Why Meta ROAS Is Vanishing in 2026 (and Which Part of the Drop Is Real)

You are not imagining it

Quick answer: Reported Meta ROAS fell for two unrelated reasons in 2026. Attribution changes in January and March made the number smaller without changing your sales. Separately, delivery got genuinely worse for a lot of fashion brands. The first drop is noise. The second one is real, and it traces back to the fact that targeting has quietly moved out of your audience settings and into your creative and your product catalog.

One apparel founder we spoke with recently was spending close to eighty thousand dollars a month on Meta three years ago. This year the number sits closer to thirteen. Not because the budget was reallocated, but because the spend stopped returning anything. Things held together through February and March, then fell off a cliff in April.

She is not a beginner. She has been running Meta ads since 2017. She has hired creative strategists, run structured briefing programs with creators, tested ABO, tested cost caps, tested bid caps, moved to lowest cost, rebuilt landing pages, upgraded photography, added video, and used competitive ad research tools. Her summary of the year was that she genuinely does not know why it stopped working, and that if new customer acquisition does not recover, the business has about a year.

If any of that sounds familiar, the useful thing to know is that the drop on your screen is actually two different drops stacked on top of each other. They have different causes and completely different fixes, and telling them apart is worth doing before you touch anything.

The part of the drop that is not real

Two reporting changes landed this year, and both of them made your numbers smaller without touching a single order.

January 12. Meta permanently removed the 7-day view and 28-day view attribution windows from its reporting API. Only 1-day view and 7-day click survived. Purchases that used to get credited days after an impression stopped being counted at all. Some advertisers lost 30 to 40 percent of reported conversions overnight.

March 3. Meta redefined what counts as a click. For years, likes, shares, saves, and comments all opened the 7-day click window. Now only a real link click does. Everything else moved into a new category called engage-through, with a 1-day window. Retargeting campaigns absorbed most of the damage, because a large share of their credited conversions had always come from those softer interactions.

Neither change touched delivery. They changed the accounting.

This part matters less because it is comforting and more because it is camouflage. If your reported ROAS fell 40 percent and 20 points of that is accounting, then you are trying to diagnose a real 20 point problem while looking at a number that says 40. Every decision you make from that number is calibrated to a loss that is bigger than the one you actually have.

The part that is real

Here is where most coverage of this topic stops, and where the actual problem starts.

Meta spent 2026 rolling out Andromeda, a rebuilt delivery and retrieval system, in phases across accounts. It leans much harder on machine selection: which creative, which audience, which product, which placement. The practical consequence is that the levers you used to pull carry less weight than they did.

Operators describe this in strikingly similar terms. Manual cost controls stopped behaving predictably. The same founder above had a bid capped ad, set to stop at 150 dollars, spend 900 dollars overnight with zero add to carts. Not a low conversion rate. Zero.

And the clicks were cheap. Forty cents. On paper that looks like a win, which is exactly what makes it dangerous. Cheap traffic that does not convert is worse than expensive traffic that does, because it burns budget while every top-line efficiency metric tells you things are going well.

That is the shape of the real problem in 2026. Not cost per click. Cost per outcome, with cost per click actively lying to you about it.

Underneath that sits genuine auction pressure. Organic reach keeps declining, which pushes more brands into paid. Large advertisers with consumer packaged goods budgets have moved into inventory that used to be dominated by DTC brands. Ultra fast fashion players spending at enormous scale sit in the same feed as brands doing a few million a year. A small apparel brand is now bidding against companies with completely different margin structures for the same impression.

None of that is a measurement artifact. That is the market.

Targeting moved, and almost nobody adjusted

This is the shift that explains most of what feels inexplicable.

Broad targeting is effectively the default now. You are not really choosing an audience anymore. The system chooses, and it decides who to show your ad to by reading the ad itself and the products behind it.

Sit with that for a second, because it changes what your job is. When targeting lived in the audience settings, your product data was just plumbing for the catalog. Now your product data is a targeting input. It is part of how the system decides whether a garment belongs in front of a 27 year old or a 45 year old, whether it reads as workwear or weekend, whether it suits one body type or another.

Most fashion catalogs were never built to answer those questions. They carry a title, a price, a few images, a size run, maybe a color. That is enough to render a product page. It is nowhere near enough to describe who a garment is for.

So the system fills the gap by guessing, and it guesses from the creative. If your model in the video is in her late twenties and your actual customer is in her forties, there is nothing in the catalog to correct the assumption. You get exactly what the founder above described: a growing suspicion that the algorithm is not reading your products correctly, and no way to verify it.

