TL;DR
Meta Andromeda is Meta’s 2025 ads delivery system overhaul, paired with Gem (the AI architecture that orchestrates ad sequences). Together they turned Facebook ads into a full-funnel distribution engine where creatives, not interests or lookalikes, decide who sees your ad. Brands that kept the old test-and-scale, interest-targeting, single-angle-per-adset playbook saw ROAS collapse through 2025. The brands that won, including the ones we operate at Skaleit, consolidated structure, diversified creative, went broad, and fed Meta the user-awareness signals it needs. This guide covers the full system, with specific client numbers across $60K to $210K spend levels.
What Meta Andromeda actually is, and why it broke your ROAS
Meta Andromeda is the rebuilt ad delivery and ranking system Meta rolled out across 2024-2025, with Gem acting as the AI brain that sequences which ads each user sees and in what order. The short version: Meta no longer optimizes the winning ad. It optimizes the winning ad sequence.
That single shift breaks almost every legacy account structure. As of 2026, Meta reorganises every creative you upload into audience and sub-audience buckets, then delivers them as a journey. If you turn off a top-of-funnel ad because its in-platform ROAS looks bad, the whole sequence below it collapses. Brands hit by this typically saw ROAS drop 30 to 50% before they understood what changed.
The accounts we audit weekly almost always show the same pattern: split campaigns for testing and scaling, one angle per adset, interest stacks layered on top, and creative iterations that only changed the headline. That stack worked before Andromeda. Post-Andromeda, it actively fights the algorithm.
Why ROAS dropped after the Andromeda update
Andromeda penalises fragmented signal. When you split creatives across many adsets, layer interests, and exclude audiences aggressively, you starve every adset of the 50 conversions per week the learning phase needs and you hide ad-sequence data from Gem.
We see three failure modes when we audit accounts that lost ROAS through 2025. First, fake winners: brands picked scaling creatives based on raw spend and purchase counts inside Ads Manager, which mixes 1-day-view attribution and existing-customer purchases into the number. Second, slow turn-offs: bad creatives ran for weeks because operators waited for “more data.” Third, copy-only iteration: changing the headline on a winning visual and expecting it to target a new audience, which Andromeda no longer treats as a new ad.
Across one client we recovered from 2.0x to 3.89x ROAS by fixing exactly these three problems, a 93% lift on roughly the same monthly spend. The strategy was not new creative volume. It was reading incremental attribution correctly, killing underperformers fast, and rebuilding the structure around how Andromeda actually distributes ads.
The new targeting model: broad, creative-led, no interests
Broad targeting is the only audience strategy that works under Andromeda. Interest targeting and lookalike audiences now restrict the data Meta uses to find your buyers, and on many accounts Meta has already removed the option to target interests exclusively, the “narrow” tick is gone.
Here is the mental model. Your creative defines your audience. Meta’s AI reads the visual, the copy, the on-screen text, the price point, the language, and decides who to show it to. If you set an interest layer on top, you are not narrowing onto your buyer, you are narrowing onto Meta’s guess of your buyer, which excludes the people Meta would otherwise have found through creative signals.
A simple test: take your ad creative, drop it into ChatGPT, and ask which audience this ad is targeting. Whatever ChatGPT reads from the image and copy is roughly what Meta reads. That is the audience that will see it. The way to target a 45-year-old skincare buyer is not an interest stack, it is a high-quality visual that shows a 45-year-old woman, copy that mentions the age range, and a visible price point that filters out low-intent clicks.
This is the same logic we apply when we scale a swimwear brand: the creative carries the demographic and price signal, the audience setting stays broad, and Meta finds the buyers.
Account structure post-Andromeda: consolidation wins
One campaign, one adset, many creatives is the structure Andromeda rewards. The legacy split between a testing campaign and a scaling campaign, with cost cap or bid cap on top, fragments learning and forces you to manage sequences manually, which Gem now does automatically.
