Meta ROAS by Industry: The Source Gap Can Be Bigger Than the Industry Gap

Beauty’s published Meta ROAS is either 1.54x or 3.2x, depending on which table you open. That 1.66x gap for one industry is wider than the gap between the highest and lowest industries in the first table. A ranking that changes that much with the publisher is, in large part, a ranking of methods.

I used to send clients industry tables as if they gave us a target. I was wrong. A ROAS figure depends on who is in the sample, which sales the platform credits to ads, how long it has to claim those sales, and what the publisher calls revenue. Line up the published figures and you can see the disagreement. Read the methods and you can see why an industry median is a poor instruction for your next budget change.

The tables agree on the label, not the number

Triple Whale’s panel covers more than 40,000 brands from August 2025 to July 2026. It reports median Meta ROAS of 1.88x, alongside median CPA of $38.99, CPM of $15.06, and conversion rate of 1.53%. Meta accounts for 66.88% of tracked ecommerce spend in that panel. This is an ecommerce read, not a target for a services business buying leads.

Landingi publishes another set of industry figures drawn from Varos data: Fashion at 2.65x, Electronics at 2.95x, Beauty at 1.57x, Food at 1.67x, and Baby at 4.39x. Eightx, using Ryze AI data, publishes a third. These are not three measurements of one controlled population. Keep that limitation beside the numbers, not in small print below them.

Read across a row before you read down a column

A dash means I am not making a cross-source comparison for that row. The paired labels are not necessarily identical category definitions, either: apparel and fashion are close enough to invite a look, not close enough to assume identical membership.

Beauty runs from 1.54x to 3.2x across publishers, a spread of 1.66x. Fashion and apparel run from 2.18x to 2.9x, a spread of 0.72x. Within Triple Whale’s own table, Media & Publishing at 1.13x and Sports & Outdoors at 2.35x sit 1.22x apart. For Beauty, the between-publisher gap exceeds that entire within-publisher range.

That does not prove which source has the right Beauty number. It tells me not to use the ranking as a budget rule. Before I act on a cell, I need to know what produced it.

The samples do not give you a shared ruler

A compilation of Triple Whale figures describes a panel of roughly 35,000 to more than 40,000 purchase-optimized ecommerce brands, depending on the cut. It covers January to December 2025 in that compilation; the live Triple Whale view covers August 2025 to July 2026. The reported median is around 1.86x to 1.88x. Even before another publisher enters the picture, the reporting period and panel description need a label.

WordStream, included in the same compilation, covers 1,180 services campaigns from April 2024 to June 2025. Its published figures include traffic CPC of $0.70, leads CPC of $1.92, and CPL of $27.66. Those are useful costs, but they are not ecommerce purchase ROAS. A form fill does not arrive with a known order value. Putting that lead cost next to store revenue without the close rate is how a tidy spreadsheet becomes nonsense.

The Ryze AI figures published by Eightx draw on 15,000 advertisers and $2.8 billion in spend, with more weight on larger DTC spenders running Advantage+. Bigger accounts and different campaign mixes can produce a different median without either publisher miscalculating its own sample. The public figures here do not let me isolate how much of the Beauty spread comes from sample mix, attribution settings, or revenue definitions. The gap is observable; its exact decomposition is not.

Three ways the ruler changes while the ads stay put

Attribution is the first place I look. Meta’s default of 7-day click plus 1-day view lets an ad claim a purchase after either a qualifying click or a qualifying view. A shorter, click-only window gives it fewer opportunities to claim the same sale. One published comparison puts reported ROAS 25% to 40% lower on 1-day click than on 7-day click plus 1-day view. Another warns that changing settings mid-quarter can make ROAS appear 30% to 50% lower without a corresponding sales change.

Those ranges are explanations of how a reported number can move, not a correction factor I can apply to every cell in the table. I have seen clients panic over a reporting drop caused by a dropdown. The fix is not to declare the old window true and the new one false. It is to stop comparing them as though nothing changed.

Measurement plumbing matters too. A published account of Conversions API says reported ROAS can rise 10% to 25% with no extra sales. Better event capture can change what Meta sees; it does not create the orders. Before calling a lift a performance win, I check whether the sales moved outside Ads Manager.

The second ruler is who and what got counted. Triple Whale’s purchase-optimized stores are not WordStream’s services leads, and the Ryze AI sample of 15,000 advertisers is not the same panel as either. Mix changes inside an account as well. Figures from Focus Digital, republished in the compilation, put retargeting at 3.61x, prospecting at 2.11x, lookalikes at 1.80x, and Advantage+ Shopping near 4.52x. A blended 2.4x account with 70% prospecting is doing a different job from a 2.4x account with 90% retargeting. One is paying to find demand; the other leans heavily on demand it already knows about. I would not give them the same marching orders.

