Meta Advantage+: What the AI Really Does

Advantage+ is the brand Meta puts on almost everything its machine learning touches: who sees your ads, where they show up, how budget moves, even how the creative looks. Open Ads Manager today and the suite is on by default. That convenience hides a real question most advertisers never answer: what does this AI actually do, where does it genuinely help, and where does it quietly cost you money? This guide breaks down the full Advantage+ suite, the Andromeda engine underneath it, the numbers Meta publishes versus what independent tests find, and the moment you should take back manual control.

Two myths to kill before we start

Two ideas dominate the forums, and both are wrong. The first: Advantage+ always beats manual. It does not. Meta’s own reported lift is an average, which means a large share of accounts do worse, and independent incrementality studies confirm it. The second: you launch Advantage+ and forget it (set and forget). That is exactly how you bleed budget. The True Classic story below shows what happens when nobody is watching. Treat Advantage+ as a powerful junior media buyer who never sleeps but has zero judgment about your brand. It executes brilliantly. It decides nothing important. That distinction is the whole article.

Andromeda: the engine you never see

Everything Advantage+ does sits on top of an engine Meta calls Andromeda, announced on the Engineering at Meta blog in December 2024. Andromeda is the retrieval layer: for every single impression, it narrows tens of millions of eligible ads down to a few thousand candidates before the auction even runs. Meta co-designed it with NVIDIA Grace Hopper hardware to handle that scale at brutal latency budgets. The headline claim is that it processes candidate ads at roughly 10,000 times the scale of its predecessor. It rolled out quietly across Facebook and Instagram, with global completion around October 2025, no press release, no UI change.

Why does this matter to you, the advertiser? Because Andromeda changes what wins. When the engine can evaluate far more ad-user combinations per impression, the bottleneck shifts away from targeting precision and toward creative volume and quality. Meta reported a +6% recall improvement to the retrieval system and +8% ads quality gains on selected segments. In plain terms: the machine got much better at finding the right person, so the lever you still control, the actual ad, matters more than the audience settings you obsess over. Andromeda is the reason Meta keeps telling you to feed it more distinct creatives. The retrieval engine is hungry.

One subtle consequence of Andromeda deserves its own warning. Because the engine now layers its influence gradually rather than through a visible toggle, your account changed without you noticing. Many advertisers who saw audience sizes balloon, CPMs shift, or delivery patterns scramble in 2025 blamed their own settings or the economy. Often the real cause was the retrieval engine reaching their account during its multi-year rollout. The practical lesson: do not over-react to month-to-month swings by re-engineering your campaigns. Some of the volatility is the platform changing under your feet. Stable creative and clean measurement are your defense against an engine you cannot directly control or even see in the interface.

The Advantage+ suite, component by component

Advantage+ is not one button. It is a family of automations you can toggle separately, and understanding each one is how you stop treating the suite as magic. There are five pieces that matter: Advantage+ Audience, Advantage+ Sales (formerly Advantage+ Shopping), Advantage+ placements, Advantage+ creative, and Advantage+ catalog. Each one hands a different decision to the algorithm. Some of those handoffs are clearly worth it. Others quietly remove control you actually wanted to keep. The skill is knowing which is which, not flipping every switch because Meta’s Opportunity Score nags you to.

Advantage+ Audience

Advantage+ Audience replaces hard targeting with a suggestion. You give Meta an audience hint, and the algorithm treats it as a soft signal, free to go beyond it whenever it predicts a better outcome. Meta’s own benchmarks claim up to 32% lower cost per acquisition and 13% lower cost per catalog sale versus manual targeting, plus figures like 28% lower CPC and 7% lower cost per website conversion in other tests. Those numbers are real but they are Meta’s, measured Meta’s way. The honest read: for broad-appeal products with healthy conversion volume, the suggestion-based approach usually wins. For narrow, regulated, or niche offers, the freedom to roam can burn spend on the wrong people.