The structure we run on most ecommerce accounts as of Q1 2026 is a single CBO (Campaign Budget Optimization) prospecting campaign, one adset inside, broad placements, no audience exclusions at the adset level, and 15 to 50 creatives stacked inside. We have seen the 50-ad cap soften on some accounts and watched single adsets keep performing past it. The principle holds either way: minimise adsets, maximise ads per adset.
Why this works: the learning phase fires at the adset level and needs 50 conversions per week to exit. If you split into five adsets with 10 creatives each, none of them stabilise. Stack everything into one adset and you compound signal. On one account we lifted ROAS from 4.30x to 6.08x on a single campaign just by adding 10 new creatives into the existing winning adset instead of launching a new one.
ABO vs CBO under Andromeda: when each still matters
ABO (Ad Set Budget Optimization) and CBO (Campaign Budget Optimization) both still work, but the use cases compressed. CBO is now the default for almost every prospecting campaign because it lets Meta allocate budget across the consolidated structure without manual interference. ABO survives in narrow cases where you want to force spend onto a specific creative or test a brand-new angle in isolation.
We covered the full decision tree in the ABO vs CBO testing and scaling guide, but the post-Andromeda summary is simple. Use CBO when you want Meta to optimise spend across the winning ad sequence. Use ABO sparingly for forced budget allocation on a single high-confidence concept or when a brand has fewer than 10 creatives ready and you need to surface signal fast.
The bigger shift is that the old “ABO testing then CBO scaling” two-campaign loop adds drag now. You are isolating creatives at the testing stage from the data they need to perform, then asking them to scale in a campaign that has never seen them in sequence. The brands we run with one CBO from launch consistently outperform the brands we run with the split.
Creative requirements: volume, diversity, and net-new concepts
Creative diversification is the single biggest lever post-Andromeda. Diversity is not changing the background colour or the CTA button. It is genuinely different concepts, formats, hooks, and angles, each with its own unique entity ID so Meta treats it as a new ad.
The format mix we run across our ecommerce clients includes street interviews, testimonial ads, split-screen ads, client story ads, “X reason why Y” ads, founder ads, VSLs, podcast-style ads, and net-new original formats for video. For statics, we mix UGC photos, before-and-after, headliners, benefits, “us versus them,” AI visuals, meme cards, sticky notes, tweet-style ads, iPhone notes, and iMessage ads. The originality matters: when our creative team produces a format people have not seen 50 times that quarter, it almost always outperforms templated work.
The net-new concept rule is the one most brands break. Copying a competitor’s “point-out arrows on a product photo” static and swapping the product gets the angle right but kills stop-scroll because viewers have seen the format too many times. We replace that with a fresh visual concept that carries the same angle. On the recovery account that went from 2.0x to 3.89x ROAS, 70% of winning creatives were statics, and every one of them used a net-new visual concept rather than a copied template. That is the same approach we use when launching fashion drops, where the concept has to feel native to the drop, not borrowed from another brand.
Iterating on winning ads: change the visual, not just the headline
Winning iterations under Andromeda require changing the visual to match the new headline and angle. Keeping the same image and only swapping the headline now targets the same audience Meta already showed it to, which is why ad fatigue accelerated through 2025.
The framework we use: when a creative wins, we ship two to four iterations that match a new sub-audience. New persona on screen, new visual context, new copy tone, same core angle if the angle was the winner, fresh angle if the angle is what we are exploring next. If the winning ad showed a salad with “lose weight with this meal plan,” the iteration for a fitness audience is not the same salad with a new headline, it is a high-protein dish with “gym bros are going crazy with this meal plan” and the on-image stats updated.
This is what lets a winning concept compound instead of decay. Every iteration extends the sequence Gem can deliver, which is the actual scaling unit on Andromeda.