The third ruler is what counts as revenue. Ecommerce ROAS usually puts gross sales in the numerator, while the business pays shipping, returns, fees, and product costs out of those sales. The break-even relationship is 1 divided by contribution margin. At a 40% contribution margin, that means 2.5x to break even; at 30%, about 3.33x; at 25%, 4x; at 20%, 5x. A 3x gross-revenue ROAS clears the hurdle at 60% margin and misses it at 30%. The same ranking tells those businesses opposite things about profit.

Then there is incrementality: how many sales the ads caused rather than merely claimed. JetFuel’s audit guide cites Haus geo-tests in which platform-reported ROAS is 1.5x to 3x measured incremental ROAS, with the widest gaps on retargeting and brand search. Email and Meta can claim the same sale under their respective attribution rules. Neither claim pays the invoice twice.

That is why I use reported ROAS to diagnose campaigns, then check it against MER, total revenue divided by total ad spend, and profit before moving serious money. No industry table can tell me the incremental return of my account from its median.

CPA helps, provided you keep its definition attached

ROAS equals average order value divided by CPA when the revenue and acquisition counts refer to the same purchases. That identity makes CPA a useful operating number: I can compare the cost of an acquired order with what that order contributes. It does not make platform-reported CPA immune to attribution. Change which purchases Meta claims and its reported CPA can move too. I still need one window and a consistent purchase definition.

Take Triple Whale’s $38.99 median CPA. At an $80 average order value, that cost implies about 2.05x ROAS. At $130 average order value and the same CPA, it implies about 3.33x. Acquisition cost has not changed; order value has. A cross-industry ROAS ranking collapses both into one number, then invites me to judge the buying from the ratio alone.

For leads, I need another step. The published $27.66 CPL becomes about $277 per customer at a 10% close rate, or $922 at a 3% close rate. Those are illustrative calculations from the same lead cost, not two benchmark results. A manager who quotes the CPL without asking what happens after the form arrives has stopped the calculation early.

The questions I ask before anyone changes a budget

When a Meta manager tells me Beauty should hit a particular ROAS, I ask four things. They belong beside the table, not buried in an appendix:

• Which attribution window produced it? A view-through sale and a click-through sale do not answer the same question.
• What did “revenue” mean? Gross store sales are not contribution after shipping, returns, and fees.
• What was the campaign mix? A retargeting-heavy median is a poor prospecting target.
• Which period and sample were measured? A category label does not guarantee comparable advertisers or months.

If the publisher or manager cannot answer, I do not fill the blanks with guesses. I also do not need to prove the table worthless. I need to decide whether it is precise enough to steer this account. Usually, it is not.

The Varos cut offers a useful warning against taking a flattering cell too far: Baby appears at 4.39x on Meta and 4.00x on Google. That comparison describes the published cut. It does not establish that moving a Baby advertiser’s budget from Google to Meta will improve profit. The account’s margins, attribution, and customer mix still have the vote.

Replace the publisher’s median with your own ceiling

I build the next decision from a trailing 90-day baseline, measured consistently. One attribution window, one revenue definition, and your own CPA ceiling give me more to work with than a borrowed industry rank.

1. Lock the reporting window. I use 7-day click only for the DTC buying comparison described here, record the setting, and do not splice figures from another window into the trend. That is a consistent choice, not a claim that one setting reveals every incremental sale.
2. Pull the trailing 90 days. Record CPA, average order value, and MER, with MER calculated as total revenue divided by total ad spend. Keep the revenue definition unchanged when you update the figures.
3. Set the acquisition ceiling from contribution. On an illustrative $90 order with 35% contribution after shipping, returns, and fees, $31.50 is available per order. Paying more than $31.50 to acquire it loses money on that order; gross-revenue ROAS below about 2.86x misses the same break-even line.

For a leads business, I work from closed customers instead of pretending every form has a purchase value. At the published $27.66 CPL, a 10% close rate implies about $277 per customer; a 4% close rate implies about $692. The lead cost did not move. The economics did.

This is why I stopped managing to industry tables and started managing to an account baseline checked against its contribution. It is also why I work with groas: the operating job is to keep the window consistent, log changes and their results, and move spend toward what beats the account’s own ceiling. A polished monthly deck cannot make an unaffordable acquisition affordable.

Keep two numbers where you can see them: allowable CPA per order and trailing 90-day CPA on the same reporting window. Check MER and profit alongside them. If CPA sits below the ceiling and MER is holding, a Beauty table reading 1.54x instead of 3.2x does not require a rescue plan. If CPA sits above it, the higher benchmark will not pay your costs. That is the decision the numbers support.