There is a quieter cost to Advantage+ Audience that the CPA numbers hide. When the algorithm roams beyond your hint, it often finds your cheapest conversions among people already close to buying, including past visitors and existing customers. That makes the reported cost per acquisition look great while the incremental value, the new demand you actually paid to create, can be thin. This is the gap between platform CPA and true business impact. Use Advantage+ Audience for prospecting at scale, but never assume a low reported CPA means you reached fresh buyers. Pair it with exclusions of recent purchasers when net-new acquisition is the real goal, and check your blended results, not just the in-platform number.

Advantage+ Sales (formerly Advantage+ Shopping)

In 2024 Meta renamed Advantage+ Shopping Campaigns to Advantage+ Sales, because the product had outgrown pure e-commerce and now spans sales, app installs, and lead generation. This is the fully automated campaign type: one campaign, minimal ad sets, the algorithm controls audience, placement, and delivery end to end. Meta also began merging the choice away, so marketers no longer pick between a manual or an Advantage+ campaign at setup. In early lead-gen testing, campaigns with Advantage+ on delivered on average 10% lower cost per qualified lead than those with it off. Powerful for scaling proven products. Dangerous if your tracking or offer is broken, because it scales the broken thing faster.

The merged setup deserves a flag of its own. By removing the explicit manual-versus-Advantage+ choice, Meta makes automation the path of least resistance, and the default is now everything on. That is convenient for a beginner and a trap for a serious operator. The danger is not that Advantage+ Sales performs badly. It is that you lose the deliberate decision of whether to automate at all. Always open every section of a new campaign before launch, confirm which automations are active, and decide each one on purpose. The campaign that launches on defaults is the campaign nobody chose, and it usually runs broader and looser than you would have set it yourself.

Advantage+ placements and creative

Advantage+ placements let the system spread your ad across every surface (Feed, Stories, Reels, Marketplace, Audience Network) and shift impressions wherever they perform. That is genuinely good: manual placement picking almost always underperforms because you cannot guess where a given creative will click. Advantage+ creative is the controversial one. It auto-applies enhancements: brightness tweaks, music, aspect-ratio changes, AI-generated variations, even swapped imagery. Since February 2025, new Sales, Leads, and App campaigns launch with these enhancements pre-selected. For a brand with tight visual standards, that default is a liability, and it is where the most damage happens.

Advantage+ catalog

Advantage+ catalog is the evolution of dynamic product ads. Connect your product feed, and the system picks which items to show each user and assembles the creative around them, pulling images, prices, and copy automatically. For retailers with hundreds or thousands of SKUs, this is where automation earns its keep: no human can hand-build an ad for every product-and-person pair. The catch is the same as everywhere else. Garbage in, garbage out. If your feed has wrong prices, missing images, or stale stock, the AI will faithfully promote the wrong thing at scale. The automation amplifies your data quality, good or bad.

The numbers: what Meta says versus what tests find

Here is the headline Meta wants you to remember. In its Q1 2025 earnings, Meta reported advertisers seeing $4.52 in revenue per $1 spent with Advantage+ campaigns, about 22% higher than manually managed ones. In Q4 2024, the Advantage+ suite passed a $20 billion annual run rate and grew 70% year over year, on total Q4 revenue of $48.39 billion. Adoption is real: roughly 35% of US retail ad spend now runs through Advantage+, up from about 19% a year earlier. Those are big, credible, official numbers. They are also exactly what a platform selling automation would publish, measured with platform attribution that tends to flatter itself.

Now the counterpoint, which Meta does not put on a slide. Incrementality testing tells a messier story. In a body of head-to-head experiments, 58% of brands saw higher incremental ROAS on manual campaigns than on Advantage+, with Advantage+ driving 12% lower DTC incremental ROAS at 18% lower daily spend. One apparel brand found that only 17% of its reported conversions were truly incremental. In another structured test, Advantage+ led by 9% at the midpoint, then faded: manual averaged 32% post-treatment lift while Advantage+ averaged only 17%. The pattern is consistent. Platform-reported ROAS overstates Advantage+ because the AI is excellent at harvesting people who would have bought anyway.