A practical iteration cadence we use across client accounts: when a creative crosses the threshold to be called a winner (consistent ROAS above account average across at least 50 conversions, ideally read on incremental new-audience attribution), we ship two iterations within seven days and a third within 14 days. One iteration explores a new persona on screen, one explores a new context or use occasion, and the optional third tests a new angle entirely while keeping the visual concept anchored. That cadence is what compounds spend on a winning concept rather than letting it die at frequency three.
Scaling without ROAS collapse: budget, audiences, and bid strategy
Scaling under Andromeda fails when your creative pool is bottom-of-funnel only. When you raise the budget, Meta has to find new audiences, and the audiences it pulls in are colder. If your ads are testimonial, “us versus them,” and “most aware” hooks, none of them speak to unaware or problem-aware buyers, and CPA blows up.
The fix is structural, not bid-related. Build creative coverage across all five user awareness stages: unaware, problem aware, solution aware, product aware, most aware. Most accounts we audit have 90% of creatives in the bottom three stages. Adding genuine unaware and problem-aware creative is what lets a brand break the $100K-per-month plateau, which we have done on multiple accounts including one we scaled past $400K per month from a $42K monthly spend baseline.
Bid strategy still matters, but it is downstream of creative coverage. We covered the manual cap mechanics in our bid cap strategy guide and the operating-level detail in the bid cap operating playbook. The short version: highest-volume bidding is the default under Andromeda because it lets Gem allocate. Cost cap and bid cap come in at scale when you need to defend CPA against the engagement-bias drift, which happens when Meta starts spending more on a creative because of high engagement rather than high efficiency.
The Andromeda recovery playbook: what to do if you got hit
If your ROAS dropped 30% or more in 2025 and has not recovered, the fix is structural, sequenced, and takes 30 to 60 days. We have run this exact recovery on accounts spending between $15K and $200K per month.
Step by step:
- Audit attribution. Switch to incremental ROAS on the new-audience segment and recalculate which creatives are actually winners. The “high spend, high purchase count” view inside Ads Manager mixes 1-day-view and existing-customer revenue, which inflates fake winners.
- Kill aggressively. Any creative spending materially with ROAS more than 30% below the campaign average gets paused. Holding bad ads on does not stabilise the account, it drags the average.
- Consolidate. Collapse testing and scaling campaigns into one CBO prospecting campaign. One adset, broad placements, no exclusions at the adset level, all creatives stacked inside.
- Set audience definitions at the account level. Engaged audience: website visitors, add-to-cart, checkout-initiated, full Klaviyo contact list, all excluding existing customers. Existing customers: 180-day pixel purchases plus a dynamic Klaviyo list of one-or-more-purchase buyers. Make both lists dynamic.
- Rebuild creative coverage. Map your existing creatives to the five user awareness stages and fill the gaps with net-new concepts, especially unaware and problem-aware.
- Set the iteration loop. Two to four visual-and-angle iterations per winner, weekly creative drops, no headline-only refreshes.
On the brand we recovered from 2.0x to 3.89x ROAS, this exact sequence ran across roughly six weeks. Spend stayed in the $17K to $23K range while ROAS doubled, which is what a real Andromeda fix looks like, structural, not media-volume-driven.
Audience setup at the account level: the mistake we see weekly
The most common Andromeda audience mistake is putting the Klaviyo list inside the engaged audience without excluding existing customers. When you do that, your existing customers sit inside both the engaged audience and the existing-customer audience, which double-counts them and breaks Meta’s full-funnel distribution.
The correct setup, as of 2026:
- Engaged audience: website visitors, add-to-cart, checkout-initiated, full Klaviyo contact list. Exclude existing customers from this audience.
- Existing customers: 180-day pixel purchase data plus a dynamic Klaviyo list filtered to “made at least one purchase.” Both lists must be dynamic so they update.
When this is set correctly and you run the consolidated CBO with no adset-level exclusions, the budget split we typically observe is roughly 65% new audience, 31% engaged audience, 3 to 4% existing customers. That is Andromeda doing the full-funnel job for you. If your split is heavily skewed toward existing customers, your audience definitions are wrong, not your campaign.
