Andromeda has no button in Ads Manager, so you cannot turn it on or configure it. What you can do is build your account in a way that makes the job easier for the system that picks your ads. This guide gathers what Meta has said publicly, what eight guides from agencies and analysts agree on, and what is in dispute. As a Meta Ads agency, we label each point so you know when you are looking at a Meta statement and when you are looking at someone's opinion.
Can you configure a campaign for Andromeda?
Not directly, because Meta announced no Andromeda setting for advertisers. According to Relevant Audience, the company describes the benefit as "better performance automatically." What you control is still the usual list: the objective, the data you send, the creative, the audiences, and the budget.
What changed is how importance is split among those pieces. Andromeda scores every eligible ad after applying your targeting, and according to Meta it reads the content of each creative. That is why almost every guide we reviewed agrees that creative weighs more than before, and fine targeting weighs less. It is a reasonable reading of the system's design, but no source has a controlled test that proves it.
What is Meta data and what is agency opinion?
Almost everything you will read online about how to set up campaigns is agency opinion. Segwise says it plainly in its own guide: Meta gives no setup recommendations in its engineering post. The figures for ad volume, minimum budgets, and refresh calendars come from each author's experience, and several cite no source.
In this article we use three labels. Meta data means the information comes from a Meta statement reported by an outlet. Agreement means several independent guides say the same thing, without Meta having confirmed it. And dispute means the sources contradict each other, and in that case we tell you how to check it in your account.
How do you structure the account step by step?
The strongest agreement is to consolidate. Distl recommends going from six ad sets to two or three, and Balistro proposes three to five, instead of 15 or more. Flighted suggests a testing campaign, a winners campaign, and, when spend justifies it, a cold audience campaign. The exact numbers are each author's judgment, but the direction is the same: fewer splits, more data per ad set.
To start, do this in order. First, one campaign per business objective, not per product or audience. Second, inside it, one or two ad sets with broad targeting. Third, a small separate campaign to test new ideas, with about 20 to 25% of spend according to Flighted and other guides. Fourth, when a test ad proves it works, move it to the main campaign.
There is a warning about merging. According to Balistro, combining ad sets restarts learning, and it is normal to see an efficiency dip of one or two weeks. If your account already performs well, consolidate gradually and not on the same day.
There is also a dispute about Advantage+ sales campaigns. Jetfuel makes them the primary campaign with 60 to 70% of budget, from about 300 dollars a day. Flighted keeps them as a separate campaign starting at 100,000 dollars a month. No Meta data resolves this, so compare the cost per purchase of each format in your account before moving money.
How much budget does each ad set need?
The budget must be enough for the ad set to produce enough results per week. According to Segwise, citing Meta's help center, an ad set exits the learning phase with about 50 results in the week after its last significant edit. Other guides repeat the figure without a source. There is no official daily minimum, and Linear Design points out that the "50 dollar daily minimum" in circulation does not appear in Meta's documentation.
A useful calculation follows from that. Multiply your cost per result by 50 and divide by 7, and you get the approximate daily spend an ad set needs to exit learning. With a cost per purchase of 30 dollars, that would be about 214 dollars a day per ad set. It is an example with invented numbers to show the math, and your real figure will differ.
If the result exceeds your budget, you have two ways out. You can concentrate everything in a single ad set. You can also optimize for an event more frequent than the purchase, such as add to cart, if that event relates to your real sales.
How many ads should you upload, and what counts as a different creative?
A creative counts as different when it changes the idea, not the format. Motion and Segwise agree that showing the same message as video, image, and carousel adds no diversity. Jetfuel proposes counting concepts and treating two ads as the same concept unless at least two of these change: format, person, setting, or benefit.
On volume, the sources do not agree, and all of it is opinion. Distl suggests five to eight concepts for accounts spending 2,000 to 10,000 dollars a month, and says it is its judgment and not a Meta rule. Segwise proposes eight to twelve concepts with two or three variations each. Jetfuel suggests 10 to 15 concepts and at least six distinct ads per ad set. Motion argues for launching 50 to 100 ads a month, and it is the most aggressive position.
Our view as a Meta Ads agency, which is opinion and not data: start with five to eight truly different ideas and scale up as your production capacity allows. Angles that work include a problem the customer recognizes, social proof, an offer, an explanation, and a testimonial. Mix polished production with native looking pieces shot on a phone.
Some analyses claim that Meta groups near identical ads and that a similarity threshold exists, with figures such as 40 or 60%. Those figures come from third parties and are not confirmed by Meta. Do not plan your production around them, but do avoid uploading ten versions of the same idea.
What happens with audiences?
Your audiences become hints, except for what Meta treats as a control. According to Linear Design, citing AdsUploader, controls are always respected: location, languages, minimum age, exclusions, and special category restrictions. Suggestions, on the other hand, include detailed targeting, custom audiences, and lookalikes, and Meta can reach people outside them if it expects better results.
This has several practical consequences. The first is that the exclusions you set as a control do work, and Linear Design calls them the most underused lever in the setup. For example, excluding recent buyers keeps you from paying to reach someone who already bought. The second is that stacking interests, ages, and lookalike audiences in layers rarely gives you the control you imagine.
The third is that Meta removed or consolidated more detailed targeting options on January 15, 2026, according to Social Media Today. Ad sets that used those options stopped delivering that day, so it is worth checking which of yours use them. Meta also reported a median 22.6% lower cost per conversion in its tests when it removed detailed targeting exclusions. That figure is from Meta and has not been audited.