This is why a suspiciously high Advantage+ ROAS is sometimes a red flag, not a trophy. If a campaign reports a 9x ROAS but your overall business revenue did not move, the AI is mostly claiming credit for existing demand, especially returning customers. Meta’s own research even found that capping existing-customer budget at a minimum of 10% can improve both ROAS and cost per result, an implicit admission that, left alone, the system over-serves people already in your funnel. The takeaway is not that Advantage+ is fake. It is that you must measure it against a holdout or geo-test, never against its own self-reported dashboard.

Real brands, real results, both directions

The wins are documented. Ray-Ban, working inside Advantage+ Sales, improved performance by layering value optimization on top: it added a consideration goal to a value-optimization campaign and saw a 9% increase in return on ad spend plus a 32% lift in average order value. That is a textbook example of human strategy steering the AI rather than replacing it. The lesson is not that Advantage+ did it. It is that a smart operator configured the objective correctly and let the engine execute. The structure was the human’s decision. The optimization was the machine’s job.

On the agency side, the scaling stories are striking. Rama Water Filter, working with the agency Socialee from November 2024, went from a 2.2x ROAS to a 6.7x ROAS by June 2025, with monthly purchase volume growing roughly 8x. Hurom, a premium juicer brand, partnered with inBeat to cut cost per acquisition by around 65% using data-tested UGC and video across Meta and Google. Notice the common thread in every win: a human supplied better creative, better offers, or better measurement. The AI scaled what already worked. None of these brands hit a button and walked away. They fed the machine well, then watched it.

It is worth being precise about who reports these wins. Meta’s earnings figures and the Ray-Ban case come from Meta and its business-partner program, so they carry an obvious incentive. The Rama and Hurom results come from the agencies that ran the accounts, who also have a stake in looking good. None of that makes the numbers false, but it tells you to read them as best-case stories with skilled human operators behind them, not as what happens when an average advertiser flips the suite on. The brands that publish failures are rarer, which is exactly why the True Classic incident, surfaced by Business Insider rather than by any vendor, is so useful. It is the unflattering data point automation marketing never volunteers.

And the failure that should haunt every set-and-forget believer: True Classic, a menswear brand, had Advantage+ creative enhancements running, reportedly without intending to. Meta swapped its top-performing ad, an attractive millennial man in a matching fleece set, for an AI-generated photo of a cheerful grandma in an armchair, holding a product the brand does not even sell. True Classic targets men aged roughly 30 to 45. The ad ran for several days before customers, not the brand, flagged it. Other advertisers told Business Insider that Meta would re-enable automatic adjustments even after they switched them off, forcing one agency to block out mornings each week just to re-check the toggles.

What the AI does not solve

Here is the hard truth Meta’s marketing skips: Advantage+ optimizes delivery, not your business. It cannot fix three things, and these three things determine most outcomes. First, broken tracking. If your Pixel and Conversions API are sending bad or sparse signal, the AI optimizes toward noise, and no amount of automation rescues a campaign that cannot see its own conversions. Second, a weak offer. The algorithm finds the people most likely to buy what you sell, but if the deal is unconvincing, it just finds them faster and watches them not convert. Third, poor creative. Andromeda made creative the lever, so mediocre ads now fail more visibly, not less.

Add a fourth, subtler failure: transparency. Because Advantage+ optimizes many variables at once, you often cannot tell which audience or placement drove a result, and Meta’s dashboard will not tell you what share of conversions came from genuinely new customers versus existing ones. The Opportunity Score makes this worse by grading your account on how many Meta features you have enabled, not on whether they help you. It measures adoption, not performance. If you optimize for a higher Opportunity Score, you optimize for Meta’s interest in selling automation, not for your incremental profit. Treat that score as a sales prompt, never as a scorecard.