Everyone responded to broad targeting by pouring resources into creative volume, because creative is the lever people can see. Creative is genuinely half the answer. The other half is the catalog, and it has been sitting untouched.

Why fashion brands are more exposed than most

Two things compound here.

The first is that a lot of apparel brands quietly gave up on dynamic product ads. The reason is almost always feed quality rather than the format itself: the wrong products surfacing, the same item repeating in a placement, ads rendering blank on the customer side, items appearing that have nothing to do with the ad they sit under. After enough of that, brands switch the format off and never come back. That removes an entire performance channel because of a data problem, and the data problem stays in place feeding everything else.

The second is that Shopify metafields are a black box for most merchants. Brands dutifully fill in the category fields, and nobody can tell them where that information goes or whether any platform reads it. So the work either does not happen or happens without confidence.

The same thin data shows up on the search side too, which we covered in why your fashion products do not show on Google Shopping. Same catalog, different symptom.

The returns layer that never makes it into the ROAS math

There is one more reason apparel ROAS is harder than it looks, and it has nothing to do with Meta.

Meta reports gross revenue. It counts the sale at checkout. It does not know the dress came back three weeks later.

Online apparel return rates commonly run between 25 and 40 percent, with fitted styles at the high end, and fit is the reason most of the time. That means the reported figure was always generous for fashion, well before any of this year's changes.

At a reported 3.2x with a 30 percent return rate, net ROAS is 2.24x. At a 45 percent gross margin, break-even sits at 2.22x. Subtract return shipping, inspection, and restocking and the campaign is underwater while the dashboard reads as healthy.

This is not an argument that your numbers were fake. It is an argument that the reported number and the number your business runs on have always been two different things, and 2026 widened the gap.

Getting your bearings

Before changing anything, it helps to know how much of your drop belongs to which category. Pull the same date range from three places: Meta reported revenue, your analytics platform, and actual store revenue.

Meta reported	Actual store revenue	What you are looking at
Down 30%	Flat	Reporting artifact. The business is fine.
Down 30%	Down 5 to 10%	Mostly accounting, some real erosion. Size the response to the smaller number.
Down 30%	Down 25 to 30%	Real. Delivery, traffic quality, and catalog signal are where to look.

If Meta shows 90 purchases and your backend shows 150, that gap is also worth sitting with. It means the system has been learning from an incomplete picture of who actually buys from you, which is its own compounding problem under broad targeting.

Then three numbers are worth more than reported ROAS on its own. Break-even ROAS, one divided by gross margin, so you know what good looks like for your business rather than for someone else's. MER, total revenue over total spend, which ignores attribution entirely and is therefore the most stable thing you have in a year when attribution keeps moving. And net ROAS after returns, tracked by category rather than store wide.

None of that fixes delivery. It tells you how big the real problem is, which is the part most operators are currently guessing at.

Where the real work sits

If targeting now runs through the ad and the catalog, then the catalog is a performance asset, not plumbing. Fit, occasion, body suitability, fabric behavior, who a garment is actually for: that is the information the system is missing and currently inventing on your behalf.

It is also the same information that decides whether the garment comes back. Every fit-driven return is a conversion you paid for twice, and unlike the auction, it is a number you can move. The true cost of fashion returns breaks down what each one costs.

You cannot out-create an auction you are losing on margin. But you can stop asking a system that decides everything to decide it with almost nothing to go on.

Sources

Meta Ads Insights API changes, January 12, 2026. On the permanent removal of the 7-day view and 28-day view attribution windows and the resulting drop in reported conversions.

Meta attribution announcement, March 3, 2026. On the redefinition of click-through attribution to link clicks only, and the creation of engage-through attribution with a 1-day conversion window.

Meta Andromeda delivery system, phased rollout through 2026. On the shift toward machine-selected audiences and creative, and the reduced predictability of manual bid and cost controls.

Interview with a DTC womenswear operator, July 2026. On monthly Meta spend falling from roughly eighty thousand dollars to thirteen thousand, an April performance collapse, and a bid capped campaign overspending its cap by six times with zero add to carts at a forty cent cost per click. Details shared with permission and anonymized.

Coresight Research and Loop Returns benchmarks. On online apparel return rates averaging around 24 to 25 percent, with women's fashion DTC commonly between 25 and 35 percent and fitted categories reaching 40 percent or higher.

PRIME AI, analysis of one million fashion returns. Finding that 67 percent of returns were driven by fit and sizing issues.

---

Meta is deciding who sees your products. It is doing it with whatever your catalog tells it. SIXFIT reads fit at the catalog level, so the data behind your ads describes who each garment is actually for. See how it works.