Vertical-specific notes: swimwear, fashion drops, toy brands
Andromeda hit some verticals harder than others, and the recovery playbook adapts to category. Swimwear, fashion drops, and toy brands all have different signal profiles, AOV ranges, and creative requirements.
Swimwear brands run continuously and live or die on creative diversity across body types and use occasions. The consolidated CBO with broad targeting works almost out of the box, the work is in the creative pool depth. We covered the structure in the swimwear scaling case study.
Fashion drop brands have a different problem: short windows, high spike, and a need to layer a promo CBO on top of the always-on prospecting structure during the drop. The fashion drop playbook details how we run that overlap.
Toy brands and kids categories tend to have higher repeat-purchase potential and tighter targeting compliance constraints. The structure stays the same, broad and consolidated, but creative leans heavier on UGC and parent-voice angles. We documented our approach in the toy brands scaling system.
Market sophistication: why “the same offer” stopped working
Market sophistication is the missing variable behind most “Andromeda killed my ROAS” complaints. Andromeda exposes weak messaging because it forces ads in front of colder, more skeptical audiences as you scale, and weak messaging cannot survive that exposure.
The five-level model from Eugene Schwartz’s Breakthrough Advertising still applies. Level one: new market, simple claims work. Level two: claims need to beat competitors. Level three: introduce a mechanism that explains why your claim works. Level four: upgrade the mechanism. Level five: identity, community, movement.
Most ecommerce categories we work in (supplements, weight loss, skincare, fashion) sit at level four or five as of 2026. If your ads still run level-two “X is better” claims, broad audiences will not buy them, and Andromeda’s full-funnel distribution will surface that problem faster than the old algorithm did. The Andromeda fix at the messaging layer is matching the message to the sophistication level, not adding another product feature to the headline.
What to expect through the rest of 2026
Andromeda and Gem are going to keep tightening, and the gap between consolidated, creative-led accounts and legacy structures is going to widen. We expect Meta to fully remove standalone interest targeting on more accounts, push Advantage+ Shopping Campaigns harder as the default, and reward ad sequences over single-ad performance even more aggressively.
What that means operationally: creative production volume becomes the primary scaling input. Account structure work compresses to a few decisions made once. Attribution literacy (incremental ROAS, new-audience segmentation) becomes the difference between scaling and stalling. Brands that run a real creative system and a real attribution layer will pull ahead. Brands that keep tweaking bid caps and audience layers will keep losing.
The brands we operate are running roughly 30 to 80 fresh creatives per month at scale, mapped across the five awareness stages, refreshed weekly, with two to four visual-and-angle iterations behind every winner. That production cadence is the new competitive moat, more than any structural setting in Ads Manager.
We also expect attribution to keep shifting. Meta’s seven-day-click, one-day-view default already misreads incrementality on most accounts, and as Andromeda routes more spend through full-funnel sequences, the gap between platform-reported ROAS and true incremental ROAS will widen. The practical implication for ecommerce brands: build a habit of reading new-audience-only ROAS weekly, layer post-purchase survey data (Fairing, KnoCommerce, or a built-in Shopify survey) to triangulate, and stop making creative kill decisions off the default Ads Manager view alone. Brands that operationalise that read across 2026 will scale faster than brands waiting for Meta to fix attribution for them.
Working with us
We run this exact playbook for ecommerce brands across fashion, swimwear, supplements, toys, and home goods, typically at $30K to $500K-plus per month in spend. If your ROAS dropped after Andromeda and you want a structural audit and recovery plan, book a consultation call with Skaleit.
About the author
Antonio Ventre is the founder of Skaleit, a Hong Kong-based ecommerce growth marketing agency specialising in Meta Ads. Skaleit manages over $10 million per year in Meta ad spend across ecommerce brands and operates the consolidated CBO and creative-diversification system described in this guide as the standard playbook across all client accounts as of 2026.