Add the broadest limit of all: Advantage+ cannot create demand that does not exist. It is an allocation and matching engine, not a demand generator. If nobody wants your product at your price, the AI will simply discover that faster and cheaper than a human would, which feels like failure but is actually accurate information. Smart advertisers use this. A quick Advantage+ Sales test on a new product is a cheap way to learn whether real demand exists before you invest in heavy creative production. The point is to read the AI as a market signal, not to expect it to manufacture interest. The machine reflects your market. It does not invent one.

When to keep manual control

Manual is not dead, it is targeted. Keep manual control in five situations. One, low conversion volume: Advantage+ works best above roughly 50 weekly conversions, so small or new accounts often need manual structure to feed the learning phase first. Two, strict brand standards: turn off Advantage+ creative enhancements when a swapped image or AI variation would embarrass you, exactly the True Classic scenario. Three, regulated or niche offers, where the audience freedom of Advantage+ Audience wastes budget on the wrong people. Four, retention or win-back campaigns that need precise audience boundaries. Five, anytime you are running an incrementality test and need a clean, controlled manual cell as your baseline.

The practical workflow is hybrid, not tribal. Run Advantage+ Sales for your proven, broad-appeal products where volume is high and the offer is strong. Keep one anchor manual ad set running in parallel as a governance and comparison cell, so you always have a non-automated baseline to judge the AI against. Cap existing-customer budget if you sell to a base that buys repeatedly, per Meta’s own 10% finding. Audit your creative enhancement toggles on a schedule, because they can re-enable themselves. And measure everything that matters against a holdout or geo-test, not against Meta’s attribution. That is what every winning case study above actually did, even when the headline credited the AI.

If you are starting from scratch, here is a sane sequence. Get tracking right first: a clean Pixel plus Conversions API, with strong event match quality, so the AI optimizes toward real conversions and not noise. Then build three to five genuinely distinct creatives, different hooks and angles, because Andromeda rewards variety, not minor tweaks of one idea. Launch one Advantage+ Sales campaign for your best offer and one anchor manual ad set beside it. Leave creative enhancements off until you have proof they help your specific brand. Let it run past the learning phase without panic edits. Then, and only then, judge it with a holdout test. That order protects you from every trap in this article.

One last reframe ties it together. The advertisers losing money to Advantage+ are rarely the ones who refuse it. They are the ones who hand it every decision and then read its own dashboard as proof it worked. The advertisers winning treat the AI as execution and keep strategy, brand, offer, and measurement firmly human. That division of labor is durable. Automation handles the millions of micro-decisions per second that no person could. You handle the handful of decisions that define the business. Andromeda and Advantage+ are tools of leverage, and leverage multiplies whatever you point it at, including a bad plan. Point it at a good one, watch it closely, and it earns its place.

The honest verdict

Advantage+ is the most powerful media-buying automation ever shipped to small advertisers, and it is genuinely worth using for most accounts. But it is not better than manual by default, and it is the opposite of set and forget. The AI handles retrieval, ranking, and delivery at a scale no human can match, thanks to Andromeda. It does not handle your tracking quality, your offer, your creative, or your brand judgment, and it will scale your mistakes as eagerly as your wins. Use it as leverage on top of good fundamentals, measure it against real incrementality, and keep your hand near the controls. The advertisers who win in 2026 are not the ones who automate the most. They are the ones who automate the right decisions and keep the important ones for themselves.

Sources

Engineering at Meta, Meta Andromeda: Supercharging Advantage+ automation (December 2024). Meta Q4 2024 and Q1 2025 earnings calls and reports (Motley Fool, CNBC, Variety). Social Media Today, Meta Advantage+ updates and AI targeting (32% ROAS, February 2025). Jon Loomer Digital, 83 Changes to Meta Advertising in 2025 and Advantage+ Sales naming. Measured, Meta Advantage+: Why High ROAS is a Red Flag. Haus.io, The Meta Report: Lessons from 640 Incrementality Experiments. Business Insider via Shopifreaks and DesignRush, True Classic AI creative incident. Ray-Ban value-optimization case study (Meta for Business). Socialee (Rama Water Filter) and inBeat Agency (Hurom) case studies. Stella Hey Stella, ASC incrementality study methodology.