With custom audiences there is a dispute. Linear Design summarizes that, with Advantage+ on, the list works as a priority and the system expands afterward, and that to restrict it you have to turn it off. Jon Loomer writes on an undated page that, with conversion objectives, audience suggestions cannot be turned off, and that in his tests those suggestions showed no measurable effect. The two descriptions do not match, so verify in your Ads Manager which options your objective allows.
On lookalikes and retargeting, the guides are split. Jetfuel says lookalike audiences are "effectively dead" and recommends broad targeting, although it admits a small account may use a 2 to 3% one temporarily. Flighted tests them inside the winners ad set, instead of opening one ad set per audience. For retargeting, Flighted proposes not separating it until you spend about 1,000 dollars a day, and limiting it to under 10% of budget. All of these are agency rules.
There are cases where strict controls make sense. They are regulated categories, such as housing, employment, or credit, and age restricted products. Linear Design adds accounts with very little conversion data and anyone who needs to exclude certain customers precisely.
Which objective and which data should you send?
Choose a single conversion event, the one closest to your revenue that happens often enough, and keep it. Jetfuel lists switching the optimization event between campaigns as a mistake. If you optimize for clicks, the system looks for people who click, and if you optimize for purchases, it looks for people who resemble those who buy.
On data, the common recommendation is to send the event through the pixel and the Conversions API at the same time, with a shared identifier to avoid duplicates. Segwise adds that it is worth sending the purchase value, the currency, and the order number. Jetfuel asks you to check event match quality in Events Manager every month, and sets a goal of 7 out of 10. That goal is the agency's, not Meta's.
This point matters more because of the November 12 date. According to PPC Land, Meta turns on a conversion measurement on the customer's device by default that day. If you do not record what your numbers looked like before, you will not know whether a jump comes from your work or from the measurement change. We explain it in detail in our pixel blog.
When should you touch the campaign and when should you leave it alone?
Leave it alone during the learning phase, unless something is broken. Each significant edit restarts the count, and checking the campaign every morning to change it is one of the surest ways to learn nothing. A weekly review is enough, and between reviews look only for errors such as rejected ads, a pixel that stopped sending, or a budget that ran out in an hour.
For refreshing creative, sources propose calendars of two to three weeks, but the figure has no solid source. Jetfuel prefers a signal: refresh when the cost per thousand impressions rises without volume rising. Segwise lists alerts, such as a weekly drop of more than 10% in click rate, a cost per result that rises more than 15%, or a frequency above 3. They are agency thresholds, without a source. Use them as a starting point and adjust with your history.
What common mistakes make results worse?
These are the mistakes that repeat most in the guides we reviewed. Over splitting heads the list: many campaigns with many ad sets and little budget in each. Uploading variations of the same idea and believing that is diversity takes second place.
Stacking interests, ages, and lookalike audiences also appears a lot, because people expect a precision the system no longer guarantees. Another mistake is setting up the Conversions API once and never auditing it, with duplicate events from integrations. Changing the objective or the event mid campaign is just as common. And last is judging by the platform's number without crossing it with your real orders.
Consider a typical case, not real data but an example of how it might look. A home decor store in Florida has six ad sets with 25 dollars a day each and two ads per set. It reduces them to two ad sets with 75 dollars a day each, uploads six different ideas, stops editing for ten days, and compares the sales Meta reports with those in its store. If the cost per real order drops, it keeps the structure. If nothing changes, it checks the data and the creative first before splitting again.
How do you check that the setup works?
Compare the result Meta reports with your real orders, and when you can, use a holdout test or a lift study. Segwise recommends validating with incrementality tests before raising budgets significantly. Meta measures its own results, so a platform number is not an independent confirmation.
A simple way is to export your store or CRM sales every week. Then calculate the ratio between what Meta reports and what was actually sold. If the ratio stays stable, you can trust the trend. If it jumps, look at what changed, whether an edit, a measurement change, or a promotion date.
When does it make sense to hire a Meta Ads agency?
When you invest enough that a poorly built structure or a data error costs real money, and you do not have time to review it every week. A Meta Ads agency can consolidate the account, audit the pixel and the Conversions API, plan a creative calendar with different ideas, and compare Meta's results with your sales. It can also tell you which of the rules in circulation hold in your account and which do not. If you want us to review your structure, you can contact the JP Director team.
This topic connects with others we have published. To understand how Meta selects ads, read our blog on Andromeda, GEM, and GenRecs. And to prepare your data before the November 12 change, review our blog on the Meta pixel.
Frequently Asked Questions
How do I turn on Andromeda in my account?
You do not. Meta announced no setting for advertisers, and the system works automatically. What you can adjust is the structure, the data, the creative, and the audiences.
How many ads should I have per campaign?
There is no official figure. Guides recommend anywhere from five to eight distinct concepts for mid size budgets up to 50 to 100 ads a month for large accounts. Start with a few truly different ideas and adjust with your results.
Did custom audiences and lookalikes stop working?
They work as hints and not as limits, and the guides disagree on how much they weigh. Exclusions are respected as a control. Check which options your objective allows in Ads Manager.
How much budget do I need per ad set?
Meta publishes no daily minimum. One reference is to multiply your cost per result by 50 and divide by 7, since about 50 results per week are needed to exit learning.
Last updated: November 9, 2026. This article will be updated if Meta publishes official setup guidance for Andromeda, clarifies how custom audiences behave, or changes the conditions of the learning phase.
