with spend synced per ad, so the ranking rests on leads you received rather than on Meta’s attributed count. The campaign setup questions cover the rest of the build.